Commit | Line | Data |
---|---|---|
d38ceaf9 AD |
1 | /* |
2 | * Copyright 2008 Advanced Micro Devices, Inc. | |
3 | * Copyright 2008 Red Hat Inc. | |
4 | * Copyright 2009 Jerome Glisse. | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a | |
7 | * copy of this software and associated documentation files (the "Software"), | |
8 | * to deal in the Software without restriction, including without limitation | |
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
10 | * and/or sell copies of the Software, and to permit persons to whom the | |
11 | * Software is furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
22 | * OTHER DEALINGS IN THE SOFTWARE. | |
23 | * | |
24 | * Authors: Dave Airlie | |
25 | * Alex Deucher | |
26 | * Jerome Glisse | |
27 | */ | |
b1ddf548 | 28 | #include <linux/power_supply.h> |
0875dc9e | 29 | #include <linux/kthread.h> |
fdf2f6c5 | 30 | #include <linux/module.h> |
d38ceaf9 AD |
31 | #include <linux/console.h> |
32 | #include <linux/slab.h> | |
4a74c38c | 33 | #include <linux/iommu.h> |
901e2be2 | 34 | #include <linux/pci.h> |
08a2fd23 | 35 | #include <linux/pci-p2pdma.h> |
d37a3929 | 36 | #include <linux/apple-gmux.h> |
fdf2f6c5 | 37 | |
b7cdb41e | 38 | #include <drm/drm_aperture.h> |
4562236b | 39 | #include <drm/drm_atomic_helper.h> |
973ad627 | 40 | #include <drm/drm_crtc_helper.h> |
45b64fd9 | 41 | #include <drm/drm_fb_helper.h> |
fcd70cd3 | 42 | #include <drm/drm_probe_helper.h> |
d38ceaf9 | 43 | #include <drm/amdgpu_drm.h> |
7b1c6263 | 44 | #include <linux/device.h> |
d38ceaf9 AD |
45 | #include <linux/vgaarb.h> |
46 | #include <linux/vga_switcheroo.h> | |
47 | #include <linux/efi.h> | |
48 | #include "amdgpu.h" | |
f4b373f4 | 49 | #include "amdgpu_trace.h" |
d38ceaf9 AD |
50 | #include "amdgpu_i2c.h" |
51 | #include "atom.h" | |
52 | #include "amdgpu_atombios.h" | |
a5bde2f9 | 53 | #include "amdgpu_atomfirmware.h" |
d0dd7f0c | 54 | #include "amd_pcie.h" |
33f34802 KW |
55 | #ifdef CONFIG_DRM_AMDGPU_SI |
56 | #include "si.h" | |
57 | #endif | |
a2e73f56 AD |
58 | #ifdef CONFIG_DRM_AMDGPU_CIK |
59 | #include "cik.h" | |
60 | #endif | |
aaa36a97 | 61 | #include "vi.h" |
460826e6 | 62 | #include "soc15.h" |
0a5b8c7b | 63 | #include "nv.h" |
d38ceaf9 | 64 | #include "bif/bif_4_1_d.h" |
bec86378 | 65 | #include <linux/firmware.h> |
89041940 | 66 | #include "amdgpu_vf_error.h" |
d38ceaf9 | 67 | |
ba997709 | 68 | #include "amdgpu_amdkfd.h" |
d2f52ac8 | 69 | #include "amdgpu_pm.h" |
d38ceaf9 | 70 | |
5183411b | 71 | #include "amdgpu_xgmi.h" |
c030f2e4 | 72 | #include "amdgpu_ras.h" |
9c7c85f7 | 73 | #include "amdgpu_pmu.h" |
bd607166 | 74 | #include "amdgpu_fru_eeprom.h" |
04442bf7 | 75 | #include "amdgpu_reset.h" |
85150626 | 76 | #include "amdgpu_virt.h" |
5183411b | 77 | |
d5ea093e | 78 | #include <linux/suspend.h> |
c6a6e2db | 79 | #include <drm/task_barrier.h> |
3f12acc8 | 80 | #include <linux/pm_runtime.h> |
d5ea093e | 81 | |
f89f8c6b AG |
82 | #include <drm/drm_drv.h> |
83 | ||
3ad5dcfe KHF |
84 | #if IS_ENABLED(CONFIG_X86) |
85 | #include <asm/intel-family.h> | |
86 | #endif | |
87 | ||
e2a75f88 | 88 | MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin"); |
3f76dced | 89 | MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin"); |
2d2e5e7e | 90 | MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin"); |
ad5a67a7 | 91 | MODULE_FIRMWARE("amdgpu/picasso_gpu_info.bin"); |
54c4d17e | 92 | MODULE_FIRMWARE("amdgpu/raven2_gpu_info.bin"); |
65e60f6e | 93 | MODULE_FIRMWARE("amdgpu/arcturus_gpu_info.bin"); |
42b325e5 | 94 | MODULE_FIRMWARE("amdgpu/navi12_gpu_info.bin"); |
e2a75f88 | 95 | |
2dc80b00 | 96 | #define AMDGPU_RESUME_MS 2000 |
7258fa31 SK |
97 | #define AMDGPU_MAX_RETRY_LIMIT 2 |
98 | #define AMDGPU_RETRY_SRIOV_RESET(r) ((r) == -EBUSY || (r) == -ETIMEDOUT || (r) == -EINVAL) | |
2dc80b00 | 99 | |
b7cdb41e ML |
100 | static const struct drm_driver amdgpu_kms_driver; |
101 | ||
050091ab | 102 | const char *amdgpu_asic_name[] = { |
da69c161 KW |
103 | "TAHITI", |
104 | "PITCAIRN", | |
105 | "VERDE", | |
106 | "OLAND", | |
107 | "HAINAN", | |
d38ceaf9 AD |
108 | "BONAIRE", |
109 | "KAVERI", | |
110 | "KABINI", | |
111 | "HAWAII", | |
112 | "MULLINS", | |
113 | "TOPAZ", | |
114 | "TONGA", | |
48299f95 | 115 | "FIJI", |
d38ceaf9 | 116 | "CARRIZO", |
139f4917 | 117 | "STONEY", |
2cc0c0b5 FC |
118 | "POLARIS10", |
119 | "POLARIS11", | |
c4642a47 | 120 | "POLARIS12", |
48ff108d | 121 | "VEGAM", |
d4196f01 | 122 | "VEGA10", |
8fab806a | 123 | "VEGA12", |
956fcddc | 124 | "VEGA20", |
2ca8a5d2 | 125 | "RAVEN", |
d6c3b24e | 126 | "ARCTURUS", |
1eee4228 | 127 | "RENOIR", |
d46b417a | 128 | "ALDEBARAN", |
852a6626 | 129 | "NAVI10", |
d0f56dc2 | 130 | "CYAN_SKILLFISH", |
87dbad02 | 131 | "NAVI14", |
9802f5d7 | 132 | "NAVI12", |
ccaf72d3 | 133 | "SIENNA_CICHLID", |
ddd8fbe7 | 134 | "NAVY_FLOUNDER", |
4f1e9a76 | 135 | "VANGOGH", |
a2468e04 | 136 | "DIMGREY_CAVEFISH", |
6f169591 | 137 | "BEIGE_GOBY", |
ee9236b7 | 138 | "YELLOW_CARP", |
3ae695d6 | 139 | "IP DISCOVERY", |
d38ceaf9 AD |
140 | "LAST", |
141 | }; | |
142 | ||
dcea6e65 KR |
143 | /** |
144 | * DOC: pcie_replay_count | |
145 | * | |
146 | * The amdgpu driver provides a sysfs API for reporting the total number | |
147 | * of PCIe replays (NAKs) | |
148 | * The file pcie_replay_count is used for this and returns the total | |
149 | * number of replays as a sum of the NAKs generated and NAKs received | |
150 | */ | |
151 | ||
152 | static ssize_t amdgpu_device_get_pcie_replay_count(struct device *dev, | |
153 | struct device_attribute *attr, char *buf) | |
154 | { | |
155 | struct drm_device *ddev = dev_get_drvdata(dev); | |
1348969a | 156 | struct amdgpu_device *adev = drm_to_adev(ddev); |
dcea6e65 KR |
157 | uint64_t cnt = amdgpu_asic_get_pcie_replay_count(adev); |
158 | ||
36000c7a | 159 | return sysfs_emit(buf, "%llu\n", cnt); |
dcea6e65 KR |
160 | } |
161 | ||
b8920e1e | 162 | static DEVICE_ATTR(pcie_replay_count, 0444, |
dcea6e65 KR |
163 | amdgpu_device_get_pcie_replay_count, NULL); |
164 | ||
4798db85 LL |
165 | /** |
166 | * DOC: board_info | |
167 | * | |
168 | * The amdgpu driver provides a sysfs API for giving board related information. | |
169 | * It provides the form factor information in the format | |
170 | * | |
171 | * type : form factor | |
172 | * | |
173 | * Possible form factor values | |
174 | * | |
175 | * - "cem" - PCIE CEM card | |
176 | * - "oam" - Open Compute Accelerator Module | |
177 | * - "unknown" - Not known | |
178 | * | |
179 | */ | |
180 | ||
76da73f0 LL |
181 | static ssize_t amdgpu_device_get_board_info(struct device *dev, |
182 | struct device_attribute *attr, | |
183 | char *buf) | |
184 | { | |
185 | struct drm_device *ddev = dev_get_drvdata(dev); | |
186 | struct amdgpu_device *adev = drm_to_adev(ddev); | |
187 | enum amdgpu_pkg_type pkg_type = AMDGPU_PKG_TYPE_CEM; | |
188 | const char *pkg; | |
189 | ||
190 | if (adev->smuio.funcs && adev->smuio.funcs->get_pkg_type) | |
191 | pkg_type = adev->smuio.funcs->get_pkg_type(adev); | |
192 | ||
193 | switch (pkg_type) { | |
194 | case AMDGPU_PKG_TYPE_CEM: | |
195 | pkg = "cem"; | |
196 | break; | |
197 | case AMDGPU_PKG_TYPE_OAM: | |
198 | pkg = "oam"; | |
199 | break; | |
200 | default: | |
201 | pkg = "unknown"; | |
202 | break; | |
203 | } | |
204 | ||
205 | return sysfs_emit(buf, "%s : %s\n", "type", pkg); | |
206 | } | |
207 | ||
208 | static DEVICE_ATTR(board_info, 0444, amdgpu_device_get_board_info, NULL); | |
209 | ||
210 | static struct attribute *amdgpu_board_attrs[] = { | |
211 | &dev_attr_board_info.attr, | |
212 | NULL, | |
213 | }; | |
214 | ||
215 | static umode_t amdgpu_board_attrs_is_visible(struct kobject *kobj, | |
216 | struct attribute *attr, int n) | |
217 | { | |
218 | struct device *dev = kobj_to_dev(kobj); | |
219 | struct drm_device *ddev = dev_get_drvdata(dev); | |
220 | struct amdgpu_device *adev = drm_to_adev(ddev); | |
221 | ||
222 | if (adev->flags & AMD_IS_APU) | |
223 | return 0; | |
224 | ||
225 | return attr->mode; | |
226 | } | |
227 | ||
228 | static const struct attribute_group amdgpu_board_attrs_group = { | |
229 | .attrs = amdgpu_board_attrs, | |
230 | .is_visible = amdgpu_board_attrs_is_visible | |
231 | }; | |
232 | ||
5494d864 AD |
233 | static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev); |
234 | ||
bd607166 | 235 | |
fd496ca8 | 236 | /** |
b98c6299 | 237 | * amdgpu_device_supports_px - Is the device a dGPU with ATPX power control |
fd496ca8 AD |
238 | * |
239 | * @dev: drm_device pointer | |
240 | * | |
b98c6299 | 241 | * Returns true if the device is a dGPU with ATPX power control, |
fd496ca8 AD |
242 | * otherwise return false. |
243 | */ | |
b98c6299 | 244 | bool amdgpu_device_supports_px(struct drm_device *dev) |
fd496ca8 AD |
245 | { |
246 | struct amdgpu_device *adev = drm_to_adev(dev); | |
247 | ||
b98c6299 | 248 | if ((adev->flags & AMD_IS_PX) && !amdgpu_is_atpx_hybrid()) |
fd496ca8 AD |
249 | return true; |
250 | return false; | |
251 | } | |
252 | ||
e3ecdffa | 253 | /** |
0330b848 | 254 | * amdgpu_device_supports_boco - Is the device a dGPU with ACPI power resources |
e3ecdffa AD |
255 | * |
256 | * @dev: drm_device pointer | |
257 | * | |
b98c6299 | 258 | * Returns true if the device is a dGPU with ACPI power control, |
e3ecdffa AD |
259 | * otherwise return false. |
260 | */ | |
31af062a | 261 | bool amdgpu_device_supports_boco(struct drm_device *dev) |
d38ceaf9 | 262 | { |
1348969a | 263 | struct amdgpu_device *adev = drm_to_adev(dev); |
d38ceaf9 | 264 | |
b98c6299 AD |
265 | if (adev->has_pr3 || |
266 | ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) | |
d38ceaf9 AD |
267 | return true; |
268 | return false; | |
269 | } | |
270 | ||
a69cba42 AD |
271 | /** |
272 | * amdgpu_device_supports_baco - Does the device support BACO | |
273 | * | |
274 | * @dev: drm_device pointer | |
275 | * | |
276 | * Returns true if the device supporte BACO, | |
277 | * otherwise return false. | |
278 | */ | |
279 | bool amdgpu_device_supports_baco(struct drm_device *dev) | |
280 | { | |
1348969a | 281 | struct amdgpu_device *adev = drm_to_adev(dev); |
a69cba42 AD |
282 | |
283 | return amdgpu_asic_supports_baco(adev); | |
284 | } | |
285 | ||
3fa8f89d S |
286 | /** |
287 | * amdgpu_device_supports_smart_shift - Is the device dGPU with | |
288 | * smart shift support | |
289 | * | |
290 | * @dev: drm_device pointer | |
291 | * | |
292 | * Returns true if the device is a dGPU with Smart Shift support, | |
293 | * otherwise returns false. | |
294 | */ | |
295 | bool amdgpu_device_supports_smart_shift(struct drm_device *dev) | |
296 | { | |
297 | return (amdgpu_device_supports_boco(dev) && | |
298 | amdgpu_acpi_is_power_shift_control_supported()); | |
299 | } | |
300 | ||
6e3cd2a9 MCC |
301 | /* |
302 | * VRAM access helper functions | |
303 | */ | |
304 | ||
e35e2b11 | 305 | /** |
048af66b | 306 | * amdgpu_device_mm_access - access vram by MM_INDEX/MM_DATA |
e35e2b11 TY |
307 | * |
308 | * @adev: amdgpu_device pointer | |
309 | * @pos: offset of the buffer in vram | |
310 | * @buf: virtual address of the buffer in system memory | |
311 | * @size: read/write size, sizeof(@buf) must > @size | |
312 | * @write: true - write to vram, otherwise - read from vram | |
313 | */ | |
048af66b KW |
314 | void amdgpu_device_mm_access(struct amdgpu_device *adev, loff_t pos, |
315 | void *buf, size_t size, bool write) | |
e35e2b11 | 316 | { |
e35e2b11 | 317 | unsigned long flags; |
048af66b KW |
318 | uint32_t hi = ~0, tmp = 0; |
319 | uint32_t *data = buf; | |
ce05ac56 | 320 | uint64_t last; |
f89f8c6b | 321 | int idx; |
ce05ac56 | 322 | |
c58a863b | 323 | if (!drm_dev_enter(adev_to_drm(adev), &idx)) |
f89f8c6b | 324 | return; |
9d11eb0d | 325 | |
048af66b KW |
326 | BUG_ON(!IS_ALIGNED(pos, 4) || !IS_ALIGNED(size, 4)); |
327 | ||
328 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
329 | for (last = pos + size; pos < last; pos += 4) { | |
330 | tmp = pos >> 31; | |
331 | ||
332 | WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)pos) | 0x80000000); | |
333 | if (tmp != hi) { | |
334 | WREG32_NO_KIQ(mmMM_INDEX_HI, tmp); | |
335 | hi = tmp; | |
336 | } | |
337 | if (write) | |
338 | WREG32_NO_KIQ(mmMM_DATA, *data++); | |
339 | else | |
340 | *data++ = RREG32_NO_KIQ(mmMM_DATA); | |
341 | } | |
342 | ||
343 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
344 | drm_dev_exit(idx); | |
345 | } | |
346 | ||
347 | /** | |
bbe04dec | 348 | * amdgpu_device_aper_access - access vram by vram aperature |
048af66b KW |
349 | * |
350 | * @adev: amdgpu_device pointer | |
351 | * @pos: offset of the buffer in vram | |
352 | * @buf: virtual address of the buffer in system memory | |
353 | * @size: read/write size, sizeof(@buf) must > @size | |
354 | * @write: true - write to vram, otherwise - read from vram | |
355 | * | |
356 | * The return value means how many bytes have been transferred. | |
357 | */ | |
358 | size_t amdgpu_device_aper_access(struct amdgpu_device *adev, loff_t pos, | |
359 | void *buf, size_t size, bool write) | |
360 | { | |
9d11eb0d | 361 | #ifdef CONFIG_64BIT |
048af66b KW |
362 | void __iomem *addr; |
363 | size_t count = 0; | |
364 | uint64_t last; | |
365 | ||
366 | if (!adev->mman.aper_base_kaddr) | |
367 | return 0; | |
368 | ||
9d11eb0d CK |
369 | last = min(pos + size, adev->gmc.visible_vram_size); |
370 | if (last > pos) { | |
048af66b KW |
371 | addr = adev->mman.aper_base_kaddr + pos; |
372 | count = last - pos; | |
9d11eb0d CK |
373 | |
374 | if (write) { | |
375 | memcpy_toio(addr, buf, count); | |
4c452b5c SS |
376 | /* Make sure HDP write cache flush happens without any reordering |
377 | * after the system memory contents are sent over PCIe device | |
378 | */ | |
9d11eb0d | 379 | mb(); |
810085dd | 380 | amdgpu_device_flush_hdp(adev, NULL); |
9d11eb0d | 381 | } else { |
810085dd | 382 | amdgpu_device_invalidate_hdp(adev, NULL); |
4c452b5c SS |
383 | /* Make sure HDP read cache is invalidated before issuing a read |
384 | * to the PCIe device | |
385 | */ | |
9d11eb0d CK |
386 | mb(); |
387 | memcpy_fromio(buf, addr, count); | |
388 | } | |
389 | ||
9d11eb0d | 390 | } |
048af66b KW |
391 | |
392 | return count; | |
393 | #else | |
394 | return 0; | |
9d11eb0d | 395 | #endif |
048af66b | 396 | } |
9d11eb0d | 397 | |
048af66b KW |
398 | /** |
399 | * amdgpu_device_vram_access - read/write a buffer in vram | |
400 | * | |
401 | * @adev: amdgpu_device pointer | |
402 | * @pos: offset of the buffer in vram | |
403 | * @buf: virtual address of the buffer in system memory | |
404 | * @size: read/write size, sizeof(@buf) must > @size | |
405 | * @write: true - write to vram, otherwise - read from vram | |
406 | */ | |
407 | void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos, | |
408 | void *buf, size_t size, bool write) | |
409 | { | |
410 | size_t count; | |
e35e2b11 | 411 | |
048af66b KW |
412 | /* try to using vram apreature to access vram first */ |
413 | count = amdgpu_device_aper_access(adev, pos, buf, size, write); | |
414 | size -= count; | |
415 | if (size) { | |
416 | /* using MM to access rest vram */ | |
417 | pos += count; | |
418 | buf += count; | |
419 | amdgpu_device_mm_access(adev, pos, buf, size, write); | |
e35e2b11 TY |
420 | } |
421 | } | |
422 | ||
d38ceaf9 | 423 | /* |
f7ee1874 | 424 | * register access helper functions. |
d38ceaf9 | 425 | */ |
56b53c0b DL |
426 | |
427 | /* Check if hw access should be skipped because of hotplug or device error */ | |
428 | bool amdgpu_device_skip_hw_access(struct amdgpu_device *adev) | |
429 | { | |
7afefb81 | 430 | if (adev->no_hw_access) |
56b53c0b DL |
431 | return true; |
432 | ||
433 | #ifdef CONFIG_LOCKDEP | |
434 | /* | |
435 | * This is a bit complicated to understand, so worth a comment. What we assert | |
436 | * here is that the GPU reset is not running on another thread in parallel. | |
437 | * | |
438 | * For this we trylock the read side of the reset semaphore, if that succeeds | |
439 | * we know that the reset is not running in paralell. | |
440 | * | |
441 | * If the trylock fails we assert that we are either already holding the read | |
442 | * side of the lock or are the reset thread itself and hold the write side of | |
443 | * the lock. | |
444 | */ | |
445 | if (in_task()) { | |
d0fb18b5 AG |
446 | if (down_read_trylock(&adev->reset_domain->sem)) |
447 | up_read(&adev->reset_domain->sem); | |
56b53c0b | 448 | else |
d0fb18b5 | 449 | lockdep_assert_held(&adev->reset_domain->sem); |
56b53c0b DL |
450 | } |
451 | #endif | |
452 | return false; | |
453 | } | |
454 | ||
e3ecdffa | 455 | /** |
f7ee1874 | 456 | * amdgpu_device_rreg - read a memory mapped IO or indirect register |
e3ecdffa AD |
457 | * |
458 | * @adev: amdgpu_device pointer | |
459 | * @reg: dword aligned register offset | |
460 | * @acc_flags: access flags which require special behavior | |
461 | * | |
462 | * Returns the 32 bit value from the offset specified. | |
463 | */ | |
f7ee1874 HZ |
464 | uint32_t amdgpu_device_rreg(struct amdgpu_device *adev, |
465 | uint32_t reg, uint32_t acc_flags) | |
d38ceaf9 | 466 | { |
f4b373f4 TSD |
467 | uint32_t ret; |
468 | ||
56b53c0b | 469 | if (amdgpu_device_skip_hw_access(adev)) |
bf36b52e AG |
470 | return 0; |
471 | ||
f7ee1874 HZ |
472 | if ((reg * 4) < adev->rmmio_size) { |
473 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && | |
474 | amdgpu_sriov_runtime(adev) && | |
d0fb18b5 | 475 | down_read_trylock(&adev->reset_domain->sem)) { |
85150626 | 476 | ret = amdgpu_kiq_rreg(adev, reg, 0); |
d0fb18b5 | 477 | up_read(&adev->reset_domain->sem); |
f7ee1874 HZ |
478 | } else { |
479 | ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); | |
480 | } | |
481 | } else { | |
482 | ret = adev->pcie_rreg(adev, reg * 4); | |
81202807 | 483 | } |
bc992ba5 | 484 | |
f7ee1874 | 485 | trace_amdgpu_device_rreg(adev->pdev->device, reg, ret); |
e78b579d | 486 | |
f4b373f4 | 487 | return ret; |
d38ceaf9 AD |
488 | } |
489 | ||
421a2a30 ML |
490 | /* |
491 | * MMIO register read with bytes helper functions | |
492 | * @offset:bytes offset from MMIO start | |
b8920e1e | 493 | */ |
421a2a30 | 494 | |
e3ecdffa AD |
495 | /** |
496 | * amdgpu_mm_rreg8 - read a memory mapped IO register | |
497 | * | |
498 | * @adev: amdgpu_device pointer | |
499 | * @offset: byte aligned register offset | |
500 | * | |
501 | * Returns the 8 bit value from the offset specified. | |
502 | */ | |
7cbbc745 AG |
503 | uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset) |
504 | { | |
56b53c0b | 505 | if (amdgpu_device_skip_hw_access(adev)) |
bf36b52e AG |
506 | return 0; |
507 | ||
421a2a30 ML |
508 | if (offset < adev->rmmio_size) |
509 | return (readb(adev->rmmio + offset)); | |
510 | BUG(); | |
511 | } | |
512 | ||
85150626 VL |
513 | |
514 | /** | |
515 | * amdgpu_device_xcc_rreg - read a memory mapped IO or indirect register with specific XCC | |
516 | * | |
517 | * @adev: amdgpu_device pointer | |
518 | * @reg: dword aligned register offset | |
519 | * @acc_flags: access flags which require special behavior | |
520 | * @xcc_id: xcc accelerated compute core id | |
521 | * | |
522 | * Returns the 32 bit value from the offset specified. | |
523 | */ | |
524 | uint32_t amdgpu_device_xcc_rreg(struct amdgpu_device *adev, | |
525 | uint32_t reg, uint32_t acc_flags, | |
526 | uint32_t xcc_id) | |
527 | { | |
528 | uint32_t ret, rlcg_flag; | |
529 | ||
530 | if (amdgpu_device_skip_hw_access(adev)) | |
531 | return 0; | |
532 | ||
533 | if ((reg * 4) < adev->rmmio_size) { | |
534 | if (amdgpu_sriov_vf(adev) && | |
535 | !amdgpu_sriov_runtime(adev) && | |
536 | adev->gfx.rlc.rlcg_reg_access_supported && | |
537 | amdgpu_virt_get_rlcg_reg_access_flag(adev, acc_flags, | |
538 | GC_HWIP, false, | |
539 | &rlcg_flag)) { | |
540 | ret = amdgpu_virt_rlcg_reg_rw(adev, reg, 0, rlcg_flag, xcc_id); | |
541 | } else if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && | |
542 | amdgpu_sriov_runtime(adev) && | |
543 | down_read_trylock(&adev->reset_domain->sem)) { | |
544 | ret = amdgpu_kiq_rreg(adev, reg, xcc_id); | |
545 | up_read(&adev->reset_domain->sem); | |
546 | } else { | |
547 | ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); | |
548 | } | |
549 | } else { | |
550 | ret = adev->pcie_rreg(adev, reg * 4); | |
551 | } | |
552 | ||
553 | return ret; | |
554 | } | |
555 | ||
421a2a30 ML |
556 | /* |
557 | * MMIO register write with bytes helper functions | |
558 | * @offset:bytes offset from MMIO start | |
559 | * @value: the value want to be written to the register | |
b8920e1e SS |
560 | */ |
561 | ||
e3ecdffa AD |
562 | /** |
563 | * amdgpu_mm_wreg8 - read a memory mapped IO register | |
564 | * | |
565 | * @adev: amdgpu_device pointer | |
566 | * @offset: byte aligned register offset | |
567 | * @value: 8 bit value to write | |
568 | * | |
569 | * Writes the value specified to the offset specified. | |
570 | */ | |
7cbbc745 AG |
571 | void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value) |
572 | { | |
56b53c0b | 573 | if (amdgpu_device_skip_hw_access(adev)) |
bf36b52e AG |
574 | return; |
575 | ||
421a2a30 ML |
576 | if (offset < adev->rmmio_size) |
577 | writeb(value, adev->rmmio + offset); | |
578 | else | |
579 | BUG(); | |
580 | } | |
581 | ||
e3ecdffa | 582 | /** |
f7ee1874 | 583 | * amdgpu_device_wreg - write to a memory mapped IO or indirect register |
e3ecdffa AD |
584 | * |
585 | * @adev: amdgpu_device pointer | |
586 | * @reg: dword aligned register offset | |
587 | * @v: 32 bit value to write to the register | |
588 | * @acc_flags: access flags which require special behavior | |
589 | * | |
590 | * Writes the value specified to the offset specified. | |
591 | */ | |
f7ee1874 HZ |
592 | void amdgpu_device_wreg(struct amdgpu_device *adev, |
593 | uint32_t reg, uint32_t v, | |
594 | uint32_t acc_flags) | |
d38ceaf9 | 595 | { |
56b53c0b | 596 | if (amdgpu_device_skip_hw_access(adev)) |
bf36b52e AG |
597 | return; |
598 | ||
f7ee1874 HZ |
599 | if ((reg * 4) < adev->rmmio_size) { |
600 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && | |
601 | amdgpu_sriov_runtime(adev) && | |
d0fb18b5 | 602 | down_read_trylock(&adev->reset_domain->sem)) { |
85150626 | 603 | amdgpu_kiq_wreg(adev, reg, v, 0); |
d0fb18b5 | 604 | up_read(&adev->reset_domain->sem); |
f7ee1874 HZ |
605 | } else { |
606 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); | |
607 | } | |
608 | } else { | |
609 | adev->pcie_wreg(adev, reg * 4, v); | |
81202807 | 610 | } |
bc992ba5 | 611 | |
f7ee1874 | 612 | trace_amdgpu_device_wreg(adev->pdev->device, reg, v); |
2e0cc4d4 | 613 | } |
d38ceaf9 | 614 | |
03f2abb0 | 615 | /** |
4cc9f86f | 616 | * amdgpu_mm_wreg_mmio_rlc - write register either with direct/indirect mmio or with RLC path if in range |
2e0cc4d4 | 617 | * |
71579346 RB |
618 | * @adev: amdgpu_device pointer |
619 | * @reg: mmio/rlc register | |
620 | * @v: value to write | |
8057a9d6 | 621 | * @xcc_id: xcc accelerated compute core id |
71579346 RB |
622 | * |
623 | * this function is invoked only for the debugfs register access | |
03f2abb0 | 624 | */ |
f7ee1874 | 625 | void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev, |
8ed49dd1 VL |
626 | uint32_t reg, uint32_t v, |
627 | uint32_t xcc_id) | |
2e0cc4d4 | 628 | { |
56b53c0b | 629 | if (amdgpu_device_skip_hw_access(adev)) |
bf36b52e AG |
630 | return; |
631 | ||
2e0cc4d4 | 632 | if (amdgpu_sriov_fullaccess(adev) && |
f7ee1874 HZ |
633 | adev->gfx.rlc.funcs && |
634 | adev->gfx.rlc.funcs->is_rlcg_access_range) { | |
2e0cc4d4 | 635 | if (adev->gfx.rlc.funcs->is_rlcg_access_range(adev, reg)) |
8ed49dd1 | 636 | return amdgpu_sriov_wreg(adev, reg, v, 0, 0, xcc_id); |
4cc9f86f TSD |
637 | } else if ((reg * 4) >= adev->rmmio_size) { |
638 | adev->pcie_wreg(adev, reg * 4, v); | |
f7ee1874 HZ |
639 | } else { |
640 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); | |
47ed4e1c | 641 | } |
d38ceaf9 AD |
642 | } |
643 | ||
85150626 VL |
644 | /** |
645 | * amdgpu_device_xcc_wreg - write to a memory mapped IO or indirect register with specific XCC | |
646 | * | |
647 | * @adev: amdgpu_device pointer | |
648 | * @reg: dword aligned register offset | |
649 | * @v: 32 bit value to write to the register | |
650 | * @acc_flags: access flags which require special behavior | |
651 | * @xcc_id: xcc accelerated compute core id | |
652 | * | |
653 | * Writes the value specified to the offset specified. | |
654 | */ | |
655 | void amdgpu_device_xcc_wreg(struct amdgpu_device *adev, | |
656 | uint32_t reg, uint32_t v, | |
657 | uint32_t acc_flags, uint32_t xcc_id) | |
658 | { | |
659 | uint32_t rlcg_flag; | |
660 | ||
661 | if (amdgpu_device_skip_hw_access(adev)) | |
662 | return; | |
663 | ||
664 | if ((reg * 4) < adev->rmmio_size) { | |
665 | if (amdgpu_sriov_vf(adev) && | |
666 | !amdgpu_sriov_runtime(adev) && | |
667 | adev->gfx.rlc.rlcg_reg_access_supported && | |
668 | amdgpu_virt_get_rlcg_reg_access_flag(adev, acc_flags, | |
669 | GC_HWIP, true, | |
670 | &rlcg_flag)) { | |
671 | amdgpu_virt_rlcg_reg_rw(adev, reg, v, rlcg_flag, xcc_id); | |
672 | } else if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && | |
673 | amdgpu_sriov_runtime(adev) && | |
674 | down_read_trylock(&adev->reset_domain->sem)) { | |
675 | amdgpu_kiq_wreg(adev, reg, v, xcc_id); | |
676 | up_read(&adev->reset_domain->sem); | |
677 | } else { | |
678 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); | |
679 | } | |
680 | } else { | |
681 | adev->pcie_wreg(adev, reg * 4, v); | |
682 | } | |
683 | } | |
684 | ||
1bba3683 HZ |
685 | /** |
686 | * amdgpu_device_indirect_rreg - read an indirect register | |
687 | * | |
688 | * @adev: amdgpu_device pointer | |
22f453fb | 689 | * @reg_addr: indirect register address to read from |
1bba3683 HZ |
690 | * |
691 | * Returns the value of indirect register @reg_addr | |
692 | */ | |
693 | u32 amdgpu_device_indirect_rreg(struct amdgpu_device *adev, | |
1bba3683 HZ |
694 | u32 reg_addr) |
695 | { | |
65ba96e9 | 696 | unsigned long flags, pcie_index, pcie_data; |
1bba3683 HZ |
697 | void __iomem *pcie_index_offset; |
698 | void __iomem *pcie_data_offset; | |
65ba96e9 HZ |
699 | u32 r; |
700 | ||
701 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); | |
702 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
1bba3683 HZ |
703 | |
704 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
705 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
706 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
707 | ||
708 | writel(reg_addr, pcie_index_offset); | |
709 | readl(pcie_index_offset); | |
710 | r = readl(pcie_data_offset); | |
711 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
712 | ||
713 | return r; | |
714 | } | |
715 | ||
0c552ed3 LM |
716 | u32 amdgpu_device_indirect_rreg_ext(struct amdgpu_device *adev, |
717 | u64 reg_addr) | |
718 | { | |
719 | unsigned long flags, pcie_index, pcie_index_hi, pcie_data; | |
720 | u32 r; | |
721 | void __iomem *pcie_index_offset; | |
722 | void __iomem *pcie_index_hi_offset; | |
723 | void __iomem *pcie_data_offset; | |
724 | ||
725 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); | |
726 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
d57e24aa | 727 | if ((reg_addr >> 32) && (adev->nbio.funcs->get_pcie_index_hi_offset)) |
0c552ed3 LM |
728 | pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev); |
729 | else | |
730 | pcie_index_hi = 0; | |
731 | ||
732 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
733 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
734 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
735 | if (pcie_index_hi != 0) | |
736 | pcie_index_hi_offset = (void __iomem *)adev->rmmio + | |
737 | pcie_index_hi * 4; | |
738 | ||
739 | writel(reg_addr, pcie_index_offset); | |
740 | readl(pcie_index_offset); | |
741 | if (pcie_index_hi != 0) { | |
742 | writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset); | |
743 | readl(pcie_index_hi_offset); | |
744 | } | |
745 | r = readl(pcie_data_offset); | |
746 | ||
747 | /* clear the high bits */ | |
748 | if (pcie_index_hi != 0) { | |
749 | writel(0, pcie_index_hi_offset); | |
750 | readl(pcie_index_hi_offset); | |
751 | } | |
752 | ||
753 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
754 | ||
755 | return r; | |
756 | } | |
757 | ||
1bba3683 HZ |
758 | /** |
759 | * amdgpu_device_indirect_rreg64 - read a 64bits indirect register | |
760 | * | |
761 | * @adev: amdgpu_device pointer | |
22f453fb | 762 | * @reg_addr: indirect register address to read from |
1bba3683 HZ |
763 | * |
764 | * Returns the value of indirect register @reg_addr | |
765 | */ | |
766 | u64 amdgpu_device_indirect_rreg64(struct amdgpu_device *adev, | |
1bba3683 HZ |
767 | u32 reg_addr) |
768 | { | |
65ba96e9 | 769 | unsigned long flags, pcie_index, pcie_data; |
1bba3683 HZ |
770 | void __iomem *pcie_index_offset; |
771 | void __iomem *pcie_data_offset; | |
65ba96e9 HZ |
772 | u64 r; |
773 | ||
774 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); | |
775 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
1bba3683 HZ |
776 | |
777 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
778 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
779 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
780 | ||
781 | /* read low 32 bits */ | |
782 | writel(reg_addr, pcie_index_offset); | |
783 | readl(pcie_index_offset); | |
784 | r = readl(pcie_data_offset); | |
785 | /* read high 32 bits */ | |
786 | writel(reg_addr + 4, pcie_index_offset); | |
787 | readl(pcie_index_offset); | |
788 | r |= ((u64)readl(pcie_data_offset) << 32); | |
789 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
790 | ||
791 | return r; | |
792 | } | |
793 | ||
a76b2870 CL |
794 | u64 amdgpu_device_indirect_rreg64_ext(struct amdgpu_device *adev, |
795 | u64 reg_addr) | |
796 | { | |
797 | unsigned long flags, pcie_index, pcie_data; | |
798 | unsigned long pcie_index_hi = 0; | |
799 | void __iomem *pcie_index_offset; | |
800 | void __iomem *pcie_index_hi_offset; | |
801 | void __iomem *pcie_data_offset; | |
802 | u64 r; | |
803 | ||
804 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); | |
805 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
806 | if ((reg_addr >> 32) && (adev->nbio.funcs->get_pcie_index_hi_offset)) | |
807 | pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev); | |
808 | ||
809 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
810 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
811 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
812 | if (pcie_index_hi != 0) | |
813 | pcie_index_hi_offset = (void __iomem *)adev->rmmio + | |
814 | pcie_index_hi * 4; | |
815 | ||
816 | /* read low 32 bits */ | |
817 | writel(reg_addr, pcie_index_offset); | |
818 | readl(pcie_index_offset); | |
819 | if (pcie_index_hi != 0) { | |
820 | writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset); | |
821 | readl(pcie_index_hi_offset); | |
822 | } | |
823 | r = readl(pcie_data_offset); | |
824 | /* read high 32 bits */ | |
825 | writel(reg_addr + 4, pcie_index_offset); | |
826 | readl(pcie_index_offset); | |
827 | if (pcie_index_hi != 0) { | |
828 | writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset); | |
829 | readl(pcie_index_hi_offset); | |
830 | } | |
831 | r |= ((u64)readl(pcie_data_offset) << 32); | |
832 | ||
833 | /* clear the high bits */ | |
834 | if (pcie_index_hi != 0) { | |
835 | writel(0, pcie_index_hi_offset); | |
836 | readl(pcie_index_hi_offset); | |
837 | } | |
838 | ||
839 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
840 | ||
841 | return r; | |
842 | } | |
843 | ||
1bba3683 HZ |
844 | /** |
845 | * amdgpu_device_indirect_wreg - write an indirect register address | |
846 | * | |
847 | * @adev: amdgpu_device pointer | |
1bba3683 HZ |
848 | * @reg_addr: indirect register offset |
849 | * @reg_data: indirect register data | |
850 | * | |
851 | */ | |
852 | void amdgpu_device_indirect_wreg(struct amdgpu_device *adev, | |
1bba3683 HZ |
853 | u32 reg_addr, u32 reg_data) |
854 | { | |
65ba96e9 | 855 | unsigned long flags, pcie_index, pcie_data; |
1bba3683 HZ |
