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> | |
fdf2f6c5 | 33 | |
4562236b | 34 | #include <drm/drm_atomic_helper.h> |
fcd70cd3 | 35 | #include <drm/drm_probe_helper.h> |
d38ceaf9 AD |
36 | #include <drm/amdgpu_drm.h> |
37 | #include <linux/vgaarb.h> | |
38 | #include <linux/vga_switcheroo.h> | |
39 | #include <linux/efi.h> | |
40 | #include "amdgpu.h" | |
f4b373f4 | 41 | #include "amdgpu_trace.h" |
d38ceaf9 AD |
42 | #include "amdgpu_i2c.h" |
43 | #include "atom.h" | |
44 | #include "amdgpu_atombios.h" | |
a5bde2f9 | 45 | #include "amdgpu_atomfirmware.h" |
d0dd7f0c | 46 | #include "amd_pcie.h" |
33f34802 KW |
47 | #ifdef CONFIG_DRM_AMDGPU_SI |
48 | #include "si.h" | |
49 | #endif | |
a2e73f56 AD |
50 | #ifdef CONFIG_DRM_AMDGPU_CIK |
51 | #include "cik.h" | |
52 | #endif | |
aaa36a97 | 53 | #include "vi.h" |
460826e6 | 54 | #include "soc15.h" |
0a5b8c7b | 55 | #include "nv.h" |
d38ceaf9 | 56 | #include "bif/bif_4_1_d.h" |
9accf2fd | 57 | #include <linux/pci.h> |
bec86378 | 58 | #include <linux/firmware.h> |
89041940 | 59 | #include "amdgpu_vf_error.h" |
d38ceaf9 | 60 | |
ba997709 | 61 | #include "amdgpu_amdkfd.h" |
d2f52ac8 | 62 | #include "amdgpu_pm.h" |
d38ceaf9 | 63 | |
5183411b | 64 | #include "amdgpu_xgmi.h" |
c030f2e4 | 65 | #include "amdgpu_ras.h" |
9c7c85f7 | 66 | #include "amdgpu_pmu.h" |
bd607166 | 67 | #include "amdgpu_fru_eeprom.h" |
5183411b | 68 | |
d5ea093e | 69 | #include <linux/suspend.h> |
c6a6e2db | 70 | #include <drm/task_barrier.h> |
3f12acc8 | 71 | #include <linux/pm_runtime.h> |
d5ea093e | 72 | |
e2a75f88 | 73 | MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin"); |
3f76dced | 74 | MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin"); |
2d2e5e7e | 75 | MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin"); |
ad5a67a7 | 76 | MODULE_FIRMWARE("amdgpu/picasso_gpu_info.bin"); |
54c4d17e | 77 | MODULE_FIRMWARE("amdgpu/raven2_gpu_info.bin"); |
65e60f6e | 78 | MODULE_FIRMWARE("amdgpu/arcturus_gpu_info.bin"); |
b51a26a0 | 79 | MODULE_FIRMWARE("amdgpu/renoir_gpu_info.bin"); |
23c6268e | 80 | MODULE_FIRMWARE("amdgpu/navi10_gpu_info.bin"); |
ed42cfe1 | 81 | MODULE_FIRMWARE("amdgpu/navi14_gpu_info.bin"); |
42b325e5 | 82 | MODULE_FIRMWARE("amdgpu/navi12_gpu_info.bin"); |
e2a75f88 | 83 | |
2dc80b00 S |
84 | #define AMDGPU_RESUME_MS 2000 |
85 | ||
050091ab | 86 | const char *amdgpu_asic_name[] = { |
da69c161 KW |
87 | "TAHITI", |
88 | "PITCAIRN", | |
89 | "VERDE", | |
90 | "OLAND", | |
91 | "HAINAN", | |
d38ceaf9 AD |
92 | "BONAIRE", |
93 | "KAVERI", | |
94 | "KABINI", | |
95 | "HAWAII", | |
96 | "MULLINS", | |
97 | "TOPAZ", | |
98 | "TONGA", | |
48299f95 | 99 | "FIJI", |
d38ceaf9 | 100 | "CARRIZO", |
139f4917 | 101 | "STONEY", |
2cc0c0b5 FC |
102 | "POLARIS10", |
103 | "POLARIS11", | |
c4642a47 | 104 | "POLARIS12", |
48ff108d | 105 | "VEGAM", |
d4196f01 | 106 | "VEGA10", |
8fab806a | 107 | "VEGA12", |
956fcddc | 108 | "VEGA20", |
2ca8a5d2 | 109 | "RAVEN", |
d6c3b24e | 110 | "ARCTURUS", |
1eee4228 | 111 | "RENOIR", |
852a6626 | 112 | "NAVI10", |
87dbad02 | 113 | "NAVI14", |
9802f5d7 | 114 | "NAVI12", |
d38ceaf9 AD |
115 | "LAST", |
116 | }; | |
117 | ||
dcea6e65 KR |
118 | /** |
119 | * DOC: pcie_replay_count | |
120 | * | |
121 | * The amdgpu driver provides a sysfs API for reporting the total number | |
122 | * of PCIe replays (NAKs) | |
123 | * The file pcie_replay_count is used for this and returns the total | |
124 | * number of replays as a sum of the NAKs generated and NAKs received | |
125 | */ | |
126 | ||
127 | static ssize_t amdgpu_device_get_pcie_replay_count(struct device *dev, | |
128 | struct device_attribute *attr, char *buf) | |
129 | { | |
130 | struct drm_device *ddev = dev_get_drvdata(dev); | |
131 | struct amdgpu_device *adev = ddev->dev_private; | |
132 | uint64_t cnt = amdgpu_asic_get_pcie_replay_count(adev); | |
133 | ||
134 | return snprintf(buf, PAGE_SIZE, "%llu\n", cnt); | |
135 | } | |
136 | ||
137 | static DEVICE_ATTR(pcie_replay_count, S_IRUGO, | |
138 | amdgpu_device_get_pcie_replay_count, NULL); | |
139 | ||
5494d864 AD |
140 | static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev); |
141 | ||
bd607166 KR |
142 | /** |
143 | * DOC: product_name | |
144 | * | |
145 | * The amdgpu driver provides a sysfs API for reporting the product name | |
146 | * for the device | |
147 | * The file serial_number is used for this and returns the product name | |
148 | * as returned from the FRU. | |
149 | * NOTE: This is only available for certain server cards | |
150 | */ | |
151 | ||
152 | static ssize_t amdgpu_device_get_product_name(struct device *dev, | |
153 | struct device_attribute *attr, char *buf) | |
154 | { | |
155 | struct drm_device *ddev = dev_get_drvdata(dev); | |
156 | struct amdgpu_device *adev = ddev->dev_private; | |
157 | ||
158 | return snprintf(buf, PAGE_SIZE, "%s\n", adev->product_name); | |
159 | } | |
160 | ||
161 | static DEVICE_ATTR(product_name, S_IRUGO, | |
162 | amdgpu_device_get_product_name, NULL); | |
163 | ||
164 | /** | |
165 | * DOC: product_number | |