856 | void __iomem *pcie_index_offset; |
857 | void __iomem *pcie_data_offset; | |
858 | ||
65ba96e9 HZ |
859 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); |
860 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
861 | ||
1bba3683 HZ |
862 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); |
863 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
864 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
865 | ||
866 | writel(reg_addr, pcie_index_offset); | |
867 | readl(pcie_index_offset); | |
868 | writel(reg_data, pcie_data_offset); | |
869 | readl(pcie_data_offset); | |
870 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
871 | } | |
872 | ||
0c552ed3 LM |
873 | void amdgpu_device_indirect_wreg_ext(struct amdgpu_device *adev, |
874 | u64 reg_addr, u32 reg_data) | |
875 | { | |
876 | unsigned long flags, pcie_index, pcie_index_hi, pcie_data; | |
877 | void __iomem *pcie_index_offset; | |
878 | void __iomem *pcie_index_hi_offset; | |
879 | void __iomem *pcie_data_offset; | |
880 | ||
881 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); | |
882 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
d57e24aa | 883 | if ((reg_addr >> 32) && (adev->nbio.funcs->get_pcie_index_hi_offset)) |
0c552ed3 LM |
884 | pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev); |
885 | else | |
886 | pcie_index_hi = 0; | |
887 | ||
888 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
889 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
890 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
891 | if (pcie_index_hi != 0) | |
892 | pcie_index_hi_offset = (void __iomem *)adev->rmmio + | |
893 | pcie_index_hi * 4; | |
894 | ||
895 | writel(reg_addr, pcie_index_offset); | |
896 | readl(pcie_index_offset); | |
897 | if (pcie_index_hi != 0) { | |
898 | writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset); | |
899 | readl(pcie_index_hi_offset); | |
900 | } | |
901 | writel(reg_data, pcie_data_offset); | |
902 | readl(pcie_data_offset); | |
903 | ||
904 | /* clear the high bits */ | |
905 | if (pcie_index_hi != 0) { | |
906 | writel(0, pcie_index_hi_offset); | |
907 | readl(pcie_index_hi_offset); | |
908 | } | |
909 | ||
910 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
911 | } | |
912 | ||
1bba3683 HZ |
913 | /** |
914 | * amdgpu_device_indirect_wreg64 - write a 64bits indirect register address | |
915 | * | |
916 | * @adev: amdgpu_device pointer | |
1bba3683 HZ |
917 | * @reg_addr: indirect register offset |
918 | * @reg_data: indirect register data | |
919 | * | |
920 | */ | |
921 | void amdgpu_device_indirect_wreg64(struct amdgpu_device *adev, | |
1bba3683 HZ |
922 | u32 reg_addr, u64 reg_data) |
923 | { | |
65ba96e9 | 924 | unsigned long flags, pcie_index, pcie_data; |
1bba3683 HZ |
925 | void __iomem *pcie_index_offset; |
926 | void __iomem *pcie_data_offset; | |
927 | ||
65ba96e9 HZ |
928 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); |
929 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
930 | ||
1bba3683 HZ |
931 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); |
932 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
933 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
934 | ||
935 | /* write low 32 bits */ | |
936 | writel(reg_addr, pcie_index_offset); | |
937 | readl(pcie_index_offset); | |
938 | writel((u32)(reg_data & 0xffffffffULL), pcie_data_offset); | |
939 | readl(pcie_data_offset); | |
940 | /* write high 32 bits */ | |
941 | writel(reg_addr + 4, pcie_index_offset); | |
942 | readl(pcie_index_offset); | |
943 | writel((u32)(reg_data >> 32), pcie_data_offset); | |
944 | readl(pcie_data_offset); | |
945 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
946 | } | |
947 | ||
a76b2870 CL |
948 | void amdgpu_device_indirect_wreg64_ext(struct amdgpu_device *adev, |
949 | u64 reg_addr, u64 reg_data) | |
950 | { | |
951 | unsigned long flags, pcie_index, pcie_data; | |
952 | unsigned long pcie_index_hi = 0; | |
953 | void __iomem *pcie_index_offset; | |
954 | void __iomem *pcie_index_hi_offset; | |
955 | void __iomem *pcie_data_offset; | |
956 | ||
957 | pcie_index = adev->nbio.funcs->get_pcie_index_offset(adev); | |
958 | pcie_data = adev->nbio.funcs->get_pcie_data_offset(adev); | |
959 | if ((reg_addr >> 32) && (adev->nbio.funcs->get_pcie_index_hi_offset)) | |
960 | pcie_index_hi = adev->nbio.funcs->get_pcie_index_hi_offset(adev); | |
961 | ||
962 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
963 | pcie_index_offset = (void __iomem *)adev->rmmio + pcie_index * 4; | |
964 | pcie_data_offset = (void __iomem *)adev->rmmio + pcie_data * 4; | |
965 | if (pcie_index_hi != 0) | |
966 | pcie_index_hi_offset = (void __iomem *)adev->rmmio + | |
967 | pcie_index_hi * 4; | |
968 | ||
969 | /* write low 32 bits */ | |
970 | writel(reg_addr, pcie_index_offset); | |
971 | readl(pcie_index_offset); | |
972 | if (pcie_index_hi != 0) { | |
973 | writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset); | |
974 | readl(pcie_index_hi_offset); | |
975 | } | |
976 | writel((u32)(reg_data & 0xffffffffULL), pcie_data_offset); | |
977 | readl(pcie_data_offset); | |
978 | /* write high 32 bits */ | |
979 | writel(reg_addr + 4, pcie_index_offset); | |
980 | readl(pcie_index_offset); | |
981 | if (pcie_index_hi != 0) { | |
982 | writel((reg_addr >> 32) & 0xff, pcie_index_hi_offset); | |
983 | readl(pcie_index_hi_offset); | |
984 | } | |
985 | writel((u32)(reg_data >> 32), pcie_data_offset); | |
986 | readl(pcie_data_offset); | |
987 | ||
988 | /* clear the high bits */ | |
989 | if (pcie_index_hi != 0) { | |
990 | writel(0, pcie_index_hi_offset); | |
991 | readl(pcie_index_hi_offset); | |
992 | } | |
993 | ||
994 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
995 | } | |
996 | ||
dabc114e HZ |
997 | /** |
998 | * amdgpu_device_get_rev_id - query device rev_id | |
999 | * | |
1000 | * @adev: amdgpu_device pointer | |
1001 | * | |
1002 | * Return device rev_id | |
1003 | */ | |
1004 | u32 amdgpu_device_get_rev_id(struct amdgpu_device *adev) | |
1005 | { | |
1006 | return adev->nbio.funcs->get_rev_id(adev); | |
1007 | } | |
1008 | ||
d38ceaf9 AD |
1009 | /** |
1010 | * amdgpu_invalid_rreg - dummy reg read function | |
1011 | * | |
982a820b | 1012 | * @adev: amdgpu_device pointer |
d38ceaf9 AD |
1013 | * @reg: offset of register |
1014 | * | |
1015 | * Dummy register read function. Used for register blocks | |
1016 | * that certain asics don't have (all asics). | |
1017 | * Returns the value in the register. | |
1018 | */ | |
1019 | static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg) | |
1020 | { | |
1021 | DRM_ERROR("Invalid callback to read register 0x%04X\n", reg); | |
1022 | BUG(); | |
1023 | return 0; | |
1024 | } | |
1025 | ||
0c552ed3 LM |
1026 | static uint32_t amdgpu_invalid_rreg_ext(struct amdgpu_device *adev, uint64_t reg) |
1027 | { | |
1028 | DRM_ERROR("Invalid callback to read register 0x%llX\n", reg); | |
1029 | BUG(); | |
1030 | return 0; | |
1031 | } | |
1032 | ||
d38ceaf9 AD |
1033 | /** |
1034 | * amdgpu_invalid_wreg - dummy reg write function | |
1035 | * | |
982a820b | 1036 | * @adev: amdgpu_device pointer |
d38ceaf9 AD |
1037 | * @reg: offset of register |
1038 | * @v: value to write to the register | |
1039 | * | |
1040 | * Dummy register read function. Used for register blocks | |
1041 | * that certain asics don't have (all asics). | |
1042 | */ | |
1043 | static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) | |
1044 | { | |
1045 | DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n", | |
1046 | reg, v); | |
1047 | BUG(); | |
1048 | } | |
1049 | ||
0c552ed3 LM |
1050 | static void amdgpu_invalid_wreg_ext(struct amdgpu_device *adev, uint64_t reg, uint32_t v) |
1051 | { | |
1052 | DRM_ERROR("Invalid callback to write register 0x%llX with 0x%08X\n", | |
1053 | reg, v); | |
1054 | BUG(); | |
1055 | } | |
1056 | ||
4fa1c6a6 TZ |
1057 | /** |
1058 | * amdgpu_invalid_rreg64 - dummy 64 bit reg read function | |
1059 | * | |
982a820b | 1060 | * @adev: amdgpu_device pointer |
4fa1c6a6 TZ |
1061 | * @reg: offset of register |
1062 | * | |
1063 | * Dummy register read function. Used for register blocks | |
1064 | * that certain asics don't have (all asics). | |
1065 | * Returns the value in the register. | |
1066 | */ | |
1067 | static uint64_t amdgpu_invalid_rreg64(struct amdgpu_device *adev, uint32_t reg) | |
1068 | { | |
1069 | DRM_ERROR("Invalid callback to read 64 bit register 0x%04X\n", reg); | |
1070 | BUG(); | |
1071 | return 0; | |
1072 | } | |
1073 | ||
a76b2870 CL |
1074 | static uint64_t amdgpu_invalid_rreg64_ext(struct amdgpu_device *adev, uint64_t reg) |
1075 | { | |
1076 | DRM_ERROR("Invalid callback to read register 0x%llX\n", reg); | |
1077 | BUG(); | |
1078 | return 0; | |
1079 | } | |
1080 | ||
4fa1c6a6 TZ |
1081 | /** |
1082 | * amdgpu_invalid_wreg64 - dummy reg write function | |
1083 | * | |
982a820b | 1084 | * @adev: amdgpu_device pointer |
4fa1c6a6 TZ |
1085 | * @reg: offset of register |
1086 | * @v: value to write to the register | |
1087 | * | |
1088 | * Dummy register read function. Used for register blocks | |
1089 | * that certain asics don't have (all asics). | |
1090 | */ | |
1091 | static void amdgpu_invalid_wreg64(struct amdgpu_device *adev, uint32_t reg, uint64_t v) | |
1092 | { | |
1093 | DRM_ERROR("Invalid callback to write 64 bit register 0x%04X with 0x%08llX\n", | |
1094 | reg, v); | |
1095 | BUG(); | |
1096 | } | |
1097 | ||
a76b2870 CL |
1098 | static void amdgpu_invalid_wreg64_ext(struct amdgpu_device *adev, uint64_t reg, uint64_t v) |
1099 | { | |
1100 | DRM_ERROR("Invalid callback to write 64 bit register 0x%llX with 0x%08llX\n", | |
1101 | reg, v); | |
1102 | BUG(); | |
1103 | } | |
1104 | ||
d38ceaf9 AD |
1105 | /** |
1106 | * amdgpu_block_invalid_rreg - dummy reg read function | |
1107 | * | |
982a820b | 1108 | * @adev: amdgpu_device pointer |
d38ceaf9 AD |
1109 | * @block: offset of instance |
1110 | * @reg: offset of register | |
1111 | * | |
1112 | * Dummy register read function. Used for register blocks | |
1113 | * that certain asics don't have (all asics). | |
1114 | * Returns the value in the register. | |
1115 | */ | |
1116 | static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev, | |
1117 | uint32_t block, uint32_t reg) | |
1118 | { | |
1119 | DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n", | |
1120 | reg, block); | |
1121 | BUG(); | |
1122 | return 0; | |
1123 | } | |
1124 | ||
1125 | /** | |
1126 | * amdgpu_block_invalid_wreg - dummy reg write function | |
1127 | * | |
982a820b | 1128 | * @adev: amdgpu_device pointer |
d38ceaf9 AD |
1129 | * @block: offset of instance |
1130 | * @reg: offset of register | |
1131 | * @v: value to write to the register | |
1132 | * | |
1133 | * Dummy register read function. Used for register blocks | |
1134 | * that certain asics don't have (all asics). | |
1135 | */ | |
1136 | static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev, | |
1137 | uint32_t block, | |
1138 | uint32_t reg, uint32_t v) | |
1139 | { | |
1140 | DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n", | |
1141 | reg, block, v); | |
1142 | BUG(); | |
1143 | } | |
1144 | ||
4d2997ab AD |
1145 | /** |
1146 | * amdgpu_device_asic_init - Wrapper for atom asic_init | |
1147 | * | |
982a820b | 1148 | * @adev: amdgpu_device pointer |
4d2997ab AD |
1149 | * |
1150 | * Does any asic specific work and then calls atom asic init. | |
1151 | */ | |
1152 | static int amdgpu_device_asic_init(struct amdgpu_device *adev) | |
1153 | { | |
15c5c5f5 LL |
1154 | int ret; |
1155 | ||
4d2997ab AD |
1156 | amdgpu_asic_pre_asic_init(adev); |
1157 | ||
4e8303cf LL |
1158 | if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) || |
1159 | amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(11, 0, 0)) { | |
15c5c5f5 LL |
1160 | amdgpu_psp_wait_for_bootloader(adev); |
1161 | ret = amdgpu_atomfirmware_asic_init(adev, true); | |
23618280 HZ |
1162 | /* TODO: check the return val and stop device initialization if boot fails */ |
1163 | amdgpu_psp_query_boot_status(adev); | |
15c5c5f5 LL |
1164 | return ret; |
1165 | } else { | |
85d1bcc6 | 1166 | return amdgpu_atom_asic_init(adev->mode_info.atom_context); |
15c5c5f5 LL |
1167 | } |
1168 | ||
1169 | return 0; | |
4d2997ab AD |
1170 | } |
1171 | ||
e3ecdffa | 1172 | /** |
7ccfd79f | 1173 | * amdgpu_device_mem_scratch_init - allocate the VRAM scratch page |
e3ecdffa | 1174 | * |
982a820b | 1175 | * @adev: amdgpu_device pointer |
e3ecdffa AD |
1176 | * |
1177 | * Allocates a scratch page of VRAM for use by various things in the | |
1178 | * driver. | |
1179 | */ | |
7ccfd79f | 1180 | static int amdgpu_device_mem_scratch_init(struct amdgpu_device *adev) |
d38ceaf9 | 1181 | { |
7ccfd79f CK |
1182 | return amdgpu_bo_create_kernel(adev, AMDGPU_GPU_PAGE_SIZE, PAGE_SIZE, |
1183 | AMDGPU_GEM_DOMAIN_VRAM | | |
1184 | AMDGPU_GEM_DOMAIN_GTT, | |
1185 | &adev->mem_scratch.robj, | |
1186 | &adev->mem_scratch.gpu_addr, | |
1187 | (void **)&adev->mem_scratch.ptr); | |
d38ceaf9 AD |
1188 | } |
1189 | ||
e3ecdffa | 1190 | /** |
7ccfd79f | 1191 | * amdgpu_device_mem_scratch_fini - Free the VRAM scratch page |
e3ecdffa | 1192 | * |
982a820b | 1193 | * @adev: amdgpu_device pointer |
e3ecdffa AD |
1194 | * |
1195 | * Frees the VRAM scratch page. | |
1196 | */ | |
7ccfd79f | 1197 | static void amdgpu_device_mem_scratch_fini(struct amdgpu_device *adev) |
d38ceaf9 | 1198 | { |
7ccfd79f | 1199 | amdgpu_bo_free_kernel(&adev->mem_scratch.robj, NULL, NULL); |
d38ceaf9 AD |
1200 | } |
1201 | ||
1202 | /** | |
9c3f2b54 | 1203 | * amdgpu_device_program_register_sequence - program an array of registers. |
d38ceaf9 AD |
1204 | * |
1205 | * @adev: amdgpu_device pointer | |
1206 | * @registers: pointer to the register array | |
1207 | * @array_size: size of the register array | |
1208 | * | |
b8920e1e | 1209 | * Programs an array or registers with and or masks. |
d38ceaf9 AD |
1210 | * This is a helper for setting golden registers. |
1211 | */ | |
9c3f2b54 AD |
1212 | void amdgpu_device_program_register_sequence(struct amdgpu_device *adev, |
1213 | const u32 *registers, | |
1214 | const u32 array_size) | |
d38ceaf9 AD |
1215 | { |
1216 | u32 tmp, reg, and_mask, or_mask; | |
1217 | int i; | |
1218 | ||
1219 | if (array_size % 3) | |
1220 | return; | |
1221 | ||
47fc644f | 1222 | for (i = 0; i < array_size; i += 3) { |
d38ceaf9 AD |
1223 | reg = registers[i + 0]; |
1224 | and_mask = registers[i + 1]; | |
1225 | or_mask = registers[i + 2]; | |
1226 | ||
1227 | if (and_mask == 0xffffffff) { | |
1228 | tmp = or_mask; | |
1229 | } else { | |
1230 | tmp = RREG32(reg); | |
1231 | tmp &= ~and_mask; | |
e0d07657 HZ |
1232 | if (adev->family >= AMDGPU_FAMILY_AI) |
1233 | tmp |= (or_mask & and_mask); | |
1234 | else | |
1235 | tmp |= or_mask; | |
d38ceaf9 AD |
1236 | } |
1237 | WREG32(reg, tmp); | |
1238 | } | |
1239 | } | |
1240 | ||
e3ecdffa AD |
1241 | /** |
1242 | * amdgpu_device_pci_config_reset - reset the GPU | |
1243 | * | |
1244 | * @adev: amdgpu_device pointer | |
1245 | * | |
1246 | * Resets the GPU using the pci config reset sequence. | |
1247 | * Only applicable to asics prior to vega10. | |
1248 | */ | |
8111c387 | 1249 | void amdgpu_device_pci_config_reset(struct amdgpu_device *adev) |
d38ceaf9 AD |
1250 | { |
1251 | pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA); | |
1252 | } | |
1253 | ||
af484df8 AD |
1254 | /** |
1255 | * amdgpu_device_pci_reset - reset the GPU using generic PCI means | |
1256 | * | |
1257 | * @adev: amdgpu_device pointer | |
1258 | * | |
1259 | * Resets the GPU using generic pci reset interfaces (FLR, SBR, etc.). | |
1260 | */ | |
1261 | int amdgpu_device_pci_reset(struct amdgpu_device *adev) | |
1262 | { | |
1263 | return pci_reset_function(adev->pdev); | |
1264 | } | |
1265 | ||
d38ceaf9 | 1266 | /* |
06ec9070 | 1267 | * amdgpu_device_wb_*() |
455a7bc2 | 1268 | * Writeback is the method by which the GPU updates special pages in memory |
ea81a173 | 1269 | * with the status of certain GPU events (fences, ring pointers,etc.). |
d38ceaf9 AD |
1270 | */ |
1271 | ||
1272 | /** | |
06ec9070 | 1273 | * amdgpu_device_wb_fini - Disable Writeback and free memory |
d38ceaf9 AD |
1274 | * |
1275 | * @adev: amdgpu_device pointer | |
1276 | * | |
1277 | * Disables Writeback and frees the Writeback memory (all asics). | |
1278 | * Used at driver shutdown. | |
1279 | */ | |
06ec9070 | 1280 | static void amdgpu_device_wb_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
1281 | { |
1282 | if (adev->wb.wb_obj) { | |
a76ed485 AD |
1283 | amdgpu_bo_free_kernel(&adev->wb.wb_obj, |
1284 | &adev->wb.gpu_addr, | |
1285 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
1286 | adev->wb.wb_obj = NULL; |
1287 | } | |
1288 | } | |
1289 | ||
1290 | /** | |
03f2abb0 | 1291 | * amdgpu_device_wb_init - Init Writeback driver info and allocate memory |
d38ceaf9 AD |
1292 | * |
1293 | * @adev: amdgpu_device pointer | |
1294 | * | |
455a7bc2 | 1295 | * Initializes writeback and allocates writeback memory (all asics). |
d38ceaf9 AD |
1296 | * Used at driver startup. |
1297 | * Returns 0 on success or an -error on failure. | |
1298 | */ | |
06ec9070 | 1299 | static int amdgpu_device_wb_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
1300 | { |
1301 | int r; | |
1302 | ||
1303 | if (adev->wb.wb_obj == NULL) { | |
97407b63 AD |
1304 | /* AMDGPU_MAX_WB * sizeof(uint32_t) * 8 = AMDGPU_MAX_WB 256bit slots */ |
1305 | r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t) * 8, | |
a76ed485 AD |
1306 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, |
1307 | &adev->wb.wb_obj, &adev->wb.gpu_addr, | |
1308 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
1309 | if (r) { |
1310 | dev_warn(adev->dev, "(%d) create WB bo failed\n", r); | |
1311 | return r; | |
1312 | } | |
d38ceaf9 AD |
1313 | |
1314 | adev->wb.num_wb = AMDGPU_MAX_WB; | |
1315 | memset(&adev->wb.used, 0, sizeof(adev->wb.used)); | |
1316 | ||
1317 | /* clear wb memory */ | |
73469585 | 1318 | memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t) * 8); |
d38ceaf9 AD |
1319 | } |
1320 | ||
1321 | return 0; | |
1322 | } | |
1323 | ||
1324 | /** | |
131b4b36 | 1325 | * amdgpu_device_wb_get - Allocate a wb entry |
d38ceaf9 AD |
1326 | * |
1327 | * @adev: amdgpu_device pointer | |
1328 | * @wb: wb index | |
1329 | * | |
1330 | * Allocate a wb slot for use by the driver (all asics). | |
1331 | * Returns 0 on success or -EINVAL on failure. | |
1332 | */ | |
131b4b36 | 1333 | int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb) |
d38ceaf9 AD |
1334 | { |
1335 | unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb); | |
d38ceaf9 | 1336 | |
97407b63 | 1337 | if (offset < adev->wb.num_wb) { |
7014285a | 1338 | __set_bit(offset, adev->wb.used); |
63ae07ca | 1339 | *wb = offset << 3; /* convert to dw offset */ |
0915fdbc ML |
1340 | return 0; |
1341 | } else { | |
1342 | return -EINVAL; | |
1343 | } | |
1344 | } | |
1345 | ||
d38ceaf9 | 1346 | /** |
131b4b36 | 1347 | * amdgpu_device_wb_free - Free a wb entry |
d38ceaf9 AD |
1348 | * |
1349 | * @adev: amdgpu_device pointer | |
1350 | * @wb: wb index | |
1351 | * | |
1352 | * Free a wb slot allocated for use by the driver (all asics) | |
1353 | */ | |
131b4b36 | 1354 | void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb) |
d38ceaf9 | 1355 | { |
73469585 | 1356 | wb >>= 3; |
d38ceaf9 | 1357 | if (wb < adev->wb.num_wb) |
73469585 | 1358 | __clear_bit(wb, adev->wb.used); |
d38ceaf9 AD |
1359 | } |
1360 | ||
d6895ad3 CK |
1361 | /** |
1362 | * amdgpu_device_resize_fb_bar - try to resize FB BAR | |
1363 | * | |
1364 | * @adev: amdgpu_device pointer | |
1365 | * | |
1366 | * Try to resize FB BAR to make all VRAM CPU accessible. We try very hard not | |
1367 | * to fail, but if any of the BARs is not accessible after the size we abort | |
1368 | * driver loading by returning -ENODEV. | |
1369 | */ | |
1370 | int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev) | |
1371 | { | |
453f617a | 1372 | int rbar_size = pci_rebar_bytes_to_size(adev->gmc.real_vram_size); |
31b8adab CK |
1373 | struct pci_bus *root; |
1374 | struct resource *res; | |
b8920e1e | 1375 | unsigned int i; |
d6895ad3 CK |
1376 | u16 cmd; |
1377 | int r; | |
1378 | ||
822130b5 AB |
1379 | if (!IS_ENABLED(CONFIG_PHYS_ADDR_T_64BIT)) |
1380 | return 0; | |
1381 | ||
0c03b912 | 1382 | /* Bypass for VF */ |
1383 | if (amdgpu_sriov_vf(adev)) | |
1384 | return 0; | |
1385 | ||
b7221f2b AD |
1386 | /* skip if the bios has already enabled large BAR */ |
1387 | if (adev->gmc.real_vram_size && | |
1388 | (pci_resource_len(adev->pdev, 0) >= adev->gmc.real_vram_size)) | |
1389 | return 0; | |
1390 | ||
31b8adab CK |
1391 | /* Check if the root BUS has 64bit memory resources */ |
1392 | root = adev->pdev->bus; | |
1393 | while (root->parent) | |
1394 | root = root->parent; | |
1395 | ||
1396 | pci_bus_for_each_resource(root, res, i) { | |
0ebb7c54 | 1397 | if (res && res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) && |
31b8adab CK |
1398 | res->start > 0x100000000ull) |
1399 | break; | |
1400 | } | |
1401 | ||
1402 | /* Trying to resize is pointless without a root hub window above 4GB */ | |
1403 | if (!res) | |
1404 | return 0; | |
1405 | ||
453f617a ND |
1406 | /* Limit the BAR size to what is available */ |
1407 | rbar_size = min(fls(pci_rebar_get_possible_sizes(adev->pdev, 0)) - 1, | |
1408 | rbar_size); | |
1409 | ||
d6895ad3 CK |
1410 | /* Disable memory decoding while we change the BAR addresses and size */ |
1411 | pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd); | |
1412 | pci_write_config_word(adev->pdev, PCI_COMMAND, | |
1413 | cmd & ~PCI_COMMAND_MEMORY); | |
1414 | ||
1415 | /* Free the VRAM and doorbell BAR, we most likely need to move both. */ | |
43c064db | 1416 | amdgpu_doorbell_fini(adev); |
d6895ad3 CK |
1417 | if (adev->asic_type >= CHIP_BONAIRE) |
1418 | pci_release_resource(adev->pdev, 2); | |
1419 | ||
1420 | pci_release_resource(adev->pdev, 0); | |
1421 | ||
1422 | r = pci_resize_resource(adev->pdev, 0, rbar_size); | |
1423 | if (r == -ENOSPC) | |
1424 | DRM_INFO("Not enough PCI address space for a large BAR."); | |
1425 | else if (r && r != -ENOTSUPP) | |
1426 | DRM_ERROR("Problem resizing BAR0 (%d).", r); | |
1427 | ||
1428 | pci_assign_unassigned_bus_resources(adev->pdev->bus); | |
1429 | ||
1430 | /* When the doorbell or fb BAR isn't available we have no chance of | |
1431 | * using the device. | |
1432 | */ | |
43c064db | 1433 | r = amdgpu_doorbell_init(adev); |
d6895ad3 CK |
1434 | if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET)) |
1435 | return -ENODEV; | |
1436 | ||
1437 | pci_write_config_word(adev->pdev, PCI_COMMAND, cmd); | |
1438 | ||
1439 | return 0; | |
1440 | } | |
a05502e5 | 1441 | |
9535a86a SZ |
1442 | static bool amdgpu_device_read_bios(struct amdgpu_device *adev) |
1443 | { | |
b8920e1e | 1444 | if (hweight32(adev->aid_mask) && (adev->flags & AMD_IS_APU)) |
9535a86a | 1445 | return false; |
9535a86a SZ |
1446 | |
1447 | return true; | |
1448 | } | |
1449 | ||
d38ceaf9 AD |
1450 | /* |
1451 | * GPU helpers function. | |
1452 | */ | |
1453 | /** | |
39c640c0 | 1454 | * amdgpu_device_need_post - check if the hw need post or not |
d38ceaf9 AD |
1455 | * |
1456 | * @adev: amdgpu_device pointer | |
1457 | * | |
c836fec5 JQ |
1458 | * Check if the asic has been initialized (all asics) at driver startup |
1459 | * or post is needed if hw reset is performed. | |
1460 | * Returns true if need or false if not. | |
d38ceaf9 | 1461 | */ |
39c640c0 | 1462 | bool amdgpu_device_need_post(struct amdgpu_device *adev) |
d38ceaf9 AD |
1463 | { |
1464 | uint32_t reg; | |
1465 | ||
bec86378 ML |
1466 | if (amdgpu_sriov_vf(adev)) |
1467 | return false; | |
1468 | ||
9535a86a SZ |
1469 | if (!amdgpu_device_read_bios(adev)) |
1470 | return false; | |
1471 | ||
bec86378 | 1472 | if (amdgpu_passthrough(adev)) { |
1da2c326 ML |
1473 | /* for FIJI: In whole GPU pass-through virtualization case, after VM reboot |
1474 | * some old smc fw still need driver do vPost otherwise gpu hang, while | |
1475 | * those smc fw version above 22.15 doesn't have this flaw, so we force | |
1476 | * vpost executed for smc version below 22.15 | |
bec86378 ML |
1477 | */ |
1478 | if (adev->asic_type == CHIP_FIJI) { | |
1479 | int err; | |
1480 | uint32_t fw_ver; | |
b8920e1e | 1481 | |
bec86378 ML |
1482 | err = request_firmware(&adev->pm.fw, "amdgpu/fiji_smc.bin", adev->dev); |
1483 | /* force vPost if error occured */ | |
1484 | if (err) | |
1485 | return true; | |
1486 | ||
1487 | fw_ver = *((uint32_t *)adev->pm.fw->data + 69); | |
1da2c326 ML |
1488 | if (fw_ver < 0x00160e00) |
1489 | return true; | |
bec86378 | 1490 | } |
bec86378 | 1491 | } |
91fe77eb | 1492 | |
e3c1b071 | 1493 | /* Don't post if we need to reset whole hive on init */ |
1494 | if (adev->gmc.xgmi.pending_reset) | |
1495 | return false; | |
1496 | ||
91fe77eb | 1497 | if (adev->has_hw_reset) { |
1498 | adev->has_hw_reset = false; | |
1499 | return true; | |
1500 | } | |
1501 | ||
1502 | /* bios scratch used on CIK+ */ | |
1503 | if (adev->asic_type >= CHIP_BONAIRE) | |
1504 | return amdgpu_atombios_scratch_need_asic_init(adev); | |
1505 | ||
1506 | /* check MEM_SIZE for older asics */ | |
1507 | reg = amdgpu_asic_get_config_memsize(adev); | |
1508 | ||
1509 | if ((reg != 0) && (reg != 0xffffffff)) | |
1510 | return false; | |
1511 | ||
1512 | return true; | |
70e64c4d ML |
1513 | } |
1514 | ||
bb0f8429 ML |
1515 | /* |
1516 | * Check whether seamless boot is supported. | |
1517 | * | |
7f4ce7b5 ML |
1518 | * So far we only support seamless boot on DCE 3.0 or later. |
1519 | * If users report that it works on older ASICS as well, we may | |
1520 | * loosen this. | |
bb0f8429 ML |
1521 | */ |
1522 | bool amdgpu_device_seamless_boot_supported(struct amdgpu_device *adev) | |
1523 | { | |
5dc270d3 ML |
1524 | switch (amdgpu_seamless) { |
1525 | case -1: | |
1526 | break; | |
1527 | case 1: | |
1528 | return true; | |
1529 | case 0: | |
1530 | return false; | |
1531 | default: | |
1532 | DRM_ERROR("Invalid value for amdgpu.seamless: %d\n", | |
1533 | amdgpu_seamless); | |
1534 | return false; | |
1535 | } | |
1536 | ||
3657a1d5 ML |
1537 | if (!(adev->flags & AMD_IS_APU)) |
1538 | return false; | |
1539 | ||
5dc270d3 ML |
1540 | if (adev->mman.keep_stolen_vga_memory) |
1541 | return false; | |
1542 | ||
7f4ce7b5 | 1543 | return adev->ip_versions[DCE_HWIP][0] >= IP_VERSION(3, 0, 0); |
bb0f8429 ML |
1544 | } |
1545 | ||
5d1eb4c4 | 1546 | /* |
2757a848 ML |
1547 | * Intel hosts such as Rocket Lake, Alder Lake, Raptor Lake and Sapphire Rapids |
1548 | * don't support dynamic speed switching. Until we have confirmation from Intel | |
1549 | * that a specific host supports it, it's safer that we keep it disabled for all. | |
5d1eb4c4 ML |
1550 | * |
1551 | * https://edc.intel.com/content/www/us/en/design/products/platforms/details/raptor-lake-s/13th-generation-core-processors-datasheet-volume-1-of-2/005/pci-express-support/ | |
1552 | * https://gitlab.freedesktop.org/drm/amd/-/issues/2663 | |
1553 | */ | |
2757a848 | 1554 | static bool amdgpu_device_pcie_dynamic_switching_supported(void) |
5d1eb4c4 ML |
1555 | { |
1556 | #if IS_ENABLED(CONFIG_X86) | |
1557 | struct cpuinfo_x86 *c = &cpu_data(0); | |
1558 | ||
1559 | if (c->x86_vendor == X86_VENDOR_INTEL) | |
1560 | return false; | |
1561 | #endif | |
1562 | return true; | |
1563 | } | |
1564 | ||
0ab5d711 ML |
1565 | /** |
1566 | * amdgpu_device_should_use_aspm - check if the device should program ASPM | |
1567 | * | |
1568 | * @adev: amdgpu_device pointer | |
1569 | * | |
1570 | * Confirm whether the module parameter and pcie bridge agree that ASPM should | |
1571 | * be set for this device. | |
1572 | * | |
1573 | * Returns true if it should be used or false if not. | |
1574 | */ | |
1575 | bool amdgpu_device_should_use_aspm(struct amdgpu_device *adev) | |
1576 | { | |
1577 | switch (amdgpu_aspm) { | |
1578 | case -1: | |
1579 | break; | |
1580 | case 0: | |
1581 | return false; | |
1582 | case 1: | |
1583 | return true; | |
1584 | default: | |
1585 | return false; | |
1586 | } | |
1a6513de ML |
1587 | if (adev->flags & AMD_IS_APU) |
1588 | return false; | |
2757a848 ML |
1589 | if (!(adev->pm.pp_feature & PP_PCIE_DPM_MASK)) |
1590 | return false; | |
0ab5d711 ML |
1591 | return pcie_aspm_enabled(adev->pdev); |
1592 | } | |
1593 | ||
d38ceaf9 AD |
1594 | /* if we get transitioned to only one device, take VGA back */ |
1595 | /** | |
06ec9070 | 1596 | * amdgpu_device_vga_set_decode - enable/disable vga decode |
d38ceaf9 | 1597 | * |
bf44e8ce | 1598 | * @pdev: PCI device pointer |
d38ceaf9 AD |
1599 | * @state: enable/disable vga decode |
1600 | * | |
1601 | * Enable/disable vga decode (all asics). | |
1602 | * Returns VGA resource flags. | |
1603 | */ | |
bf44e8ce CH |
1604 | static unsigned int amdgpu_device_vga_set_decode(struct pci_dev *pdev, |
1605 | bool state) | |
d38ceaf9 | 1606 | { |
bf44e8ce | 1607 | struct amdgpu_device *adev = drm_to_adev(pci_get_drvdata(pdev)); |
b8920e1e | 1608 | |
d38ceaf9 AD |
1609 | amdgpu_asic_set_vga_state(adev, state); |
1610 | if (state) | |
1611 | return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM | | |
1612 | VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
1613 | else | |
1614 | return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
1615 | } | |
1616 | ||
e3ecdffa AD |
1617 | /** |
1618 | * amdgpu_device_check_block_size - validate the vm block size | |
1619 | * | |
1620 | * @adev: amdgpu_device pointer | |
1621 | * | |
1622 | * Validates the vm block size specified via module parameter. | |
1623 | * The vm block size defines number of bits in page table versus page directory, | |
1624 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
1625 | * page table and the remaining bits are in the page directory. | |
1626 | */ | |
06ec9070 | 1627 | static void amdgpu_device_check_block_size(struct amdgpu_device *adev) |
a1adf8be CZ |
1628 | { |
1629 | /* defines number of bits in page table versus page directory, | |
1630 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
b8920e1e SS |
1631 | * page table and the remaining bits are in the page directory |
1632 | */ | |
bab4fee7 JZ |
1633 | if (amdgpu_vm_block_size == -1) |
1634 | return; | |
a1adf8be | 1635 | |
bab4fee7 | 1636 | if (amdgpu_vm_block_size < 9) { |
a1adf8be CZ |
1637 | dev_warn(adev->dev, "VM page table size (%d) too small\n", |
1638 | amdgpu_vm_block_size); | |
97489129 | 1639 | amdgpu_vm_block_size = -1; |
a1adf8be | 1640 | } |
a1adf8be CZ |
1641 | } |
1642 | ||
e3ecdffa AD |
1643 | /** |
1644 | * amdgpu_device_check_vm_size - validate the vm size | |
1645 | * | |
1646 | * @adev: amdgpu_device pointer | |
1647 | * | |
1648 | * Validates the vm size in GB specified via module parameter. | |
1649 | * The VM size is the size of the GPU virtual memory space in GB. | |
1650 | */ | |
06ec9070 | 1651 | static void amdgpu_device_check_vm_size(struct amdgpu_device *adev) |
83ca145d | 1652 | { |
64dab074 AD |
1653 | /* no need to check the default value */ |
1654 | if (amdgpu_vm_size == -1) | |
1655 | return; | |
1656 | ||
83ca145d ZJ |
1657 | if (amdgpu_vm_size < 1) { |
1658 | dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n", | |
1659 | amdgpu_vm_size); | |
f3368128 | 1660 | amdgpu_vm_size = -1; |
83ca145d | 1661 | } |
83ca145d ZJ |
1662 | } |
1663 | ||
7951e376 RZ |
1664 | static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev) |
1665 | { | |
1666 | struct sysinfo si; | |
a9d4fe2f | 1667 | bool is_os_64 = (sizeof(void *) == 8); |
7951e376 RZ |
1668 | uint64_t total_memory; |
1669 | uint64_t dram_size_seven_GB = 0x1B8000000; | |
1670 | uint64_t dram_size_three_GB = 0xB8000000; | |
1671 | ||
1672 | if (amdgpu_smu_memory_pool_size == 0) | |
1673 | return; | |
1674 | ||
1675 | if (!is_os_64) { | |
1676 | DRM_WARN("Not 64-bit OS, feature not supported\n"); | |
1677 | goto def_value; | |
1678 | } | |
1679 | si_meminfo(&si); | |
1680 | total_memory = (uint64_t)si.totalram * si.mem_unit; | |
1681 | ||