166 | * | |
167 | * The amdgpu driver provides a sysfs API for reporting the part number | |
168 | * for the device | |
169 | * The file serial_number is used for this and returns the part number | |
170 | * as returned from the FRU. | |
171 | * NOTE: This is only available for certain server cards | |
172 | */ | |
173 | ||
174 | static ssize_t amdgpu_device_get_product_number(struct device *dev, | |
175 | struct device_attribute *attr, char *buf) | |
176 | { | |
177 | struct drm_device *ddev = dev_get_drvdata(dev); | |
178 | struct amdgpu_device *adev = ddev->dev_private; | |
179 | ||
180 | return snprintf(buf, PAGE_SIZE, "%s\n", adev->product_number); | |
181 | } | |
182 | ||
183 | static DEVICE_ATTR(product_number, S_IRUGO, | |
184 | amdgpu_device_get_product_number, NULL); | |
185 | ||
186 | /** | |
187 | * DOC: serial_number | |
188 | * | |
189 | * The amdgpu driver provides a sysfs API for reporting the serial number | |
190 | * for the device | |
191 | * The file serial_number is used for this and returns the serial number | |
192 | * as returned from the FRU. | |
193 | * NOTE: This is only available for certain server cards | |
194 | */ | |
195 | ||
196 | static ssize_t amdgpu_device_get_serial_number(struct device *dev, | |
197 | struct device_attribute *attr, char *buf) | |
198 | { | |
199 | struct drm_device *ddev = dev_get_drvdata(dev); | |
200 | struct amdgpu_device *adev = ddev->dev_private; | |
201 | ||
202 | return snprintf(buf, PAGE_SIZE, "%s\n", adev->serial); | |
203 | } | |
204 | ||
205 | static DEVICE_ATTR(serial_number, S_IRUGO, | |
206 | amdgpu_device_get_serial_number, NULL); | |
207 | ||
e3ecdffa | 208 | /** |
31af062a | 209 | * amdgpu_device_supports_boco - Is the device a dGPU with HG/PX power control |
e3ecdffa AD |
210 | * |
211 | * @dev: drm_device pointer | |
212 | * | |
213 | * Returns true if the device is a dGPU with HG/PX power control, | |
214 | * otherwise return false. | |
215 | */ | |
31af062a | 216 | bool amdgpu_device_supports_boco(struct drm_device *dev) |
d38ceaf9 AD |
217 | { |
218 | struct amdgpu_device *adev = dev->dev_private; | |
219 | ||
2f7d10b3 | 220 | if (adev->flags & AMD_IS_PX) |
d38ceaf9 AD |
221 | return true; |
222 | return false; | |
223 | } | |
224 | ||
a69cba42 AD |
225 | /** |
226 | * amdgpu_device_supports_baco - Does the device support BACO | |
227 | * | |
228 | * @dev: drm_device pointer | |
229 | * | |
230 | * Returns true if the device supporte BACO, | |
231 | * otherwise return false. | |
232 | */ | |
233 | bool amdgpu_device_supports_baco(struct drm_device *dev) | |
234 | { | |
235 | struct amdgpu_device *adev = dev->dev_private; | |
236 | ||
237 | return amdgpu_asic_supports_baco(adev); | |
238 | } | |
239 | ||
e35e2b11 TY |
240 | /** |
241 | * VRAM access helper functions. | |
242 | * | |
243 | * amdgpu_device_vram_access - read/write a buffer in vram | |
244 | * | |
245 | * @adev: amdgpu_device pointer | |
246 | * @pos: offset of the buffer in vram | |
247 | * @buf: virtual address of the buffer in system memory | |
248 | * @size: read/write size, sizeof(@buf) must > @size | |
249 | * @write: true - write to vram, otherwise - read from vram | |
250 | */ | |
251 | void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos, | |
252 | uint32_t *buf, size_t size, bool write) | |
253 | { | |
e35e2b11 | 254 | unsigned long flags; |
ce05ac56 CK |
255 | uint32_t hi = ~0; |
256 | uint64_t last; | |
257 | ||
258 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
259 | for (last = pos + size; pos < last; pos += 4) { | |
260 | uint32_t tmp = pos >> 31; | |
e35e2b11 | 261 | |
e35e2b11 | 262 | WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)pos) | 0x80000000); |
ce05ac56 CK |
263 | if (tmp != hi) { |
264 | WREG32_NO_KIQ(mmMM_INDEX_HI, tmp); | |
265 | hi = tmp; | |
266 | } | |
e35e2b11 TY |
267 | if (write) |
268 | WREG32_NO_KIQ(mmMM_DATA, *buf++); | |
269 | else | |
270 | *buf++ = RREG32_NO_KIQ(mmMM_DATA); | |
e35e2b11 | 271 | } |
ce05ac56 | 272 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); |
e35e2b11 TY |
273 | } |
274 | ||
d38ceaf9 | 275 | /* |
2eee0229 | 276 | * device register access helper functions. |
d38ceaf9 | 277 | */ |
e3ecdffa | 278 | /** |
2eee0229 | 279 | * amdgpu_device_rreg - read a register |
e3ecdffa AD |
280 | * |
281 | * @adev: amdgpu_device pointer | |
282 | * @reg: dword aligned register offset | |
283 | * @acc_flags: access flags which require special behavior | |
284 | * | |
285 | * Returns the 32 bit value from the offset specified. | |
286 | */ | |
2eee0229 HZ |
287 | uint32_t amdgpu_device_rreg(struct amdgpu_device *adev, uint32_t reg, |
288 | uint32_t acc_flags) | |
d38ceaf9 | 289 | { |
f4b373f4 TSD |
290 | uint32_t ret; |
291 | ||
f384ff95 | 292 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) |
d33a99c4 | 293 | return amdgpu_kiq_rreg(adev, reg); |
bc992ba5 | 294 | |
ec59847e | 295 | if ((reg * 4) < adev->rmmio_size) |
f4b373f4 | 296 | ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); |
2eee0229 HZ |
297 | else |
298 | ret = adev->pcie_rreg(adev, (reg * 4)); | |
299 | trace_amdgpu_device_rreg(adev->pdev->device, reg, ret); | |
f4b373f4 | 300 | return ret; |
d38ceaf9 AD |
301 | } |
302 | ||
421a2a30 ML |
303 | /* |
304 | * MMIO register read with bytes helper functions | |
305 | * @offset:bytes offset from MMIO start | |
306 | * | |
307 | */ | |
308 | ||
e3ecdffa AD |
309 | /** |
310 | * amdgpu_mm_rreg8 - read a memory mapped IO register | |
311 | * | |
312 | * @adev: amdgpu_device pointer | |
313 | * @offset: byte aligned register offset | |
314 | * | |
315 | * Returns the 8 bit value from the offset specified. | |
316 | */ | |
421a2a30 ML |
317 | uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset) { |
318 | if (offset < adev->rmmio_size) | |
319 | return (readb(adev->rmmio + offset)); | |
320 | BUG(); | |
321 | } | |
322 | ||
323 | /* | |
324 | * MMIO register write with bytes helper functions | |
325 | * @offset:bytes offset from MMIO start | |
326 | * @value: the value want to be written to the register | |
327 | * | |
328 | */ | |
e3ecdffa AD |
329 | /** |
330 | * amdgpu_mm_wreg8 - read a memory mapped IO register | |
331 | * | |
332 | * @adev: amdgpu_device pointer | |
333 | * @offset: byte aligned register offset | |
334 | * @value: 8 bit value to write | |
335 | * | |
336 | * Writes the value specified to the offset specified. | |
337 | */ | |
421a2a30 ML |
338 | void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value) { |
339 | if (offset < adev->rmmio_size) | |
340 | writeb(value, adev->rmmio + offset); | |
341 | else | |
342 | BUG(); | |
343 | } | |
344 | ||
2eee0229 HZ |
345 | void static inline amdgpu_device_wreg_no_kiq(struct amdgpu_device *adev, uint32_t reg, |
346 | uint32_t v, uint32_t acc_flags) | |
2e0cc4d4 | 347 | { |
2eee0229 | 348 | trace_amdgpu_device_wreg(adev->pdev->device, reg, v); |
2e0cc4d4 | 349 | |
ec59847e | 350 | if ((reg * 4) < adev->rmmio_size) |
2e0cc4d4 | 351 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); |
2eee0229 HZ |
352 | else |
353 | adev->pcie_wreg(adev, (reg * 4), v); | |
2e0cc4d4 ML |
354 | } |
355 | ||
e3ecdffa | 356 | /** |
2eee0229 | 357 | * amdgpu_device_wreg - write to a register |
e3ecdffa AD |
358 | * |
359 | * @adev: amdgpu_device pointer | |
360 | * @reg: dword aligned register offset | |
361 | * @v: 32 bit value to write to the register | |
362 | * @acc_flags: access flags which require special behavior | |
363 | * | |
364 | * Writes the value specified to the offset specified. | |
365 | */ | |
2eee0229 HZ |
366 | void amdgpu_device_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, |
367 | uint32_t acc_flags) | |
d38ceaf9 | 368 | { |
f384ff95 | 369 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) |
d33a99c4 | 370 | return amdgpu_kiq_wreg(adev, reg, v); |
bc992ba5 | 371 | |
2eee0229 | 372 | amdgpu_device_wreg_no_kiq(adev, reg, v, acc_flags); |
2e0cc4d4 | 373 | } |
d38ceaf9 | 374 | |
2e0cc4d4 ML |
375 | /* |
376 | * amdgpu_mm_wreg_mmio_rlc - write register either with mmio or with RLC path if in range | |
377 | * | |
378 | * this function is invoked only the debugfs register access | |
379 | * */ | |
380 | void amdgpu_mm_wreg_mmio_rlc(struct amdgpu_device *adev, uint32_t reg, uint32_t v, | |
381 | uint32_t acc_flags) | |
382 | { | |
383 | if (amdgpu_sriov_fullaccess(adev) && | |
384 | adev->gfx.rlc.funcs && | |
385 | adev->gfx.rlc.funcs->is_rlcg_access_range) { | |
47ed4e1c | 386 | |
2e0cc4d4 ML |
387 | if (adev->gfx.rlc.funcs->is_rlcg_access_range(adev, reg)) |
388 | return adev->gfx.rlc.funcs->rlcg_wreg(adev, reg, v); | |
47ed4e1c | 389 | } |
2e0cc4d4 | 390 | |
2eee0229 | 391 | amdgpu_device_wreg_no_kiq(adev, reg, v, acc_flags); |
d38ceaf9 AD |
392 | } |
393 | ||
e3ecdffa AD |
394 | /** |
395 | * amdgpu_io_rreg - read an IO register | |
396 | * | |
397 | * @adev: amdgpu_device pointer | |
398 | * @reg: dword aligned register offset | |
399 | * | |
400 | * Returns the 32 bit value from the offset specified. | |
401 | */ | |
d38ceaf9 AD |
402 | u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg) |
403 | { | |
404 | if ((reg * 4) < adev->rio_mem_size) | |
405 | return ioread32(adev->rio_mem + (reg * 4)); | |
406 | else { | |
407 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
408 | return ioread32(adev->rio_mem + (mmMM_DATA * 4)); | |
409 | } | |
410 | } | |
411 | ||
e3ecdffa AD |
412 | /** |
413 | * amdgpu_io_wreg - write to an IO register | |
414 | * | |
415 | * @adev: amdgpu_device pointer | |
416 | * @reg: dword aligned register offset | |
417 | * @v: 32 bit value to write to the register | |
418 | * | |
419 | * Writes the value specified to the offset specified. | |
420 | */ | |
d38ceaf9 AD |
421 | void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v) |
422 | { | |
d38ceaf9 AD |
423 | if ((reg * 4) < adev->rio_mem_size) |
424 | iowrite32(v, adev->rio_mem + (reg * 4)); | |
425 | else { | |
426 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
427 | iowrite32(v, adev->rio_mem + (mmMM_DATA * 4)); | |
428 | } | |
429 | } | |
430 | ||
431 | /** | |