1682 | if ((amdgpu_smu_memory_pool_size == 1) || | |
1683 | (amdgpu_smu_memory_pool_size == 2)) { | |
1684 | if (total_memory < dram_size_three_GB) | |
1685 | goto def_value1; | |
1686 | } else if ((amdgpu_smu_memory_pool_size == 4) || | |
1687 | (amdgpu_smu_memory_pool_size == 8)) { | |
1688 | if (total_memory < dram_size_seven_GB) | |
1689 | goto def_value1; | |
1690 | } else { | |
1691 | DRM_WARN("Smu memory pool size not supported\n"); | |
1692 | goto def_value; | |
1693 | } | |
1694 | adev->pm.smu_prv_buffer_size = amdgpu_smu_memory_pool_size << 28; | |
1695 | ||
1696 | return; | |
1697 | ||
1698 | def_value1: | |
1699 | DRM_WARN("No enough system memory\n"); | |
1700 | def_value: | |
1701 | adev->pm.smu_prv_buffer_size = 0; | |
1702 | } | |
1703 | ||
9f6a7857 HR |
1704 | static int amdgpu_device_init_apu_flags(struct amdgpu_device *adev) |
1705 | { | |
1706 | if (!(adev->flags & AMD_IS_APU) || | |
1707 | adev->asic_type < CHIP_RAVEN) | |
1708 | return 0; | |
1709 | ||
1710 | switch (adev->asic_type) { | |
1711 | case CHIP_RAVEN: | |
1712 | if (adev->pdev->device == 0x15dd) | |
1713 | adev->apu_flags |= AMD_APU_IS_RAVEN; | |
1714 | if (adev->pdev->device == 0x15d8) | |
1715 | adev->apu_flags |= AMD_APU_IS_PICASSO; | |
1716 | break; | |
1717 | case CHIP_RENOIR: | |
1718 | if ((adev->pdev->device == 0x1636) || | |
1719 | (adev->pdev->device == 0x164c)) | |
1720 | adev->apu_flags |= AMD_APU_IS_RENOIR; | |
1721 | else | |
1722 | adev->apu_flags |= AMD_APU_IS_GREEN_SARDINE; | |
1723 | break; | |
1724 | case CHIP_VANGOGH: | |
1725 | adev->apu_flags |= AMD_APU_IS_VANGOGH; | |
1726 | break; | |
1727 | case CHIP_YELLOW_CARP: | |
1728 | break; | |
d0f56dc2 | 1729 | case CHIP_CYAN_SKILLFISH: |
dfcc3e8c AD |
1730 | if ((adev->pdev->device == 0x13FE) || |
1731 | (adev->pdev->device == 0x143F)) | |
d0f56dc2 TZ |
1732 | adev->apu_flags |= AMD_APU_IS_CYAN_SKILLFISH2; |
1733 | break; | |
9f6a7857 | 1734 | default: |
4eaf21b7 | 1735 | break; |
9f6a7857 HR |
1736 | } |
1737 | ||
1738 | return 0; | |
1739 | } | |
1740 | ||
d38ceaf9 | 1741 | /** |
06ec9070 | 1742 | * amdgpu_device_check_arguments - validate module params |
d38ceaf9 AD |
1743 | * |
1744 | * @adev: amdgpu_device pointer | |
1745 | * | |
1746 | * Validates certain module parameters and updates | |
1747 | * the associated values used by the driver (all asics). | |
1748 | */ | |
912dfc84 | 1749 | static int amdgpu_device_check_arguments(struct amdgpu_device *adev) |
d38ceaf9 | 1750 | { |
5b011235 CZ |
1751 | if (amdgpu_sched_jobs < 4) { |
1752 | dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n", | |
1753 | amdgpu_sched_jobs); | |
1754 | amdgpu_sched_jobs = 4; | |
47fc644f | 1755 | } else if (!is_power_of_2(amdgpu_sched_jobs)) { |
5b011235 CZ |
1756 | dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n", |
1757 | amdgpu_sched_jobs); | |
1758 | amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs); | |
1759 | } | |
d38ceaf9 | 1760 | |
83e74db6 | 1761 | if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) { |
f9321cc4 CK |
1762 | /* gart size must be greater or equal to 32M */ |
1763 | dev_warn(adev->dev, "gart size (%d) too small\n", | |
1764 | amdgpu_gart_size); | |
83e74db6 | 1765 | amdgpu_gart_size = -1; |
d38ceaf9 AD |
1766 | } |
1767 | ||
36d38372 | 1768 | if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) { |
c4e1a13a | 1769 | /* gtt size must be greater or equal to 32M */ |
36d38372 CK |
1770 | dev_warn(adev->dev, "gtt size (%d) too small\n", |
1771 | amdgpu_gtt_size); | |
1772 | amdgpu_gtt_size = -1; | |
d38ceaf9 AD |
1773 | } |
1774 | ||
d07f14be RH |
1775 | /* valid range is between 4 and 9 inclusive */ |
1776 | if (amdgpu_vm_fragment_size != -1 && | |
1777 | (amdgpu_vm_fragment_size > 9 || amdgpu_vm_fragment_size < 4)) { | |
1778 | dev_warn(adev->dev, "valid range is between 4 and 9\n"); | |
1779 | amdgpu_vm_fragment_size = -1; | |
1780 | } | |
1781 | ||
5d5bd5e3 KW |
1782 | if (amdgpu_sched_hw_submission < 2) { |
1783 | dev_warn(adev->dev, "sched hw submission jobs (%d) must be at least 2\n", | |
1784 | amdgpu_sched_hw_submission); | |
1785 | amdgpu_sched_hw_submission = 2; | |
1786 | } else if (!is_power_of_2(amdgpu_sched_hw_submission)) { | |
1787 | dev_warn(adev->dev, "sched hw submission jobs (%d) must be a power of 2\n", | |
1788 | amdgpu_sched_hw_submission); | |
1789 | amdgpu_sched_hw_submission = roundup_pow_of_two(amdgpu_sched_hw_submission); | |
1790 | } | |
1791 | ||
2656fd23 AG |
1792 | if (amdgpu_reset_method < -1 || amdgpu_reset_method > 4) { |
1793 | dev_warn(adev->dev, "invalid option for reset method, reverting to default\n"); | |
1794 | amdgpu_reset_method = -1; | |
1795 | } | |
1796 | ||
7951e376 RZ |
1797 | amdgpu_device_check_smu_prv_buffer_size(adev); |
1798 | ||
06ec9070 | 1799 | amdgpu_device_check_vm_size(adev); |
d38ceaf9 | 1800 | |
06ec9070 | 1801 | amdgpu_device_check_block_size(adev); |
6a7f76e7 | 1802 | |
19aede77 | 1803 | adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type); |
912dfc84 | 1804 | |
e3c00faa | 1805 | return 0; |
d38ceaf9 AD |
1806 | } |
1807 | ||
1808 | /** | |
1809 | * amdgpu_switcheroo_set_state - set switcheroo state | |
1810 | * | |
1811 | * @pdev: pci dev pointer | |
1694467b | 1812 | * @state: vga_switcheroo state |
d38ceaf9 | 1813 | * |
12024b17 | 1814 | * Callback for the switcheroo driver. Suspends or resumes |
d38ceaf9 AD |
1815 | * the asics before or after it is powered up using ACPI methods. |
1816 | */ | |
8aba21b7 LT |
1817 | static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, |
1818 | enum vga_switcheroo_state state) | |
d38ceaf9 AD |
1819 | { |
1820 | struct drm_device *dev = pci_get_drvdata(pdev); | |
de185019 | 1821 | int r; |
d38ceaf9 | 1822 | |
b98c6299 | 1823 | if (amdgpu_device_supports_px(dev) && state == VGA_SWITCHEROO_OFF) |
d38ceaf9 AD |
1824 | return; |
1825 | ||
1826 | if (state == VGA_SWITCHEROO_ON) { | |
dd4fa6c1 | 1827 | pr_info("switched on\n"); |
d38ceaf9 AD |
1828 | /* don't suspend or resume card normally */ |
1829 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
1830 | ||
8f66090b TZ |
1831 | pci_set_power_state(pdev, PCI_D0); |
1832 | amdgpu_device_load_pci_state(pdev); | |
1833 | r = pci_enable_device(pdev); | |
de185019 AD |
1834 | if (r) |
1835 | DRM_WARN("pci_enable_device failed (%d)\n", r); | |
1836 | amdgpu_device_resume(dev, true); | |
d38ceaf9 | 1837 | |
d38ceaf9 | 1838 | dev->switch_power_state = DRM_SWITCH_POWER_ON; |
d38ceaf9 | 1839 | } else { |
dd4fa6c1 | 1840 | pr_info("switched off\n"); |
d38ceaf9 | 1841 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; |
5095d541 | 1842 | amdgpu_device_prepare(dev); |
de185019 | 1843 | amdgpu_device_suspend(dev, true); |
8f66090b | 1844 | amdgpu_device_cache_pci_state(pdev); |
de185019 | 1845 | /* Shut down the device */ |
8f66090b TZ |
1846 | pci_disable_device(pdev); |
1847 | pci_set_power_state(pdev, PCI_D3cold); | |
d38ceaf9 AD |
1848 | dev->switch_power_state = DRM_SWITCH_POWER_OFF; |
1849 | } | |
1850 | } | |
1851 | ||
1852 | /** | |
1853 | * amdgpu_switcheroo_can_switch - see if switcheroo state can change | |
1854 | * | |
1855 | * @pdev: pci dev pointer | |
1856 | * | |
1857 | * Callback for the switcheroo driver. Check of the switcheroo | |
1858 | * state can be changed. | |
1859 | * Returns true if the state can be changed, false if not. | |
1860 | */ | |
1861 | static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev) | |
1862 | { | |
1863 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1864 | ||
b8920e1e | 1865 | /* |
d38ceaf9 AD |
1866 | * FIXME: open_count is protected by drm_global_mutex but that would lead to |
1867 | * locking inversion with the driver load path. And the access here is | |
1868 | * completely racy anyway. So don't bother with locking for now. | |
1869 | */ | |
7e13ad89 | 1870 | return atomic_read(&dev->open_count) == 0; |
d38ceaf9 AD |
1871 | } |
1872 | ||
1873 | static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = { | |
1874 | .set_gpu_state = amdgpu_switcheroo_set_state, | |
1875 | .reprobe = NULL, | |
1876 | .can_switch = amdgpu_switcheroo_can_switch, | |
1877 | }; | |
1878 | ||
e3ecdffa AD |
1879 | /** |
1880 | * amdgpu_device_ip_set_clockgating_state - set the CG state | |
1881 | * | |
87e3f136 | 1882 | * @dev: amdgpu_device pointer |
e3ecdffa AD |
1883 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) |
1884 | * @state: clockgating state (gate or ungate) | |
1885 | * | |
1886 | * Sets the requested clockgating state for all instances of | |
1887 | * the hardware IP specified. | |
1888 | * Returns the error code from the last instance. | |
1889 | */ | |
43fa561f | 1890 | int amdgpu_device_ip_set_clockgating_state(void *dev, |
2990a1fc AD |
1891 | enum amd_ip_block_type block_type, |
1892 | enum amd_clockgating_state state) | |
d38ceaf9 | 1893 | { |
43fa561f | 1894 | struct amdgpu_device *adev = dev; |
d38ceaf9 AD |
1895 | int i, r = 0; |
1896 | ||
1897 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1898 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1899 | continue; |
c722865a RZ |
1900 | if (adev->ip_blocks[i].version->type != block_type) |
1901 | continue; | |
1902 | if (!adev->ip_blocks[i].version->funcs->set_clockgating_state) | |
1903 | continue; | |
1904 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state( | |
1905 | (void *)adev, state); | |
1906 | if (r) | |
1907 | DRM_ERROR("set_clockgating_state of IP block <%s> failed %d\n", | |
1908 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1909 | } |
1910 | return r; | |
1911 | } | |
1912 | ||
e3ecdffa AD |
1913 | /** |
1914 | * amdgpu_device_ip_set_powergating_state - set the PG state | |
1915 | * | |
87e3f136 | 1916 | * @dev: amdgpu_device pointer |
e3ecdffa AD |
1917 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) |
1918 | * @state: powergating state (gate or ungate) | |
1919 | * | |
1920 | * Sets the requested powergating state for all instances of | |
1921 | * the hardware IP specified. | |
1922 | * Returns the error code from the last instance. | |
1923 | */ | |
43fa561f | 1924 | int amdgpu_device_ip_set_powergating_state(void *dev, |
2990a1fc AD |
1925 | enum amd_ip_block_type block_type, |
1926 | enum amd_powergating_state state) | |
d38ceaf9 | 1927 | { |
43fa561f | 1928 | struct amdgpu_device *adev = dev; |
d38ceaf9 AD |
1929 | int i, r = 0; |
1930 | ||
1931 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1932 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1933 | continue; |
c722865a RZ |
1934 | if (adev->ip_blocks[i].version->type != block_type) |
1935 | continue; | |
1936 | if (!adev->ip_blocks[i].version->funcs->set_powergating_state) | |
1937 | continue; | |
1938 | r = adev->ip_blocks[i].version->funcs->set_powergating_state( | |
1939 | (void *)adev, state); | |
1940 | if (r) | |
1941 | DRM_ERROR("set_powergating_state of IP block <%s> failed %d\n", | |
1942 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1943 | } |
1944 | return r; | |
1945 | } | |
1946 | ||
e3ecdffa AD |
1947 | /** |
1948 | * amdgpu_device_ip_get_clockgating_state - get the CG state | |
1949 | * | |
1950 | * @adev: amdgpu_device pointer | |
1951 | * @flags: clockgating feature flags | |
1952 | * | |
1953 | * Walks the list of IPs on the device and updates the clockgating | |
1954 | * flags for each IP. | |
1955 | * Updates @flags with the feature flags for each hardware IP where | |
1956 | * clockgating is enabled. | |
1957 | */ | |
2990a1fc | 1958 | void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev, |
25faeddc | 1959 | u64 *flags) |
6cb2d4e4 HR |
1960 | { |
1961 | int i; | |
1962 | ||
1963 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1964 | if (!adev->ip_blocks[i].status.valid) | |
1965 | continue; | |
1966 | if (adev->ip_blocks[i].version->funcs->get_clockgating_state) | |
1967 | adev->ip_blocks[i].version->funcs->get_clockgating_state((void *)adev, flags); | |
1968 | } | |
1969 | } | |
1970 | ||
e3ecdffa AD |
1971 | /** |
1972 | * amdgpu_device_ip_wait_for_idle - wait for idle | |
1973 | * | |
1974 | * @adev: amdgpu_device pointer | |
1975 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) | |
1976 | * | |
1977 | * Waits for the request hardware IP to be idle. | |
1978 | * Returns 0 for success or a negative error code on failure. | |
1979 | */ | |
2990a1fc AD |
1980 | int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev, |
1981 | enum amd_ip_block_type block_type) | |
5dbbb60b AD |
1982 | { |
1983 | int i, r; | |
1984 | ||
1985 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1986 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1987 | continue; |
a1255107 AD |
1988 | if (adev->ip_blocks[i].version->type == block_type) { |
1989 | r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev); | |
5dbbb60b AD |
1990 | if (r) |
1991 | return r; | |
1992 | break; | |
1993 | } | |
1994 | } | |
1995 | return 0; | |
1996 | ||
1997 | } | |
1998 | ||
e3ecdffa AD |
1999 | /** |
2000 | * amdgpu_device_ip_is_idle - is the hardware IP idle | |
2001 | * | |
2002 | * @adev: amdgpu_device pointer | |
2003 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) | |
2004 | * | |
2005 | * Check if the hardware IP is idle or not. | |
2006 | * Returns true if it the IP is idle, false if not. | |
2007 | */ | |
2990a1fc AD |
2008 | bool amdgpu_device_ip_is_idle(struct amdgpu_device *adev, |
2009 | enum amd_ip_block_type block_type) | |
5dbbb60b AD |
2010 | { |
2011 | int i; | |
2012 | ||
2013 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 2014 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 2015 | continue; |
a1255107 AD |
2016 | if (adev->ip_blocks[i].version->type == block_type) |
2017 | return adev->ip_blocks[i].version->funcs->is_idle((void *)adev); | |
5dbbb60b AD |
2018 | } |
2019 | return true; | |
2020 | ||
2021 | } | |
2022 | ||
e3ecdffa AD |
2023 | /** |
2024 | * amdgpu_device_ip_get_ip_block - get a hw IP pointer | |
2025 | * | |
2026 | * @adev: amdgpu_device pointer | |
87e3f136 | 2027 | * @type: Type of hardware IP (SMU, GFX, UVD, etc.) |
e3ecdffa AD |
2028 | * |
2029 | * Returns a pointer to the hardware IP block structure | |
2030 | * if it exists for the asic, otherwise NULL. | |
2031 | */ | |
2990a1fc AD |
2032 | struct amdgpu_ip_block * |
2033 | amdgpu_device_ip_get_ip_block(struct amdgpu_device *adev, | |
2034 | enum amd_ip_block_type type) | |
d38ceaf9 AD |
2035 | { |
2036 | int i; | |
2037 | ||
2038 | for (i = 0; i < adev->num_ip_blocks; i++) | |
a1255107 | 2039 | if (adev->ip_blocks[i].version->type == type) |
d38ceaf9 AD |
2040 | return &adev->ip_blocks[i]; |
2041 | ||
2042 | return NULL; | |
2043 | } | |
2044 | ||
2045 | /** | |
2990a1fc | 2046 | * amdgpu_device_ip_block_version_cmp |
d38ceaf9 AD |
2047 | * |
2048 | * @adev: amdgpu_device pointer | |
5fc3aeeb | 2049 | * @type: enum amd_ip_block_type |
d38ceaf9 AD |
2050 | * @major: major version |
2051 | * @minor: minor version | |
2052 | * | |
2053 | * return 0 if equal or greater | |
2054 | * return 1 if smaller or the ip_block doesn't exist | |
2055 | */ | |
2990a1fc AD |
2056 | int amdgpu_device_ip_block_version_cmp(struct amdgpu_device *adev, |
2057 | enum amd_ip_block_type type, | |
2058 | u32 major, u32 minor) | |
d38ceaf9 | 2059 | { |
2990a1fc | 2060 | struct amdgpu_ip_block *ip_block = amdgpu_device_ip_get_ip_block(adev, type); |
d38ceaf9 | 2061 | |
a1255107 AD |
2062 | if (ip_block && ((ip_block->version->major > major) || |
2063 | ((ip_block->version->major == major) && | |
2064 | (ip_block->version->minor >= minor)))) | |
d38ceaf9 AD |
2065 | return 0; |
2066 | ||
2067 | return 1; | |
2068 | } | |
2069 | ||
a1255107 | 2070 | /** |
2990a1fc | 2071 | * amdgpu_device_ip_block_add |
a1255107 AD |
2072 | * |
2073 | * @adev: amdgpu_device pointer | |
2074 | * @ip_block_version: pointer to the IP to add | |
2075 | * | |
2076 | * Adds the IP block driver information to the collection of IPs | |
2077 | * on the asic. | |
2078 | */ | |
2990a1fc AD |
2079 | int amdgpu_device_ip_block_add(struct amdgpu_device *adev, |
2080 | const struct amdgpu_ip_block_version *ip_block_version) | |
a1255107 AD |
2081 | { |
2082 | if (!ip_block_version) | |
2083 | return -EINVAL; | |
2084 | ||
7bd939d0 LG |
2085 | switch (ip_block_version->type) { |
2086 | case AMD_IP_BLOCK_TYPE_VCN: | |
2087 | if (adev->harvest_ip_mask & AMD_HARVEST_IP_VCN_MASK) | |
2088 | return 0; | |
2089 | break; | |
2090 | case AMD_IP_BLOCK_TYPE_JPEG: | |
2091 | if (adev->harvest_ip_mask & AMD_HARVEST_IP_JPEG_MASK) | |
2092 | return 0; | |
2093 | break; | |
2094 | default: | |
2095 | break; | |
2096 | } | |
2097 | ||
e966a725 | 2098 | DRM_INFO("add ip block number %d <%s>\n", adev->num_ip_blocks, |
a0bae357 HR |
2099 | ip_block_version->funcs->name); |
2100 | ||
a1255107 AD |
2101 | adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version; |
2102 | ||
2103 | return 0; | |
2104 | } | |
2105 | ||
e3ecdffa AD |
2106 | /** |
2107 | * amdgpu_device_enable_virtual_display - enable virtual display feature | |
2108 | * | |
2109 | * @adev: amdgpu_device pointer | |
2110 | * | |
2111 | * Enabled the virtual display feature if the user has enabled it via | |
2112 | * the module parameter virtual_display. This feature provides a virtual | |
2113 | * display hardware on headless boards or in virtualized environments. | |
2114 | * This function parses and validates the configuration string specified by | |
2115 | * the user and configues the virtual display configuration (number of | |
2116 | * virtual connectors, crtcs, etc.) specified. | |
2117 | */ | |
483ef985 | 2118 | static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev) |
9accf2fd ED |
2119 | { |
2120 | adev->enable_virtual_display = false; | |
2121 | ||
2122 | if (amdgpu_virtual_display) { | |
8f66090b | 2123 | const char *pci_address_name = pci_name(adev->pdev); |
0f66356d | 2124 | char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname; |
9accf2fd ED |
2125 | |
2126 | pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL); | |
2127 | pciaddstr_tmp = pciaddstr; | |
0f66356d ED |
2128 | while ((pciaddname_tmp = strsep(&pciaddstr_tmp, ";"))) { |
2129 | pciaddname = strsep(&pciaddname_tmp, ","); | |
967de2a9 YT |
2130 | if (!strcmp("all", pciaddname) |
2131 | || !strcmp(pci_address_name, pciaddname)) { | |
0f66356d ED |
2132 | long num_crtc; |
2133 | int res = -1; | |
2134 | ||
9accf2fd | 2135 | adev->enable_virtual_display = true; |
0f66356d ED |
2136 | |
2137 | if (pciaddname_tmp) | |
2138 | res = kstrtol(pciaddname_tmp, 10, | |
2139 | &num_crtc); | |
2140 | ||
2141 | if (!res) { | |
2142 | if (num_crtc < 1) | |
2143 | num_crtc = 1; | |
2144 | if (num_crtc > 6) | |
2145 | num_crtc = 6; | |
2146 | adev->mode_info.num_crtc = num_crtc; | |
2147 | } else { | |
2148 | adev->mode_info.num_crtc = 1; | |
2149 | } | |
9accf2fd ED |
2150 | break; |
2151 | } | |
2152 | } | |
2153 | ||
0f66356d ED |
2154 | DRM_INFO("virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n", |
2155 | amdgpu_virtual_display, pci_address_name, | |
2156 | adev->enable_virtual_display, adev->mode_info.num_crtc); | |
9accf2fd ED |
2157 | |
2158 | kfree(pciaddstr); | |
2159 | } | |
2160 | } | |
2161 | ||
25263da3 AD |
2162 | void amdgpu_device_set_sriov_virtual_display(struct amdgpu_device *adev) |
2163 | { | |
2164 | if (amdgpu_sriov_vf(adev) && !adev->enable_virtual_display) { | |
2165 | adev->mode_info.num_crtc = 1; | |
2166 | adev->enable_virtual_display = true; | |
2167 | DRM_INFO("virtual_display:%d, num_crtc:%d\n", | |
2168 | adev->enable_virtual_display, adev->mode_info.num_crtc); | |
2169 | } | |
2170 | } | |
2171 | ||
e3ecdffa AD |
2172 | /** |
2173 | * amdgpu_device_parse_gpu_info_fw - parse gpu info firmware | |
2174 | * | |
2175 | * @adev: amdgpu_device pointer | |
2176 | * | |
2177 | * Parses the asic configuration parameters specified in the gpu info | |
2178 | * firmware and makes them availale to the driver for use in configuring | |
2179 | * the asic. | |
2180 | * Returns 0 on success, -EINVAL on failure. | |
2181 | */ | |
e2a75f88 AD |
2182 | static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) |
2183 | { | |
e2a75f88 | 2184 | const char *chip_name; |
c0a43457 | 2185 | char fw_name[40]; |
e2a75f88 AD |
2186 | int err; |
2187 | const struct gpu_info_firmware_header_v1_0 *hdr; | |
2188 | ||
ab4fe3e1 HR |
2189 | adev->firmware.gpu_info_fw = NULL; |
2190 | ||
72de33f8 | 2191 | if (adev->mman.discovery_bin) { |
cc375d8c TY |
2192 | /* |
2193 | * FIXME: The bounding box is still needed by Navi12, so | |
e24d0e91 | 2194 | * temporarily read it from gpu_info firmware. Should be dropped |
cc375d8c TY |
2195 | * when DAL no longer needs it. |
2196 | */ | |
2197 | if (adev->asic_type != CHIP_NAVI12) | |
2198 | return 0; | |
258620d0 AD |
2199 | } |
2200 | ||
e2a75f88 | 2201 | switch (adev->asic_type) { |
e2a75f88 AD |
2202 | default: |
2203 | return 0; | |
2204 | case CHIP_VEGA10: | |
2205 | chip_name = "vega10"; | |
2206 | break; | |
3f76dced AD |
2207 | case CHIP_VEGA12: |
2208 | chip_name = "vega12"; | |
2209 | break; | |
2d2e5e7e | 2210 | case CHIP_RAVEN: |
54f78a76 | 2211 | if (adev->apu_flags & AMD_APU_IS_RAVEN2) |
54c4d17e | 2212 | chip_name = "raven2"; |
54f78a76 | 2213 | else if (adev->apu_flags & AMD_APU_IS_PICASSO) |
741deade | 2214 | chip_name = "picasso"; |
54c4d17e FX |
2215 | else |
2216 | chip_name = "raven"; | |
2d2e5e7e | 2217 | break; |
65e60f6e LM |
2218 | case CHIP_ARCTURUS: |
2219 | chip_name = "arcturus"; | |
2220 | break; | |
42b325e5 XY |
2221 | case CHIP_NAVI12: |
2222 | chip_name = "navi12"; | |
2223 | break; | |
e2a75f88 AD |
2224 | } |
2225 | ||
2226 | snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name); | |
b31d3063 | 2227 | err = amdgpu_ucode_request(adev, &adev->firmware.gpu_info_fw, fw_name); |
e2a75f88 AD |
2228 | if (err) { |
2229 | dev_err(adev->dev, | |
b31d3063 | 2230 | "Failed to get gpu_info firmware \"%s\"\n", |
e2a75f88 AD |
2231 | fw_name); |
2232 | goto out; | |
2233 | } | |
2234 | ||
ab4fe3e1 | 2235 | hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data; |
e2a75f88 AD |
2236 | amdgpu_ucode_print_gpu_info_hdr(&hdr->header); |
2237 | ||
2238 | switch (hdr->version_major) { | |
2239 | case 1: | |
2240 | { | |
2241 | const struct gpu_info_firmware_v1_0 *gpu_info_fw = | |
ab4fe3e1 | 2242 | (const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data + |
e2a75f88 AD |
2243 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); |
2244 | ||
cc375d8c TY |
2245 | /* |
2246 | * Should be droped when DAL no longer needs it. | |
2247 | */ | |
2248 | if (adev->asic_type == CHIP_NAVI12) | |
ec51d3fa XY |
2249 | goto parse_soc_bounding_box; |
2250 | ||
b5ab16bf AD |
2251 | adev->gfx.config.max_shader_engines = le32_to_cpu(gpu_info_fw->gc_num_se); |
2252 | adev->gfx.config.max_cu_per_sh = le32_to_cpu(gpu_info_fw->gc_num_cu_per_sh); | |
2253 | adev->gfx.config.max_sh_per_se = le32_to_cpu(gpu_info_fw->gc_num_sh_per_se); | |
2254 | adev->gfx.config.max_backends_per_se = le32_to_cpu(gpu_info_fw->gc_num_rb_per_se); | |
e2a75f88 | 2255 | adev->gfx.config.max_texture_channel_caches = |
b5ab16bf AD |
2256 | le32_to_cpu(gpu_info_fw->gc_num_tccs); |
2257 | adev->gfx.config.max_gprs = le32_to_cpu(gpu_info_fw->gc_num_gprs); | |
2258 | adev->gfx.config.max_gs_threads = le32_to_cpu(gpu_info_fw->gc_num_max_gs_thds); | |
2259 | adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gpu_info_fw->gc_gs_table_depth); | |
2260 | adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gpu_info_fw->gc_gsprim_buff_depth); | |
e2a75f88 | 2261 | adev->gfx.config.double_offchip_lds_buf = |
b5ab16bf AD |
2262 | le32_to_cpu(gpu_info_fw->gc_double_offchip_lds_buffer); |
2263 | adev->gfx.cu_info.wave_front_size = le32_to_cpu(gpu_info_fw->gc_wave_size); | |
51fd0370 HZ |
2264 | adev->gfx.cu_info.max_waves_per_simd = |
2265 | le32_to_cpu(gpu_info_fw->gc_max_waves_per_simd); | |
2266 | adev->gfx.cu_info.max_scratch_slots_per_cu = | |
2267 | le32_to_cpu(gpu_info_fw->gc_max_scratch_slots_per_cu); | |
2268 | adev->gfx.cu_info.lds_size = le32_to_cpu(gpu_info_fw->gc_lds_size); | |
48321c3d | 2269 | if (hdr->version_minor >= 1) { |
35c2e910 HZ |
2270 | const struct gpu_info_firmware_v1_1 *gpu_info_fw = |
2271 | (const struct gpu_info_firmware_v1_1 *)(adev->firmware.gpu_info_fw->data + | |
2272 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); | |
2273 | adev->gfx.config.num_sc_per_sh = | |
2274 | le32_to_cpu(gpu_info_fw->num_sc_per_sh); | |
2275 | adev->gfx.config.num_packer_per_sc = | |
2276 | le32_to_cpu(gpu_info_fw->num_packer_per_sc); | |
2277 | } | |
ec51d3fa XY |
2278 | |
2279 | parse_soc_bounding_box: | |
ec51d3fa XY |
2280 | /* |
2281 | * soc bounding box info is not integrated in disocovery table, | |
258620d0 | 2282 | * we always need to parse it from gpu info firmware if needed. |
ec51d3fa | 2283 | */ |
48321c3d HW |
2284 | if (hdr->version_minor == 2) { |
2285 | const struct gpu_info_firmware_v1_2 *gpu_info_fw = | |
2286 | (const struct gpu_info_firmware_v1_2 *)(adev->firmware.gpu_info_fw->data + | |
2287 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); | |
2288 | adev->dm.soc_bounding_box = &gpu_info_fw->soc_bounding_box; | |
2289 | } | |
e2a75f88 AD |
2290 | break; |
2291 | } | |
2292 | default: | |
2293 | dev_err(adev->dev, | |
2294 | "Unsupported gpu_info table %d\n", hdr->header.ucode_version); | |
2295 | err = -EINVAL; | |
2296 | goto out; | |
2297 | } | |
2298 | out: | |
e2a75f88 AD |
2299 | return err; |
2300 | } | |
2301 | ||
e3ecdffa AD |
2302 | /** |
2303 | * amdgpu_device_ip_early_init - run early init for hardware IPs | |
2304 | * | |
2305 | * @adev: amdgpu_device pointer | |
2306 | * | |
2307 | * Early initialization pass for hardware IPs. The hardware IPs that make | |
2308 | * up each asic are discovered each IP's early_init callback is run. This | |
2309 | * is the first stage in initializing the asic. | |
2310 | * Returns 0 on success, negative error code on failure. | |
2311 | */ | |
06ec9070 | 2312 | static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) |
d38ceaf9 | 2313 | { |
901e2be2 | 2314 | struct pci_dev *parent; |
aaa36a97 | 2315 | int i, r; |
ced69502 | 2316 | bool total; |
d38ceaf9 | 2317 | |
483ef985 | 2318 | amdgpu_device_enable_virtual_display(adev); |
a6be7570 | 2319 | |
00a979f3 | 2320 | if (amdgpu_sriov_vf(adev)) { |
00a979f3 | 2321 | r = amdgpu_virt_request_full_gpu(adev, true); |
aaa36a97 AD |
2322 | if (r) |
2323 | return r; | |
00a979f3 WS |
2324 | } |
2325 | ||
d38ceaf9 | 2326 | switch (adev->asic_type) { |
33f34802 KW |
2327 | #ifdef CONFIG_DRM_AMDGPU_SI |
2328 | case CHIP_VERDE: | |
2329 | case CHIP_TAHITI: | |
2330 | case CHIP_PITCAIRN: | |
2331 | case CHIP_OLAND: | |
2332 | case CHIP_HAINAN: | |
295d0daf | 2333 | adev->family = AMDGPU_FAMILY_SI; |
33f34802 KW |
2334 | r = si_set_ip_blocks(adev); |
2335 | if (r) | |
2336 | return r; | |
2337 | break; | |
2338 | #endif | |
a2e73f56 AD |
2339 | #ifdef CONFIG_DRM_AMDGPU_CIK |
2340 | case CHIP_BONAIRE: | |
2341 | case CHIP_HAWAII: | |
2342 | case CHIP_KAVERI: | |
2343 | case CHIP_KABINI: | |
2344 | case CHIP_MULLINS: | |
e1ad2d53 | 2345 | if (adev->flags & AMD_IS_APU) |
a2e73f56 | 2346 | adev->family = AMDGPU_FAMILY_KV; |
e1ad2d53 AD |
2347 | else |
2348 | adev->family = AMDGPU_FAMILY_CI; | |
a2e73f56 AD |
2349 | |
2350 | r = cik_set_ip_blocks(adev); | |
2351 | if (r) | |
2352 | return r; | |
2353 | break; | |
2354 | #endif | |
da87c30b AD |
2355 | case CHIP_TOPAZ: |
2356 | case CHIP_TONGA: | |
2357 | case CHIP_FIJI: | |
2358 | case CHIP_POLARIS10: | |
2359 | case CHIP_POLARIS11: | |
2360 | case CHIP_POLARIS12: | |
2361 | case CHIP_VEGAM: | |
2362 | case CHIP_CARRIZO: | |
2363 | case CHIP_STONEY: | |
2364 | if (adev->flags & AMD_IS_APU) | |
2365 | adev->family = AMDGPU_FAMILY_CZ; | |
2366 | else | |
2367 | adev->family = AMDGPU_FAMILY_VI; | |
2368 | ||
2369 | r = vi_set_ip_blocks(adev); | |
2370 | if (r) | |
2371 | return r; | |
2372 | break; | |
d38ceaf9 | 2373 | default: |
63352b7f AD |
2374 | r = amdgpu_discovery_set_ip_blocks(adev); |
2375 | if (r) | |
2376 | return r; | |
2377 | break; | |
d38ceaf9 AD |
2378 | } |
2379 | ||
901e2be2 AD |
2380 | if (amdgpu_has_atpx() && |
2381 | (amdgpu_is_atpx_hybrid() || | |
2382 | amdgpu_has_atpx_dgpu_power_cntl()) && | |
2383 | ((adev->flags & AMD_IS_APU) == 0) && | |
7b1c6263 | 2384 | !dev_is_removable(&adev->pdev->dev)) |
901e2be2 AD |
2385 | adev->flags |= AMD_IS_PX; |
2386 | ||
85ac2021 | 2387 | if (!(adev->flags & AMD_IS_APU)) { |
c4c8955b | 2388 | parent = pcie_find_root_port(adev->pdev); |
85ac2021 AD |
2389 | adev->has_pr3 = parent ? pci_pr3_present(parent) : false; |
2390 | } | |
901e2be2 | 2391 | |
1884734a | 2392 | |
3b94fb10 | 2393 | adev->pm.pp_feature = amdgpu_pp_feature_mask; |
a35ad98b | 2394 | if (amdgpu_sriov_vf(adev) || sched_policy == KFD_SCHED_POLICY_NO_HWS) |
00544006 | 2395 | adev->pm.pp_feature &= ~PP_GFXOFF_MASK; |
4215a119 HC |
2396 | if (amdgpu_sriov_vf(adev) && adev->asic_type == CHIP_SIENNA_CICHLID) |
2397 | adev->pm.pp_feature &= ~PP_OVERDRIVE_MASK; | |
fbf1035b ML |
2398 | if (!amdgpu_device_pcie_dynamic_switching_supported()) |
2399 | adev->pm.pp_feature &= ~PP_PCIE_DPM_MASK; | |
00f54b97 | 2400 | |
ced69502 | 2401 | total = true; |
d38ceaf9 AD |
2402 | for (i = 0; i < adev->num_ip_blocks; i++) { |
2403 | if ((amdgpu_ip_block_mask & (1 << i)) == 0) { | |
0c451baf | 2404 | DRM_WARN("disabled ip block: %d <%s>\n", |
ed8cf00c | 2405 | i, adev->ip_blocks[i].version->funcs->name); |
a1255107 | 2406 | adev->ip_blocks[i].status.valid = false; |
d38ceaf9 | 2407 | } else { |
a1255107 AD |
2408 | if (adev->ip_blocks[i].version->funcs->early_init) { |
2409 | r = adev->ip_blocks[i].version->funcs->early_init((void *)adev); | |
2c1a2784 | 2410 | if (r == -ENOENT) { |
a1255107 | 2411 | adev->ip_blocks[i].status.valid = false; |
2c1a2784 | 2412 | } else if (r) { |
a1255107 AD |
2413 | DRM_ERROR("early_init of IP block <%s> failed %d\n", |
2414 | adev->ip_blocks[i].version->funcs->name, r); | |
ced69502 | 2415 | total = false; |
2c1a2784 | 2416 | } else { |
a1255107 | 2417 | adev->ip_blocks[i].status.valid = true; |
2c1a2784 | 2418 | } |
974e6b64 | 2419 | } else { |
a1255107 | 2420 | adev->ip_blocks[i].status.valid = true; |
d38ceaf9 | 2421 | } |
d38ceaf9 | 2422 | } |
21a249ca AD |
2423 | /* get the vbios after the asic_funcs are set up */ |
2424 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON) { | |
6e29c227 AD |
2425 | r = amdgpu_device_parse_gpu_info_fw(adev); |
2426 | if (r) | |
2427 | return r; | |
2428 | ||
21a249ca | 2429 | /* Read BIOS */ |
9535a86a SZ |
2430 | if (amdgpu_device_read_bios(adev)) { |
2431 | if (!amdgpu_get_bios(adev)) | |
2432 | return -EINVAL; | |
21a249ca | 2433 | |
9535a86a SZ |
2434 | r = amdgpu_atombios_init(adev); |
2435 | if (r) { | |
2436 | dev_err(adev->dev, "amdgpu_atombios_init failed\n"); | |
2437 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0); | |
2438 | return r; | |
2439 | } | |
21a249ca | 2440 | } |
77eabc6f PJZ |
2441 | |
2442 | /*get pf2vf msg info at it's earliest time*/ | |
2443 | if (amdgpu_sriov_vf(adev)) | |
2444 | amdgpu_virt_init_data_exchange(adev); | |
2445 | ||
21a249ca | 2446 | } |
d38ceaf9 | 2447 | } |
ced69502 ML |
2448 | if (!total) |
2449 | return -ENODEV; | |
d38ceaf9 | 2450 | |
00fa4035 | 2451 | amdgpu_amdkfd_device_probe(adev); |
395d1fb9 NH |
2452 | adev->cg_flags &= amdgpu_cg_mask; |
2453 | adev->pg_flags &= amdgpu_pg_mask; | |
2454 | ||
d38ceaf9 AD |
2455 | return 0; |
2456 | } | |
2457 | ||
0a4f2520 RZ |
2458 | static int amdgpu_device_ip_hw_init_phase1(struct amdgpu_device *adev) |
2459 | { | |
2460 | int i, r; | |
2461 | ||
2462 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
2463 | if (!adev->ip_blocks[i].status.sw) | |
2464 | continue; | |
2465 | if (adev->ip_blocks[i].status.hw) | |
2466 | continue; | |
2467 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || | |
2d11fd3f | 2468 | (amdgpu_sriov_vf(adev) && (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP)) || |
0a4f2520 RZ |
2469 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) { |