432 | * amdgpu_mm_rdoorbell - read a doorbell dword | |
433 | * | |
434 | * @adev: amdgpu_device pointer | |
435 | * @index: doorbell index | |
436 | * | |
437 | * Returns the value in the doorbell aperture at the | |
438 | * requested doorbell index (CIK). | |
439 | */ | |
440 | u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index) | |
441 | { | |
442 | if (index < adev->doorbell.num_doorbells) { | |
443 | return readl(adev->doorbell.ptr + index); | |
444 | } else { | |
445 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
446 | return 0; | |
447 | } | |
448 | } | |
449 | ||
450 | /** | |
451 | * amdgpu_mm_wdoorbell - write a doorbell dword | |
452 | * | |
453 | * @adev: amdgpu_device pointer | |
454 | * @index: doorbell index | |
455 | * @v: value to write | |
456 | * | |
457 | * Writes @v to the doorbell aperture at the | |
458 | * requested doorbell index (CIK). | |
459 | */ | |
460 | void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v) | |
461 | { | |
462 | if (index < adev->doorbell.num_doorbells) { | |
463 | writel(v, adev->doorbell.ptr + index); | |
464 | } else { | |
465 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
466 | } | |
467 | } | |
468 | ||
832be404 KW |
469 | /** |
470 | * amdgpu_mm_rdoorbell64 - read a doorbell Qword | |
471 | * | |
472 | * @adev: amdgpu_device pointer | |
473 | * @index: doorbell index | |
474 | * | |
475 | * Returns the value in the doorbell aperture at the | |
476 | * requested doorbell index (VEGA10+). | |
477 | */ | |
478 | u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index) | |
479 | { | |
480 | if (index < adev->doorbell.num_doorbells) { | |
481 | return atomic64_read((atomic64_t *)(adev->doorbell.ptr + index)); | |
482 | } else { | |
483 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
484 | return 0; | |
485 | } | |
486 | } | |
487 | ||
488 | /** | |
489 | * amdgpu_mm_wdoorbell64 - write a doorbell Qword | |
490 | * | |
491 | * @adev: amdgpu_device pointer | |
492 | * @index: doorbell index | |
493 | * @v: value to write | |
494 | * | |
495 | * Writes @v to the doorbell aperture at the | |
496 | * requested doorbell index (VEGA10+). | |
497 | */ | |
498 | void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v) | |
499 | { | |
500 | if (index < adev->doorbell.num_doorbells) { | |
501 | atomic64_set((atomic64_t *)(adev->doorbell.ptr + index), v); | |
502 | } else { | |
503 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
504 | } | |
505 | } | |
506 | ||
d38ceaf9 AD |
507 | /** |
508 | * amdgpu_invalid_rreg - dummy reg read function | |
509 | * | |
510 | * @adev: amdgpu device pointer | |
511 | * @reg: offset of register | |
512 | * | |
513 | * Dummy register read function. Used for register blocks | |
514 | * that certain asics don't have (all asics). | |
515 | * Returns the value in the register. | |
516 | */ | |
517 | static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg) | |
518 | { | |
519 | DRM_ERROR("Invalid callback to read register 0x%04X\n", reg); | |
520 | BUG(); | |
521 | return 0; | |
522 | } | |
523 | ||
524 | /** | |
525 | * amdgpu_invalid_wreg - dummy reg write function | |
526 | * | |
527 | * @adev: amdgpu device pointer | |
528 | * @reg: offset of register | |
529 | * @v: value to write to the register | |
530 | * | |
531 | * Dummy register read function. Used for register blocks | |
532 | * that certain asics don't have (all asics). | |
533 | */ | |
534 | static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) | |
535 | { | |
536 | DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n", | |
537 | reg, v); | |
538 | BUG(); | |
539 | } | |
540 | ||
4fa1c6a6 TZ |
541 | /** |
542 | * amdgpu_invalid_rreg64 - dummy 64 bit reg read function | |
543 | * | |
544 | * @adev: amdgpu device pointer | |
545 | * @reg: offset of register | |
546 | * | |
547 | * Dummy register read function. Used for register blocks | |
548 | * that certain asics don't have (all asics). | |
549 | * Returns the value in the register. | |
550 | */ | |
551 | static uint64_t amdgpu_invalid_rreg64(struct amdgpu_device *adev, uint32_t reg) | |
552 | { | |
553 | DRM_ERROR("Invalid callback to read 64 bit register 0x%04X\n", reg); | |
554 | BUG(); | |
555 | return 0; | |
556 | } | |
557 | ||
558 | /** | |
559 | * amdgpu_invalid_wreg64 - dummy reg write function | |
560 | * | |
561 | * @adev: amdgpu device pointer | |
562 | * @reg: offset of register | |
563 | * @v: value to write to the register | |
564 | * | |
565 | * Dummy register read function. Used for register blocks | |
566 | * that certain asics don't have (all asics). | |
567 | */ | |
568 | static void amdgpu_invalid_wreg64(struct amdgpu_device *adev, uint32_t reg, uint64_t v) | |
569 | { | |
570 | DRM_ERROR("Invalid callback to write 64 bit register 0x%04X with 0x%08llX\n", | |
571 | reg, v); | |
572 | BUG(); | |
573 | } | |
574 | ||
d38ceaf9 AD |
575 | /** |
576 | * amdgpu_block_invalid_rreg - dummy reg read function | |
577 | * | |
578 | * @adev: amdgpu device pointer | |
579 | * @block: offset of instance | |
580 | * @reg: offset of register | |
581 | * | |
582 | * Dummy register read function. Used for register blocks | |
583 | * that certain asics don't have (all asics). | |
584 | * Returns the value in the register. | |
585 | */ | |
586 | static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev, | |
587 | uint32_t block, uint32_t reg) | |
588 | { | |
589 | DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n", | |
590 | reg, block); | |
591 | BUG(); | |
592 | return 0; | |
593 | } | |
594 | ||
595 | /** | |
596 | * amdgpu_block_invalid_wreg - dummy reg write function | |
597 | * | |
598 | * @adev: amdgpu device pointer | |
599 | * @block: offset of instance | |
600 | * @reg: offset of register | |
601 | * @v: value to write to the register | |
602 | * | |
603 | * Dummy register read function. Used for register blocks | |
604 | * that certain asics don't have (all asics). | |
605 | */ | |
606 | static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev, | |
607 | uint32_t block, | |
608 | uint32_t reg, uint32_t v) | |
609 | { | |
610 | DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n", | |
611 | reg, block, v); | |
612 | BUG(); | |
613 | } | |
614 | ||
e3ecdffa AD |
615 | /** |
616 | * amdgpu_device_vram_scratch_init - allocate the VRAM scratch page | |
617 | * | |
618 | * @adev: amdgpu device pointer | |
619 | * | |
620 | * Allocates a scratch page of VRAM for use by various things in the | |
621 | * driver. | |
622 | */ | |
06ec9070 | 623 | static int amdgpu_device_vram_scratch_init(struct amdgpu_device *adev) |
d38ceaf9 | 624 | { |
a4a02777 CK |
625 | return amdgpu_bo_create_kernel(adev, AMDGPU_GPU_PAGE_SIZE, |
626 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, | |
627 | &adev->vram_scratch.robj, | |
628 | &adev->vram_scratch.gpu_addr, | |
629 | (void **)&adev->vram_scratch.ptr); | |
d38ceaf9 AD |
630 | } |
631 | ||
e3ecdffa AD |
632 | /** |
633 | * amdgpu_device_vram_scratch_fini - Free the VRAM scratch page | |
634 | * | |
635 | * @adev: amdgpu device pointer | |
636 | * | |
637 | * Frees the VRAM scratch page. | |
638 | */ | |
06ec9070 | 639 | static void amdgpu_device_vram_scratch_fini(struct amdgpu_device *adev) |
d38ceaf9 | 640 | { |
078af1a3 | 641 | amdgpu_bo_free_kernel(&adev->vram_scratch.robj, NULL, NULL); |
d38ceaf9 AD |
642 | } |
643 | ||
644 | /** | |
9c3f2b54 | 645 | * amdgpu_device_program_register_sequence - program an array of registers. |
d38ceaf9 AD |
646 | * |
647 | * @adev: amdgpu_device pointer | |
648 | * @registers: pointer to the register array | |
649 | * @array_size: size of the register array | |
650 | * | |
651 | * Programs an array or registers with and and or masks. | |
652 | * This is a helper for setting golden registers. | |
653 | */ | |
9c3f2b54 AD |
654 | void amdgpu_device_program_register_sequence(struct amdgpu_device *adev, |
655 | const u32 *registers, | |
656 | const u32 array_size) | |
d38ceaf9 AD |
657 | { |
658 | u32 tmp, reg, and_mask, or_mask; | |
659 | int i; | |
660 | ||
661 | if (array_size % 3) | |
662 | return; | |
663 | ||
664 | for (i = 0; i < array_size; i +=3) { | |
665 | reg = registers[i + 0]; | |
666 | and_mask = registers[i + 1]; | |
667 | or_mask = registers[i + 2]; | |
668 | ||
669 | if (and_mask == 0xffffffff) { | |
670 | tmp = or_mask; | |
671 | } else { | |
672 | tmp = RREG32(reg); | |
673 | tmp &= ~and_mask; | |
e0d07657 HZ |
674 | if (adev->family >= AMDGPU_FAMILY_AI) |
675 | tmp |= (or_mask & and_mask); | |
676 | else | |
677 | tmp |= or_mask; | |
d38ceaf9 AD |
678 | } |
679 | WREG32(reg, tmp); | |
680 | } | |
681 | } | |
682 | ||
e3ecdffa AD |
683 | /** |
684 | * amdgpu_device_pci_config_reset - reset the GPU | |
685 | * | |
686 | * @adev: amdgpu_device pointer | |
687 | * | |
688 | * Resets the GPU using the pci config reset sequence. | |
689 | * Only applicable to asics prior to vega10. | |
690 | */ | |
8111c387 | 691 | void amdgpu_device_pci_config_reset(struct amdgpu_device *adev) |
d38ceaf9 AD |
692 | { |
693 | pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA); | |
694 | } | |
695 | ||
696 | /* | |
697 | * GPU doorbell aperture helpers function. | |
698 | */ | |
699 | /** | |
06ec9070 | 700 | * amdgpu_device_doorbell_init - Init doorbell driver information. |
d38ceaf9 AD |
701 | * |
702 | * @adev: amdgpu_device pointer | |
703 | * | |
704 | * Init doorbell driver information (CIK) | |
705 | * Returns 0 on success, error on failure. | |
706 | */ | |
06ec9070 | 707 | static int amdgpu_device_doorbell_init(struct amdgpu_device *adev) |
d38ceaf9 | 708 | { |
6585661d | 709 | |
705e519e CK |
710 | /* No doorbell on SI hardware generation */ |
711 | if (adev->asic_type < CHIP_BONAIRE) { | |
712 | adev->doorbell.base = 0; | |
713 | adev->doorbell.size = 0; | |
714 | adev->doorbell.num_doorbells = 0; | |
715 | adev->doorbell.ptr = NULL; | |
716 | return 0; | |
717 | } | |
718 | ||
d6895ad3 CK |
719 | if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET) |
720 | return -EINVAL; | |
721 | ||
22357775 AD |
722 | amdgpu_asic_init_doorbell_index(adev); |