2470 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
2471 | if (r) { | |
2472 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
2473 | adev->ip_blocks[i].version->funcs->name, r); | |
2474 | return r; | |
2475 | } | |
2476 | adev->ip_blocks[i].status.hw = true; | |
2477 | } | |
2478 | } | |
2479 | ||
2480 | return 0; | |
2481 | } | |
2482 | ||
2483 | static int amdgpu_device_ip_hw_init_phase2(struct amdgpu_device *adev) | |
2484 | { | |
2485 | int i, r; | |
2486 | ||
2487 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
2488 | if (!adev->ip_blocks[i].status.sw) | |
2489 | continue; | |
2490 | if (adev->ip_blocks[i].status.hw) | |
2491 | continue; | |
2492 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
2493 | if (r) { | |
2494 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
2495 | adev->ip_blocks[i].version->funcs->name, r); | |
2496 | return r; | |
2497 | } | |
2498 | adev->ip_blocks[i].status.hw = true; | |
2499 | } | |
2500 | ||
2501 | return 0; | |
2502 | } | |
2503 | ||
7a3e0bb2 RZ |
2504 | static int amdgpu_device_fw_loading(struct amdgpu_device *adev) |
2505 | { | |
2506 | int r = 0; | |
2507 | int i; | |
80f41f84 | 2508 | uint32_t smu_version; |
7a3e0bb2 RZ |
2509 | |
2510 | if (adev->asic_type >= CHIP_VEGA10) { | |
2511 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
482f0e53 ML |
2512 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_PSP) |
2513 | continue; | |
2514 | ||
e3c1b071 | 2515 | if (!adev->ip_blocks[i].status.sw) |
2516 | continue; | |
2517 | ||
482f0e53 ML |
2518 | /* no need to do the fw loading again if already done*/ |
2519 | if (adev->ip_blocks[i].status.hw == true) | |
2520 | break; | |
2521 | ||
53b3f8f4 | 2522 | if (amdgpu_in_reset(adev) || adev->in_suspend) { |
482f0e53 ML |
2523 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
2524 | if (r) { | |
2525 | DRM_ERROR("resume of IP block <%s> failed %d\n", | |
7a3e0bb2 | 2526 | adev->ip_blocks[i].version->funcs->name, r); |
482f0e53 ML |
2527 | return r; |
2528 | } | |
2529 | } else { | |
2530 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
2531 | if (r) { | |
2532 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
2533 | adev->ip_blocks[i].version->funcs->name, r); | |
2534 | return r; | |
7a3e0bb2 | 2535 | } |
7a3e0bb2 | 2536 | } |
482f0e53 ML |
2537 | |
2538 | adev->ip_blocks[i].status.hw = true; | |
2539 | break; | |
7a3e0bb2 RZ |
2540 | } |
2541 | } | |
482f0e53 | 2542 | |
8973d9ec ED |
2543 | if (!amdgpu_sriov_vf(adev) || adev->asic_type == CHIP_TONGA) |
2544 | r = amdgpu_pm_load_smu_firmware(adev, &smu_version); | |
7a3e0bb2 | 2545 | |
80f41f84 | 2546 | return r; |
7a3e0bb2 RZ |
2547 | } |
2548 | ||
5fd8518d AG |
2549 | static int amdgpu_device_init_schedulers(struct amdgpu_device *adev) |
2550 | { | |
2551 | long timeout; | |
2552 | int r, i; | |
2553 | ||
2554 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
2555 | struct amdgpu_ring *ring = adev->rings[i]; | |
2556 | ||
2557 | /* No need to setup the GPU scheduler for rings that don't need it */ | |
2558 | if (!ring || ring->no_scheduler) | |
2559 | continue; | |
2560 | ||
2561 | switch (ring->funcs->type) { | |
2562 | case AMDGPU_RING_TYPE_GFX: | |
2563 | timeout = adev->gfx_timeout; | |
2564 | break; | |
2565 | case AMDGPU_RING_TYPE_COMPUTE: | |
2566 | timeout = adev->compute_timeout; | |
2567 | break; | |
2568 | case AMDGPU_RING_TYPE_SDMA: | |
2569 | timeout = adev->sdma_timeout; | |
2570 | break; | |
2571 | default: | |
2572 | timeout = adev->video_timeout; | |
2573 | break; | |
2574 | } | |
2575 | ||
2576 | r = drm_sched_init(&ring->sched, &amdgpu_sched_ops, | |
56e44960 | 2577 | DRM_SCHED_PRIORITY_COUNT, |
11f25c84 | 2578 | ring->num_hw_submission, 0, |
8ab62eda JG |
2579 | timeout, adev->reset_domain->wq, |
2580 | ring->sched_score, ring->name, | |
2581 | adev->dev); | |
5fd8518d AG |
2582 | if (r) { |
2583 | DRM_ERROR("Failed to create scheduler on ring %s.\n", | |
2584 | ring->name); | |
2585 | return r; | |
2586 | } | |
037b98a2 AD |
2587 | r = amdgpu_uvd_entity_init(adev, ring); |
2588 | if (r) { | |
2589 | DRM_ERROR("Failed to create UVD scheduling entity on ring %s.\n", | |
2590 | ring->name); | |
2591 | return r; | |
2592 | } | |
2593 | r = amdgpu_vce_entity_init(adev, ring); | |
2594 | if (r) { | |
2595 | DRM_ERROR("Failed to create VCE scheduling entity on ring %s.\n", | |
2596 | ring->name); | |
2597 | return r; | |
2598 | } | |
5fd8518d AG |
2599 | } |
2600 | ||
d425c6f4 JZ |
2601 | amdgpu_xcp_update_partition_sched_list(adev); |
2602 | ||
5fd8518d AG |
2603 | return 0; |
2604 | } | |
2605 | ||
2606 | ||
e3ecdffa AD |
2607 | /** |
2608 | * amdgpu_device_ip_init - run init for hardware IPs | |
2609 | * | |
2610 | * @adev: amdgpu_device pointer | |
2611 | * | |
2612 | * Main initialization pass for hardware IPs. The list of all the hardware | |
2613 | * IPs that make up the asic is walked and the sw_init and hw_init callbacks | |
2614 | * are run. sw_init initializes the software state associated with each IP | |
2615 | * and hw_init initializes the hardware associated with each IP. | |
2616 | * Returns 0 on success, negative error code on failure. | |
2617 | */ | |
06ec9070 | 2618 | static int amdgpu_device_ip_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
2619 | { |
2620 | int i, r; | |
2621 | ||
c030f2e4 | 2622 | r = amdgpu_ras_init(adev); |
2623 | if (r) | |
2624 | return r; | |
2625 | ||
d38ceaf9 | 2626 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 2627 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 2628 | continue; |
a1255107 | 2629 | r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev); |
2c1a2784 | 2630 | if (r) { |
a1255107 AD |
2631 | DRM_ERROR("sw_init of IP block <%s> failed %d\n", |
2632 | adev->ip_blocks[i].version->funcs->name, r); | |
72d3f592 | 2633 | goto init_failed; |
2c1a2784 | 2634 | } |
a1255107 | 2635 | adev->ip_blocks[i].status.sw = true; |
bfca0289 | 2636 | |
c1c39032 AD |
2637 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON) { |
2638 | /* need to do common hw init early so everything is set up for gmc */ | |
2639 | r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); | |
2640 | if (r) { | |
2641 | DRM_ERROR("hw_init %d failed %d\n", i, r); | |
2642 | goto init_failed; | |
2643 | } | |
2644 | adev->ip_blocks[i].status.hw = true; | |
2645 | } else if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { | |
2646 | /* need to do gmc hw init early so we can allocate gpu mem */ | |
892deb48 VS |
2647 | /* Try to reserve bad pages early */ |
2648 | if (amdgpu_sriov_vf(adev)) | |
2649 | amdgpu_virt_exchange_data(adev); | |
2650 | ||
7ccfd79f | 2651 | r = amdgpu_device_mem_scratch_init(adev); |
2c1a2784 | 2652 | if (r) { |
7ccfd79f | 2653 | DRM_ERROR("amdgpu_mem_scratch_init failed %d\n", r); |
72d3f592 | 2654 | goto init_failed; |
2c1a2784 | 2655 | } |
a1255107 | 2656 | r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); |
2c1a2784 AD |
2657 | if (r) { |
2658 | DRM_ERROR("hw_init %d failed %d\n", i, r); | |
72d3f592 | 2659 | goto init_failed; |
2c1a2784 | 2660 | } |
06ec9070 | 2661 | r = amdgpu_device_wb_init(adev); |
2c1a2784 | 2662 | if (r) { |
06ec9070 | 2663 | DRM_ERROR("amdgpu_device_wb_init failed %d\n", r); |
72d3f592 | 2664 | goto init_failed; |
2c1a2784 | 2665 | } |
a1255107 | 2666 | adev->ip_blocks[i].status.hw = true; |
2493664f ML |
2667 | |
2668 | /* right after GMC hw init, we create CSA */ | |
02ff519e | 2669 | if (adev->gfx.mcbp) { |
1e256e27 | 2670 | r = amdgpu_allocate_static_csa(adev, &adev->virt.csa_obj, |
58ab2c08 CK |
2671 | AMDGPU_GEM_DOMAIN_VRAM | |
2672 | AMDGPU_GEM_DOMAIN_GTT, | |
2673 | AMDGPU_CSA_SIZE); | |
2493664f ML |
2674 | if (r) { |
2675 | DRM_ERROR("allocate CSA failed %d\n", r); | |
72d3f592 | 2676 | goto init_failed; |
2493664f ML |
2677 | } |
2678 | } | |
c8031019 APS |
2679 | |
2680 | r = amdgpu_seq64_init(adev); | |
2681 | if (r) { | |
2682 | DRM_ERROR("allocate seq64 failed %d\n", r); | |
2683 | goto init_failed; | |
2684 | } | |
d38ceaf9 AD |
2685 | } |
2686 | } | |
2687 | ||
c9ffa427 | 2688 | if (amdgpu_sriov_vf(adev)) |
22c16d25 | 2689 | amdgpu_virt_init_data_exchange(adev); |
c9ffa427 | 2690 | |
533aed27 AG |
2691 | r = amdgpu_ib_pool_init(adev); |
2692 | if (r) { | |
2693 | dev_err(adev->dev, "IB initialization failed (%d).\n", r); | |
2694 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r); | |
2695 | goto init_failed; | |
2696 | } | |
2697 | ||
c8963ea4 RZ |
2698 | r = amdgpu_ucode_create_bo(adev); /* create ucode bo when sw_init complete*/ |
2699 | if (r) | |
72d3f592 | 2700 | goto init_failed; |
0a4f2520 RZ |
2701 | |
2702 | r = amdgpu_device_ip_hw_init_phase1(adev); | |
2703 | if (r) | |
72d3f592 | 2704 | goto init_failed; |
0a4f2520 | 2705 | |
7a3e0bb2 RZ |
2706 | r = amdgpu_device_fw_loading(adev); |
2707 | if (r) | |
72d3f592 | 2708 | goto init_failed; |
7a3e0bb2 | 2709 | |
0a4f2520 RZ |
2710 | r = amdgpu_device_ip_hw_init_phase2(adev); |
2711 | if (r) | |
72d3f592 | 2712 | goto init_failed; |
d38ceaf9 | 2713 | |
121a2bc6 AG |
2714 | /* |
2715 | * retired pages will be loaded from eeprom and reserved here, | |
2716 | * it should be called after amdgpu_device_ip_hw_init_phase2 since | |
2717 | * for some ASICs the RAS EEPROM code relies on SMU fully functioning | |
2718 | * for I2C communication which only true at this point. | |
b82e65a9 GC |
2719 | * |
2720 | * amdgpu_ras_recovery_init may fail, but the upper only cares the | |
2721 | * failure from bad gpu situation and stop amdgpu init process | |
2722 | * accordingly. For other failed cases, it will still release all | |
2723 | * the resource and print error message, rather than returning one | |
2724 | * negative value to upper level. | |
121a2bc6 AG |
2725 | * |
2726 | * Note: theoretically, this should be called before all vram allocations | |
2727 | * to protect retired page from abusing | |
2728 | */ | |
b82e65a9 GC |
2729 | r = amdgpu_ras_recovery_init(adev); |
2730 | if (r) | |
2731 | goto init_failed; | |
121a2bc6 | 2732 | |
cfbb6b00 AG |
2733 | /** |
2734 | * In case of XGMI grab extra reference for reset domain for this device | |
2735 | */ | |
a4c63caf | 2736 | if (adev->gmc.xgmi.num_physical_nodes > 1) { |
cfbb6b00 | 2737 | if (amdgpu_xgmi_add_device(adev) == 0) { |
46c67660 | 2738 | if (!amdgpu_sriov_vf(adev)) { |
2efc30f0 VC |
2739 | struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); |
2740 | ||
dfd0287b LH |
2741 | if (WARN_ON(!hive)) { |
2742 | r = -ENOENT; | |
2743 | goto init_failed; | |
2744 | } | |
2745 | ||
46c67660 | 2746 | if (!hive->reset_domain || |
2747 | !amdgpu_reset_get_reset_domain(hive->reset_domain)) { | |
2748 | r = -ENOENT; | |
2749 | amdgpu_put_xgmi_hive(hive); | |
2750 | goto init_failed; | |
2751 | } | |
2752 | ||
2753 | /* Drop the early temporary reset domain we created for device */ | |
2754 | amdgpu_reset_put_reset_domain(adev->reset_domain); | |
2755 | adev->reset_domain = hive->reset_domain; | |
9dfa4860 | 2756 | amdgpu_put_xgmi_hive(hive); |
cfbb6b00 | 2757 | } |
a4c63caf AG |
2758 | } |
2759 | } | |
2760 | ||
5fd8518d AG |
2761 | r = amdgpu_device_init_schedulers(adev); |
2762 | if (r) | |
2763 | goto init_failed; | |
e3c1b071 | 2764 | |
b7043800 AD |
2765 | if (adev->mman.buffer_funcs_ring->sched.ready) |
2766 | amdgpu_ttm_set_buffer_funcs_status(adev, true); | |
2767 | ||
e3c1b071 | 2768 | /* Don't init kfd if whole hive need to be reset during init */ |
84b4dd3f PY |
2769 | if (!adev->gmc.xgmi.pending_reset) { |
2770 | kgd2kfd_init_zone_device(adev); | |
e3c1b071 | 2771 | amdgpu_amdkfd_device_init(adev); |
84b4dd3f | 2772 | } |
c6332b97 | 2773 | |
bd607166 KR |
2774 | amdgpu_fru_get_product_info(adev); |
2775 | ||
72d3f592 | 2776 | init_failed: |
c6332b97 | 2777 | |
72d3f592 | 2778 | return r; |
d38ceaf9 AD |
2779 | } |
2780 | ||
e3ecdffa AD |
2781 | /** |
2782 | * amdgpu_device_fill_reset_magic - writes reset magic to gart pointer | |
2783 | * | |
2784 | * @adev: amdgpu_device pointer | |
2785 | * | |
2786 | * Writes a reset magic value to the gart pointer in VRAM. The driver calls | |
2787 | * this function before a GPU reset. If the value is retained after a | |
2788 | * GPU reset, VRAM has not been lost. Some GPU resets may destry VRAM contents. | |
2789 | */ | |
06ec9070 | 2790 | static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev) |
0c49e0b8 CZ |
2791 | { |
2792 | memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM); | |
2793 | } | |
2794 | ||
e3ecdffa AD |
2795 | /** |
2796 | * amdgpu_device_check_vram_lost - check if vram is valid | |
2797 | * | |
2798 | * @adev: amdgpu_device pointer | |
2799 | * | |
2800 | * Checks the reset magic value written to the gart pointer in VRAM. | |
2801 | * The driver calls this after a GPU reset to see if the contents of | |
2802 | * VRAM is lost or now. | |
2803 | * returns true if vram is lost, false if not. | |
2804 | */ | |
06ec9070 | 2805 | static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev) |
0c49e0b8 | 2806 | { |
dadce777 EQ |
2807 | if (memcmp(adev->gart.ptr, adev->reset_magic, |
2808 | AMDGPU_RESET_MAGIC_NUM)) | |
2809 | return true; | |
2810 | ||
53b3f8f4 | 2811 | if (!amdgpu_in_reset(adev)) |
dadce777 EQ |
2812 | return false; |
2813 | ||
2814 | /* | |
2815 | * For all ASICs with baco/mode1 reset, the VRAM is | |
2816 | * always assumed to be lost. | |
2817 | */ | |
2818 | switch (amdgpu_asic_reset_method(adev)) { | |
2819 | case AMD_RESET_METHOD_BACO: | |
2820 | case AMD_RESET_METHOD_MODE1: | |
2821 | return true; | |
2822 | default: | |
2823 | return false; | |
2824 | } | |
0c49e0b8 CZ |
2825 | } |
2826 | ||
e3ecdffa | 2827 | /** |
1112a46b | 2828 | * amdgpu_device_set_cg_state - set clockgating for amdgpu device |
e3ecdffa AD |
2829 | * |
2830 | * @adev: amdgpu_device pointer | |
b8b72130 | 2831 | * @state: clockgating state (gate or ungate) |
e3ecdffa | 2832 | * |
e3ecdffa | 2833 | * The list of all the hardware IPs that make up the asic is walked and the |
1112a46b RZ |
2834 | * set_clockgating_state callbacks are run. |
2835 | * Late initialization pass enabling clockgating for hardware IPs. | |
2836 | * Fini or suspend, pass disabling clockgating for hardware IPs. | |
e3ecdffa AD |
2837 | * Returns 0 on success, negative error code on failure. |
2838 | */ | |
fdd34271 | 2839 | |
5d89bb2d LL |
2840 | int amdgpu_device_set_cg_state(struct amdgpu_device *adev, |
2841 | enum amd_clockgating_state state) | |
d38ceaf9 | 2842 | { |
1112a46b | 2843 | int i, j, r; |
d38ceaf9 | 2844 | |
4a2ba394 SL |
2845 | if (amdgpu_emu_mode == 1) |
2846 | return 0; | |
2847 | ||
1112a46b RZ |
2848 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2849 | i = state == AMD_CG_STATE_GATE ? j : adev->num_ip_blocks - j - 1; | |
a2d31dc3 | 2850 | if (!adev->ip_blocks[i].status.late_initialized) |
d38ceaf9 | 2851 | continue; |
47198eb7 | 2852 | /* skip CG for GFX, SDMA on S0ix */ |
5d70a549 | 2853 | if (adev->in_s0ix && |
47198eb7 AD |
2854 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX || |
2855 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SDMA)) | |
5d70a549 | 2856 | continue; |
4a446d55 | 2857 | /* skip CG for VCE/UVD, it's handled specially */ |
a1255107 | 2858 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && |
57716327 | 2859 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE && |
34319b32 | 2860 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN && |
52f2e779 | 2861 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG && |
57716327 | 2862 | adev->ip_blocks[i].version->funcs->set_clockgating_state) { |
4a446d55 | 2863 | /* enable clockgating to save power */ |
a1255107 | 2864 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev, |
1112a46b | 2865 | state); |
4a446d55 AD |
2866 | if (r) { |
2867 | DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n", | |
a1255107 | 2868 | adev->ip_blocks[i].version->funcs->name, r); |
4a446d55 AD |
2869 | return r; |
2870 | } | |
b0b00ff1 | 2871 | } |
d38ceaf9 | 2872 | } |
06b18f61 | 2873 | |
c9f96fd5 RZ |
2874 | return 0; |
2875 | } | |
2876 | ||
5d89bb2d LL |
2877 | int amdgpu_device_set_pg_state(struct amdgpu_device *adev, |
2878 | enum amd_powergating_state state) | |
c9f96fd5 | 2879 | { |
1112a46b | 2880 | int i, j, r; |
06b18f61 | 2881 | |
c9f96fd5 RZ |
2882 | if (amdgpu_emu_mode == 1) |
2883 | return 0; | |
2884 | ||
1112a46b RZ |
2885 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2886 | i = state == AMD_PG_STATE_GATE ? j : adev->num_ip_blocks - j - 1; | |
a2d31dc3 | 2887 | if (!adev->ip_blocks[i].status.late_initialized) |
c9f96fd5 | 2888 | continue; |
47198eb7 | 2889 | /* skip PG for GFX, SDMA on S0ix */ |
5d70a549 | 2890 | if (adev->in_s0ix && |
47198eb7 AD |
2891 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX || |
2892 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SDMA)) | |
5d70a549 | 2893 | continue; |
c9f96fd5 RZ |
2894 | /* skip CG for VCE/UVD, it's handled specially */ |
2895 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && | |
2896 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE && | |
2897 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN && | |
52f2e779 | 2898 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG && |
c9f96fd5 RZ |
2899 | adev->ip_blocks[i].version->funcs->set_powergating_state) { |
2900 | /* enable powergating to save power */ | |
2901 | r = adev->ip_blocks[i].version->funcs->set_powergating_state((void *)adev, | |
1112a46b | 2902 | state); |
c9f96fd5 RZ |
2903 | if (r) { |
2904 | DRM_ERROR("set_powergating_state(gate) of IP block <%s> failed %d\n", | |
2905 | adev->ip_blocks[i].version->funcs->name, r); | |
2906 | return r; | |
2907 | } | |
2908 | } | |
2909 | } | |
2dc80b00 S |
2910 | return 0; |
2911 | } | |
2912 | ||
beff74bc AD |
2913 | static int amdgpu_device_enable_mgpu_fan_boost(void) |
2914 | { | |
2915 | struct amdgpu_gpu_instance *gpu_ins; | |
2916 | struct amdgpu_device *adev; | |
2917 | int i, ret = 0; | |
2918 | ||
2919 | mutex_lock(&mgpu_info.mutex); | |
2920 | ||
2921 | /* | |
2922 | * MGPU fan boost feature should be enabled | |
2923 | * only when there are two or more dGPUs in | |
2924 | * the system | |
2925 | */ | |
2926 | if (mgpu_info.num_dgpu < 2) | |
2927 | goto out; | |
2928 | ||
2929 | for (i = 0; i < mgpu_info.num_dgpu; i++) { | |
2930 | gpu_ins = &(mgpu_info.gpu_ins[i]); | |
2931 | adev = gpu_ins->adev; | |
2932 | if (!(adev->flags & AMD_IS_APU) && | |
f10bb940 | 2933 | !gpu_ins->mgpu_fan_enabled) { |
beff74bc AD |
2934 | ret = amdgpu_dpm_enable_mgpu_fan_boost(adev); |
2935 | if (ret) | |
2936 | break; | |
2937 | ||
2938 | gpu_ins->mgpu_fan_enabled = 1; | |
2939 | } | |
2940 | } | |
2941 | ||
2942 | out: | |
2943 | mutex_unlock(&mgpu_info.mutex); | |
2944 | ||
2945 | return ret; | |
2946 | } | |
2947 | ||
e3ecdffa AD |
2948 | /** |
2949 | * amdgpu_device_ip_late_init - run late init for hardware IPs | |
2950 | * | |
2951 | * @adev: amdgpu_device pointer | |
2952 | * | |
2953 | * Late initialization pass for hardware IPs. The list of all the hardware | |
2954 | * IPs that make up the asic is walked and the late_init callbacks are run. | |
2955 | * late_init covers any special initialization that an IP requires | |
2956 | * after all of the have been initialized or something that needs to happen | |
2957 | * late in the init process. | |
2958 | * Returns 0 on success, negative error code on failure. | |
2959 | */ | |
06ec9070 | 2960 | static int amdgpu_device_ip_late_init(struct amdgpu_device *adev) |
2dc80b00 | 2961 | { |
60599a03 | 2962 | struct amdgpu_gpu_instance *gpu_instance; |
2dc80b00 S |
2963 | int i = 0, r; |
2964 | ||
2965 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
73f847db | 2966 | if (!adev->ip_blocks[i].status.hw) |
2dc80b00 S |
2967 | continue; |
2968 | if (adev->ip_blocks[i].version->funcs->late_init) { | |
2969 | r = adev->ip_blocks[i].version->funcs->late_init((void *)adev); | |
2970 | if (r) { | |
2971 | DRM_ERROR("late_init of IP block <%s> failed %d\n", | |
2972 | adev->ip_blocks[i].version->funcs->name, r); | |
2973 | return r; | |
2974 | } | |
2dc80b00 | 2975 | } |
73f847db | 2976 | adev->ip_blocks[i].status.late_initialized = true; |
2dc80b00 S |
2977 | } |
2978 | ||
867e24ca | 2979 | r = amdgpu_ras_late_init(adev); |
2980 | if (r) { | |
2981 | DRM_ERROR("amdgpu_ras_late_init failed %d", r); | |
2982 | return r; | |
2983 | } | |
2984 | ||
a891d239 DL |
2985 | amdgpu_ras_set_error_query_ready(adev, true); |
2986 | ||
1112a46b RZ |
2987 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE); |
2988 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE); | |
916ac57f | 2989 | |
06ec9070 | 2990 | amdgpu_device_fill_reset_magic(adev); |
d38ceaf9 | 2991 | |
beff74bc AD |
2992 | r = amdgpu_device_enable_mgpu_fan_boost(); |
2993 | if (r) | |
2994 | DRM_ERROR("enable mgpu fan boost failed (%d).\n", r); | |
2995 | ||
4da8b639 | 2996 | /* For passthrough configuration on arcturus and aldebaran, enable special handling SBR */ |
47fc644f SS |
2997 | if (amdgpu_passthrough(adev) && |
2998 | ((adev->asic_type == CHIP_ARCTURUS && adev->gmc.xgmi.num_physical_nodes > 1) || | |
2999 | adev->asic_type == CHIP_ALDEBARAN)) | |
bc143d8b | 3000 | amdgpu_dpm_handle_passthrough_sbr(adev, true); |
60599a03 EQ |
3001 | |
3002 | if (adev->gmc.xgmi.num_physical_nodes > 1) { | |
3003 | mutex_lock(&mgpu_info.mutex); | |
3004 | ||
3005 | /* | |
3006 | * Reset device p-state to low as this was booted with high. | |
3007 | * | |
3008 | * This should be performed only after all devices from the same | |
3009 | * hive get initialized. | |
3010 | * | |
3011 | * However, it's unknown how many device in the hive in advance. | |
3012 | * As this is counted one by one during devices initializations. | |
3013 | * | |
3014 | * So, we wait for all XGMI interlinked devices initialized. | |
3015 | * This may bring some delays as those devices may come from | |
3016 | * different hives. But that should be OK. | |
3017 | */ | |
3018 | if (mgpu_info.num_dgpu == adev->gmc.xgmi.num_physical_nodes) { | |
3019 | for (i = 0; i < mgpu_info.num_gpu; i++) { | |
3020 | gpu_instance = &(mgpu_info.gpu_ins[i]); | |
3021 | if (gpu_instance->adev->flags & AMD_IS_APU) | |
3022 | continue; | |
3023 | ||
d84a430d JK |
3024 | r = amdgpu_xgmi_set_pstate(gpu_instance->adev, |
3025 | AMDGPU_XGMI_PSTATE_MIN); | |
60599a03 EQ |
3026 | if (r) { |
3027 | DRM_ERROR("pstate setting failed (%d).\n", r); | |
3028 | break; | |
3029 | } | |
3030 | } | |
3031 | } | |
3032 | ||
3033 | mutex_unlock(&mgpu_info.mutex); | |
3034 | } | |
3035 | ||
d38ceaf9 AD |
3036 | return 0; |
3037 | } | |
3038 | ||
613aa3ea LY |
3039 | /** |
3040 | * amdgpu_device_smu_fini_early - smu hw_fini wrapper | |
3041 | * | |
3042 | * @adev: amdgpu_device pointer | |
3043 | * | |
3044 | * For ASICs need to disable SMC first | |
3045 | */ | |
3046 | static void amdgpu_device_smu_fini_early(struct amdgpu_device *adev) | |
3047 | { | |
3048 | int i, r; | |
3049 | ||
4e8303cf | 3050 | if (amdgpu_ip_version(adev, GC_HWIP, 0) > IP_VERSION(9, 0, 0)) |
613aa3ea LY |
3051 | return; |
3052 | ||
3053 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
3054 | if (!adev->ip_blocks[i].status.hw) | |
3055 | continue; | |
3056 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { | |
3057 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); | |
3058 | /* XXX handle errors */ | |
3059 | if (r) { | |
3060 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", | |
3061 | adev->ip_blocks[i].version->funcs->name, r); | |
3062 | } | |
3063 | adev->ip_blocks[i].status.hw = false; | |
3064 | break; | |
3065 | } | |
3066 | } | |
3067 | } | |
3068 | ||
e9669fb7 | 3069 | static int amdgpu_device_ip_fini_early(struct amdgpu_device *adev) |
d38ceaf9 AD |
3070 | { |
3071 | int i, r; | |
3072 | ||
e9669fb7 AG |
3073 | for (i = 0; i < adev->num_ip_blocks; i++) { |
3074 | if (!adev->ip_blocks[i].version->funcs->early_fini) | |
3075 | continue; | |
5278a159 | 3076 | |
e9669fb7 AG |
3077 | r = adev->ip_blocks[i].version->funcs->early_fini((void *)adev); |
3078 | if (r) { | |
3079 | DRM_DEBUG("early_fini of IP block <%s> failed %d\n", | |
3080 | adev->ip_blocks[i].version->funcs->name, r); | |
3081 | } | |
3082 | } | |
c030f2e4 | 3083 | |
05df1f01 | 3084 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); |
fdd34271 RZ |
3085 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); |
3086 | ||
7270e895 TY |
3087 | amdgpu_amdkfd_suspend(adev, false); |
3088 | ||
613aa3ea LY |
3089 | /* Workaroud for ASICs need to disable SMC first */ |
3090 | amdgpu_device_smu_fini_early(adev); | |
3e96dbfd | 3091 | |
d38ceaf9 | 3092 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 3093 | if (!adev->ip_blocks[i].status.hw) |
d38ceaf9 | 3094 | continue; |
8201a67a | 3095 | |
a1255107 | 3096 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); |
d38ceaf9 | 3097 | /* XXX handle errors */ |
2c1a2784 | 3098 | if (r) { |
a1255107 AD |
3099 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", |
3100 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 3101 | } |
8201a67a | 3102 | |
a1255107 | 3103 | adev->ip_blocks[i].status.hw = false; |
d38ceaf9 AD |
3104 | } |
3105 | ||
6effad8a GC |
3106 | if (amdgpu_sriov_vf(adev)) { |
3107 | if (amdgpu_virt_release_full_gpu(adev, false)) | |
3108 | DRM_ERROR("failed to release exclusive mode on fini\n"); | |
3109 | } | |
3110 | ||
e9669fb7 AG |
3111 | return 0; |
3112 | } | |
3113 | ||
3114 | /** | |
3115 | * amdgpu_device_ip_fini - run fini for hardware IPs | |
3116 | * | |
3117 | * @adev: amdgpu_device pointer | |
3118 | * | |
3119 | * Main teardown pass for hardware IPs. The list of all the hardware | |
3120 | * IPs that make up the asic is walked and the hw_fini and sw_fini callbacks | |
3121 | * are run. hw_fini tears down the hardware associated with each IP | |
3122 | * and sw_fini tears down any software state associated with each IP. | |
3123 | * Returns 0 on success, negative error code on failure. | |
3124 | */ | |
3125 | static int amdgpu_device_ip_fini(struct amdgpu_device *adev) | |
3126 | { | |
3127 | int i, r; | |
3128 | ||
3129 | if (amdgpu_sriov_vf(adev) && adev->virt.ras_init_done) | |
3130 | amdgpu_virt_release_ras_err_handler_data(adev); | |
3131 | ||
e9669fb7 AG |
3132 | if (adev->gmc.xgmi.num_physical_nodes > 1) |
3133 | amdgpu_xgmi_remove_device(adev); | |
3134 | ||
c004d44e | 3135 | amdgpu_amdkfd_device_fini_sw(adev); |
9950cda2 | 3136 | |
d38ceaf9 | 3137 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 3138 | if (!adev->ip_blocks[i].status.sw) |
d38ceaf9 | 3139 | continue; |
c12aba3a ML |
3140 | |
3141 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { | |
c8963ea4 | 3142 | amdgpu_ucode_free_bo(adev); |
1e256e27 | 3143 | amdgpu_free_static_csa(&adev->virt.csa_obj); |
c12aba3a | 3144 | amdgpu_device_wb_fini(adev); |
7ccfd79f | 3145 | amdgpu_device_mem_scratch_fini(adev); |
533aed27 | 3146 | amdgpu_ib_pool_fini(adev); |
c8031019 | 3147 | amdgpu_seq64_fini(adev); |
c12aba3a ML |
3148 | } |
3149 | ||
a1255107 | 3150 | r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev); |
d38ceaf9 | 3151 | /* XXX handle errors */ |
2c1a2784 | 3152 | if (r) { |
a1255107 AD |
3153 | DRM_DEBUG("sw_fini of IP block <%s> failed %d\n", |
3154 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 3155 | } |
a1255107 AD |
3156 | adev->ip_blocks[i].status.sw = false; |
3157 | adev->ip_blocks[i].status.valid = false; | |
d38ceaf9 AD |
3158 | } |
3159 | ||
a6dcfd9c | 3160 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 3161 | if (!adev->ip_blocks[i].status.late_initialized) |
8a2eef1d | 3162 | continue; |
a1255107 AD |
3163 | if (adev->ip_blocks[i].version->funcs->late_fini) |
3164 | adev->ip_blocks[i].version->funcs->late_fini((void *)adev); | |
3165 | adev->ip_blocks[i].status.late_initialized = false; | |
a6dcfd9c ML |
3166 | } |
3167 | ||
c030f2e4 | 3168 | amdgpu_ras_fini(adev); |
3169 | ||
d38ceaf9 AD |
3170 | return 0; |
3171 | } | |
3172 | ||
e3ecdffa | 3173 | /** |
beff74bc | 3174 | * amdgpu_device_delayed_init_work_handler - work handler for IB tests |
e3ecdffa | 3175 | * |
1112a46b | 3176 | * @work: work_struct. |
e3ecdffa | 3177 | */ |
beff74bc | 3178 | static void amdgpu_device_delayed_init_work_handler(struct work_struct *work) |
2dc80b00 S |
3179 | { |
3180 | struct amdgpu_device *adev = | |
beff74bc | 3181 | container_of(work, struct amdgpu_device, delayed_init_work.work); |
916ac57f RZ |
3182 | int r; |
3183 | ||
3184 | r = amdgpu_ib_ring_tests(adev); | |
3185 | if (r) | |
3186 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
2dc80b00 S |
3187 | } |
3188 | ||
1e317b99 RZ |
3189 | static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work) |
3190 | { | |
3191 | struct amdgpu_device *adev = | |
3192 | container_of(work, struct amdgpu_device, gfx.gfx_off_delay_work.work); | |
3193 | ||
90a92662 MD |
3194 | WARN_ON_ONCE(adev->gfx.gfx_off_state); |
3195 | WARN_ON_ONCE(adev->gfx.gfx_off_req_count); | |
3196 | ||
3197 | if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true)) | |
3198 | adev->gfx.gfx_off_state = true; | |
1e317b99 RZ |
3199 | } |
3200 | ||
e3ecdffa | 3201 | /** |
e7854a03 | 3202 | * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1) |
e3ecdffa AD |
3203 | * |
3204 | * @adev: amdgpu_device pointer | |
3205 | * | |
3206 | * Main suspend function for hardware IPs. The list of all the hardware | |
3207 | * IPs that make up the asic is walked, clockgating is disabled and the | |
3208 | * suspend callbacks are run. suspend puts the hardware and software state | |
3209 | * in each IP into a state suitable for suspend. | |
3210 | * Returns 0 on success, negative error code on failure. | |
3211 | */ | |
e7854a03 AD |
3212 | static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev) |
3213 | { | |
3214 | int i, r; | |
3215 | ||
50ec83f0 AD |
3216 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); |
3217 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); | |
05df1f01 | 3218 | |
b31d6ada EQ |
3219 | /* |
3220 | * Per PMFW team's suggestion, driver needs to handle gfxoff | |
3221 | * and df cstate features disablement for gpu reset(e.g. Mode1Reset) | |
3222 | * scenario. Add the missing df cstate disablement here. | |
3223 | */ | |
3224 | if (amdgpu_dpm_set_df_cstate(adev, DF_CSTATE_DISALLOW)) | |
3225 | dev_warn(adev->dev, "Failed to disallow df cstate"); | |
3226 | ||
e7854a03 AD |
3227 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
3228 | if (!adev->ip_blocks[i].status.valid) | |
3229 | continue; | |
2b9f7848 | 3230 | |
e7854a03 | 3231 | /* displays are handled separately */ |
2b9f7848 ND |
3232 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_DCE) |
3233 | continue; | |
3234 | ||
3235 | /* XXX handle errors */ | |
3236 | r = adev->ip_blocks[i].version->funcs->suspend(adev); | |
3237 | /* XXX handle errors */ | |
3238 | if (r) { | |
3239 | DRM_ERROR("suspend of IP block <%s> failed %d\n", | |
3240 | adev->ip_blocks[i].version->funcs->name, r); | |
3241 | return r; | |
e7854a03 | 3242 | } |
2b9f7848 ND |
3243 | |
3244 | adev->ip_blocks[i].status.hw = false; | |
e7854a03 AD |
3245 | } |
3246 | ||
e7854a03 AD |
3247 | return 0; |
3248 | } | |
3249 | ||
3250 | /** | |
3251 | * amdgpu_device_ip_suspend_phase2 - run suspend for hardware IPs (phase 2) | |
3252 | * | |
3253 | * @adev: amdgpu_device pointer | |
3254 | * | |
3255 | * Main suspend function for hardware IPs. The list of all the hardware | |
3256 | * IPs that make up the asic is walked, clockgating is disabled and the | |
3257 | * suspend callbacks are run. suspend puts the hardware and software state | |
3258 | * in each IP into a state suitable for suspend. | |
3259 | * Returns 0 on success, negative error code on failure. | |
3260 | */ | |
3261 | static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev) | |
d38ceaf9 AD |
3262 | { |
3263 | int i, r; | |
3264 | ||
557f42a2 | 3265 | if (adev->in_s0ix) |
bc143d8b | 3266 | amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D3Entry); |
34416931 | 3267 | |