723 | ||
d38ceaf9 AD |
724 | /* doorbell bar mapping */ |
725 | adev->doorbell.base = pci_resource_start(adev->pdev, 2); | |
726 | adev->doorbell.size = pci_resource_len(adev->pdev, 2); | |
727 | ||
edf600da | 728 | adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32), |
9564f192 | 729 | adev->doorbell_index.max_assignment+1); |
d38ceaf9 AD |
730 | if (adev->doorbell.num_doorbells == 0) |
731 | return -EINVAL; | |
732 | ||
ec3db8a6 | 733 | /* For Vega, reserve and map two pages on doorbell BAR since SDMA |
88dc26e4 OZ |
734 | * paging queue doorbell use the second page. The |
735 | * AMDGPU_DOORBELL64_MAX_ASSIGNMENT definition assumes all the | |
736 | * doorbells are in the first page. So with paging queue enabled, | |
737 | * the max num_doorbells should + 1 page (0x400 in dword) | |
ec3db8a6 PY |
738 | */ |
739 | if (adev->asic_type >= CHIP_VEGA10) | |
88dc26e4 | 740 | adev->doorbell.num_doorbells += 0x400; |
ec3db8a6 | 741 | |
8972e5d2 CK |
742 | adev->doorbell.ptr = ioremap(adev->doorbell.base, |
743 | adev->doorbell.num_doorbells * | |
744 | sizeof(u32)); | |
745 | if (adev->doorbell.ptr == NULL) | |
d38ceaf9 | 746 | return -ENOMEM; |
d38ceaf9 AD |
747 | |
748 | return 0; | |
749 | } | |
750 | ||
751 | /** | |
06ec9070 | 752 | * amdgpu_device_doorbell_fini - Tear down doorbell driver information. |
d38ceaf9 AD |
753 | * |
754 | * @adev: amdgpu_device pointer | |
755 | * | |
756 | * Tear down doorbell driver information (CIK) | |
757 | */ | |
06ec9070 | 758 | static void amdgpu_device_doorbell_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
759 | { |
760 | iounmap(adev->doorbell.ptr); | |
761 | adev->doorbell.ptr = NULL; | |
762 | } | |
763 | ||
22cb0164 | 764 | |
d38ceaf9 AD |
765 | |
766 | /* | |
06ec9070 | 767 | * amdgpu_device_wb_*() |
455a7bc2 | 768 | * Writeback is the method by which the GPU updates special pages in memory |
ea81a173 | 769 | * with the status of certain GPU events (fences, ring pointers,etc.). |
d38ceaf9 AD |
770 | */ |
771 | ||
772 | /** | |
06ec9070 | 773 | * amdgpu_device_wb_fini - Disable Writeback and free memory |
d38ceaf9 AD |
774 | * |
775 | * @adev: amdgpu_device pointer | |
776 | * | |
777 | * Disables Writeback and frees the Writeback memory (all asics). | |
778 | * Used at driver shutdown. | |
779 | */ | |
06ec9070 | 780 | static void amdgpu_device_wb_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
781 | { |
782 | if (adev->wb.wb_obj) { | |
a76ed485 AD |
783 | amdgpu_bo_free_kernel(&adev->wb.wb_obj, |
784 | &adev->wb.gpu_addr, | |
785 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
786 | adev->wb.wb_obj = NULL; |
787 | } | |
788 | } | |
789 | ||
790 | /** | |
06ec9070 | 791 | * amdgpu_device_wb_init- Init Writeback driver info and allocate memory |
d38ceaf9 AD |
792 | * |
793 | * @adev: amdgpu_device pointer | |
794 | * | |
455a7bc2 | 795 | * Initializes writeback and allocates writeback memory (all asics). |
d38ceaf9 AD |
796 | * Used at driver startup. |
797 | * Returns 0 on success or an -error on failure. | |
798 | */ | |
06ec9070 | 799 | static int amdgpu_device_wb_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
800 | { |
801 | int r; | |
802 | ||
803 | if (adev->wb.wb_obj == NULL) { | |
97407b63 AD |
804 | /* AMDGPU_MAX_WB * sizeof(uint32_t) * 8 = AMDGPU_MAX_WB 256bit slots */ |
805 | r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t) * 8, | |
a76ed485 AD |
806 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, |
807 | &adev->wb.wb_obj, &adev->wb.gpu_addr, | |
808 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
809 | if (r) { |
810 | dev_warn(adev->dev, "(%d) create WB bo failed\n", r); | |
811 | return r; | |
812 | } | |
d38ceaf9 AD |
813 | |
814 | adev->wb.num_wb = AMDGPU_MAX_WB; | |
815 | memset(&adev->wb.used, 0, sizeof(adev->wb.used)); | |
816 | ||
817 | /* clear wb memory */ | |
73469585 | 818 | memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t) * 8); |
d38ceaf9 AD |
819 | } |
820 | ||
821 | return 0; | |
822 | } | |
823 | ||
824 | /** | |
131b4b36 | 825 | * amdgpu_device_wb_get - Allocate a wb entry |
d38ceaf9 AD |
826 | * |
827 | * @adev: amdgpu_device pointer | |
828 | * @wb: wb index | |
829 | * | |
830 | * Allocate a wb slot for use by the driver (all asics). | |
831 | * Returns 0 on success or -EINVAL on failure. | |
832 | */ | |
131b4b36 | 833 | int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb) |
d38ceaf9 AD |
834 | { |
835 | unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb); | |
d38ceaf9 | 836 | |
97407b63 | 837 | if (offset < adev->wb.num_wb) { |
7014285a | 838 | __set_bit(offset, adev->wb.used); |
63ae07ca | 839 | *wb = offset << 3; /* convert to dw offset */ |
0915fdbc ML |
840 | return 0; |
841 | } else { | |
842 | return -EINVAL; | |
843 | } | |
844 | } | |
845 | ||
d38ceaf9 | 846 | /** |
131b4b36 | 847 | * amdgpu_device_wb_free - Free a wb entry |
d38ceaf9 AD |
848 | * |
849 | * @adev: amdgpu_device pointer | |
850 | * @wb: wb index | |
851 | * | |
852 | * Free a wb slot allocated for use by the driver (all asics) | |
853 | */ | |
131b4b36 | 854 | void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb) |
d38ceaf9 | 855 | { |
73469585 | 856 | wb >>= 3; |
d38ceaf9 | 857 | if (wb < adev->wb.num_wb) |
73469585 | 858 | __clear_bit(wb, adev->wb.used); |
d38ceaf9 AD |
859 | } |
860 | ||
d6895ad3 CK |
861 | /** |
862 | * amdgpu_device_resize_fb_bar - try to resize FB BAR | |
863 | * | |
864 | * @adev: amdgpu_device pointer | |
865 | * | |
866 | * Try to resize FB BAR to make all VRAM CPU accessible. We try very hard not | |
867 | * to fail, but if any of the BARs is not accessible after the size we abort | |
868 | * driver loading by returning -ENODEV. | |
869 | */ | |
870 | int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev) | |
871 | { | |
770d13b1 | 872 | u64 space_needed = roundup_pow_of_two(adev->gmc.real_vram_size); |
d6895ad3 | 873 | u32 rbar_size = order_base_2(((space_needed >> 20) | 1)) - 1; |
31b8adab CK |
874 | struct pci_bus *root; |
875 | struct resource *res; | |
876 | unsigned i; | |
d6895ad3 CK |
877 | u16 cmd; |
878 | int r; | |
879 | ||
0c03b912 | 880 | /* Bypass for VF */ |
881 | if (amdgpu_sriov_vf(adev)) | |
882 | return 0; | |
883 | ||
31b8adab CK |
884 | /* Check if the root BUS has 64bit memory resources */ |
885 | root = adev->pdev->bus; | |
886 | while (root->parent) | |
887 | root = root->parent; | |
888 | ||
889 | pci_bus_for_each_resource(root, res, i) { | |
0ebb7c54 | 890 | if (res && res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) && |
31b8adab CK |
891 | res->start > 0x100000000ull) |
892 | break; | |
893 | } | |
894 | ||
895 | /* Trying to resize is pointless without a root hub window above 4GB */ | |
896 | if (!res) | |
897 | return 0; | |
898 | ||
d6895ad3 CK |
899 | /* Disable memory decoding while we change the BAR addresses and size */ |
900 | pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd); | |
901 | pci_write_config_word(adev->pdev, PCI_COMMAND, | |
902 | cmd & ~PCI_COMMAND_MEMORY); | |
903 | ||
904 | /* Free the VRAM and doorbell BAR, we most likely need to move both. */ | |
06ec9070 | 905 | amdgpu_device_doorbell_fini(adev); |
d6895ad3 CK |
906 | if (adev->asic_type >= CHIP_BONAIRE) |
907 | pci_release_resource(adev->pdev, 2); | |
908 | ||
909 | pci_release_resource(adev->pdev, 0); | |
910 | ||
911 | r = pci_resize_resource(adev->pdev, 0, rbar_size); | |
912 | if (r == -ENOSPC) | |
913 | DRM_INFO("Not enough PCI address space for a large BAR."); | |
914 | else if (r && r != -ENOTSUPP) | |
915 | DRM_ERROR("Problem resizing BAR0 (%d).", r); | |
916 | ||
917 | pci_assign_unassigned_bus_resources(adev->pdev->bus); | |
918 | ||
919 | /* When the doorbell or fb BAR isn't available we have no chance of | |
920 | * using the device. | |
921 | */ | |
06ec9070 | 922 | r = amdgpu_device_doorbell_init(adev); |
d6895ad3 CK |
923 | if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET)) |
924 | return -ENODEV; | |
925 | ||
926 | pci_write_config_word(adev->pdev, PCI_COMMAND, cmd); | |
927 | ||
928 | return 0; | |
929 | } | |
a05502e5 | 930 | |
d38ceaf9 AD |
931 | /* |
932 | * GPU helpers function. | |
933 | */ | |
934 | /** | |
39c640c0 | 935 | * amdgpu_device_need_post - check if the hw need post or not |
d38ceaf9 AD |
936 | * |
937 | * @adev: amdgpu_device pointer | |
938 | * | |
c836fec5 JQ |
939 | * Check if the asic has been initialized (all asics) at driver startup |
940 | * or post is needed if hw reset is performed. | |
941 | * Returns true if need or false if not. | |
d38ceaf9 | 942 | */ |
39c640c0 | 943 | bool amdgpu_device_need_post(struct amdgpu_device *adev) |
d38ceaf9 AD |
944 | { |
945 | uint32_t reg; | |
946 | ||
bec86378 ML |
947 | if (amdgpu_sriov_vf(adev)) |
948 | return false; | |
949 | ||
950 | if (amdgpu_passthrough(adev)) { | |
1da2c326 ML |
951 | /* for FIJI: In whole GPU pass-through virtualization case, after VM reboot |
952 | * some old smc fw still need driver do vPost otherwise gpu hang, while | |
953 | * those smc fw version above 22.15 doesn't have this flaw, so we force | |
954 | * vpost executed for smc version below 22.15 | |
bec86378 ML |
955 | */ |
956 | if (adev->asic_type == CHIP_FIJI) { | |
957 | int err; | |
958 | uint32_t fw_ver; | |
959 | err = request_firmware(&adev->pm.fw, "amdgpu/fiji_smc.bin", adev->dev); | |
960 | /* force vPost if error occured */ | |
961 | if (err) | |
962 | return true; | |
963 | ||
964 | fw_ver = *((uint32_t *)adev->pm.fw->data + 69); | |
1da2c326 ML |
965 | if (fw_ver < 0x00160e00) |
966 | return true; | |
bec86378 | 967 | } |
bec86378 | 968 | } |
91fe77eb | 969 | |
970 | if (adev->has_hw_reset) { | |
971 | adev->has_hw_reset = false; | |
972 | return true; | |
973 | } | |
974 | ||
975 | /* bios scratch used on CIK+ */ | |
976 | if (adev->asic_type >= CHIP_BONAIRE) | |
977 | return amdgpu_atombios_scratch_need_asic_init(adev); | |
978 | ||
979 | /* check MEM_SIZE for older asics */ | |
980 | reg = amdgpu_asic_get_config_memsize(adev); | |
981 | ||
982 | if ((reg != 0) && (reg != 0xffffffff)) | |