d38ceaf9 | 3268 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 3269 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 3270 | continue; |
e7854a03 AD |
3271 | /* displays are handled in phase1 */ |
3272 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) | |
3273 | continue; | |
bff77e86 LM |
3274 | /* PSP lost connection when err_event_athub occurs */ |
3275 | if (amdgpu_ras_intr_triggered() && | |
3276 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) { | |
3277 | adev->ip_blocks[i].status.hw = false; | |
3278 | continue; | |
3279 | } | |
e3c1b071 | 3280 | |
3281 | /* skip unnecessary suspend if we do not initialize them yet */ | |
3282 | if (adev->gmc.xgmi.pending_reset && | |
3283 | !(adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || | |
3284 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC || | |
3285 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || | |
3286 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH)) { | |
3287 | adev->ip_blocks[i].status.hw = false; | |
3288 | continue; | |
3289 | } | |
557f42a2 | 3290 | |
afa6646b | 3291 | /* skip suspend of gfx/mes and psp for S0ix |
32ff160d AD |
3292 | * gfx is in gfxoff state, so on resume it will exit gfxoff just |
3293 | * like at runtime. PSP is also part of the always on hardware | |
3294 | * so no need to suspend it. | |
3295 | */ | |
557f42a2 | 3296 | if (adev->in_s0ix && |
32ff160d | 3297 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP || |
afa6646b AD |
3298 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GFX || |
3299 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_MES)) | |
557f42a2 AD |
3300 | continue; |
3301 | ||
2a7798ea AD |
3302 | /* SDMA 5.x+ is part of GFX power domain so it's covered by GFXOFF */ |
3303 | if (adev->in_s0ix && | |
4e8303cf LL |
3304 | (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >= |
3305 | IP_VERSION(5, 0, 0)) && | |
3306 | (adev->ip_blocks[i].version->type == | |
3307 | AMD_IP_BLOCK_TYPE_SDMA)) | |
2a7798ea AD |
3308 | continue; |
3309 | ||
e11c7750 TH |
3310 | /* Once swPSP provides the IMU, RLC FW binaries to TOS during cold-boot. |
3311 | * These are in TMR, hence are expected to be reused by PSP-TOS to reload | |
3312 | * from this location and RLC Autoload automatically also gets loaded | |
3313 | * from here based on PMFW -> PSP message during re-init sequence. | |
3314 | * Therefore, the psp suspend & resume should be skipped to avoid destroy | |
3315 | * the TMR and reload FWs again for IMU enabled APU ASICs. | |
3316 | */ | |
3317 | if (amdgpu_in_reset(adev) && | |
3318 | (adev->flags & AMD_IS_APU) && adev->gfx.imu.funcs && | |
3319 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) | |
3320 | continue; | |
3321 | ||
d38ceaf9 | 3322 | /* XXX handle errors */ |
a1255107 | 3323 | r = adev->ip_blocks[i].version->funcs->suspend(adev); |
d38ceaf9 | 3324 | /* XXX handle errors */ |
2c1a2784 | 3325 | if (r) { |
a1255107 AD |
3326 | DRM_ERROR("suspend of IP block <%s> failed %d\n", |
3327 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 3328 | } |
876923fb | 3329 | adev->ip_blocks[i].status.hw = false; |
a3a09142 | 3330 | /* handle putting the SMC in the appropriate state */ |
47fc644f | 3331 | if (!amdgpu_sriov_vf(adev)) { |
86b93fd6 JZ |
3332 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { |
3333 | r = amdgpu_dpm_set_mp1_state(adev, adev->mp1_state); | |
3334 | if (r) { | |
3335 | DRM_ERROR("SMC failed to set mp1 state %d, %d\n", | |
3336 | adev->mp1_state, r); | |
3337 | return r; | |
3338 | } | |
a3a09142 AD |
3339 | } |
3340 | } | |
d38ceaf9 AD |
3341 | } |
3342 | ||
3343 | return 0; | |
3344 | } | |
3345 | ||
e7854a03 AD |
3346 | /** |
3347 | * amdgpu_device_ip_suspend - run suspend for hardware IPs | |
3348 | * | |
3349 | * @adev: amdgpu_device pointer | |
3350 | * | |
3351 | * Main suspend function for hardware IPs. The list of all the hardware | |
3352 | * IPs that make up the asic is walked, clockgating is disabled and the | |
3353 | * suspend callbacks are run. suspend puts the hardware and software state | |
3354 | * in each IP into a state suitable for suspend. | |
3355 | * Returns 0 on success, negative error code on failure. | |
3356 | */ | |
3357 | int amdgpu_device_ip_suspend(struct amdgpu_device *adev) | |
3358 | { | |
3359 | int r; | |
3360 | ||
3c73683c JC |
3361 | if (amdgpu_sriov_vf(adev)) { |
3362 | amdgpu_virt_fini_data_exchange(adev); | |
e7819644 | 3363 | amdgpu_virt_request_full_gpu(adev, false); |
3c73683c | 3364 | } |
e7819644 | 3365 | |
b7043800 AD |
3366 | amdgpu_ttm_set_buffer_funcs_status(adev, false); |
3367 | ||
e7854a03 AD |
3368 | r = amdgpu_device_ip_suspend_phase1(adev); |
3369 | if (r) | |
3370 | return r; | |
3371 | r = amdgpu_device_ip_suspend_phase2(adev); | |
3372 | ||
e7819644 YT |
3373 | if (amdgpu_sriov_vf(adev)) |
3374 | amdgpu_virt_release_full_gpu(adev, false); | |
3375 | ||
e7854a03 AD |
3376 | return r; |
3377 | } | |
3378 | ||
06ec9070 | 3379 | static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
3380 | { |
3381 | int i, r; | |
3382 | ||
2cb681b6 | 3383 | static enum amd_ip_block_type ip_order[] = { |
2cb681b6 | 3384 | AMD_IP_BLOCK_TYPE_COMMON, |
c1c39032 | 3385 | AMD_IP_BLOCK_TYPE_GMC, |
39186aef | 3386 | AMD_IP_BLOCK_TYPE_PSP, |
2cb681b6 ML |
3387 | AMD_IP_BLOCK_TYPE_IH, |
3388 | }; | |
a90ad3c2 | 3389 | |
95ea3dbc | 3390 | for (i = 0; i < adev->num_ip_blocks; i++) { |
2cb681b6 ML |
3391 | int j; |
3392 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 3393 | |
4cd2a96d J |
3394 | block = &adev->ip_blocks[i]; |
3395 | block->status.hw = false; | |
2cb681b6 | 3396 | |
4cd2a96d | 3397 | for (j = 0; j < ARRAY_SIZE(ip_order); j++) { |
2cb681b6 | 3398 | |
4cd2a96d | 3399 | if (block->version->type != ip_order[j] || |
2cb681b6 ML |
3400 | !block->status.valid) |
3401 | continue; | |
3402 | ||
3403 | r = block->version->funcs->hw_init(adev); | |
0aaeefcc | 3404 | DRM_INFO("RE-INIT-early: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); |
c41d1cf6 ML |
3405 | if (r) |
3406 | return r; | |
482f0e53 | 3407 | block->status.hw = true; |
a90ad3c2 ML |
3408 | } |
3409 | } | |
3410 | ||
3411 | return 0; | |
3412 | } | |
3413 | ||
06ec9070 | 3414 | static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
3415 | { |
3416 | int i, r; | |
3417 | ||
2cb681b6 ML |
3418 | static enum amd_ip_block_type ip_order[] = { |
3419 | AMD_IP_BLOCK_TYPE_SMC, | |
3420 | AMD_IP_BLOCK_TYPE_DCE, | |
3421 | AMD_IP_BLOCK_TYPE_GFX, | |
3422 | AMD_IP_BLOCK_TYPE_SDMA, | |
ec64350d | 3423 | AMD_IP_BLOCK_TYPE_MES, |
257deb8c | 3424 | AMD_IP_BLOCK_TYPE_UVD, |
d83c7a07 | 3425 | AMD_IP_BLOCK_TYPE_VCE, |
d2cdc014 YZ |
3426 | AMD_IP_BLOCK_TYPE_VCN, |
3427 | AMD_IP_BLOCK_TYPE_JPEG | |
2cb681b6 | 3428 | }; |
a90ad3c2 | 3429 | |
2cb681b6 ML |
3430 | for (i = 0; i < ARRAY_SIZE(ip_order); i++) { |
3431 | int j; | |
3432 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 3433 | |
2cb681b6 ML |
3434 | for (j = 0; j < adev->num_ip_blocks; j++) { |
3435 | block = &adev->ip_blocks[j]; | |
3436 | ||
3437 | if (block->version->type != ip_order[i] || | |
482f0e53 ML |
3438 | !block->status.valid || |
3439 | block->status.hw) | |
2cb681b6 ML |
3440 | continue; |
3441 | ||
895bd048 JZ |
3442 | if (block->version->type == AMD_IP_BLOCK_TYPE_SMC) |
3443 | r = block->version->funcs->resume(adev); | |
3444 | else | |
3445 | r = block->version->funcs->hw_init(adev); | |
3446 | ||
0aaeefcc | 3447 | DRM_INFO("RE-INIT-late: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); |
c41d1cf6 ML |
3448 | if (r) |
3449 | return r; | |
482f0e53 | 3450 | block->status.hw = true; |
a90ad3c2 ML |
3451 | } |
3452 | } | |
3453 | ||
3454 | return 0; | |
3455 | } | |
3456 | ||
e3ecdffa AD |
3457 | /** |
3458 | * amdgpu_device_ip_resume_phase1 - run resume for hardware IPs | |
3459 | * | |
3460 | * @adev: amdgpu_device pointer | |
3461 | * | |
3462 | * First resume function for hardware IPs. The list of all the hardware | |
3463 | * IPs that make up the asic is walked and the resume callbacks are run for | |
3464 | * COMMON, GMC, and IH. resume puts the hardware into a functional state | |
3465 | * after a suspend and updates the software state as necessary. This | |
3466 | * function is also used for restoring the GPU after a GPU reset. | |
3467 | * Returns 0 on success, negative error code on failure. | |
3468 | */ | |
06ec9070 | 3469 | static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev) |
d38ceaf9 AD |
3470 | { |
3471 | int i, r; | |
3472 | ||
a90ad3c2 | 3473 | for (i = 0; i < adev->num_ip_blocks; i++) { |
482f0e53 | 3474 | if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw) |
a90ad3c2 | 3475 | continue; |
a90ad3c2 | 3476 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
e3ecdffa | 3477 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || |
d7274ec7 BZ |
3478 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH || |
3479 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP && amdgpu_sriov_vf(adev))) { | |
482f0e53 | 3480 | |
fcf0649f CZ |
3481 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
3482 | if (r) { | |
3483 | DRM_ERROR("resume of IP block <%s> failed %d\n", | |
3484 | adev->ip_blocks[i].version->funcs->name, r); | |
3485 | return r; | |
3486 | } | |
482f0e53 | 3487 | adev->ip_blocks[i].status.hw = true; |
a90ad3c2 ML |
3488 | } |
3489 | } | |
3490 | ||
3491 | return 0; | |
3492 | } | |
3493 | ||
e3ecdffa AD |
3494 | /** |
3495 | * amdgpu_device_ip_resume_phase2 - run resume for hardware IPs | |
3496 | * | |
3497 | * @adev: amdgpu_device pointer | |
3498 | * | |
3499 | * First resume function for hardware IPs. The list of all the hardware | |
3500 | * IPs that make up the asic is walked and the resume callbacks are run for | |
3501 | * all blocks except COMMON, GMC, and IH. resume puts the hardware into a | |
3502 | * functional state after a suspend and updates the software state as | |
3503 | * necessary. This function is also used for restoring the GPU after a GPU | |
3504 | * reset. | |
3505 | * Returns 0 on success, negative error code on failure. | |
3506 | */ | |
06ec9070 | 3507 | static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev) |
d38ceaf9 AD |
3508 | { |
3509 | int i, r; | |
3510 | ||
3511 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
482f0e53 | 3512 | if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw) |
d38ceaf9 | 3513 | continue; |
fcf0649f | 3514 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
e3ecdffa | 3515 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || |
7a3e0bb2 RZ |
3516 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH || |
3517 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) | |
fcf0649f | 3518 | continue; |
a1255107 | 3519 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
2c1a2784 | 3520 | if (r) { |
a1255107 AD |
3521 | DRM_ERROR("resume of IP block <%s> failed %d\n", |
3522 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 3523 | return r; |
2c1a2784 | 3524 | } |
482f0e53 | 3525 | adev->ip_blocks[i].status.hw = true; |
d38ceaf9 AD |
3526 | } |
3527 | ||
3528 | return 0; | |
3529 | } | |
3530 | ||
e3ecdffa AD |
3531 | /** |
3532 | * amdgpu_device_ip_resume - run resume for hardware IPs | |
3533 | * | |
3534 | * @adev: amdgpu_device pointer | |
3535 | * | |
3536 | * Main resume function for hardware IPs. The hardware IPs | |
3537 | * are split into two resume functions because they are | |
b8920e1e | 3538 | * also used in recovering from a GPU reset and some additional |
e3ecdffa AD |
3539 | * steps need to be take between them. In this case (S3/S4) they are |
3540 | * run sequentially. | |
3541 | * Returns 0 on success, negative error code on failure. | |
3542 | */ | |
06ec9070 | 3543 | static int amdgpu_device_ip_resume(struct amdgpu_device *adev) |
fcf0649f CZ |
3544 | { |
3545 | int r; | |
3546 | ||
06ec9070 | 3547 | r = amdgpu_device_ip_resume_phase1(adev); |
fcf0649f CZ |
3548 | if (r) |
3549 | return r; | |
7a3e0bb2 RZ |
3550 | |
3551 | r = amdgpu_device_fw_loading(adev); | |
3552 | if (r) | |
3553 | return r; | |
3554 | ||
06ec9070 | 3555 | r = amdgpu_device_ip_resume_phase2(adev); |
fcf0649f | 3556 | |
b7043800 AD |
3557 | if (adev->mman.buffer_funcs_ring->sched.ready) |
3558 | amdgpu_ttm_set_buffer_funcs_status(adev, true); | |
3559 | ||
fcf0649f CZ |
3560 | return r; |
3561 | } | |
3562 | ||
e3ecdffa AD |
3563 | /** |
3564 | * amdgpu_device_detect_sriov_bios - determine if the board supports SR-IOV | |
3565 | * | |
3566 | * @adev: amdgpu_device pointer | |
3567 | * | |
3568 | * Query the VBIOS data tables to determine if the board supports SR-IOV. | |
3569 | */ | |
4e99a44e | 3570 | static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev) |
048765ad | 3571 | { |
6867e1b5 ML |
3572 | if (amdgpu_sriov_vf(adev)) { |
3573 | if (adev->is_atom_fw) { | |
58ff791a | 3574 | if (amdgpu_atomfirmware_gpu_virtualization_supported(adev)) |
6867e1b5 ML |
3575 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; |
3576 | } else { | |
3577 | if (amdgpu_atombios_has_gpu_virtualization_table(adev)) | |
3578 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; | |
3579 | } | |
3580 | ||
3581 | if (!(adev->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS)) | |
3582 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0); | |
a5bde2f9 | 3583 | } |
048765ad AR |
3584 | } |
3585 | ||
e3ecdffa AD |
3586 | /** |
3587 | * amdgpu_device_asic_has_dc_support - determine if DC supports the asic | |
3588 | * | |
3589 | * @asic_type: AMD asic type | |
3590 | * | |
3591 | * Check if there is DC (new modesetting infrastructre) support for an asic. | |
3592 | * returns true if DC has support, false if not. | |
3593 | */ | |
4562236b HW |
3594 | bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type) |
3595 | { | |
3596 | switch (asic_type) { | |
0637d417 AD |
3597 | #ifdef CONFIG_DRM_AMDGPU_SI |
3598 | case CHIP_HAINAN: | |
3599 | #endif | |
3600 | case CHIP_TOPAZ: | |
3601 | /* chips with no display hardware */ | |
3602 | return false; | |
4562236b | 3603 | #if defined(CONFIG_DRM_AMD_DC) |
64200c46 MR |
3604 | case CHIP_TAHITI: |
3605 | case CHIP_PITCAIRN: | |
3606 | case CHIP_VERDE: | |
3607 | case CHIP_OLAND: | |
2d32ffd6 AD |
3608 | /* |
3609 | * We have systems in the wild with these ASICs that require | |
3610 | * LVDS and VGA support which is not supported with DC. | |
3611 | * | |
3612 | * Fallback to the non-DC driver here by default so as not to | |
3613 | * cause regressions. | |
3614 | */ | |
3615 | #if defined(CONFIG_DRM_AMD_DC_SI) | |
3616 | return amdgpu_dc > 0; | |
3617 | #else | |
3618 | return false; | |
64200c46 | 3619 | #endif |
4562236b | 3620 | case CHIP_BONAIRE: |
0d6fbccb | 3621 | case CHIP_KAVERI: |
367e6687 AD |
3622 | case CHIP_KABINI: |
3623 | case CHIP_MULLINS: | |
d9fda248 HW |
3624 | /* |
3625 | * We have systems in the wild with these ASICs that require | |
b5a0168e | 3626 | * VGA support which is not supported with DC. |
d9fda248 HW |
3627 | * |
3628 | * Fallback to the non-DC driver here by default so as not to | |
3629 | * cause regressions. | |
3630 | */ | |
3631 | return amdgpu_dc > 0; | |
f7f12b25 | 3632 | default: |
fd187853 | 3633 | return amdgpu_dc != 0; |
f7f12b25 | 3634 | #else |
4562236b | 3635 | default: |
93b09a9a | 3636 | if (amdgpu_dc > 0) |
b8920e1e | 3637 | DRM_INFO_ONCE("Display Core has been requested via kernel parameter but isn't supported by ASIC, ignoring\n"); |
4562236b | 3638 | return false; |
f7f12b25 | 3639 | #endif |
4562236b HW |
3640 | } |
3641 | } | |
3642 | ||
3643 | /** | |
3644 | * amdgpu_device_has_dc_support - check if dc is supported | |
3645 | * | |
982a820b | 3646 | * @adev: amdgpu_device pointer |
4562236b HW |
3647 | * |
3648 | * Returns true for supported, false for not supported | |
3649 | */ | |
3650 | bool amdgpu_device_has_dc_support(struct amdgpu_device *adev) | |
3651 | { | |
25263da3 | 3652 | if (adev->enable_virtual_display || |
abaf210c | 3653 | (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK)) |
2555039d XY |
3654 | return false; |
3655 | ||
4562236b HW |
3656 | return amdgpu_device_asic_has_dc_support(adev->asic_type); |
3657 | } | |
3658 | ||
d4535e2c AG |
3659 | static void amdgpu_device_xgmi_reset_func(struct work_struct *__work) |
3660 | { | |
3661 | struct amdgpu_device *adev = | |
3662 | container_of(__work, struct amdgpu_device, xgmi_reset_work); | |
d95e8e97 | 3663 | struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev); |
d4535e2c | 3664 | |
c6a6e2db AG |
3665 | /* It's a bug to not have a hive within this function */ |
3666 | if (WARN_ON(!hive)) | |
3667 | return; | |
3668 | ||
3669 | /* | |
3670 | * Use task barrier to synchronize all xgmi reset works across the | |
3671 | * hive. task_barrier_enter and task_barrier_exit will block | |
3672 | * until all the threads running the xgmi reset works reach | |
3673 | * those points. task_barrier_full will do both blocks. | |
3674 | */ | |
3675 | if (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) { | |
3676 | ||
3677 | task_barrier_enter(&hive->tb); | |
4a580877 | 3678 | adev->asic_reset_res = amdgpu_device_baco_enter(adev_to_drm(adev)); |
c6a6e2db AG |
3679 | |
3680 | if (adev->asic_reset_res) | |
3681 | goto fail; | |
3682 | ||
3683 | task_barrier_exit(&hive->tb); | |
4a580877 | 3684 | adev->asic_reset_res = amdgpu_device_baco_exit(adev_to_drm(adev)); |
c6a6e2db AG |
3685 | |
3686 | if (adev->asic_reset_res) | |
3687 | goto fail; | |
43c4d576 | 3688 | |
21226f02 | 3689 | amdgpu_ras_reset_error_count(adev, AMDGPU_RAS_BLOCK__MMHUB); |
c6a6e2db AG |
3690 | } else { |
3691 | ||
3692 | task_barrier_full(&hive->tb); | |
3693 | adev->asic_reset_res = amdgpu_asic_reset(adev); | |
3694 | } | |
ce316fa5 | 3695 | |
c6a6e2db | 3696 | fail: |
d4535e2c | 3697 | if (adev->asic_reset_res) |
fed184e9 | 3698 | DRM_WARN("ASIC reset failed with error, %d for drm dev, %s", |
4a580877 | 3699 | adev->asic_reset_res, adev_to_drm(adev)->unique); |
d95e8e97 | 3700 | amdgpu_put_xgmi_hive(hive); |
d4535e2c AG |
3701 | } |
3702 | ||
71f98027 AD |
3703 | static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev) |
3704 | { | |
3705 | char *input = amdgpu_lockup_timeout; | |
3706 | char *timeout_setting = NULL; | |
3707 | int index = 0; | |
3708 | long timeout; | |
3709 | int ret = 0; | |
3710 | ||
3711 | /* | |
67387dfe AD |
3712 | * By default timeout for non compute jobs is 10000 |
3713 | * and 60000 for compute jobs. | |
71f98027 | 3714 | * In SR-IOV or passthrough mode, timeout for compute |
b7b2a316 | 3715 | * jobs are 60000 by default. |
71f98027 AD |
3716 | */ |
3717 | adev->gfx_timeout = msecs_to_jiffies(10000); | |
3718 | adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; | |
9882e278 ED |
3719 | if (amdgpu_sriov_vf(adev)) |
3720 | adev->compute_timeout = amdgpu_sriov_is_pp_one_vf(adev) ? | |
3721 | msecs_to_jiffies(60000) : msecs_to_jiffies(10000); | |
71f98027 | 3722 | else |
67387dfe | 3723 | adev->compute_timeout = msecs_to_jiffies(60000); |
71f98027 | 3724 | |
f440ff44 | 3725 | if (strnlen(input, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH)) { |
71f98027 | 3726 | while ((timeout_setting = strsep(&input, ",")) && |
f440ff44 | 3727 | strnlen(timeout_setting, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH)) { |
71f98027 AD |
3728 | ret = kstrtol(timeout_setting, 0, &timeout); |
3729 | if (ret) | |
3730 | return ret; | |
3731 | ||
3732 | if (timeout == 0) { | |
3733 | index++; | |
3734 | continue; | |
3735 | } else if (timeout < 0) { | |
3736 | timeout = MAX_SCHEDULE_TIMEOUT; | |
127aedf9 CK |
3737 | dev_warn(adev->dev, "lockup timeout disabled"); |
3738 | add_taint(TAINT_SOFTLOCKUP, LOCKDEP_STILL_OK); | |
71f98027 AD |
3739 | } else { |
3740 | timeout = msecs_to_jiffies(timeout); | |
3741 | } | |
3742 | ||
3743 | switch (index++) { | |
3744 | case 0: | |
3745 | adev->gfx_timeout = timeout; | |
3746 | break; | |
3747 | case 1: | |
3748 | adev->compute_timeout = timeout; | |
3749 | break; | |
3750 | case 2: | |
3751 | adev->sdma_timeout = timeout; | |
3752 | break; | |
3753 | case 3: | |
3754 | adev->video_timeout = timeout; | |
3755 | break; | |
3756 | default: | |
3757 | break; | |
3758 | } | |
3759 | } | |
3760 | /* | |
3761 | * There is only one value specified and | |
3762 | * it should apply to all non-compute jobs. | |
3763 | */ | |
bcccee89 | 3764 | if (index == 1) { |
71f98027 | 3765 | adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; |
bcccee89 ED |
3766 | if (amdgpu_sriov_vf(adev) || amdgpu_passthrough(adev)) |
3767 | adev->compute_timeout = adev->gfx_timeout; | |
3768 | } | |
71f98027 AD |
3769 | } |
3770 | ||
3771 | return ret; | |
3772 | } | |
d4535e2c | 3773 | |
4a74c38c PY |
3774 | /** |
3775 | * amdgpu_device_check_iommu_direct_map - check if RAM direct mapped to GPU | |
3776 | * | |
3777 | * @adev: amdgpu_device pointer | |
3778 | * | |
3779 | * RAM direct mapped to GPU if IOMMU is not enabled or is pass through mode | |
3780 | */ | |
3781 | static void amdgpu_device_check_iommu_direct_map(struct amdgpu_device *adev) | |
3782 | { | |
3783 | struct iommu_domain *domain; | |
3784 | ||
3785 | domain = iommu_get_domain_for_dev(adev->dev); | |
3786 | if (!domain || domain->type == IOMMU_DOMAIN_IDENTITY) | |
3787 | adev->ram_is_direct_mapped = true; | |
3788 | } | |
3789 | ||
77f3a5cd | 3790 | static const struct attribute *amdgpu_dev_attributes[] = { |
77f3a5cd ND |
3791 | &dev_attr_pcie_replay_count.attr, |
3792 | NULL | |
3793 | }; | |
3794 | ||
02ff519e AD |
3795 | static void amdgpu_device_set_mcbp(struct amdgpu_device *adev) |
3796 | { | |
3797 | if (amdgpu_mcbp == 1) | |
3798 | adev->gfx.mcbp = true; | |
1e9e15dc JZ |
3799 | else if (amdgpu_mcbp == 0) |
3800 | adev->gfx.mcbp = false; | |
4e8303cf LL |
3801 | else if ((amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(9, 0, 0)) && |
3802 | (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(10, 0, 0)) && | |
1e9e15dc | 3803 | adev->gfx.num_gfx_rings) |
50a7c876 AD |
3804 | adev->gfx.mcbp = true; |
3805 | ||
02ff519e AD |
3806 | if (amdgpu_sriov_vf(adev)) |
3807 | adev->gfx.mcbp = true; | |
3808 | ||
3809 | if (adev->gfx.mcbp) | |
3810 | DRM_INFO("MCBP is enabled\n"); | |
3811 | } | |
3812 | ||
d38ceaf9 AD |
3813 | /** |
3814 | * amdgpu_device_init - initialize the driver | |
3815 | * | |
3816 | * @adev: amdgpu_device pointer | |
d38ceaf9 AD |
3817 | * @flags: driver flags |
3818 | * | |
3819 | * Initializes the driver info and hw (all asics). | |
3820 | * Returns 0 for success or an error on failure. | |
3821 | * Called at driver startup. | |
3822 | */ | |
3823 | int amdgpu_device_init(struct amdgpu_device *adev, | |
d38ceaf9 AD |
3824 | uint32_t flags) |
3825 | { | |
8aba21b7 LT |
3826 | struct drm_device *ddev = adev_to_drm(adev); |
3827 | struct pci_dev *pdev = adev->pdev; | |
d38ceaf9 | 3828 | int r, i; |
b98c6299 | 3829 | bool px = false; |
95844d20 | 3830 | u32 max_MBps; |
59e9fff1 | 3831 | int tmp; |
d38ceaf9 AD |
3832 | |
3833 | adev->shutdown = false; | |
d38ceaf9 | 3834 | adev->flags = flags; |
4e66d7d2 YZ |
3835 | |
3836 | if (amdgpu_force_asic_type >= 0 && amdgpu_force_asic_type < CHIP_LAST) | |
3837 | adev->asic_type = amdgpu_force_asic_type; | |
3838 | else | |
3839 | adev->asic_type = flags & AMD_ASIC_MASK; | |
3840 | ||
d38ceaf9 | 3841 | adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT; |
593aa2d2 | 3842 | if (amdgpu_emu_mode == 1) |
8bdab6bb | 3843 | adev->usec_timeout *= 10; |
770d13b1 | 3844 | adev->gmc.gart_size = 512 * 1024 * 1024; |
d38ceaf9 AD |
3845 | adev->accel_working = false; |
3846 | adev->num_rings = 0; | |
68ce8b24 | 3847 | RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub()); |
d38ceaf9 AD |
3848 | adev->mman.buffer_funcs = NULL; |
3849 | adev->mman.buffer_funcs_ring = NULL; | |
3850 | adev->vm_manager.vm_pte_funcs = NULL; | |
0c88b430 | 3851 | adev->vm_manager.vm_pte_num_scheds = 0; |
132f34e4 | 3852 | adev->gmc.gmc_funcs = NULL; |
7bd939d0 | 3853 | adev->harvest_ip_mask = 0x0; |
f54d1867 | 3854 | adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS); |
b8866c26 | 3855 | bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); |
d38ceaf9 AD |
3856 | |
3857 | adev->smc_rreg = &amdgpu_invalid_rreg; | |
3858 | adev->smc_wreg = &amdgpu_invalid_wreg; | |
3859 | adev->pcie_rreg = &amdgpu_invalid_rreg; | |
3860 | adev->pcie_wreg = &amdgpu_invalid_wreg; | |
0c552ed3 LM |
3861 | adev->pcie_rreg_ext = &amdgpu_invalid_rreg_ext; |
3862 | adev->pcie_wreg_ext = &amdgpu_invalid_wreg_ext; | |
36b9a952 HR |
3863 | adev->pciep_rreg = &amdgpu_invalid_rreg; |
3864 | adev->pciep_wreg = &amdgpu_invalid_wreg; | |
4fa1c6a6 TZ |
3865 | adev->pcie_rreg64 = &amdgpu_invalid_rreg64; |
3866 | adev->pcie_wreg64 = &amdgpu_invalid_wreg64; | |
a76b2870 CL |
3867 | adev->pcie_rreg64_ext = &amdgpu_invalid_rreg64_ext; |
3868 | adev->pcie_wreg64_ext = &amdgpu_invalid_wreg64_ext; | |
d38ceaf9 AD |
3869 | adev->uvd_ctx_rreg = &amdgpu_invalid_rreg; |
3870 | adev->uvd_ctx_wreg = &amdgpu_invalid_wreg; | |
3871 | adev->didt_rreg = &amdgpu_invalid_rreg; | |
3872 | adev->didt_wreg = &amdgpu_invalid_wreg; | |
ccdbb20a RZ |
3873 | adev->gc_cac_rreg = &amdgpu_invalid_rreg; |
3874 | adev->gc_cac_wreg = &amdgpu_invalid_wreg; | |
d38ceaf9 AD |
3875 | adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg; |
3876 | adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg; | |
3877 | ||
3e39ab90 AD |
3878 | DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n", |
3879 | amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device, | |
3880 | pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision); | |
d38ceaf9 AD |
3881 | |
3882 | /* mutex initialization are all done here so we | |
b8920e1e SS |
3883 | * can recall function without having locking issues |
3884 | */ | |
0e5ca0d1 | 3885 | mutex_init(&adev->firmware.mutex); |
d38ceaf9 AD |
3886 | mutex_init(&adev->pm.mutex); |
3887 | mutex_init(&adev->gfx.gpu_clock_mutex); | |
3888 | mutex_init(&adev->srbm_mutex); | |
b8866c26 | 3889 | mutex_init(&adev->gfx.pipe_reserve_mutex); |
d23ee13f | 3890 | mutex_init(&adev->gfx.gfx_off_mutex); |
98a54e88 | 3891 | mutex_init(&adev->gfx.partition_mutex); |
d38ceaf9 | 3892 | mutex_init(&adev->grbm_idx_mutex); |
d38ceaf9 | 3893 | mutex_init(&adev->mn_lock); |
e23b74aa | 3894 | mutex_init(&adev->virt.vf_errors.lock); |
d38ceaf9 | 3895 | hash_init(adev->mn_hash); |
32eaeae0 | 3896 | mutex_init(&adev->psp.mutex); |
bd052211 | 3897 | mutex_init(&adev->notifier_lock); |
8cda7a4f | 3898 | mutex_init(&adev->pm.stable_pstate_ctx_lock); |
f113cc32 | 3899 | mutex_init(&adev->benchmark_mutex); |
d38ceaf9 | 3900 | |
ab3b9de6 | 3901 | amdgpu_device_init_apu_flags(adev); |
9f6a7857 | 3902 | |
912dfc84 EQ |
3903 | r = amdgpu_device_check_arguments(adev); |
3904 | if (r) | |
3905 | return r; | |
d38ceaf9 | 3906 | |
d38ceaf9 AD |
3907 | spin_lock_init(&adev->mmio_idx_lock); |
3908 | spin_lock_init(&adev->smc_idx_lock); | |
3909 | spin_lock_init(&adev->pcie_idx_lock); | |
3910 | spin_lock_init(&adev->uvd_ctx_idx_lock); | |
3911 | spin_lock_init(&adev->didt_idx_lock); | |
ccdbb20a | 3912 | spin_lock_init(&adev->gc_cac_idx_lock); |
16abb5d2 | 3913 | spin_lock_init(&adev->se_cac_idx_lock); |
d38ceaf9 | 3914 | spin_lock_init(&adev->audio_endpt_idx_lock); |
95844d20 | 3915 | spin_lock_init(&adev->mm_stats.lock); |
d38ceaf9 | 3916 | |
0c4e7fa5 CZ |
3917 | INIT_LIST_HEAD(&adev->shadow_list); |
3918 | mutex_init(&adev->shadow_list_lock); | |
3919 | ||
655ce9cb | 3920 | INIT_LIST_HEAD(&adev->reset_list); |
3921 | ||
6492e1b0 | 3922 | INIT_LIST_HEAD(&adev->ras_list); |
3923 | ||
3e38b634 EQ |
3924 | INIT_LIST_HEAD(&adev->pm.od_kobj_list); |
3925 | ||
beff74bc AD |
3926 | INIT_DELAYED_WORK(&adev->delayed_init_work, |
3927 | amdgpu_device_delayed_init_work_handler); | |
1e317b99 RZ |
3928 | INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work, |
3929 | amdgpu_device_delay_enable_gfx_off); | |
2dc80b00 | 3930 | |
d4535e2c AG |
3931 | INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func); |
3932 | ||
d23ee13f | 3933 | adev->gfx.gfx_off_req_count = 1; |
0ad7347a AA |
3934 | adev->gfx.gfx_off_residency = 0; |
3935 | adev->gfx.gfx_off_entrycount = 0; | |
b6e79d9a | 3936 | adev->pm.ac_power = power_supply_is_system_supplied() > 0; |
b1ddf548 | 3937 | |
b265bdbd EQ |
3938 | atomic_set(&adev->throttling_logging_enabled, 1); |
3939 | /* | |
3940 | * If throttling continues, logging will be performed every minute | |
3941 | * to avoid log flooding. "-1" is subtracted since the thermal | |
3942 | * throttling interrupt comes every second. Thus, the total logging | |
3943 | * interval is 59 seconds(retelimited printk interval) + 1(waiting | |
3944 | * for throttling interrupt) = 60 seconds. | |
3945 | */ | |
3946 | ratelimit_state_init(&adev->throttling_logging_rs, (60 - 1) * HZ, 1); | |
3947 | ratelimit_set_flags(&adev->throttling_logging_rs, RATELIMIT_MSG_ON_RELEASE); | |
3948 | ||
0fa49558 AX |
3949 | /* Registers mapping */ |
3950 | /* TODO: block userspace mapping of io register */ | |
da69c161 KW |
3951 | if (adev->asic_type >= CHIP_BONAIRE) { |
3952 | adev->rmmio_base = pci_resource_start(adev->pdev, 5); | |
3953 | adev->rmmio_size = pci_resource_len(adev->pdev, 5); | |
3954 | } else { | |
3955 | adev->rmmio_base = pci_resource_start(adev->pdev, 2); | |
3956 | adev->rmmio_size = pci_resource_len(adev->pdev, 2); | |
3957 | } | |
d38ceaf9 | 3958 | |
6c08e0ef EQ |
3959 | for (i = 0; i < AMD_IP_BLOCK_TYPE_NUM; i++) |
3960 | atomic_set(&adev->pm.pwr_state[i], POWER_STATE_UNKNOWN); | |
3961 | ||
d38ceaf9 | 3962 | adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size); |
b8920e1e | 3963 | if (!adev->rmmio) |
d38ceaf9 | 3964 | return -ENOMEM; |
b8920e1e | 3965 | |
d38ceaf9 | 3966 | DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base); |
b8920e1e | 3967 | DRM_INFO("register mmio size: %u\n", (unsigned int)adev->rmmio_size); |
d38ceaf9 | 3968 | |
436afdfa PY |
3969 | /* |
3970 | * Reset domain needs to be present early, before XGMI hive discovered | |
3971 | * (if any) and intitialized to use reset sem and in_gpu reset flag | |
3972 | * early on during init and before calling to RREG32. | |
3973 | */ | |
3974 | adev->reset_domain = amdgpu_reset_create_reset_domain(SINGLE_DEVICE, "amdgpu-reset-dev"); | |
3975 | if (!adev->reset_domain) | |
3976 | return -ENOMEM; | |
3977 | ||
3aa0115d ML |
3978 | /* detect hw virtualization here */ |
3979 | amdgpu_detect_virtualization(adev); | |
3980 | ||
04e85958 TL |
3981 | amdgpu_device_get_pcie_info(adev); |
3982 | ||
dffa11b4 ML |
3983 | r = amdgpu_device_get_job_timeout_settings(adev); |
3984 | if (r) { | |
3985 | dev_err(adev->dev, "invalid lockup_timeout parameter syntax\n"); | |
4ef87d8f | 3986 | return r; |
a190d1c7 XY |
3987 | } |
3988 | ||
d38ceaf9 | 3989 | /* early init functions */ |
06ec9070 | 3990 | r = amdgpu_device_ip_early_init(adev); |
d38ceaf9 | 3991 | if (r) |
4ef87d8f | 3992 | return r; |
d38ceaf9 | 3993 | |
02ff519e AD |
3994 | amdgpu_device_set_mcbp(adev); |
3995 | ||
b7cdb41e ML |
3996 | /* Get rid of things like offb */ |
3997 | r = drm_aperture_remove_conflicting_pci_framebuffers(adev->pdev, &amdgpu_kms_driver); | |
3998 | if (r) | |
3999 | return r; | |
4000 | ||
4d33e704 SK |
4001 | /* Enable TMZ based on IP_VERSION */ |
4002 | amdgpu_gmc_tmz_set(adev); | |
4003 | ||
957b0787 | 4004 | amdgpu_gmc_noretry_set(adev); |
4a0165f0 VS |
4005 | /* Need to get xgmi info early to decide the reset behavior*/ |
4006 | if (adev->gmc.xgmi.supported) { | |
4007 | r = adev->gfxhub.funcs->get_xgmi_info(adev); | |
4008 | if (r) | |
4009 | return r; | |
4010 | } | |
4011 | ||
8e6d0b69 | 4012 | /* enable PCIE atomic ops */ |
b4520bfd GW |
4013 | if (amdgpu_sriov_vf(adev)) { |
4014 | if (adev->virt.fw_reserve.p_pf2vf) | |
4015 | adev->have_atomics_support = ((struct amd_sriov_msg_pf2vf_info *) | |
4016 | adev->virt.fw_reserve.p_pf2vf)->pcie_atomic_ops_support_flags == | |
4017 | (PCI_EXP_DEVCAP2_ATOMIC_COMP32 | PCI_EXP_DEVCAP2_ATOMIC_COMP64); | |
0e768043 YZ |
4018 | /* APUs w/ gfx9 onwards doesn't reply on PCIe atomics, rather it is a |
4019 | * internal path natively support atomics, set have_atomics_support to true. | |
4020 | */ | |
b4520bfd | 4021 | } else if ((adev->flags & AMD_IS_APU) && |
4e8303cf LL |
4022 | (amdgpu_ip_version(adev, GC_HWIP, 0) > |
4023 | IP_VERSION(9, 0, 0))) { | |
0e768043 | 4024 | adev->have_atomics_support = true; |
b4520bfd | 4025 | } else { |
8e6d0b69 | 4026 | adev->have_atomics_support = |
4027 | !pci_enable_atomic_ops_to_root(adev->pdev, | |
4028 | PCI_EXP_DEVCAP2_ATOMIC_COMP32 | | |
4029 | PCI_EXP_DEVCAP2_ATOMIC_COMP64); | |
b4520bfd GW |
4030 | } |
4031 | ||
8e6d0b69 | 4032 | if (!adev->have_atomics_support) |
4033 | dev_info(adev->dev, "PCIE atomic ops is not supported\n"); | |
4034 | ||
6585661d | 4035 | /* doorbell bar mapping and doorbell index init*/ |
43c064db | 4036 | amdgpu_doorbell_init(adev); |
6585661d | 4037 | |
9475a943 SL |
4038 | if (amdgpu_emu_mode == 1) { |
4039 | /* post the asic on emulation mode */ | |
4040 | emu_soc_asic_init(adev); | |
bfca0289 | 4041 | goto fence_driver_init; |
9475a943 | 4042 | } |
bfca0289 | 4043 | |
04442bf7 LL |
4044 | amdgpu_reset_init(adev); |
4045 | ||
4e99a44e | 4046 | /* detect if we are with an SRIOV vbios */ |
b4520bfd GW |
4047 | if (adev->bios) |
4048 | amdgpu_device_detect_sriov_bios(adev); | |
048765ad | 4049 | |
95e8e59e AD |
4050 | /* check if we need to reset the asic |
4051 | * E.g., driver was not cleanly unloaded previously, etc. | |
4052 | */ | |
f14899fd | 4053 | if (!amdgpu_sriov_vf(adev) && amdgpu_asic_need_reset_on_init(adev)) { |
e3c1b071 | 4054 | if (adev->gmc.xgmi.num_physical_nodes) { |
4055 | dev_info(adev->dev, "Pending hive reset.\n"); | |
4056 | adev->gmc.xgmi.pending_reset = true; | |
4057 | /* Only need to init necessary block for SMU to handle the reset */ | |
4058 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
4059 | if (!adev->ip_blocks[i].status.valid) | |
4060 | continue; | |
4061 | if (!(adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || | |
4062 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || | |
4063 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH || | |
4064 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC)) { | |
751f43e7 | 4065 | DRM_DEBUG("IP %s disabled for hw_init.\n", |
e3c1b071 | 4066 | adev->ip_blocks[i].version->funcs->name); |
4067 | adev->ip_blocks[i].status.hw = true; | |
4068 | } | |
4069 | } | |
4070 | } else { | |
5f38ac54 KF |
4071 | switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) { |
4072 | case IP_VERSION(13, 0, 0): | |
4073 | case IP_VERSION(13, 0, 7): | |
4074 | case IP_VERSION(13, 0, 10): | |
4075 | r = psp_gpu_reset(adev); | |
4076 | break; | |
4077 | default: | |
4078 | tmp = amdgpu_reset_method; | |
4079 | /* It should do a default reset when loading or reloading the driver, | |
4080 | * regardless of the module parameter reset_method. | |
4081 | */ | |
4082 | amdgpu_reset_method = AMD_RESET_METHOD_NONE; | |
4083 | r = amdgpu_asic_reset(adev); | |
4084 | amdgpu_reset_method = tmp; | |
4085 | break; | |
4086 | } | |
4087 | ||
e3c1b071 | 4088 | if (r) { |
4089 | dev_err(adev->dev, "asic reset on init failed\n"); | |
4090 | goto failed; | |
4091 | } | |
95e8e59e AD |
4092 | } |
4093 | } | |
4094 | ||
d38ceaf9 | 4095 | /* Post card if necessary */ |
39c640c0 | 4096 | if (amdgpu_device_need_post(adev)) { |
d38ceaf9 | 4097 | if (!adev->bios) { |
bec86378 | 4098 | dev_err(adev->dev, "no vBIOS found\n"); |
83ba126a AD |
4099 | r = -EINVAL; |
4100 | goto failed; | |
d38ceaf9 | 4101 | } |
bec86378 | 4102 | DRM_INFO("GPU posting now...\n"); |
4d2997ab | 4103 | r = amdgpu_device_asic_init(adev); |
4e99a44e ML |
4104 | if (r) { |
4105 | dev_err(adev->dev, "gpu post error!\n"); | |
4106 | goto failed; | |
4107 | } | |
d38ceaf9 AD |
4108 | } |
4109 | ||
9535a86a SZ |
4110 | if (adev->bios) { |
4111 | if (adev->is_atom_fw) { | |
4112 | /* Initialize clocks */ | |
4113 | r = amdgpu_atomfirmware_get_clock_info(adev); | |
4114 | if (r) { | |
4115 | dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n"); | |
4116 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); | |
4117 | goto failed; | |
4118 | } | |
4119 | } else { | |
4120 | /* Initialize clocks */ | |
4121 | r = amdgpu_atombios_get_clock_info(adev); | |
4122 | if (r) { | |
4123 | dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n"); | |
4124 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); | |
4125 | goto failed; | |
4126 | } | |
4127 | /* init i2c buses */ | |
4128 | if (!amdgpu_device_has_dc_support(adev)) | |
4129 | amdgpu_atombios_i2c_init(adev); | |
a5bde2f9 | 4130 | } |
2c1a2784 | 4131 | } |
d38ceaf9 | 4132 | |
bfca0289 | 4133 | fence_driver_init: |
d38ceaf9 | 4134 | /* Fence driver */ |
067f44c8 | 4135 | r = amdgpu_fence_driver_sw_init(adev); |
2c1a2784 | 4136 | if (r) { |
067f44c8 | 4137 | dev_err(adev->dev, "amdgpu_fence_driver_sw_init failed\n"); |
e23b74aa | 4138 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0); |
83ba126a | 4139 | goto failed; |
2c1a2784 | 4140 | } |
d38ceaf9 AD |
4141 | |
4142 | /* init the mode config */ | |
4a580877 | 4143 | drm_mode_config_init(adev_to_drm(adev)); |
d38ceaf9 | 4144 | |
06ec9070 | 4145 | r = amdgpu_device_ip_init(adev); |
d38ceaf9 | 4146 | if (r) { |
06ec9070 | 4147 | dev_err(adev->dev, "amdgpu_device_ip_init failed\n"); |
e23b74aa | 4148 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0); |
970fd197 | 4149 | goto release_ras_con; |
d38ceaf9 AD |
4150 | } |
4151 | ||
8d35a259 LG |
4152 | amdgpu_fence_driver_hw_init(adev); |
4153 | ||
d69b8971 YZ |
4154 | dev_info(adev->dev, |
4155 | "SE %d, SH per SE %d, CU per SH %d, active_cu_number %d\n", | |
d7f72fe4 YZ |
4156 | adev->gfx.config.max_shader_engines, |
4157 | adev->gfx.config.max_sh_per_se, | |
4158 | adev->gfx.config.max_cu_per_sh, | |
4159 | adev->gfx.cu_info.number); | |
4160 | ||
d38ceaf9 AD |
4161 | adev->accel_working = true; |
4162 | ||
e59c0205 AX |
4163 | amdgpu_vm_check_compute_bug(adev); |
4164 | ||
95844d20 MO |
4165 | /* Initialize the buffer migration limit. */ |
4166 | if (amdgpu_moverate >= 0) | |
4167 | max_MBps = amdgpu_moverate; | |
4168 | else | |
4169 | max_MBps = 8; /* Allow 8 MB/s. */ | |
4170 | /* Get a log2 for easy divisions. */ | |
4171 | adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps)); | |
4172 | ||
b0adca4d EQ |
4173 | /* |
4174 | * Register gpu instance before amdgpu_device_enable_mgpu_fan_boost. | |
4175 | * Otherwise the mgpu fan boost feature will be skipped due to the | |
4176 | * gpu instance is counted less. | |
4177 | */ | |
4178 | amdgpu_register_gpu_instance(adev); | |
4179 | ||
d38ceaf9 AD |
4180 | /* enable clockgating, etc. after ib tests, etc. since some blocks require |
4181 | * explicit gating rather than handling it automatically. | |
4182 | */ | |
e3c1b071 | 4183 | if (!adev->gmc.xgmi.pending_reset) { |
4184 | r = amdgpu_device_ip_late_init(adev); | |
4185 | if (r) { | |
4186 | dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n"); | |
4187 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r); | |
970fd197 | 4188 | goto release_ras_con; |
e3c1b071 | 4189 | } |
4190 | /* must succeed. */ | |
4191 | amdgpu_ras_resume(adev); | |
4192 | queue_delayed_work(system_wq, &adev->delayed_init_work, | |
4193 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
2c1a2784 | 4194 | } |
d38ceaf9 | 4195 | |
38eecbe0 CL |
4196 | if (amdgpu_sriov_vf(adev)) { |
4197 | amdgpu_virt_release_full_gpu(adev, true); | |
2c738637 | 4198 | flush_delayed_work(&adev->delayed_init_work); |
38eecbe0 | 4199 | } |
2c738637 | 4200 | |
90bcb9b5 EQ |
4201 | /* |
4202 | * Place those sysfs registering after `late_init`. As some of those | |
4203 | * operations performed in `late_init` might affect the sysfs | |
4204 | * interfaces creating. | |
4205 | */ | |
4206 | r = amdgpu_atombios_sysfs_init(adev); | |
4207 | if (r) | |
4208 | drm_err(&adev->ddev, | |
4209 | "registering atombios sysfs failed (%d).\n", r); | |
4210 | ||
4211 | r = amdgpu_pm_sysfs_init(adev); | |
4212 | if (r) | |
4213 | DRM_ERROR("registering pm sysfs failed (%d).\n", r); | |
4214 | ||
4215 | r = amdgpu_ucode_sysfs_init(adev); | |
4216 | if (r) { | |
4217 | adev->ucode_sysfs_en = false; | |
4218 | DRM_ERROR("Creating firmware sysfs failed (%d).\n", r); | |
4219 | } else | |
4220 | adev->ucode_sysfs_en = true; | |
4221 | ||
77f3a5cd | 4222 | r = sysfs_create_files(&adev->dev->kobj, amdgpu_dev_attributes); |
5aea5327 | 4223 | if (r) |
77f3a5cd | 4224 | dev_err(adev->dev, "Could not create amdgpu device attr\n"); |
bd607166 | 4225 | |
76da73f0 LL |
4226 | r = devm_device_add_group(adev->dev, &amdgpu_board_attrs_group); |
4227 | if (r) | |
4228 | dev_err(adev->dev, | |
4229 | "Could not create amdgpu board attributes\n"); | |
4230 | ||
7957ec80 LL |
4231 | amdgpu_fru_sysfs_init(adev); |
4232 | ||
d155bef0 AB |
4233 | if (IS_ENABLED(CONFIG_PERF_EVENTS)) |
4234 | r = amdgpu_pmu_init(adev); | |
9c7c85f7 JK |
4235 | if (r) |
4236 | dev_err(adev->dev, "amdgpu_pmu_init failed\n"); | |
4237 | ||
c1dd4aa6 AG |
4238 | /* Have stored pci confspace at hand for restore in sudden PCI error */ |
4239 | if (amdgpu_device_cache_pci_state(adev->pdev)) | |
4240 | pci_restore_state(pdev); | |
4241 | ||
8c3dd61c KHF |
4242 | /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */ |
4243 | /* this will fail for cards that aren't VGA class devices, just | |
b8920e1e SS |
4244 | * ignore it |
4245 | */ | |
8c3dd61c | 4246 | if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) |
bf44e8ce | 4247 | vga_client_register(adev->pdev, amdgpu_device_vga_set_decode); |
8c3dd61c | 4248 | |
d37a3929 OC |
4249 | px = amdgpu_device_supports_px(ddev); |
4250 | ||
7b1c6263 | 4251 | if (px || (!dev_is_removable(&adev->pdev->dev) && |
d37a3929 | 4252 | apple_gmux_detect(NULL, NULL))) |
8c3dd61c KHF |
4253 | vga_switcheroo_register_client(adev->pdev, |
4254 | &amdgpu_switcheroo_ops, px); | |
d37a3929 OC |
4255 | |
4256 | if (px) | |
8c3dd61c | 4257 | vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain); |
8c3dd61c | 4258 | |
e3c1b071 | 4259 | if (adev->gmc.xgmi.pending_reset) |
4260 | queue_delayed_work(system_wq, &mgpu_info.delayed_reset_work, | |
4261 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
4262 | ||
4a74c38c PY |
4263 | amdgpu_device_check_iommu_direct_map(adev); |
4264 | ||
d38ceaf9 | 4265 | return 0; |
83ba126a | 4266 | |
970fd197 | 4267 | release_ras_con: |
38eecbe0 CL |
4268 | if (amdgpu_sriov_vf(adev)) |
4269 | amdgpu_virt_release_full_gpu(adev, true); | |
4270 | ||
4271 | /* failed in exclusive mode due to timeout */ | |
4272 | if (amdgpu_sriov_vf(adev) && | |
4273 | !amdgpu_sriov_runtime(adev) && | |
4274 | amdgpu_virt_mmio_blocked(adev) && | |
4275 | !amdgpu_virt_wait_reset(adev)) { | |
4276 | dev_err(adev->dev, "VF exclusive mode timeout\n"); | |
4277 | /* Don't send request since VF is inactive. */ | |
4278 | adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME; | |
4279 | adev->virt.ops = NULL; | |
4280 | r = -EAGAIN; | |
4281 | } | |
970fd197 SY |
4282 | amdgpu_release_ras_context(adev); |
4283 | ||
83ba126a | 4284 | failed: |
89041940 | 4285 | amdgpu_vf_error_trans_all(adev); |
8840a387 | 4286 | |
83ba126a | 4287 | return r; |
d38ceaf9 AD |
4288 | } |
4289 | ||
07775fc1 AG |
4290 | static void amdgpu_device_unmap_mmio(struct amdgpu_device *adev) |
4291 | { | |
62d5f9f7 | 4292 | |
07775fc1 AG |
4293 | /* Clear all CPU mappings pointing to this device */ |
4294 | unmap_mapping_range(adev->ddev.anon_inode->i_mapping, 0, 0, 1); | |
4295 | ||
4296 | /* Unmap all mapped bars - Doorbell, registers and VRAM */ | |
43c064db | 4297 | amdgpu_doorbell_fini(adev); |
07775fc1 AG |
4298 | |
4299 | iounmap(adev->rmmio); | |
4300 | adev->rmmio = NULL; | |
4301 | if (adev->mman.aper_base_kaddr) | |
4302 | iounmap(adev->mman.aper_base_kaddr); | |
4303 | adev->mman.aper_base_kaddr = NULL; | |
4304 | ||
4305 | /* Memory manager related */ | |
a0ba1279 | 4306 | if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) { |
07775fc1 AG |
4307 | arch_phys_wc_del(adev->gmc.vram_mtrr); |
4308 | arch_io_free_memtype_wc(adev->gmc.aper_base, adev->gmc.aper_size); | |
4309 | } | |
4310 | } | |
4311 | ||
d38ceaf9 | 4312 | /** |
bbe04dec | 4313 | * amdgpu_device_fini_hw - tear down the driver |
d38ceaf9 AD |
4314 | * |
4315 | * @adev: amdgpu_device pointer | |
4316 | * | |
4317 | * Tear down the driver info (all asics). | |
4318 | * Called at driver shutdown. | |
4319 | */ | |
72c8c97b | 4320 | void amdgpu_device_fini_hw(struct amdgpu_device *adev) |
d38ceaf9 | 4321 | { |
aac89168 | 4322 | dev_info(adev->dev, "amdgpu: finishing device.\n"); |
9f875167 | 4323 | flush_delayed_work(&adev->delayed_init_work); |
d0d13fe8 | 4324 | adev->shutdown = true; |
9f875167 | 4325 | |
752c683d ML |
4326 | /* make sure IB test finished before entering exclusive mode |
4327 | * to avoid preemption on IB test | |
b8920e1e | 4328 | */ |
519b8b76 | 4329 | if (amdgpu_sriov_vf(adev)) { |
752c683d | 4330 | amdgpu_virt_request_full_gpu(adev, false); |
519b8b76 BZ |
4331 | amdgpu_virt_fini_data_exchange(adev); |
4332 | } | |
752c683d | 4333 | |
e5b03032 ML |
4334 | /* disable all interrupts */ |
4335 | amdgpu_irq_disable_all(adev); | |
47fc644f | 4336 | if (adev->mode_info.mode_config_initialized) { |
1053b9c9 | 4337 | if (!drm_drv_uses_atomic_modeset(adev_to_drm(adev))) |
4a580877 | 4338 | drm_helper_force_disable_all(adev_to_drm(adev)); |
ff97cba8 | 4339 | else |
4a580877 | 4340 | drm_atomic_helper_shutdown(adev_to_drm(adev)); |
ff97cba8 | 4341 | } |
8d35a259 | 4342 | amdgpu_fence_driver_hw_fini(adev); |
72c8c97b | 4343 | |
cd3a8a59 | 4344 | if (adev->mman.initialized) |
9bff18d1 | 4345 | drain_workqueue(adev->mman.bdev.wq); |
98f56188 | 4346 | |
53e9d836 | 4347 | if (adev->pm.sysfs_initialized) |
7c868b59 | 4348 | amdgpu_pm_sysfs_fini(adev); |
72c8c97b AG |
4349 | if (adev->ucode_sysfs_en) |
4350 | amdgpu_ucode_sysfs_fini(adev); | |
4351 | sysfs_remove_files(&adev->dev->kobj, amdgpu_dev_attributes); | |
7957ec80 | 4352 | amdgpu_fru_sysfs_fini(adev); |
72c8c97b | 4353 | |
232d1d43 SY |
4354 | /* disable ras feature must before hw fini */ |
4355 | amdgpu_ras_pre_fini(adev); | |
4356 | ||
b7043800 AD |
4357 | amdgpu_ttm_set_buffer_funcs_status(adev, false); |
4358 | ||
e9669fb7 | 4359 | amdgpu_device_ip_fini_early(adev); |
d10d0daa | 4360 | |
a3848df6 YW |
4361 | amdgpu_irq_fini_hw(adev); |
4362 | ||
b6fd6e0f SK |
4363 | if (adev->mman.initialized) |
4364 | ttm_device_clear_dma_mappings(&adev->mman.bdev); | |
894c6890 | 4365 | |
d10d0daa | 4366 | amdgpu_gart_dummy_page_fini(adev); |
07775fc1 | 4367 | |
39934d3e VP |
4368 | if (drm_dev_is_unplugged(adev_to_drm(adev))) |
4369 | amdgpu_device_unmap_mmio(adev); | |
87172e89 | 4370 | |
72c8c97b AG |
4371 | } |
4372 | ||
4373 | void amdgpu_device_fini_sw(struct amdgpu_device *adev) | |
4374 | { | |
62d5f9f7 | 4375 | int idx; |
d37a3929 | 4376 | bool px; |
62d5f9f7 | 4377 | |
8d35a259 | 4378 | amdgpu_fence_driver_sw_fini(adev); |
a5c5d8d5 | 4379 | amdgpu_device_ip_fini(adev); |
b31d3063 | 4380 | amdgpu_ucode_release(&adev->firmware.gpu_info_fw); |
d38ceaf9 | 4381 | adev->accel_working = false; |
68ce8b24 | 4382 | dma_fence_put(rcu_dereference_protected(adev->gang_submit, true)); |
04442bf7 LL |
4383 | |
4384 | amdgpu_reset_fini(adev); | |
4385 | ||
d38ceaf9 | 4386 | /* free i2c buses */ |
4562236b HW |
4387 | if (!amdgpu_device_has_dc_support(adev)) |
4388 | amdgpu_i2c_fini(adev); | |
bfca0289 SL |
4389 | |
4390 | if (amdgpu_emu_mode != 1) | |
4391 | amdgpu_atombios_fini(adev); | |
4392 | ||
d38ceaf9 AD |
4393 | kfree(adev->bios); |
4394 | adev->bios = NULL; | |
d37a3929 | 4395 | |
8a2b5139 LL |
4396 | kfree(adev->fru_info); |
4397 | adev->fru_info = NULL; | |
4398 | ||
d37a3929 OC |
4399 | px = amdgpu_device_supports_px(adev_to_drm(adev)); |
4400 | ||
7b1c6263 | 4401 | if (px || (!dev_is_removable(&adev->pdev->dev) && |
d37a3929 | 4402 | apple_gmux_detect(NULL, NULL))) |
84c8b22e | 4403 | vga_switcheroo_unregister_client(adev->pdev); |
d37a3929 OC |
4404 | |
4405 | if (px) | |
83ba126a | 4406 | vga_switcheroo_fini_domain_pm_ops(adev->dev); |
d37a3929 | 4407 | |
38d6be81 | 4408 | if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) |
b8779475 | 4409 | vga_client_unregister(adev->pdev); |
e9bc1bf7 | 4410 | |
62d5f9f7 LS |
4411 | if (drm_dev_enter(adev_to_drm(adev), &idx)) { |
4412 | ||
4413 | iounmap(adev->rmmio); | |
4414 | adev->rmmio = NULL; | |
43c064db | 4415 | amdgpu_doorbell_fini(adev); |
62d5f9f7 LS |
4416 | drm_dev_exit(idx); |
4417 | } | |
4418 | ||
d155bef0 AB |
4419 | if (IS_ENABLED(CONFIG_PERF_EVENTS)) |
4420 | amdgpu_pmu_fini(adev); | |
72de33f8 | 4421 | if (adev->mman.discovery_bin) |
a190d1c7 | 4422 | amdgpu_discovery_fini(adev); |
72c8c97b | 4423 | |
cfbb6b00 AG |
4424 | amdgpu_reset_put_reset_domain(adev->reset_domain); |
4425 | adev->reset_domain = NULL; | |
4426 | ||
72c8c97b AG |
4427 | kfree(adev->pci_state); |
4428 | ||
d38ceaf9 AD |
4429 | } |
4430 | ||
58144d28 ND |
4431 | /** |
4432 | * amdgpu_device_evict_resources - evict device resources | |
4433 | * @adev: amdgpu device object | |
4434 | * | |
4435 | * Evicts all ttm device resources(vram BOs, gart table) from the lru list | |
4436 | * of the vram memory type. Mainly used for evicting device resources | |
4437 | * at suspend time. | |
4438 | * | |
4439 | */ | |
7863c155 | 4440 | static int amdgpu_device_evict_resources(struct amdgpu_device *adev) |
58144d28 | 4441 | { |
7863c155 ML |
4442 | int ret; |
4443 | ||
e53d9665 ML |
4444 | /* No need to evict vram on APUs for suspend to ram or s2idle */ |
4445 | if ((adev->in_s3 || adev->in_s0ix) && (adev->flags & AMD_IS_APU)) | |
7863c155 | 4446 | return 0; |
58144d28 | 4447 | |
7863c155 ML |
4448 | ret = amdgpu_ttm_evict_resources(adev, TTM_PL_VRAM); |
4449 | if (ret) | |
58144d28 | 4450 | DRM_WARN("evicting device resources failed\n"); |
7863c155 | 4451 | return ret; |
58144d28 | 4452 | } |
d38ceaf9 AD |
4453 | |
4454 | /* | |
4455 | * Suspend & resume. | |
4456 | */ | |
5095d541 ML |
4457 | /** |
4458 | * amdgpu_device_prepare - prepare for device suspend | |
4459 | * | |
4460 | * @dev: drm dev pointer | |
4461 | * | |
4462 | * Prepare to put the hw in the suspend state (all asics). | |
4463 | * Returns 0 for success or an error on failure. | |
4464 | * Called at driver suspend. | |
4465 | */ | |
4466 | int amdgpu_device_prepare(struct drm_device *dev) | |
4467 | { | |
4468 | struct amdgpu_device *adev = drm_to_adev(dev); | |
cb11ca32 | 4469 | int i, r; |
5095d541 ML |
4470 | |
4471 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
4472 | return 0; | |
4473 | ||
4474 | /* Evict the majority of BOs before starting suspend sequence */ | |
4475 | r = amdgpu_device_evict_resources(adev); | |
4476 | if (r) | |
4477 | return r; | |
4478 | ||
cb11ca32 ML |
4479 | for (i = 0; i < adev->num_ip_blocks; i++) { |
4480 | if (!adev->ip_blocks[i].status.valid) | |
4481 | continue; | |
4482 | if (!adev->ip_blocks[i].version->funcs->prepare_suspend) | |
4483 | continue; | |
4484 | r = adev->ip_blocks[i].version->funcs->prepare_suspend((void *)adev); | |
4485 | if (r) | |
4486 | return r; | |
4487 | } | |
4488 | ||
5095d541 ML |
4489 | return 0; |
4490 | } | |
4491 | ||
d38ceaf9 | 4492 | /** |
810ddc3a | 4493 | * amdgpu_device_suspend - initiate device suspend |
d38ceaf9 | 4494 | * |
87e3f136 | 4495 | * @dev: drm dev pointer |
87e3f136 | 4496 | * @fbcon : notify the fbdev of suspend |
d38ceaf9 AD |
4497 | * |
4498 | * Puts the hw in the suspend state (all asics). | |
4499 | * Returns 0 for success or an error on failure. | |
4500 | * Called at driver suspend. | |
4501 | */ | |
de185019 | 4502 | int amdgpu_device_suspend(struct drm_device *dev, bool fbcon) |
d38ceaf9 | 4503 | { |
a2e15b0e | 4504 | struct amdgpu_device *adev = drm_to_adev(dev); |
d7274ec7 | 4505 | int r = 0; |
d38ceaf9 | 4506 | |
d38ceaf9 AD |
4507 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) |
4508 | return 0; | |
4509 | ||
44779b43 | 4510 | adev->in_suspend = true; |
3fa8f89d | 4511 | |
d7274ec7 BZ |
4512 | if (amdgpu_sriov_vf(adev)) { |
4513 | amdgpu_virt_fini_data_exchange(adev); | |
4514 | r = amdgpu_virt_request_full_gpu(adev, false); | |
4515 | if (r) | |
4516 | return r; | |
4517 | } | |
4518 | ||
3fa8f89d S |
4519 | if (amdgpu_acpi_smart_shift_update(dev, AMDGPU_SS_DEV_D3)) |
4520 | DRM_WARN("smart shift update failed\n"); | |
4521 | ||
5f818173 | 4522 | if (fbcon) |
087451f3 | 4523 | drm_fb_helper_set_suspend_unlocked(adev_to_drm(adev)->fb_helper, true); |
5f818173 | 4524 | |
beff74bc | 4525 | cancel_delayed_work_sync(&adev->delayed_init_work); |
0dee7263 | 4526 | flush_delayed_work(&adev->gfx.gfx_off_delay_work); |
a5459475 | 4527 | |
5e6932fe | 4528 | amdgpu_ras_suspend(adev); |
4529 | ||
b7043800 AD |
4530 | amdgpu_ttm_set_buffer_funcs_status(adev, false); |
4531 | ||
2196927b | 4532 | amdgpu_device_ip_suspend_phase1(adev); |
fe1053b7 | 4533 | |
c004d44e | 4534 | if (!adev->in_s0ix) |
5d3a2d95 | 4535 | amdgpu_amdkfd_suspend(adev, adev->in_runpm); |
94fa5660 | 4536 | |
7863c155 ML |
4537 | r = amdgpu_device_evict_resources(adev); |
4538 | if (r) | |
4539 | return r; | |
d38ceaf9 | 4540 | |
8d35a259 | 4541 | amdgpu_fence_driver_hw_fini(adev); |
d38ceaf9 | 4542 | |
2196927b | 4543 | amdgpu_device_ip_suspend_phase2(adev); |
d38ceaf9 | 4544 | |
d7274ec7 BZ |
4545 | if (amdgpu_sriov_vf(adev)) |
4546 | amdgpu_virt_release_full_gpu(adev, false); | |
4547 | ||
d38ceaf9 AD |
4548 | return 0; |
4549 | } | |
4550 | ||
4551 | /** | |
810ddc3a | 4552 | * amdgpu_device_resume - initiate device resume |
d38ceaf9 | 4553 | * |
87e3f136 | 4554 | * @dev: drm dev pointer |
87e3f136 | 4555 | * @fbcon : notify the fbdev of resume |
d38ceaf9 AD |
4556 | * |
4557 | * Bring the hw back to operating state (all asics). | |
4558 | * Returns 0 for success or an error on failure. | |
4559 | * Called at driver resume. | |
4560 | */ | |
de185019 | 4561 | int amdgpu_device_resume(struct drm_device *dev, bool fbcon) |
d38ceaf9 | 4562 | { |
1348969a | 4563 | struct amdgpu_device *adev = drm_to_adev(dev); |
03161a6e | 4564 | int r = 0; |
d38ceaf9 | 4565 | |
d7274ec7 BZ |
4566 | if (amdgpu_sriov_vf(adev)) { |
4567 | r = amdgpu_virt_request_full_gpu(adev, true); | |
4568 | if (r) | |
4569 | return r; | |
4570 | } | |
4571 | ||
d38ceaf9 AD |
4572 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) |
4573 | return 0; | |
4574 | ||
62498733 | 4575 | if (adev->in_s0ix) |
bc143d8b | 4576 | amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D0Entry); |
628c36d7 | 4577 | |
d38ceaf9 | 4578 | /* post card */ |
39c640c0 | 4579 | if (amdgpu_device_need_post(adev)) { |
4d2997ab | 4580 | r = amdgpu_device_asic_init(adev); |
74b0b157 | 4581 | if (r) |
aac89168 | 4582 | dev_err(adev->dev, "amdgpu asic init failed\n"); |
74b0b157 | 4583 | } |
d38ceaf9 | 4584 | |
06ec9070 | 4585 | r = amdgpu_device_ip_resume(adev); |
d7274ec7 | 4586 | |
e6707218 | 4587 | if (r) { |
aac89168 | 4588 | dev_err(adev->dev, "amdgpu_device_ip_resume failed (%d).\n", r); |
3c22c1ea | 4589 | goto exit; |
e6707218 | 4590 | } |
8d35a259 | 4591 | amdgpu_fence_driver_hw_init(adev); |
5ceb54c6 | 4592 | |
c004d44e | 4593 | if (!adev->in_s0ix) { |
5d3a2d95 AD |
4594 | r = amdgpu_amdkfd_resume(adev, adev->in_runpm); |
4595 | if (r) | |
3c22c1ea | 4596 | goto exit; |
5d3a2d95 | 4597 | } |
756e6880 | 4598 | |
8ed79c40 TH |
4599 | r = amdgpu_device_ip_late_init(adev); |
4600 | if (r) | |
4601 | goto exit; | |
4602 | ||
4603 | queue_delayed_work(system_wq, &adev->delayed_init_work, | |
4604 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
3c22c1ea SF |
4605 | exit: |
4606 | if (amdgpu_sriov_vf(adev)) { | |
4607 | amdgpu_virt_init_data_exchange(adev); | |
4608 | amdgpu_virt_release_full_gpu(adev, true); | |
4609 | } | |
4610 | ||
4611 | if (r) | |
4612 | return r; | |
4613 | ||
96a5d8d4 | 4614 | /* Make sure IB tests flushed */ |
beff74bc | 4615 | flush_delayed_work(&adev->delayed_init_work); |
96a5d8d4 | 4616 | |
a2e15b0e | 4617 | if (fbcon) |
087451f3 | 4618 | drm_fb_helper_set_suspend_unlocked(adev_to_drm(adev)->fb_helper, false); |
d38ceaf9 | 4619 | |
5e6932fe | 4620 | amdgpu_ras_resume(adev); |
4621 | ||
d09ef243 AD |
4622 | if (adev->mode_info.num_crtc) { |
4623 | /* | |
4624 | * Most of the connector probing functions try to acquire runtime pm | |
4625 | * refs to ensure that the GPU is powered on when connector polling is | |
4626 | * performed. Since we're calling this from a runtime PM callback, | |
4627 | * trying to acquire rpm refs will cause us to deadlock. | |
4628 | * | |
4629 | * Since we're guaranteed to be holding the rpm lock, it's safe to | |
4630 | * temporarily disable the rpm helpers so this doesn't deadlock us. | |
4631 | */ | |
23a1a9e5 | 4632 | #ifdef CONFIG_PM |
d09ef243 | 4633 | dev->dev->power.disable_depth++; |
23a1a9e5 | 4634 | #endif |
d09ef243 AD |
4635 | if (!adev->dc_enabled) |
4636 | drm_helper_hpd_irq_event(dev); | |
4637 | else | |
4638 | drm_kms_helper_hotplug_event(dev); | |
23a1a9e5 | 4639 | #ifdef CONFIG_PM |
d09ef243 | 4640 | dev->dev->power.disable_depth--; |
23a1a9e5 | 4641 | #endif |
d09ef243 | 4642 | } |
44779b43 RZ |
4643 | adev->in_suspend = false; |
4644 | ||
dc907c9d JX |
4645 | if (adev->enable_mes) |
4646 | amdgpu_mes_self_test(adev); | |
4647 | ||
3fa8f89d S |
4648 | if (amdgpu_acpi_smart_shift_update(dev, AMDGPU_SS_DEV_D0)) |
4649 | DRM_WARN("smart shift update failed\n"); | |
4650 | ||
4d3b9ae5 | 4651 | return 0; |
d38ceaf9 AD |
4652 | } |
4653 | ||
e3ecdffa AD |
4654 | /** |
4655 | * amdgpu_device_ip_check_soft_reset - did soft reset succeed | |
4656 | * | |
4657 | * @adev: amdgpu_device pointer | |
4658 | * | |
4659 | * The list of all the hardware IPs that make up the asic is walked and | |
4660 | * the check_soft_reset callbacks are run. check_soft_reset determines | |
4661 | * if the asic is still hung or not. | |
4662 | * Returns true if any of the IPs are still in a hung state, false if not. | |
4663 | */ | |
06ec9070 | 4664 | static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev) |
63fbf42f CZ |
4665 | { |
4666 | int i; | |
4667 | bool asic_hang = false; | |
4668 | ||
f993d628 ML |
4669 | if (amdgpu_sriov_vf(adev)) |
4670 | return true; | |
4671 | ||
8bc04c29 AD |
4672 | if (amdgpu_asic_need_full_reset(adev)) |
4673 | return true; | |
4674 | ||
63fbf42f | 4675 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 4676 | if (!adev->ip_blocks[i].status.valid) |
63fbf42f | 4677 | continue; |
a1255107 AD |
4678 | if (adev->ip_blocks[i].version->funcs->check_soft_reset) |
4679 | adev->ip_blocks[i].status.hang = | |
4680 | adev->ip_blocks[i].version->funcs->check_soft_reset(adev); | |
4681 | if (adev->ip_blocks[i].status.hang) { | |
aac89168 | 4682 | dev_info(adev->dev, "IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name); |
63fbf42f CZ |
4683 | asic_hang = true; |
4684 | } | |
4685 | } | |
4686 | return asic_hang; | |
4687 | } | |
4688 | ||
e3ecdffa AD |
4689 | /** |
4690 | * amdgpu_device_ip_pre_soft_reset - prepare for soft reset | |
4691 | * | |
4692 | * @adev: amdgpu_device pointer | |
4693 | * | |
4694 | * The list of all the hardware IPs that make up the asic is walked and the | |
4695 | * pre_soft_reset callbacks are run if the block is hung. pre_soft_reset | |
4696 | * handles any IP specific hardware or software state changes that are | |
4697 | * necessary for a soft reset to succeed. | |
4698 | * Returns 0 on success, negative error code on failure. | |
4699 | */ | |
06ec9070 | 4700 | static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev) |
d31a501e CZ |
4701 | { |
4702 | int i, r = 0; | |
4703 | ||
4704 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 4705 | if (!adev->ip_blocks[i].status.valid) |
d31a501e | 4706 | continue; |
a1255107 AD |
4707 | if (adev->ip_blocks[i].status.hang && |
4708 | adev->ip_blocks[i].version->funcs->pre_soft_reset) { | |
4709 | r = adev->ip_blocks[i].version->funcs->pre_soft_reset(adev); | |
d31a501e CZ |
4710 | if (r) |
4711 | return r; | |
4712 | } | |
4713 | } | |
4714 | ||
4715 | return 0; | |
4716 | } | |
4717 | ||
e3ecdffa AD |
4718 | /** |
4719 | * amdgpu_device_ip_need_full_reset - check if a full asic reset is needed | |
4720 | * | |
4721 | * @adev: amdgpu_device pointer | |
4722 | * | |
4723 | * Some hardware IPs cannot be soft reset. If they are hung, a full gpu | |
4724 | * reset is necessary to recover. | |
4725 | * Returns true if a full asic reset is required, false if not. | |
4726 | */ | |
06ec9070 | 4727 | static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev) |
35d782fe | 4728 | { |
da146d3b AD |
4729 | int i; |
4730 | ||
8bc04c29 AD |
4731 | if (amdgpu_asic_need_full_reset(adev)) |
4732 | return true; | |
4733 | ||
da146d3b | 4734 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 4735 | if (!adev->ip_blocks[i].status.valid) |
da146d3b | 4736 | continue; |
a1255107 AD |
4737 | if ((adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) || |
4738 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) || | |
4739 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_ACP) || | |
98512bb8 KW |
4740 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) || |
4741 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) { | |
a1255107 | 4742 | if (adev->ip_blocks[i].status.hang) { |
aac89168 | 4743 | dev_info(adev->dev, "Some block need full reset!\n"); |
da146d3b AD |
4744 | return true; |
4745 | } | |
4746 | } | |
35d782fe CZ |
4747 | } |
4748 | return false; | |
4749 | } | |
4750 | ||
e3ecdffa AD |
4751 | /** |
4752 | * amdgpu_device_ip_soft_reset - do a soft reset | |
4753 | * | |
4754 | * @adev: amdgpu_device pointer | |
4755 | * | |
4756 | * The list of all the hardware IPs that make up the asic is walked and the | |
4757 | * soft_reset callbacks are run if the block is hung. soft_reset handles any | |
4758 | * IP specific hardware or software state changes that are necessary to soft | |
4759 | * reset the IP. | |
4760 | * Returns 0 on success, negative error code on failure. | |
4761 | */ | |
06ec9070 | 4762 | static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
4763 | { |
4764 | int i, r = 0; | |
4765 | ||
4766 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 4767 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 4768 | continue; |
a1255107 AD |
4769 | if (adev->ip_blocks[i].status.hang && |
4770 | adev->ip_blocks[i].version->funcs->soft_reset) { | |
4771 | r = adev->ip_blocks[i].version->funcs->soft_reset(adev); | |
35d782fe CZ |
4772 | if (r) |
4773 | return r; | |
4774 | } | |
4775 | } | |
4776 | ||
4777 | return 0; | |
4778 | } | |
4779 | ||
e3ecdffa AD |
4780 | /** |
4781 | * amdgpu_device_ip_post_soft_reset - clean up from soft reset | |
4782 | * | |
4783 | * @adev: amdgpu_device pointer | |
4784 | * | |
4785 | * The list of all the hardware IPs that make up the asic is walked and the | |
4786 | * post_soft_reset callbacks are run if the asic was hung. post_soft_reset | |
4787 | * handles any IP specific hardware or software state changes that are | |
4788 | * necessary after the IP has been soft reset. | |
4789 | * Returns 0 on success, negative error code on failure. | |
4790 | */ | |
06ec9070 | 4791 | static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
4792 | { |
4793 | int i, r = 0; | |
4794 | ||
4795 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 4796 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 4797 | continue; |
a1255107 AD |
4798 | if (adev->ip_blocks[i].status.hang && |
4799 | adev->ip_blocks[i].version->funcs->post_soft_reset) | |
4800 | r = adev->ip_blocks[i].version->funcs->post_soft_reset(adev); | |
35d782fe CZ |
4801 | if (r) |
4802 | return r; | |
4803 | } | |
4804 | ||
4805 | return 0; | |
4806 | } | |
4807 | ||
e3ecdffa | 4808 | /** |
c33adbc7 | 4809 | * amdgpu_device_recover_vram - Recover some VRAM contents |
e3ecdffa AD |
4810 | * |
4811 | * @adev: amdgpu_device pointer | |
4812 | * | |
4813 | * Restores the contents of VRAM buffers from the shadows in GTT. Used to | |
4814 | * restore things like GPUVM page tables after a GPU reset where | |
4815 | * the contents of VRAM might be lost. | |
403009bf CK |
4816 | * |
4817 | * Returns: | |
4818 | * 0 on success, negative error code on failure. | |
e3ecdffa | 4819 | */ |
c33adbc7 | 4820 | static int amdgpu_device_recover_vram(struct amdgpu_device *adev) |
c41d1cf6 | 4821 | { |
c41d1cf6 | 4822 | struct dma_fence *fence = NULL, *next = NULL; |
403009bf | 4823 | struct amdgpu_bo *shadow; |
e18aaea7 | 4824 | struct amdgpu_bo_vm *vmbo; |
403009bf | 4825 | long r = 1, tmo; |
c41d1cf6 ML |
4826 | |
4827 | if (amdgpu_sriov_runtime(adev)) | |
b045d3af | 4828 | tmo = msecs_to_jiffies(8000); |
c41d1cf6 ML |
4829 | else |
4830 | tmo = msecs_to_jiffies(100); | |
4831 | ||
aac89168 | 4832 | dev_info(adev->dev, "recover vram bo from shadow start\n"); |
c41d1cf6 | 4833 | mutex_lock(&adev->shadow_list_lock); |
e18aaea7 | 4834 | list_for_each_entry(vmbo, &adev->shadow_list, shadow_list) { |
4994d1f0 LC |
4835 | /* If vm is compute context or adev is APU, shadow will be NULL */ |
4836 | if (!vmbo->shadow) | |
4837 | continue; | |
4838 | shadow = vmbo->shadow; | |
4839 | ||
403009bf | 4840 | /* No need to recover an evicted BO */ |
d3116756 CK |
4841 | if (shadow->tbo.resource->mem_type != TTM_PL_TT || |
4842 | shadow->tbo.resource->start == AMDGPU_BO_INVALID_OFFSET || | |
4843 | shadow->parent->tbo.resource->mem_type != TTM_PL_VRAM) | |
403009bf CK |
4844 | continue; |
4845 | ||
4846 | r = amdgpu_bo_restore_shadow(shadow, &next); | |
4847 | if (r) | |
4848 | break; | |
4849 | ||
c41d1cf6 | 4850 | if (fence) { |
1712fb1a | 4851 | tmo = dma_fence_wait_timeout(fence, false, tmo); |
403009bf CK |
4852 | dma_fence_put(fence); |
4853 | fence = next; | |
1712fb1a | 4854 | if (tmo == 0) { |
4855 | r = -ETIMEDOUT; | |
c41d1cf6 | 4856 | break; |
1712fb1a | 4857 | } else if (tmo < 0) { |
4858 | r = tmo; | |
4859 | break; | |
4860 | } | |
403009bf CK |
4861 | } else { |
4862 | fence = next; | |
c41d1cf6 | 4863 | } |
c41d1cf6 ML |
4864 | } |
4865 | mutex_unlock(&adev->shadow_list_lock); | |
4866 | ||
403009bf CK |
4867 | if (fence) |
4868 | tmo = dma_fence_wait_timeout(fence, false, tmo); | |
c41d1cf6 ML |
4869 | dma_fence_put(fence); |
4870 | ||
1712fb1a | 4871 | if (r < 0 || tmo <= 0) { |
aac89168 | 4872 | dev_err(adev->dev, "recover vram bo from shadow failed, r is %ld, tmo is %ld\n", r, tmo); |
403009bf CK |
4873 | return -EIO; |
4874 | } | |
c41d1cf6 | 4875 | |
aac89168 | 4876 | dev_info(adev->dev, "recover vram bo from shadow done\n"); |
403009bf | 4877 | return 0; |
c41d1cf6 ML |
4878 | } |
4879 | ||
a90ad3c2 | 4880 | |
e3ecdffa | 4881 | /** |
06ec9070 | 4882 | * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf |
5740682e | 4883 | * |
982a820b | 4884 | * @adev: amdgpu_device pointer |
87e3f136 | 4885 | * @from_hypervisor: request from hypervisor |
5740682e ML |
4886 | * |
4887 | * do VF FLR and reinitialize Asic | |
3f48c681 | 4888 | * return 0 means succeeded otherwise failed |
e3ecdffa AD |
4889 | */ |
4890 | static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, | |
4891 | bool from_hypervisor) | |
5740682e ML |
4892 | { |
4893 | int r; | |
a5f67c93 | 4894 | struct amdgpu_hive_info *hive = NULL; |
7258fa31 | 4895 | int retry_limit = 0; |
5740682e | 4896 | |
7258fa31 | 4897 | retry: |
c004d44e | 4898 | amdgpu_amdkfd_pre_reset(adev); |
428890a3 | 4899 | |
5740682e ML |
4900 | if (from_hypervisor) |
4901 | r = amdgpu_virt_request_full_gpu(adev, true); | |
4902 | else | |
4903 | r = amdgpu_virt_reset_gpu(adev); | |
4904 | if (r) | |
4905 | return r; | |
f734b213 | 4906 | amdgpu_irq_gpu_reset_resume_helper(adev); |
a90ad3c2 | 4907 | |
83f24a8f HC |
4908 | /* some sw clean up VF needs to do before recover */ |
4909 | amdgpu_virt_post_reset(adev); | |
4910 | ||
a90ad3c2 | 4911 | /* Resume IP prior to SMC */ |
06ec9070 | 4912 | r = amdgpu_device_ip_reinit_early_sriov(adev); |
5740682e ML |
4913 | if (r) |
4914 | goto error; | |
a90ad3c2 | 4915 | |
c9ffa427 | 4916 | amdgpu_virt_init_data_exchange(adev); |
a90ad3c2 | 4917 | |
7a3e0bb2 RZ |
4918 | r = amdgpu_device_fw_loading(adev); |
4919 | if (r) | |
4920 | return r; | |
4921 | ||
a90ad3c2 | 4922 | /* now we are okay to resume SMC/CP/SDMA */ |
06ec9070 | 4923 | r = amdgpu_device_ip_reinit_late_sriov(adev); |
5740682e ML |
4924 | if (r) |
4925 | goto error; | |
a90ad3c2 | 4926 | |
a5f67c93 ZL |
4927 | hive = amdgpu_get_xgmi_hive(adev); |
4928 | /* Update PSP FW topology after reset */ | |
4929 | if (hive && adev->gmc.xgmi.num_physical_nodes > 1) | |
4930 | r = amdgpu_xgmi_update_topology(hive, adev); | |
4931 | ||
4932 | if (hive) | |
4933 | amdgpu_put_xgmi_hive(hive); | |
4934 | ||
4935 | if (!r) { | |
a5f67c93 | 4936 | r = amdgpu_ib_ring_tests(adev); |
9c12f5cd | 4937 | |
c004d44e | 4938 | amdgpu_amdkfd_post_reset(adev); |
a5f67c93 | 4939 | } |
a90ad3c2 | 4940 | |
abc34253 | 4941 | error: |
c41d1cf6 | 4942 | if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) { |
e3526257 | 4943 | amdgpu_inc_vram_lost(adev); |
c33adbc7 | 4944 | r = amdgpu_device_recover_vram(adev); |
a90ad3c2 | 4945 | } |
437f3e0b | 4946 | amdgpu_virt_release_full_gpu(adev, true); |
a90ad3c2 | 4947 | |
7258fa31 SK |
4948 | if (AMDGPU_RETRY_SRIOV_RESET(r)) { |
4949 | if (retry_limit < AMDGPU_MAX_RETRY_LIMIT) { | |
4950 | retry_limit++; | |
4951 | goto retry; | |
4952 | } else | |
4953 | DRM_ERROR("GPU reset retry is beyond the retry limit\n"); | |
4954 | } | |
4955 | ||
a90ad3c2 ML |
4956 | return r; |
4957 | } | |
4958 | ||
9a1cddd6 | 4959 | /** |
4960 | * amdgpu_device_has_job_running - check if there is any job in mirror list | |
4961 | * | |
982a820b | 4962 | * @adev: amdgpu_device pointer |
9a1cddd6 | 4963 | * |
4964 | * check if there is any job in mirror list | |
4965 | */ | |
4966 | bool amdgpu_device_has_job_running(struct amdgpu_device *adev) | |
4967 | { | |
4968 | int i; | |
4969 | struct drm_sched_job *job; | |
4970 | ||
4971 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
4972 | struct amdgpu_ring *ring = adev->rings[i]; | |
4973 | ||
4974 | if (!ring || !ring->sched.thread) | |
4975 | continue; | |
4976 | ||
4977 | spin_lock(&ring->sched.job_list_lock); | |
6efa4b46 LT |
4978 | job = list_first_entry_or_null(&ring->sched.pending_list, |
4979 | struct drm_sched_job, list); | |
9a1cddd6 | 4980 | spin_unlock(&ring->sched.job_list_lock); |
4981 | if (job) | |
4982 | return true; | |
4983 | } | |
4984 | return false; | |
4985 | } | |
4986 | ||
12938fad CK |
4987 | /** |
4988 | * amdgpu_device_should_recover_gpu - check if we should try GPU recovery | |
4989 | * | |
982a820b | 4990 | * @adev: amdgpu_device pointer |
12938fad CK |
4991 | * |
4992 | * Check amdgpu_gpu_recovery and SRIOV status to see if we should try to recover | |
4993 | * a hung GPU. | |
4994 | */ | |
4995 | bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev) | |
4996 | { | |
12938fad | 4997 | |
3ba7b418 AG |
4998 | if (amdgpu_gpu_recovery == 0) |
4999 | goto disabled; | |
5000 | ||
1a11a65d YC |
5001 | /* Skip soft reset check in fatal error mode */ |
5002 | if (!amdgpu_ras_is_poison_mode_supported(adev)) | |
5003 | return true; | |
5004 | ||
3ba7b418 AG |
5005 | if (amdgpu_sriov_vf(adev)) |
5006 | return true; | |
5007 | ||
5008 | if (amdgpu_gpu_recovery == -1) { | |
5009 | switch (adev->asic_type) { | |
b3523c45 AD |
5010 | #ifdef CONFIG_DRM_AMDGPU_SI |
5011 | case CHIP_VERDE: | |
5012 | case CHIP_TAHITI: | |
5013 | case CHIP_PITCAIRN: | |
5014 | case CHIP_OLAND: | |
5015 | case CHIP_HAINAN: | |
5016 | #endif | |
5017 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
5018 | case CHIP_KAVERI: | |
5019 | case CHIP_KABINI: | |
5020 | case CHIP_MULLINS: | |
5021 | #endif | |
5022 | case CHIP_CARRIZO: | |
5023 | case CHIP_STONEY: | |
5024 | case CHIP_CYAN_SKILLFISH: | |
3ba7b418 | 5025 | goto disabled; |
b3523c45 AD |
5026 | default: |
5027 | break; | |
3ba7b418 | 5028 | } |
12938fad CK |
5029 | } |
5030 | ||
5031 | return true; | |
3ba7b418 AG |
5032 | |
5033 | disabled: | |
aac89168 | 5034 | dev_info(adev->dev, "GPU recovery disabled.\n"); |
3ba7b418 | 5035 | return false; |
12938fad CK |
5036 | } |
5037 | ||
5c03e584 FX |
5038 | int amdgpu_device_mode1_reset(struct amdgpu_device *adev) |
5039 | { | |
47fc644f SS |
5040 | u32 i; |
5041 | int ret = 0; | |
5c03e584 | 5042 | |
47fc644f | 5043 | amdgpu_atombios_scratch_regs_engine_hung(adev, true); |
5c03e584 | 5044 | |
47fc644f | 5045 | dev_info(adev->dev, "GPU mode1 reset\n"); |
5c03e584 | 5046 | |
47fc644f SS |
5047 | /* disable BM */ |
5048 | pci_clear_master(adev->pdev); | |
5c03e584 | 5049 | |
47fc644f | 5050 | amdgpu_device_cache_pci_state(adev->pdev); |
5c03e584 | 5051 | |
47fc644f SS |
5052 | if (amdgpu_dpm_is_mode1_reset_supported(adev)) { |
5053 | dev_info(adev->dev, "GPU smu mode1 reset\n"); | |
5054 | ret = amdgpu_dpm_mode1_reset(adev); | |
5055 | } else { | |
5056 | dev_info(adev->dev, "GPU psp mode1 reset\n"); | |
5057 | ret = psp_gpu_reset(adev); | |
5058 | } | |
5c03e584 | 5059 | |
47fc644f | 5060 | if (ret) |
2c0f880a | 5061 | goto mode1_reset_failed; |
5c03e584 | 5062 | |
47fc644f | 5063 | amdgpu_device_load_pci_state(adev->pdev); |
15c5c5f5 LL |
5064 | ret = amdgpu_psp_wait_for_bootloader(adev); |
5065 | if (ret) | |
2c0f880a | 5066 | goto mode1_reset_failed; |
5c03e584 | 5067 | |
47fc644f SS |
5068 | /* wait for asic to come out of reset */ |
5069 | for (i = 0; i < adev->usec_timeout; i++) { | |
5070 | u32 memsize = adev->nbio.funcs->get_memsize(adev); | |
5c03e584 | 5071 | |
47fc644f SS |
5072 | if (memsize != 0xffffffff) |
5073 | break; | |
5074 | udelay(1); | |
5075 | } | |
5c03e584 | 5076 | |
2c0f880a HZ |
5077 | if (i >= adev->usec_timeout) { |
5078 | ret = -ETIMEDOUT; | |
5079 | goto mode1_reset_failed; | |
5080 | } | |
5081 | ||
47fc644f | 5082 | amdgpu_atombios_scratch_regs_engine_hung(adev, false); |
15c5c5f5 | 5083 | |
2c0f880a HZ |
5084 | return 0; |
5085 | ||
5086 | mode1_reset_failed: | |
5087 | dev_err(adev->dev, "GPU mode1 reset failed\n"); | |
47fc644f | 5088 | return ret; |
5c03e584 | 5089 | } |
5c6dd71e | 5090 | |
e3c1b071 | 5091 | int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, |
04442bf7 | 5092 | struct amdgpu_reset_context *reset_context) |
26bc5340 | 5093 | { |
5c1e6fa4 | 5094 | int i, r = 0; |
04442bf7 LL |
5095 | struct amdgpu_job *job = NULL; |
5096 | bool need_full_reset = | |
5097 | test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); | |
5098 | ||
5099 | if (reset_context->reset_req_dev == adev) | |
5100 | job = reset_context->job; | |
71182665 | 5101 | |
b602ca5f TZ |
5102 | if (amdgpu_sriov_vf(adev)) { |
5103 | /* stop the data exchange thread */ | |
5104 | amdgpu_virt_fini_data_exchange(adev); | |
5105 | } | |
5106 | ||
9e225fb9 AG |
5107 | amdgpu_fence_driver_isr_toggle(adev, true); |
5108 | ||
71182665 | 5109 | /* block all schedulers and reset given job's ring */ |
0875dc9e CZ |
5110 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
5111 | struct amdgpu_ring *ring = adev->rings[i]; | |
5112 | ||
51687759 | 5113 | if (!ring || !ring->sched.thread) |
0875dc9e | 5114 | continue; |
5740682e | 5115 | |
b8920e1e SS |
5116 | /* Clear job fence from fence drv to avoid force_completion |
5117 | * leave NULL and vm flush fence in fence drv | |
5118 | */ | |
5c1e6fa4 | 5119 | amdgpu_fence_driver_clear_job_fences(ring); |
c530b02f | 5120 | |
2f9d4084 ML |
5121 | /* after all hw jobs are reset, hw fence is meaningless, so force_completion */ |
5122 | amdgpu_fence_driver_force_completion(ring); | |
0875dc9e | 5123 | } |
d38ceaf9 | 5124 | |
9e225fb9 AG |
5125 | amdgpu_fence_driver_isr_toggle(adev, false); |
5126 | ||
ff99849b | 5127 | if (job && job->vm) |
222b5f04 AG |
5128 | drm_sched_increase_karma(&job->base); |
5129 | ||
04442bf7 | 5130 | r = amdgpu_reset_prepare_hwcontext(adev, reset_context); |
404b277b | 5131 | /* If reset handler not implemented, continue; otherwise return */ |
b8920e1e | 5132 | if (r == -EOPNOTSUPP) |
404b277b LL |
5133 | r = 0; |
5134 | else | |
04442bf7 LL |
5135 | return r; |
5136 | ||
1d721ed6 | 5137 | /* Don't suspend on bare metal if we are not going to HW reset the ASIC */ |
26bc5340 AG |
5138 | if (!amdgpu_sriov_vf(adev)) { |
5139 | ||
5140 | if (!need_full_reset) | |
5141 | need_full_reset = amdgpu_device_ip_need_full_reset(adev); | |
5142 | ||
360cd081 LG |
5143 | if (!need_full_reset && amdgpu_gpu_recovery && |
5144 | amdgpu_device_ip_check_soft_reset(adev)) { | |
26bc5340 AG |
5145 | amdgpu_device_ip_pre_soft_reset(adev); |
5146 | r = amdgpu_device_ip_soft_reset(adev); | |
5147 | amdgpu_device_ip_post_soft_reset(adev); | |
5148 | if (r || amdgpu_device_ip_check_soft_reset(adev)) { | |
aac89168 | 5149 | dev_info(adev->dev, "soft reset failed, will fallback to full reset!\n"); |
26bc5340 AG |
5150 | need_full_reset = true; |
5151 | } | |
5152 | } | |
5153 | ||
5154 | if (need_full_reset) | |
5155 | r = amdgpu_device_ip_suspend(adev); | |
04442bf7 LL |
5156 | if (need_full_reset) |
5157 | set_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); | |
5158 | else | |
5159 | clear_bit(AMDGPU_NEED_FULL_RESET, | |
5160 | &reset_context->flags); | |
26bc5340 AG |
5161 | } |
5162 | ||
5163 | return r; | |
5164 | } | |
5165 | ||
15fd09a0 SA |
5166 | static int amdgpu_reset_reg_dumps(struct amdgpu_device *adev) |
5167 | { | |
15fd09a0 SA |
5168 | int i; |
5169 | ||
38a15ad9 | 5170 | lockdep_assert_held(&adev->reset_domain->sem); |
15fd09a0 | 5171 | |
2d6a2a28 AA |
5172 | for (i = 0; i < adev->reset_info.num_regs; i++) { |
5173 | adev->reset_info.reset_dump_reg_value[i] = | |
5174 | RREG32(adev->reset_info.reset_dump_reg_list[i]); | |
5175 | ||
5176 | trace_amdgpu_reset_reg_dumps(adev->reset_info.reset_dump_reg_list[i], | |
5177 | adev->reset_info.reset_dump_reg_value[i]); | |
15fd09a0 SA |
5178 | } |
5179 | ||
5180 | return 0; | |
5181 | } | |
5182 | ||
04442bf7 LL |
5183 | int amdgpu_do_asic_reset(struct list_head *device_list_handle, |
5184 | struct amdgpu_reset_context *reset_context) | |
26bc5340 AG |
5185 | { |
5186 | struct amdgpu_device *tmp_adev = NULL; | |
04442bf7 | 5187 | bool need_full_reset, skip_hw_reset, vram_lost = false; |
26bc5340 | 5188 | int r = 0; |
f5c7e779 | 5189 | bool gpu_reset_for_dev_remove = 0; |
26bc5340 | 5190 | |
04442bf7 LL |
5191 | /* Try reset handler method first */ |
5192 | tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, | |
5193 | reset_list); | |
15fd09a0 | 5194 | amdgpu_reset_reg_dumps(tmp_adev); |
0a83bb35 LL |
5195 | |
5196 | reset_context->reset_device_list = device_list_handle; | |
04442bf7 | 5197 | r = amdgpu_reset_perform_reset(tmp_adev, reset_context); |
404b277b | 5198 | /* If reset handler not implemented, continue; otherwise return */ |
b8920e1e | 5199 | if (r == -EOPNOTSUPP) |
404b277b LL |
5200 | r = 0; |
5201 | else | |
04442bf7 LL |
5202 | return r; |
5203 | ||
5204 | /* Reset handler not implemented, use the default method */ | |
5205 | need_full_reset = | |
5206 | test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); | |
5207 | skip_hw_reset = test_bit(AMDGPU_SKIP_HW_RESET, &reset_context->flags); | |
5208 | ||
f5c7e779 YC |
5209 | gpu_reset_for_dev_remove = |
5210 | test_bit(AMDGPU_RESET_FOR_DEVICE_REMOVE, &reset_context->flags) && | |
5211 | test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); | |
5212 | ||
26bc5340 | 5213 | /* |
655ce9cb | 5214 | * ASIC reset has to be done on all XGMI hive nodes ASAP |
26bc5340 AG |
5215 | * to allow proper links negotiation in FW (within 1 sec) |
5216 | */ | |
7ac71382 | 5217 | if (!skip_hw_reset && need_full_reset) { |
655ce9cb | 5218 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
041a62bc | 5219 | /* For XGMI run all resets in parallel to speed up the process */ |
d4535e2c | 5220 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { |
e3c1b071 | 5221 | tmp_adev->gmc.xgmi.pending_reset = false; |
c96cf282 | 5222 | if (!queue_work(system_unbound_wq, &tmp_adev->xgmi_reset_work)) |
d4535e2c AG |
5223 | r = -EALREADY; |
5224 | } else | |
5225 | r = amdgpu_asic_reset(tmp_adev); | |
d4535e2c | 5226 | |
041a62bc | 5227 | if (r) { |
aac89168 | 5228 | dev_err(tmp_adev->dev, "ASIC reset failed with error, %d for drm dev, %s", |
4a580877 | 5229 | r, adev_to_drm(tmp_adev)->unique); |
19349072 | 5230 | goto out; |
ce316fa5 LM |
5231 | } |
5232 | } | |
5233 | ||
041a62bc AG |
5234 | /* For XGMI wait for all resets to complete before proceed */ |
5235 | if (!r) { | |
655ce9cb | 5236 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
ce316fa5 LM |
5237 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { |
5238 | flush_work(&tmp_adev->xgmi_reset_work); | |
5239 | r = tmp_adev->asic_reset_res; | |
5240 | if (r) | |
5241 | break; | |
ce316fa5 LM |
5242 | } |
5243 | } | |
5244 | } | |
ce316fa5 | 5245 | } |
26bc5340 | 5246 | |
43c4d576 | 5247 | if (!r && amdgpu_ras_intr_triggered()) { |
655ce9cb | 5248 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
21226f02 | 5249 | amdgpu_ras_reset_error_count(tmp_adev, AMDGPU_RAS_BLOCK__MMHUB); |
43c4d576 JC |
5250 | } |
5251 | ||
00eaa571 | 5252 | amdgpu_ras_intr_cleared(); |
43c4d576 | 5253 | } |
00eaa571 | 5254 | |
f5c7e779 YC |
5255 | /* Since the mode1 reset affects base ip blocks, the |
5256 | * phase1 ip blocks need to be resumed. Otherwise there | |
5257 | * will be a BIOS signature error and the psp bootloader | |
5258 | * can't load kdb on the next amdgpu install. | |
5259 | */ | |
5260 | if (gpu_reset_for_dev_remove) { | |
5261 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) | |
5262 | amdgpu_device_ip_resume_phase1(tmp_adev); | |
5263 | ||
5264 | goto end; | |
5265 | } | |
5266 | ||
655ce9cb | 5267 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
26bc5340 AG |
5268 | if (need_full_reset) { |
5269 | /* post card */ | |
e3c1b071 | 5270 | r = amdgpu_device_asic_init(tmp_adev); |
5271 | if (r) { | |
aac89168 | 5272 | dev_warn(tmp_adev->dev, "asic atom init failed!"); |
e3c1b071 | 5273 | } else { |
26bc5340 | 5274 | dev_info(tmp_adev->dev, "GPU reset succeeded, trying to resume\n"); |
9cec53c1 | 5275 | |
26bc5340 AG |
5276 | r = amdgpu_device_ip_resume_phase1(tmp_adev); |
5277 | if (r) | |
5278 | goto out; | |
5279 | ||
5280 | vram_lost = amdgpu_device_check_vram_lost(tmp_adev); | |
a7691785 AA |
5281 | |
5282 | amdgpu_coredump(tmp_adev, vram_lost, reset_context); | |
5283 | ||
26bc5340 | 5284 | if (vram_lost) { |
77e7f829 | 5285 | DRM_INFO("VRAM is lost due to GPU reset!\n"); |
e3526257 | 5286 | amdgpu_inc_vram_lost(tmp_adev); |
26bc5340 AG |
5287 | } |
5288 | ||
26bc5340 AG |
5289 | r = amdgpu_device_fw_loading(tmp_adev); |
5290 | if (r) | |
5291 | return r; | |
5292 | ||
c45e38f2 LL |
5293 | r = amdgpu_xcp_restore_partition_mode( |
5294 | tmp_adev->xcp_mgr); | |
5295 | if (r) | |
5296 | goto out; | |
5297 | ||
26bc5340 AG |
5298 | r = amdgpu_device_ip_resume_phase2(tmp_adev); |
5299 | if (r) | |
5300 | goto out; | |
5301 | ||
b7043800 AD |
5302 | if (tmp_adev->mman.buffer_funcs_ring->sched.ready) |
5303 | amdgpu_ttm_set_buffer_funcs_status(tmp_adev, true); | |
5304 | ||
26bc5340 AG |
5305 | if (vram_lost) |
5306 | amdgpu_device_fill_reset_magic(tmp_adev); | |
5307 | ||
fdafb359 EQ |
5308 | /* |
5309 | * Add this ASIC as tracked as reset was already | |
5310 | * complete successfully. | |
5311 | */ | |
5312 | amdgpu_register_gpu_instance(tmp_adev); | |
5313 | ||
04442bf7 LL |
5314 | if (!reset_context->hive && |
5315 | tmp_adev->gmc.xgmi.num_physical_nodes > 1) | |
e3c1b071 | 5316 | amdgpu_xgmi_add_device(tmp_adev); |
5317 | ||
7c04ca50 | 5318 | r = amdgpu_device_ip_late_init(tmp_adev); |
5319 | if (r) | |
5320 | goto out; | |
5321 | ||
087451f3 | 5322 | drm_fb_helper_set_suspend_unlocked(adev_to_drm(tmp_adev)->fb_helper, false); |
565d1941 | 5323 | |
e8fbaf03 GC |
5324 | /* |
5325 | * The GPU enters bad state once faulty pages | |
5326 | * by ECC has reached the threshold, and ras | |
5327 | * recovery is scheduled next. So add one check | |
5328 | * here to break recovery if it indeed exceeds | |
5329 | * bad page threshold, and remind user to | |
5330 | * retire this GPU or setting one bigger | |
5331 | * bad_page_threshold value to fix this once | |
5332 | * probing driver again. | |
5333 | */ | |
11003c68 | 5334 | if (!amdgpu_ras_eeprom_check_err_threshold(tmp_adev)) { |
e8fbaf03 GC |
5335 | /* must succeed. */ |
5336 | amdgpu_ras_resume(tmp_adev); | |
5337 | } else { | |
5338 | r = -EINVAL; | |
5339 | goto out; | |
5340 | } | |
e79a04d5 | 5341 | |
26bc5340 | 5342 | /* Update PSP FW topology after reset */ |
04442bf7 LL |
5343 | if (reset_context->hive && |
5344 | tmp_adev->gmc.xgmi.num_physical_nodes > 1) | |
5345 | r = amdgpu_xgmi_update_topology( | |
5346 | reset_context->hive, tmp_adev); | |
26bc5340 AG |
5347 | } |
5348 | } | |
5349 | ||
26bc5340 AG |
5350 | out: |
5351 | if (!r) { | |
5352 | amdgpu_irq_gpu_reset_resume_helper(tmp_adev); | |
5353 | r = amdgpu_ib_ring_tests(tmp_adev); | |
5354 | if (r) { | |
5355 | dev_err(tmp_adev->dev, "ib ring test failed (%d).\n", r); | |
26bc5340 AG |
5356 | need_full_reset = true; |
5357 | r = -EAGAIN; | |
5358 | goto end; | |
5359 | } | |
5360 | } | |
5361 | ||
5362 | if (!r) | |
5363 | r = amdgpu_device_recover_vram(tmp_adev); | |
5364 | else | |
5365 | tmp_adev->asic_reset_res = r; | |
5366 | } | |
5367 | ||
5368 | end: | |
04442bf7 LL |
5369 | if (need_full_reset) |
5370 | set_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); | |
5371 | else | |
5372 | clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); | |
26bc5340 AG |
5373 | return r; |
5374 | } | |
5375 | ||
e923be99 | 5376 | static void amdgpu_device_set_mp1_state(struct amdgpu_device *adev) |
26bc5340 | 5377 | { |
5740682e | 5378 | |
a3a09142 AD |
5379 | switch (amdgpu_asic_reset_method(adev)) { |
5380 | case AMD_RESET_METHOD_MODE1: | |
5381 | adev->mp1_state = PP_MP1_STATE_SHUTDOWN; | |
5382 | break; | |
5383 | case AMD_RESET_METHOD_MODE2: | |
5384 | adev->mp1_state = PP_MP1_STATE_RESET; | |
5385 | break; | |
5386 | default: | |
5387 | adev->mp1_state = PP_MP1_STATE_NONE; | |
5388 | break; | |
5389 | } | |
26bc5340 | 5390 | } |
d38ceaf9 | 5391 | |
e923be99 | 5392 | static void amdgpu_device_unset_mp1_state(struct amdgpu_device *adev) |
26bc5340 | 5393 | { |
89041940 | 5394 | amdgpu_vf_error_trans_all(adev); |
a3a09142 | 5395 | adev->mp1_state = PP_MP1_STATE_NONE; |
91fb309d HC |
5396 | } |
5397 | ||
3f12acc8 EQ |
5398 | static void amdgpu_device_resume_display_audio(struct amdgpu_device *adev) |
5399 | { | |
5400 | struct pci_dev *p = NULL; | |
5401 | ||
5402 | p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus), | |
5403 | adev->pdev->bus->number, 1); | |
5404 | if (p) { | |
5405 | pm_runtime_enable(&(p->dev)); | |
5406 | pm_runtime_resume(&(p->dev)); | |
5407 | } | |
b85e285e YY |
5408 | |
5409 | pci_dev_put(p); | |
3f12acc8 EQ |
5410 | } |
5411 | ||
5412 | static int amdgpu_device_suspend_display_audio(struct amdgpu_device *adev) | |
5413 | { | |
5414 | enum amd_reset_method reset_method; | |
5415 | struct pci_dev *p = NULL; | |
5416 | u64 expires; | |
5417 | ||
5418 | /* | |
5419 | * For now, only BACO and mode1 reset are confirmed | |
5420 | * to suffer the audio issue without proper suspended. | |
5421 | */ | |
5422 | reset_method = amdgpu_asic_reset_method(adev); | |
5423 | if ((reset_method != AMD_RESET_METHOD_BACO) && | |
5424 | (reset_method != AMD_RESET_METHOD_MODE1)) | |
5425 | return -EINVAL; | |
5426 | ||
5427 | p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus), | |
5428 | adev->pdev->bus->number, 1); | |
5429 | if (!p) | |
5430 | return -ENODEV; | |
5431 | ||
5432 | expires = pm_runtime_autosuspend_expiration(&(p->dev)); | |
5433 | if (!expires) | |
5434 | /* | |
5435 | * If we cannot get the audio device autosuspend delay, | |
5436 | * a fixed 4S interval will be used. Considering 3S is | |
5437 | * the audio controller default autosuspend delay setting. | |
5438 | * 4S used here is guaranteed to cover that. | |
5439 | */ | |
54b7feb9 | 5440 | expires = ktime_get_mono_fast_ns() + NSEC_PER_SEC * 4ULL; |
3f12acc8 EQ |
5441 | |
5442 | while (!pm_runtime_status_suspended(&(p->dev))) { | |
5443 | if (!pm_runtime_suspend(&(p->dev))) | |
5444 | break; | |
5445 | ||
5446 | if (expires < ktime_get_mono_fast_ns()) { | |
5447 | dev_warn(adev->dev, "failed to suspend display audio\n"); | |
b85e285e | 5448 | pci_dev_put(p); |
3f12acc8 EQ |
5449 | /* TODO: abort the succeeding gpu reset? */ |
5450 | return -ETIMEDOUT; | |
5451 | } | |
5452 | } | |
5453 | ||
5454 | pm_runtime_disable(&(p->dev)); | |
5455 | ||
b85e285e | 5456 | pci_dev_put(p); |
3f12acc8 EQ |
5457 | return 0; |
5458 | } | |
5459 | ||
d193b12b | 5460 | static inline void amdgpu_device_stop_pending_resets(struct amdgpu_device *adev) |
247c7b0d AG |
5461 | { |
5462 | struct amdgpu_ras *con = amdgpu_ras_get_context(adev); | |
5463 | ||
5464 | #if defined(CONFIG_DEBUG_FS) | |
5465 | if (!amdgpu_sriov_vf(adev)) | |
5466 | cancel_work(&adev->reset_work); | |
5467 | #endif | |
5468 | ||
5469 | if (adev->kfd.dev) | |
5470 | cancel_work(&adev->kfd.reset_work); | |
5471 | ||
5472 | if (amdgpu_sriov_vf(adev)) | |
5473 | cancel_work(&adev->virt.flr_work); | |
5474 | ||
5475 | if (con && adev->ras_enabled) | |
5476 | cancel_work(&con->recovery_work); | |
5477 | ||
5478 | } | |
5479 | ||
26bc5340 | 5480 | /** |
6e9c65f7 | 5481 | * amdgpu_device_gpu_recover - reset the asic and recover scheduler |
26bc5340 | 5482 | * |
982a820b | 5483 | * @adev: amdgpu_device pointer |
26bc5340 | 5484 | * @job: which job trigger hang |
80bd2de1 | 5485 | * @reset_context: amdgpu reset context pointer |
26bc5340 AG |
5486 | * |
5487 | * Attempt to reset the GPU if it has hung (all asics). | |
5488 | * Attempt to do soft-reset or full-reset and reinitialize Asic | |
5489 | * Returns 0 for success or an error on failure. | |
5490 | */ | |
5491 | ||
cf727044 | 5492 | int amdgpu_device_gpu_recover(struct amdgpu_device *adev, |
f1549c09 LG |
5493 | struct amdgpu_job *job, |
5494 | struct amdgpu_reset_context *reset_context) | |
26bc5340 | 5495 | { |
1d721ed6 | 5496 | struct list_head device_list, *device_list_handle = NULL; |
7dd8c205 | 5497 | bool job_signaled = false; |
26bc5340 | 5498 | struct amdgpu_hive_info *hive = NULL; |
26bc5340 | 5499 | struct amdgpu_device *tmp_adev = NULL; |
1d721ed6 | 5500 | int i, r = 0; |
bb5c7235 | 5501 | bool need_emergency_restart = false; |
3f12acc8 | 5502 | bool audio_suspended = false; |
f5c7e779 YC |
5503 | bool gpu_reset_for_dev_remove = false; |
5504 | ||
5505 | gpu_reset_for_dev_remove = | |
5506 | test_bit(AMDGPU_RESET_FOR_DEVICE_REMOVE, &reset_context->flags) && | |
5507 | test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); | |
26bc5340 | 5508 | |
6e3cd2a9 | 5509 | /* |
bb5c7235 WS |
5510 | * Special case: RAS triggered and full reset isn't supported |
5511 | */ | |
5512 | need_emergency_restart = amdgpu_ras_need_emergency_restart(adev); | |
5513 | ||
d5ea093e AG |
5514 | /* |
5515 | * Flush RAM to disk so that after reboot | |
5516 | * the user can read log and see why the system rebooted. | |
5517 | */ | |
80285ae1 SY |
5518 | if (need_emergency_restart && amdgpu_ras_get_context(adev) && |
5519 | amdgpu_ras_get_context(adev)->reboot) { | |
d5ea093e AG |
5520 | DRM_WARN("Emergency reboot."); |
5521 | ||
5522 | ksys_sync_helper(); | |
5523 | emergency_restart(); | |
5524 | } | |
5525 | ||
b823821f | 5526 | dev_info(adev->dev, "GPU %s begin!\n", |
bb5c7235 | 5527 | need_emergency_restart ? "jobs stop":"reset"); |
26bc5340 | 5528 | |
175ac6ec ZL |
5529 | if (!amdgpu_sriov_vf(adev)) |
5530 | hive = amdgpu_get_xgmi_hive(adev); | |
681260df | 5531 | if (hive) |
53b3f8f4 | 5532 | mutex_lock(&hive->hive_lock); |
26bc5340 | 5533 | |
f1549c09 LG |
5534 | reset_context->job = job; |
5535 | reset_context->hive = hive; | |
9e94d22c EQ |
5536 | /* |
5537 | * Build list of devices to reset. | |
5538 | * In case we are in XGMI hive mode, resort the device list | |
5539 | * to put adev in the 1st position. | |
5540 | */ | |
5541 | INIT_LIST_HEAD(&device_list); | |
175ac6ec | 5542 | if (!amdgpu_sriov_vf(adev) && (adev->gmc.xgmi.num_physical_nodes > 1)) { |
83d29a5f | 5543 | list_for_each_entry(tmp_adev, &hive->device_list, gmc.xgmi.head) { |
655ce9cb | 5544 | list_add_tail(&tmp_adev->reset_list, &device_list); |
83d29a5f YC |
5545 | if (gpu_reset_for_dev_remove && adev->shutdown) |
5546 | tmp_adev->shutdown = true; | |
5547 | } | |
655ce9cb | 5548 | if (!list_is_first(&adev->reset_list, &device_list)) |
5549 | list_rotate_to_front(&adev->reset_list, &device_list); | |
5550 | device_list_handle = &device_list; | |
26bc5340 | 5551 | } else { |
655ce9cb | 5552 | list_add_tail(&adev->reset_list, &device_list); |
26bc5340 AG |
5553 | device_list_handle = &device_list; |
5554 | } | |
5555 | ||
e923be99 AG |
5556 | /* We need to lock reset domain only once both for XGMI and single device */ |
5557 | tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, | |
5558 | reset_list); | |
3675c2f2 | 5559 | amdgpu_device_lock_reset_domain(tmp_adev->reset_domain); |
e923be99 | 5560 | |
1d721ed6 | 5561 | /* block all schedulers and reset given job's ring */ |
655ce9cb | 5562 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
f287a3c5 | 5563 | |
e923be99 | 5564 | amdgpu_device_set_mp1_state(tmp_adev); |
f287a3c5 | 5565 | |
3f12acc8 EQ |
5566 | /* |
5567 | * Try to put the audio codec into suspend state | |
5568 | * before gpu reset started. | |
5569 | * | |
5570 | * Due to the power domain of the graphics device | |
5571 | * is shared with AZ power domain. Without this, | |
5572 | * we may change the audio hardware from behind | |
5573 | * the audio driver's back. That will trigger | |
5574 | * some audio codec errors. | |
5575 | */ | |
5576 | if (!amdgpu_device_suspend_display_audio(tmp_adev)) | |
5577 | audio_suspended = true; | |
5578 | ||
9e94d22c EQ |
5579 | amdgpu_ras_set_error_query_ready(tmp_adev, false); |
5580 | ||
52fb44cf EQ |
5581 | cancel_delayed_work_sync(&tmp_adev->delayed_init_work); |
5582 | ||
c004d44e | 5583 | if (!amdgpu_sriov_vf(tmp_adev)) |
428890a3 | 5584 | amdgpu_amdkfd_pre_reset(tmp_adev); |
9e94d22c | 5585 | |
12ffa55d AG |
5586 | /* |
5587 | * Mark these ASICs to be reseted as untracked first | |
5588 | * And add them back after reset completed | |
5589 | */ | |
5590 | amdgpu_unregister_gpu_instance(tmp_adev); | |
5591 | ||
163d4cd2 | 5592 | drm_fb_helper_set_suspend_unlocked(adev_to_drm(tmp_adev)->fb_helper, true); |
565d1941 | 5593 | |
f1c1314b | 5594 | /* disable ras on ALL IPs */ |
bb5c7235 | 5595 | if (!need_emergency_restart && |
b823821f | 5596 | amdgpu_device_ip_need_full_reset(tmp_adev)) |
f1c1314b | 5597 | amdgpu_ras_suspend(tmp_adev); |
5598 | ||
1d721ed6 AG |
5599 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
5600 | struct amdgpu_ring *ring = tmp_adev->rings[i]; | |
5601 | ||
5602 | if (!ring || !ring->sched.thread) | |
5603 | continue; | |
5604 | ||
0b2d2c2e | 5605 | drm_sched_stop(&ring->sched, job ? &job->base : NULL); |
7c6e68c7 | 5606 | |
bb5c7235 | 5607 | if (need_emergency_restart) |
7c6e68c7 | 5608 | amdgpu_job_stop_all_jobs_on_sched(&ring->sched); |
1d721ed6 | 5609 | } |
8f8c80f4 | 5610 | atomic_inc(&tmp_adev->gpu_reset_counter); |
1d721ed6 AG |
5611 | } |
5612 | ||
bb5c7235 | 5613 | if (need_emergency_restart) |
7c6e68c7 AG |
5614 | goto skip_sched_resume; |
5615 | ||
1d721ed6 AG |
5616 | /* |
5617 | * Must check guilty signal here since after this point all old | |
5618 | * HW fences are force signaled. | |
5619 | * | |
5620 | * job->base holds a reference to parent fence | |
5621 | */ | |
f6a3f660 | 5622 | if (job && dma_fence_is_signaled(&job->hw_fence)) { |
1d721ed6 | 5623 | job_signaled = true; |
1d721ed6 AG |
5624 | dev_info(adev->dev, "Guilty job already signaled, skipping HW reset"); |
5625 | goto skip_hw_reset; | |
5626 | } | |
5627 | ||
26bc5340 | 5628 | retry: /* Rest of adevs pre asic reset from XGMI hive. */ |
655ce9cb | 5629 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
f5c7e779 YC |
5630 | if (gpu_reset_for_dev_remove) { |
5631 | /* Workaroud for ASICs need to disable SMC first */ | |
5632 | amdgpu_device_smu_fini_early(tmp_adev); | |
5633 | } | |
f1549c09 | 5634 | r = amdgpu_device_pre_asic_reset(tmp_adev, reset_context); |
26bc5340 AG |
5635 | /*TODO Should we stop ?*/ |
5636 | if (r) { | |
aac89168 | 5637 | dev_err(tmp_adev->dev, "GPU pre asic reset failed with err, %d for drm dev, %s ", |
4a580877 | 5638 | r, adev_to_drm(tmp_adev)->unique); |
26bc5340 AG |
5639 | tmp_adev->asic_reset_res = r; |
5640 | } | |
247c7b0d AG |
5641 | |
5642 | /* | |
5643 | * Drop all pending non scheduler resets. Scheduler resets | |
5644 | * were already dropped during drm_sched_stop | |
5645 | */ | |
d193b12b | 5646 | amdgpu_device_stop_pending_resets(tmp_adev); |
26bc5340 AG |
5647 | } |
5648 | ||
5649 | /* Actual ASIC resets if needed.*/ | |
4f30d920 | 5650 | /* Host driver will handle XGMI hive reset for SRIOV */ |
26bc5340 AG |
5651 | if (amdgpu_sriov_vf(adev)) { |
5652 | r = amdgpu_device_reset_sriov(adev, job ? false : true); | |
5653 | if (r) | |
5654 | adev->asic_reset_res = r; | |
950d6425 | 5655 | |
28606c4e | 5656 | /* Aldebaran and gfx_11_0_3 support ras in SRIOV, so need resume ras during reset */ |
4e8303cf LL |
5657 | if (amdgpu_ip_version(adev, GC_HWIP, 0) == |
5658 | IP_VERSION(9, 4, 2) || | |
5659 | amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(11, 0, 3)) | |
950d6425 | 5660 | amdgpu_ras_resume(adev); |
26bc5340 | 5661 | } else { |
f1549c09 | 5662 | r = amdgpu_do_asic_reset(device_list_handle, reset_context); |
b98a1648 | 5663 | if (r && r == -EAGAIN) |
26bc5340 | 5664 | goto retry; |
f5c7e779 YC |
5665 | |
5666 | if (!r && gpu_reset_for_dev_remove) | |
5667 | goto recover_end; | |
26bc5340 AG |
5668 | } |
5669 | ||
1d721ed6 AG |
5670 | skip_hw_reset: |
5671 | ||
26bc5340 | 5672 | /* Post ASIC reset for all devs .*/ |
655ce9cb | 5673 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
7c6e68c7 | 5674 | |
1d721ed6 AG |
5675 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
5676 | struct amdgpu_ring *ring = tmp_adev->rings[i]; | |
5677 | ||
5678 | if (!ring || !ring->sched.thread) | |
5679 | continue; | |
5680 | ||
6868a2c4 | 5681 | drm_sched_start(&ring->sched, true); |
1d721ed6 AG |
5682 | } |
5683 | ||
b8920e1e | 5684 | if (!drm_drv_uses_atomic_modeset(adev_to_drm(tmp_adev)) && !job_signaled) |
4a580877 | 5685 | drm_helper_resume_force_mode(adev_to_drm(tmp_adev)); |
1d721ed6 | 5686 | |
7258fa31 SK |
5687 | if (tmp_adev->asic_reset_res) |
5688 | r = tmp_adev->asic_reset_res; | |
5689 | ||
1d721ed6 | 5690 | tmp_adev->asic_reset_res = 0; |
26bc5340 AG |
5691 | |
5692 | if (r) { | |
5693 | /* bad news, how to tell it to userspace ? */ | |
12ffa55d | 5694 | dev_info(tmp_adev->dev, "GPU reset(%d) failed\n", atomic_read(&tmp_adev->gpu_reset_counter)); |
26bc5340 AG |
5695 | amdgpu_vf_error_put(tmp_adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r); |
5696 | } else { | |
12ffa55d | 5697 | dev_info(tmp_adev->dev, "GPU reset(%d) succeeded!\n", atomic_read(&tmp_adev->gpu_reset_counter)); |
3fa8f89d S |
5698 | if (amdgpu_acpi_smart_shift_update(adev_to_drm(tmp_adev), AMDGPU_SS_DEV_D0)) |
5699 | DRM_WARN("smart shift update failed\n"); | |
26bc5340 | 5700 | } |
7c6e68c7 | 5701 | } |
26bc5340 | 5702 | |
7c6e68c7 | 5703 | skip_sched_resume: |
655ce9cb | 5704 | list_for_each_entry(tmp_adev, device_list_handle, reset_list) { |
428890a3 | 5705 | /* unlock kfd: SRIOV would do it separately */ |
c004d44e | 5706 | if (!need_emergency_restart && !amdgpu_sriov_vf(tmp_adev)) |
428890a3 | 5707 | amdgpu_amdkfd_post_reset(tmp_adev); |
8e2712e7 | 5708 | |
5709 | /* kfd_post_reset will do nothing if kfd device is not initialized, | |
5710 | * need to bring up kfd here if it's not be initialized before | |
5711 | */ | |
5712 | if (!adev->kfd.init_complete) | |
5713 | amdgpu_amdkfd_device_init(adev); | |
5714 | ||
3f12acc8 EQ |
5715 | if (audio_suspended) |
5716 | amdgpu_device_resume_display_audio(tmp_adev); | |
e923be99 AG |
5717 | |
5718 | amdgpu_device_unset_mp1_state(tmp_adev); | |
d293470e YC |
5719 | |
5720 | amdgpu_ras_set_error_query_ready(tmp_adev, true); | |
26bc5340 AG |
5721 | } |
5722 | ||
f5c7e779 | 5723 | recover_end: |
e923be99 AG |
5724 | tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, |
5725 | reset_list); | |
5726 | amdgpu_device_unlock_reset_domain(tmp_adev->reset_domain); | |
5727 | ||
9e94d22c | 5728 | if (hive) { |
9e94d22c | 5729 | mutex_unlock(&hive->hive_lock); |
d95e8e97 | 5730 | amdgpu_put_xgmi_hive(hive); |
9e94d22c | 5731 | } |
26bc5340 | 5732 | |
f287a3c5 | 5733 | if (r) |
26bc5340 | 5734 | dev_info(adev->dev, "GPU reset end with ret = %d\n", r); |
ab9a0b1f AG |
5735 | |
5736 | atomic_set(&adev->reset_domain->reset_res, r); | |
d38ceaf9 AD |
5737 | return r; |
5738 | } | |
5739 | ||
466a7d11 ML |
5740 | /** |
5741 | * amdgpu_device_partner_bandwidth - find the bandwidth of appropriate partner | |
5742 | * | |
5743 | * @adev: amdgpu_device pointer | |
5744 | * @speed: pointer to the speed of the link | |
5745 | * @width: pointer to the width of the link | |
5746 | * | |
5747 | * Evaluate the hierarchy to find the speed and bandwidth capabilities of the | |
5748 | * first physical partner to an AMD dGPU. | |
5749 | * This will exclude any virtual switches and links. | |
5750 | */ | |
5751 | static void amdgpu_device_partner_bandwidth(struct amdgpu_device *adev, | |
5752 | enum pci_bus_speed *speed, | |
5753 | enum pcie_link_width *width) | |
5754 | { | |
5755 | struct pci_dev *parent = adev->pdev; | |
5756 | ||
5757 | if (!speed || !width) | |
5758 | return; | |
5759 | ||
5760 | *speed = PCI_SPEED_UNKNOWN; | |
5761 | *width = PCIE_LNK_WIDTH_UNKNOWN; | |
5762 | ||
5763 | while ((parent = pci_upstream_bridge(parent))) { | |
5764 | /* skip upstream/downstream switches internal to dGPU*/ | |
5765 | if (parent->vendor == PCI_VENDOR_ID_ATI) | |
5766 | continue; | |
5767 | *speed = pcie_get_speed_cap(parent); | |
5768 | *width = pcie_get_width_cap(parent); | |
5769 | break; | |
5770 | } | |
5771 | } | |
5772 | ||
e3ecdffa AD |
5773 | /** |
5774 | * amdgpu_device_get_pcie_info - fence pcie info about the PCIE slot | |
5775 | * | |
5776 | * @adev: amdgpu_device pointer | |
5777 | * | |
5778 | * Fetchs and stores in the driver the PCIE capabilities (gen speed | |
5779 | * and lanes) of the slot the device is in. Handles APUs and | |
5780 | * virtualized environments where PCIE config space may not be available. | |
5781 | */ | |
5494d864 | 5782 | static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) |
d0dd7f0c | 5783 | { |
5d9a6330 | 5784 | struct pci_dev *pdev; |
c5313457 HK |
5785 | enum pci_bus_speed speed_cap, platform_speed_cap; |
5786 | enum pcie_link_width platform_link_width; | |
d0dd7f0c | 5787 | |
cd474ba0 AD |
5788 | if (amdgpu_pcie_gen_cap) |
5789 | adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap; | |
d0dd7f0c | 5790 | |
cd474ba0 AD |
5791 | if (amdgpu_pcie_lane_cap) |
5792 | adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap; | |
d0dd7f0c | 5793 | |
cd474ba0 | 5794 | /* covers APUs as well */ |
04e85958 | 5795 | if (pci_is_root_bus(adev->pdev->bus) && !amdgpu_passthrough(adev)) { |
cd474ba0 AD |
5796 | if (adev->pm.pcie_gen_mask == 0) |
5797 | adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; | |
5798 | if (adev->pm.pcie_mlw_mask == 0) | |
5799 | adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; | |
d0dd7f0c | 5800 | return; |
cd474ba0 | 5801 | } |
d0dd7f0c | 5802 | |
c5313457 HK |
5803 | if (adev->pm.pcie_gen_mask && adev->pm.pcie_mlw_mask) |
5804 | return; | |
5805 | ||
466a7d11 ML |
5806 | amdgpu_device_partner_bandwidth(adev, &platform_speed_cap, |
5807 | &platform_link_width); | |
c5313457 | 5808 | |
cd474ba0 | 5809 | if (adev->pm.pcie_gen_mask == 0) { |
5d9a6330 AD |
5810 | /* asic caps */ |
5811 | pdev = adev->pdev; | |
5812 | speed_cap = pcie_get_speed_cap(pdev); | |
5813 | if (speed_cap == PCI_SPEED_UNKNOWN) { | |
5814 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
cd474ba0 AD |
5815 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | |
5816 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
cd474ba0 | 5817 | } else { |
2b3a1f51 FX |
5818 | if (speed_cap == PCIE_SPEED_32_0GT) |
5819 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
5820 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
5821 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
5822 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4 | | |
5823 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN5); | |
5824 | else if (speed_cap == PCIE_SPEED_16_0GT) | |
5d9a6330 AD |
5825 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
5826 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
5827 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
5828 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4); | |
5829 | else if (speed_cap == PCIE_SPEED_8_0GT) | |
5830 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
5831 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
5832 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
5833 | else if (speed_cap == PCIE_SPEED_5_0GT) | |
5834 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
5835 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
5836 | else | |
5837 | adev->pm.pcie_gen_mask |= CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
5838 | } | |
5839 | /* platform caps */ | |
c5313457 | 5840 | if (platform_speed_cap == PCI_SPEED_UNKNOWN) { |
5d9a6330 AD |
5841 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
5842 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
5843 | } else { | |
2b3a1f51 FX |
5844 | if (platform_speed_cap == PCIE_SPEED_32_0GT) |
5845 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
5846 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
5847 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
5848 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4 | | |
5849 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN5); | |
5850 | else if (platform_speed_cap == PCIE_SPEED_16_0GT) | |
5d9a6330 AD |
5851 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
5852 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
5853 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
5854 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4); | |
c5313457 | 5855 | else if (platform_speed_cap == PCIE_SPEED_8_0GT) |
5d9a6330 AD |
5856 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
5857 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
5858 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
c5313457 | 5859 | else if (platform_speed_cap == PCIE_SPEED_5_0GT) |
5d9a6330 AD |
5860 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
5861 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
5862 | else | |
5863 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
5864 | ||
cd474ba0 AD |
5865 | } |
5866 | } | |
5867 | if (adev->pm.pcie_mlw_mask == 0) { | |
c5313457 | 5868 | if (platform_link_width == PCIE_LNK_WIDTH_UNKNOWN) { |
5d9a6330 AD |
5869 | adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_PCIE_MLW_MASK; |
5870 | } else { | |
c5313457 | 5871 | switch (platform_link_width) { |
5d9a6330 | 5872 | case PCIE_LNK_X32: |
cd474ba0 AD |
5873 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 | |
5874 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | | |
5875 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
5876 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
5877 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
5878 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
5879 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
5880 | break; | |
5d9a6330 | 5881 | case PCIE_LNK_X16: |
cd474ba0 AD |
5882 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | |
5883 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
5884 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
5885 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
5886 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
5887 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
5888 | break; | |
5d9a6330 | 5889 | case PCIE_LNK_X12: |
cd474ba0 AD |
5890 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | |
5891 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
5892 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
5893 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
5894 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
5895 | break; | |
5d9a6330 | 5896 | case PCIE_LNK_X8: |
cd474ba0 AD |
5897 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | |
5898 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
5899 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
5900 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
5901 | break; | |
5d9a6330 | 5902 | case PCIE_LNK_X4: |
cd474ba0 AD |
5903 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | |
5904 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
5905 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
5906 | break; | |
5d9a6330 | 5907 | case PCIE_LNK_X2: |
cd474ba0 AD |
5908 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | |
5909 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
5910 | break; | |
5d9a6330 | 5911 | case PCIE_LNK_X1: |
cd474ba0 AD |
5912 | adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1; |
5913 | break; | |
5914 | default: | |
5915 | break; | |
5916 | } | |
d0dd7f0c AD |
5917 | } |
5918 | } | |
5919 | } | |
d38ceaf9 | 5920 | |
08a2fd23 RE |
5921 | /** |
5922 | * amdgpu_device_is_peer_accessible - Check peer access through PCIe BAR | |
5923 | * | |
5924 | * @adev: amdgpu_device pointer | |
5925 | * @peer_adev: amdgpu_device pointer for peer device trying to access @adev | |
5926 | * | |
5927 | * Return true if @peer_adev can access (DMA) @adev through the PCIe | |
5928 | * BAR, i.e. @adev is "large BAR" and the BAR matches the DMA mask of | |
5929 | * @peer_adev. | |
5930 | */ | |
5931 | bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev, | |
5932 | struct amdgpu_device *peer_adev) | |
5933 | { | |
5934 | #ifdef CONFIG_HSA_AMD_P2P | |
5935 | uint64_t address_mask = peer_adev->dev->dma_mask ? | |
5936 | ~*peer_adev->dev->dma_mask : ~((1ULL << 32) - 1); | |
5937 | resource_size_t aper_limit = | |
5938 | adev->gmc.aper_base + adev->gmc.aper_size - 1; | |
bb66ecbf LL |
5939 | bool p2p_access = |
5940 | !adev->gmc.xgmi.connected_to_cpu && | |
5941 | !(pci_p2pdma_distance(adev->pdev, peer_adev->dev, false) < 0); | |
08a2fd23 RE |
5942 | |
5943 | return pcie_p2p && p2p_access && (adev->gmc.visible_vram_size && | |
5944 | adev->gmc.real_vram_size == adev->gmc.visible_vram_size && | |
5945 | !(adev->gmc.aper_base & address_mask || | |
5946 | aper_limit & address_mask)); | |
5947 | #else | |
5948 | return false; | |
5949 | #endif | |
5950 | } | |
5951 | ||
361dbd01 AD |
5952 | int amdgpu_device_baco_enter(struct drm_device *dev) |
5953 | { | |
1348969a | 5954 | struct amdgpu_device *adev = drm_to_adev(dev); |
7a22677b | 5955 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
361dbd01 | 5956 | |
6ab68650 | 5957 | if (!amdgpu_device_supports_baco(dev)) |
361dbd01 AD |
5958 | return -ENOTSUPP; |
5959 | ||
8ab0d6f0 | 5960 | if (ras && adev->ras_enabled && |
acdae216 | 5961 | adev->nbio.funcs->enable_doorbell_interrupt) |
7a22677b LM |
5962 | adev->nbio.funcs->enable_doorbell_interrupt(adev, false); |
5963 | ||
9530273e | 5964 | return amdgpu_dpm_baco_enter(adev); |
361dbd01 AD |
5965 | } |
5966 | ||
5967 | int amdgpu_device_baco_exit(struct drm_device *dev) | |
5968 | { | |
1348969a | 5969 | struct amdgpu_device *adev = drm_to_adev(dev); |
7a22677b | 5970 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
9530273e | 5971 | int ret = 0; |
361dbd01 | 5972 | |
6ab68650 | 5973 | if (!amdgpu_device_supports_baco(dev)) |
361dbd01 AD |
5974 | return -ENOTSUPP; |
5975 | ||
9530273e EQ |
5976 | ret = amdgpu_dpm_baco_exit(adev); |
5977 | if (ret) | |
5978 | return ret; | |
7a22677b | 5979 | |
8ab0d6f0 | 5980 | if (ras && adev->ras_enabled && |
acdae216 | 5981 | adev->nbio.funcs->enable_doorbell_interrupt) |
7a22677b LM |
5982 | adev->nbio.funcs->enable_doorbell_interrupt(adev, true); |
5983 | ||
1bece222 CL |
5984 | if (amdgpu_passthrough(adev) && |
5985 | adev->nbio.funcs->clear_doorbell_interrupt) | |
5986 | adev->nbio.funcs->clear_doorbell_interrupt(adev); | |
5987 | ||
7a22677b | 5988 | return 0; |
361dbd01 | 5989 | } |
c9a6b82f AG |
5990 | |
5991 | /** | |
5992 | * amdgpu_pci_error_detected - Called when a PCI error is detected. | |
5993 | * @pdev: PCI device struct | |
5994 | * @state: PCI channel state | |
5995 | * | |
5996 | * Description: Called when a PCI error is detected. | |
5997 | * | |
5998 | * Return: PCI_ERS_RESULT_NEED_RESET or PCI_ERS_RESULT_DISCONNECT. | |
5999 | */ | |
6000 | pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state) | |
6001 | { | |
6002 | struct drm_device *dev = pci_get_drvdata(pdev); | |
6003 | struct amdgpu_device *adev = drm_to_adev(dev); | |
acd89fca | 6004 | int i; |
c9a6b82f AG |
6005 | |
6006 | DRM_INFO("PCI error: detected callback, state(%d)!!\n", state); | |
6007 | ||
6894305c AG |
6008 | if (adev->gmc.xgmi.num_physical_nodes > 1) { |
6009 | DRM_WARN("No support for XGMI hive yet..."); | |
6010 | return PCI_ERS_RESULT_DISCONNECT; | |
6011 | } | |
6012 | ||
e17e27f9 GC |
6013 | adev->pci_channel_state = state; |
6014 | ||
c9a6b82f AG |
6015 | switch (state) { |
6016 | case pci_channel_io_normal: | |
6017 | return PCI_ERS_RESULT_CAN_RECOVER; | |
acd89fca | 6018 | /* Fatal error, prepare for slot reset */ |
8a11d283 TZ |
6019 | case pci_channel_io_frozen: |
6020 | /* | |
d0fb18b5 | 6021 | * Locking adev->reset_domain->sem will prevent any external access |
acd89fca AG |
6022 | * to GPU during PCI error recovery |
6023 | */ | |
3675c2f2 | 6024 | amdgpu_device_lock_reset_domain(adev->reset_domain); |
e923be99 | 6025 | amdgpu_device_set_mp1_state(adev); |
acd89fca AG |
6026 | |
6027 | /* | |
6028 | * Block any work scheduling as we do for regular GPU reset | |
6029 | * for the duration of the recovery | |
6030 | */ | |
6031 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
6032 | struct amdgpu_ring *ring = adev->rings[i]; | |
6033 | ||
6034 | if (!ring || !ring->sched.thread) | |
6035 | continue; | |
6036 | ||
6037 | drm_sched_stop(&ring->sched, NULL); | |
6038 | } | |
8f8c80f4 | 6039 | atomic_inc(&adev->gpu_reset_counter); |
c9a6b82f AG |
6040 | return PCI_ERS_RESULT_NEED_RESET; |
6041 | case pci_channel_io_perm_failure: | |
6042 | /* Permanent error, prepare for device removal */ | |
6043 | return PCI_ERS_RESULT_DISCONNECT; | |
6044 | } | |
6045 | ||
6046 | return PCI_ERS_RESULT_NEED_RESET; | |
6047 | } | |
6048 | ||
6049 | /** | |
6050 | * amdgpu_pci_mmio_enabled - Enable MMIO and dump debug registers | |
6051 | * @pdev: pointer to PCI device | |
6052 | */ | |
6053 | pci_ers_result_t amdgpu_pci_mmio_enabled(struct pci_dev *pdev) | |
6054 | { | |
6055 | ||
6056 | DRM_INFO("PCI error: mmio enabled callback!!\n"); | |
6057 | ||
6058 | /* TODO - dump whatever for debugging purposes */ | |
6059 | ||
6060 | /* This called only if amdgpu_pci_error_detected returns | |
6061 | * PCI_ERS_RESULT_CAN_RECOVER. Read/write to the device still | |
6062 | * works, no need to reset slot. | |
6063 | */ | |
6064 | ||
6065 | return PCI_ERS_RESULT_RECOVERED; | |
6066 | } | |
6067 | ||
6068 | /** | |
6069 | * amdgpu_pci_slot_reset - Called when PCI slot has been reset. | |
6070 | * @pdev: PCI device struct | |
6071 | * | |
6072 | * Description: This routine is called by the pci error recovery | |
6073 | * code after the PCI slot has been reset, just before we | |
6074 | * should resume normal operations. | |
6075 | */ | |
6076 | pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev) | |
6077 | { | |
6078 | struct drm_device *dev = pci_get_drvdata(pdev); | |
6079 | struct amdgpu_device *adev = drm_to_adev(dev); | |
362c7b91 | 6080 | int r, i; |
04442bf7 | 6081 | struct amdgpu_reset_context reset_context; |
362c7b91 | 6082 | u32 memsize; |
7ac71382 | 6083 | struct list_head device_list; |
c9a6b82f AG |
6084 | |
6085 | DRM_INFO("PCI error: slot reset callback!!\n"); | |
6086 | ||
04442bf7 LL |
6087 | memset(&reset_context, 0, sizeof(reset_context)); |
6088 | ||
7ac71382 | 6089 | INIT_LIST_HEAD(&device_list); |
655ce9cb | 6090 | list_add_tail(&adev->reset_list, &device_list); |
7ac71382 | 6091 | |
362c7b91 AG |
6092 | /* wait for asic to come out of reset */ |
6093 | msleep(500); | |
6094 | ||
7ac71382 | 6095 | /* Restore PCI confspace */ |
c1dd4aa6 | 6096 | amdgpu_device_load_pci_state(pdev); |
c9a6b82f | 6097 | |
362c7b91 AG |
6098 | /* confirm ASIC came out of reset */ |
6099 | for (i = 0; i < adev->usec_timeout; i++) { | |
6100 | memsize = amdgpu_asic_get_config_memsize(adev); | |
6101 | ||
6102 | if (memsize != 0xffffffff) | |
6103 | break; | |
6104 | udelay(1); | |
6105 | } | |
6106 | if (memsize == 0xffffffff) { | |
6107 | r = -ETIME; | |
6108 | goto out; | |
6109 | } | |
6110 | ||
04442bf7 LL |
6111 | reset_context.method = AMD_RESET_METHOD_NONE; |
6112 | reset_context.reset_req_dev = adev; | |
6113 | set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); | |
6114 | set_bit(AMDGPU_SKIP_HW_RESET, &reset_context.flags); | |
6115 | ||
7afefb81 | 6116 | adev->no_hw_access = true; |
04442bf7 | 6117 | r = amdgpu_device_pre_asic_reset(adev, &reset_context); |
7afefb81 | 6118 | adev->no_hw_access = false; |
c9a6b82f AG |
6119 | if (r) |
6120 | goto out; | |
6121 | ||
04442bf7 | 6122 | r = amdgpu_do_asic_reset(&device_list, &reset_context); |
c9a6b82f AG |
6123 | |
6124 | out: | |
c9a6b82f | 6125 | if (!r) { |
c1dd4aa6 AG |
6126 | if (amdgpu_device_cache_pci_state(adev->pdev)) |
6127 | pci_restore_state(adev->pdev); | |
6128 | ||
c9a6b82f AG |
6129 | DRM_INFO("PCIe error recovery succeeded\n"); |
6130 | } else { | |
6131 | DRM_ERROR("PCIe error recovery failed, err:%d", r); | |
e923be99 AG |
6132 | amdgpu_device_unset_mp1_state(adev); |
6133 | amdgpu_device_unlock_reset_domain(adev->reset_domain); | |
c9a6b82f AG |
6134 | } |
6135 | ||
6136 | return r ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; | |
6137 | } | |
6138 | ||
6139 | /** | |
6140 | * amdgpu_pci_resume() - resume normal ops after PCI reset | |
6141 | * @pdev: pointer to PCI device | |
6142 | * | |
6143 | * Called when the error recovery driver tells us that its | |
505199a3 | 6144 | * OK to resume normal operation. |
c9a6b82f AG |
6145 | */ |
6146 | void amdgpu_pci_resume(struct pci_dev *pdev) | |
6147 | { | |
6148 | struct drm_device *dev = pci_get_drvdata(pdev); | |
6149 | struct amdgpu_device *adev = drm_to_adev(dev); | |
acd89fca | 6150 | int i; |
c9a6b82f | 6151 | |
c9a6b82f AG |
6152 | |
6153 | DRM_INFO("PCI error: resume callback!!\n"); | |
acd89fca | 6154 | |
e17e27f9 GC |
6155 | /* Only continue execution for the case of pci_channel_io_frozen */ |
6156 | if (adev->pci_channel_state != pci_channel_io_frozen) | |
6157 | return; | |
6158 | ||
acd89fca AG |
6159 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
6160 | struct amdgpu_ring *ring = adev->rings[i]; | |
6161 | ||
6162 | if (!ring || !ring->sched.thread) | |
6163 | continue; | |
6164 | ||
acd89fca AG |
6165 | drm_sched_start(&ring->sched, true); |
6166 | } | |
6167 | ||
e923be99 AG |
6168 | amdgpu_device_unset_mp1_state(adev); |
6169 | amdgpu_device_unlock_reset_domain(adev->reset_domain); | |
c9a6b82f | 6170 | } |
c1dd4aa6 AG |
6171 | |
6172 | bool amdgpu_device_cache_pci_state(struct pci_dev *pdev) | |
6173 | { | |
6174 | struct drm_device *dev = pci_get_drvdata(pdev); | |
6175 | struct amdgpu_device *adev = drm_to_adev(dev); | |
6176 | int r; | |
6177 | ||
6178 | r = pci_save_state(pdev); | |
6179 | if (!r) { | |
6180 | kfree(adev->pci_state); | |
6181 | ||
6182 | adev->pci_state = pci_store_saved_state(pdev); | |
6183 | ||
6184 | if (!adev->pci_state) { | |
6185 | DRM_ERROR("Failed to store PCI saved state"); | |
6186 | return false; | |
6187 | } | |
6188 | } else { | |
6189 | DRM_WARN("Failed to save PCI state, err:%d\n", r); | |
6190 | return false; | |
6191 | } | |
6192 | ||
6193 | return true; | |
6194 | } | |
6195 | ||
6196 | bool amdgpu_device_load_pci_state(struct pci_dev *pdev) | |
6197 | { | |
6198 | struct drm_device *dev = pci_get_drvdata(pdev); | |
6199 | struct amdgpu_device *adev = drm_to_adev(dev); | |
6200 | int r; | |
6201 | ||
6202 | if (!adev->pci_state) | |
6203 | return false; | |
6204 | ||
6205 | r = pci_load_saved_state(pdev, adev->pci_state); | |
6206 | ||
6207 | if (!r) { | |
6208 | pci_restore_state(pdev); | |
6209 | } else { | |
6210 | DRM_WARN("Failed to load PCI state, err:%d\n", r); | |
6211 | return false; | |
6212 | } | |
6213 | ||
6214 | return true; | |
6215 | } | |
6216 | ||
810085dd EH |
6217 | void amdgpu_device_flush_hdp(struct amdgpu_device *adev, |
6218 | struct amdgpu_ring *ring) | |
6219 | { | |
6220 | #ifdef CONFIG_X86_64 | |
b818a5d3 | 6221 | if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev)) |
810085dd EH |
6222 | return; |
6223 | #endif | |
6224 | if (adev->gmc.xgmi.connected_to_cpu) | |
6225 | return; | |
6226 | ||
6227 | if (ring && ring->funcs->emit_hdp_flush) | |
6228 | amdgpu_ring_emit_hdp_flush(ring); | |
6229 | else | |
6230 | amdgpu_asic_flush_hdp(adev, ring); | |
6231 | } | |
c1dd4aa6 | 6232 | |
810085dd EH |
6233 | void amdgpu_device_invalidate_hdp(struct amdgpu_device *adev, |
6234 | struct amdgpu_ring *ring) | |
6235 | { | |
6236 | #ifdef CONFIG_X86_64 | |
b818a5d3 | 6237 | if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev)) |
810085dd EH |
6238 | return; |
6239 | #endif | |
6240 | if (adev->gmc.xgmi.connected_to_cpu) | |
6241 | return; | |
c1dd4aa6 | 6242 | |
810085dd EH |
6243 | amdgpu_asic_invalidate_hdp(adev, ring); |
6244 | } | |
34f3a4a9 | 6245 | |
89a7a870 AG |
6246 | int amdgpu_in_reset(struct amdgpu_device *adev) |
6247 | { | |
6248 | return atomic_read(&adev->reset_domain->in_gpu_reset); | |
53a17b6b TZ |
6249 | } |
6250 | ||
34f3a4a9 LY |
6251 | /** |
6252 | * amdgpu_device_halt() - bring hardware to some kind of halt state | |
6253 | * | |
6254 | * @adev: amdgpu_device pointer | |
6255 | * | |
6256 | * Bring hardware to some kind of halt state so that no one can touch it | |
6257 | * any more. It will help to maintain error context when error occurred. | |
6258 | * Compare to a simple hang, the system will keep stable at least for SSH | |
6259 | * access. Then it should be trivial to inspect the hardware state and | |
6260 | * see what's going on. Implemented as following: | |
6261 | * | |
6262 | * 1. drm_dev_unplug() makes device inaccessible to user space(IOCTLs, etc), | |
6263 | * clears all CPU mappings to device, disallows remappings through page faults | |
6264 | * 2. amdgpu_irq_disable_all() disables all interrupts | |
6265 | * 3. amdgpu_fence_driver_hw_fini() signals all HW fences | |
6266 | * 4. set adev->no_hw_access to avoid potential crashes after setp 5 | |
6267 | * 5. amdgpu_device_unmap_mmio() clears all MMIO mappings | |
6268 | * 6. pci_disable_device() and pci_wait_for_pending_transaction() | |
6269 | * flush any in flight DMA operations | |
6270 | */ | |
6271 | void amdgpu_device_halt(struct amdgpu_device *adev) | |
6272 | { | |
6273 | struct pci_dev *pdev = adev->pdev; | |
e0f943b4 | 6274 | struct drm_device *ddev = adev_to_drm(adev); |
34f3a4a9 | 6275 | |
2c1c7ba4 | 6276 | amdgpu_xcp_dev_unplug(adev); |
34f3a4a9 LY |
6277 | drm_dev_unplug(ddev); |
6278 | ||
6279 | amdgpu_irq_disable_all(adev); | |
6280 | ||
6281 | amdgpu_fence_driver_hw_fini(adev); | |
6282 | ||
6283 | adev->no_hw_access = true; | |
6284 | ||
6285 | amdgpu_device_unmap_mmio(adev); | |
6286 | ||
6287 | pci_disable_device(pdev); | |
6288 | pci_wait_for_pending_transaction(pdev); | |
6289 | } | |
86700a40 XD |
6290 | |
6291 | u32 amdgpu_device_pcie_port_rreg(struct amdgpu_device *adev, | |
6292 | u32 reg) | |
6293 | { | |
6294 | unsigned long flags, address, data; | |
6295 | u32 r; | |
6296 | ||
6297 | address = adev->nbio.funcs->get_pcie_port_index_offset(adev); | |
6298 | data = adev->nbio.funcs->get_pcie_port_data_offset(adev); | |
6299 | ||
6300 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
6301 | WREG32(address, reg * 4); | |
6302 | (void)RREG32(address); | |
6303 | r = RREG32(data); | |
6304 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
6305 | return r; | |
6306 | } | |
6307 | ||
6308 | void amdgpu_device_pcie_port_wreg(struct amdgpu_device *adev, | |
6309 | u32 reg, u32 v) | |
6310 | { | |
6311 | unsigned long flags, address, data; | |
6312 | ||
6313 | address = adev->nbio.funcs->get_pcie_port_index_offset(adev); | |
6314 | data = adev->nbio.funcs->get_pcie_port_data_offset(adev); | |
6315 | ||
6316 | spin_lock_irqsave(&adev->pcie_idx_lock, flags); | |
6317 | WREG32(address, reg * 4); | |
6318 | (void)RREG32(address); | |
6319 | WREG32(data, v); | |
6320 | (void)RREG32(data); | |
6321 | spin_unlock_irqrestore(&adev->pcie_idx_lock, flags); | |
6322 | } | |
68ce8b24 CK |
6323 | |
6324 | /** | |
6325 | * amdgpu_device_switch_gang - switch to a new gang | |
6326 | * @adev: amdgpu_device pointer | |
6327 | * @gang: the gang to switch to | |
6328 | * | |
6329 | * Try to switch to a new gang. | |
6330 | * Returns: NULL if we switched to the new gang or a reference to the current | |
6331 | * gang leader. | |
6332 | */ | |
6333 | struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev, | |
6334 | struct dma_fence *gang) | |
6335 | { | |
6336 | struct dma_fence *old = NULL; | |
6337 | ||
6338 | do { | |
6339 | dma_fence_put(old); | |
6340 | rcu_read_lock(); | |
6341 | old = dma_fence_get_rcu_safe(&adev->gang_submit); | |
6342 | rcu_read_unlock(); | |
6343 | ||
6344 | if (old == gang) | |
6345 | break; | |
6346 | ||
6347 | if (!dma_fence_is_signaled(old)) | |
6348 | return old; | |
6349 | ||
6350 | } while (cmpxchg((struct dma_fence __force **)&adev->gang_submit, | |
6351 | old, gang) != old); | |
6352 | ||
6353 | dma_fence_put(old); | |
6354 | return NULL; | |
6355 | } | |
220c8cc8 AD |
6356 | |
6357 | bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev) | |
6358 | { | |
6359 | switch (adev->asic_type) { | |
6360 | #ifdef CONFIG_DRM_AMDGPU_SI | |
6361 | case CHIP_HAINAN: | |
6362 | #endif | |
6363 | case CHIP_TOPAZ: | |
6364 | /* chips with no display hardware */ | |
6365 | return false; | |
6366 | #ifdef CONFIG_DRM_AMDGPU_SI | |
6367 | case CHIP_TAHITI: | |
6368 | case CHIP_PITCAIRN: | |
6369 | case CHIP_VERDE: | |
6370 | case CHIP_OLAND: | |
6371 | #endif | |
6372 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
6373 | case CHIP_BONAIRE: | |
6374 | case CHIP_HAWAII: | |
6375 | case CHIP_KAVERI: | |
6376 | case CHIP_KABINI: | |
6377 | case CHIP_MULLINS: | |
6378 | #endif | |
6379 | case CHIP_TONGA: | |
6380 | case CHIP_FIJI: | |
6381 | case CHIP_POLARIS10: | |
6382 | case CHIP_POLARIS11: | |
6383 | case CHIP_POLARIS12: | |
6384 | case CHIP_VEGAM: | |
6385 | case CHIP_CARRIZO: | |
6386 | case CHIP_STONEY: | |
6387 | /* chips with display hardware */ | |
6388 | return true; | |
6389 | default: | |
6390 | /* IP discovery */ | |
4e8303cf | 6391 | if (!amdgpu_ip_version(adev, DCE_HWIP, 0) || |
220c8cc8 AD |
6392 | (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK)) |
6393 | return false; | |
6394 | return true; | |
6395 | } | |
6396 | } | |
81283fee JZ |
6397 | |
6398 | uint32_t amdgpu_device_wait_on_rreg(struct amdgpu_device *adev, | |
6399 | uint32_t inst, uint32_t reg_addr, char reg_name[], | |
6400 | uint32_t expected_value, uint32_t mask) | |
6401 | { | |
6402 | uint32_t ret = 0; | |
6403 | uint32_t old_ = 0; | |
6404 | uint32_t tmp_ = RREG32(reg_addr); | |
6405 | uint32_t loop = adev->usec_timeout; | |
6406 | ||
6407 | while ((tmp_ & (mask)) != (expected_value)) { | |
6408 | if (old_ != tmp_) { | |
6409 | loop = adev->usec_timeout; | |
6410 | old_ = tmp_; | |
6411 | } else | |
6412 | udelay(1); | |
6413 | tmp_ = RREG32(reg_addr); | |
6414 | loop--; | |
6415 | if (!loop) { | |
6416 | DRM_WARN("Register(%d) [%s] failed to reach value 0x%08x != 0x%08xn", | |
6417 | inst, reg_name, (uint32_t)expected_value, | |
6418 | (uint32_t)(tmp_ & (mask))); | |
6419 | ret = -ETIMEDOUT; | |
6420 | break; | |
6421 | } | |
6422 | } | |
6423 | return ret; | |
6424 | } |