983 | return false; | |
984 | ||
985 | return true; | |
bec86378 ML |
986 | } |
987 | ||
d38ceaf9 AD |
988 | /* if we get transitioned to only one device, take VGA back */ |
989 | /** | |
06ec9070 | 990 | * amdgpu_device_vga_set_decode - enable/disable vga decode |
d38ceaf9 AD |
991 | * |
992 | * @cookie: amdgpu_device pointer | |
993 | * @state: enable/disable vga decode | |
994 | * | |
995 | * Enable/disable vga decode (all asics). | |
996 | * Returns VGA resource flags. | |
997 | */ | |
06ec9070 | 998 | static unsigned int amdgpu_device_vga_set_decode(void *cookie, bool state) |
d38ceaf9 AD |
999 | { |
1000 | struct amdgpu_device *adev = cookie; | |
1001 | amdgpu_asic_set_vga_state(adev, state); | |
1002 | if (state) | |
1003 | return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM | | |
1004 | VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
1005 | else | |
1006 | return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
1007 | } | |
1008 | ||
e3ecdffa AD |
1009 | /** |
1010 | * amdgpu_device_check_block_size - validate the vm block size | |
1011 | * | |
1012 | * @adev: amdgpu_device pointer | |
1013 | * | |
1014 | * Validates the vm block size specified via module parameter. | |
1015 | * The vm block size defines number of bits in page table versus page directory, | |
1016 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
1017 | * page table and the remaining bits are in the page directory. | |
1018 | */ | |
06ec9070 | 1019 | static void amdgpu_device_check_block_size(struct amdgpu_device *adev) |
a1adf8be CZ |
1020 | { |
1021 | /* defines number of bits in page table versus page directory, | |
1022 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
1023 | * page table and the remaining bits are in the page directory */ | |
bab4fee7 JZ |
1024 | if (amdgpu_vm_block_size == -1) |
1025 | return; | |
a1adf8be | 1026 | |
bab4fee7 | 1027 | if (amdgpu_vm_block_size < 9) { |
a1adf8be CZ |
1028 | dev_warn(adev->dev, "VM page table size (%d) too small\n", |
1029 | amdgpu_vm_block_size); | |
97489129 | 1030 | amdgpu_vm_block_size = -1; |
a1adf8be | 1031 | } |
a1adf8be CZ |
1032 | } |
1033 | ||
e3ecdffa AD |
1034 | /** |
1035 | * amdgpu_device_check_vm_size - validate the vm size | |
1036 | * | |
1037 | * @adev: amdgpu_device pointer | |
1038 | * | |
1039 | * Validates the vm size in GB specified via module parameter. | |
1040 | * The VM size is the size of the GPU virtual memory space in GB. | |
1041 | */ | |
06ec9070 | 1042 | static void amdgpu_device_check_vm_size(struct amdgpu_device *adev) |
83ca145d | 1043 | { |
64dab074 AD |
1044 | /* no need to check the default value */ |
1045 | if (amdgpu_vm_size == -1) | |
1046 | return; | |
1047 | ||
83ca145d ZJ |
1048 | if (amdgpu_vm_size < 1) { |
1049 | dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n", | |
1050 | amdgpu_vm_size); | |
f3368128 | 1051 | amdgpu_vm_size = -1; |
83ca145d | 1052 | } |
83ca145d ZJ |
1053 | } |
1054 | ||
7951e376 RZ |
1055 | static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev) |
1056 | { | |
1057 | struct sysinfo si; | |
a9d4fe2f | 1058 | bool is_os_64 = (sizeof(void *) == 8); |
7951e376 RZ |
1059 | uint64_t total_memory; |
1060 | uint64_t dram_size_seven_GB = 0x1B8000000; | |
1061 | uint64_t dram_size_three_GB = 0xB8000000; | |
1062 | ||
1063 | if (amdgpu_smu_memory_pool_size == 0) | |
1064 | return; | |
1065 | ||
1066 | if (!is_os_64) { | |
1067 | DRM_WARN("Not 64-bit OS, feature not supported\n"); | |
1068 | goto def_value; | |
1069 | } | |
1070 | si_meminfo(&si); | |
1071 | total_memory = (uint64_t)si.totalram * si.mem_unit; | |
1072 | ||
1073 | if ((amdgpu_smu_memory_pool_size == 1) || | |
1074 | (amdgpu_smu_memory_pool_size == 2)) { | |
1075 | if (total_memory < dram_size_three_GB) | |
1076 | goto def_value1; | |
1077 | } else if ((amdgpu_smu_memory_pool_size == 4) || | |
1078 | (amdgpu_smu_memory_pool_size == 8)) { | |
1079 | if (total_memory < dram_size_seven_GB) | |
1080 | goto def_value1; | |
1081 | } else { | |
1082 | DRM_WARN("Smu memory pool size not supported\n"); | |
1083 | goto def_value; | |
1084 | } | |
1085 | adev->pm.smu_prv_buffer_size = amdgpu_smu_memory_pool_size << 28; | |
1086 | ||
1087 | return; | |
1088 | ||
1089 | def_value1: | |
1090 | DRM_WARN("No enough system memory\n"); | |
1091 | def_value: | |
1092 | adev->pm.smu_prv_buffer_size = 0; | |
1093 | } | |
1094 | ||
d38ceaf9 | 1095 | /** |
06ec9070 | 1096 | * amdgpu_device_check_arguments - validate module params |
d38ceaf9 AD |
1097 | * |
1098 | * @adev: amdgpu_device pointer | |
1099 | * | |
1100 | * Validates certain module parameters and updates | |
1101 | * the associated values used by the driver (all asics). | |
1102 | */ | |
912dfc84 | 1103 | static int amdgpu_device_check_arguments(struct amdgpu_device *adev) |
d38ceaf9 | 1104 | { |
5b011235 CZ |
1105 | if (amdgpu_sched_jobs < 4) { |
1106 | dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n", | |
1107 | amdgpu_sched_jobs); | |
1108 | amdgpu_sched_jobs = 4; | |
76117507 | 1109 | } else if (!is_power_of_2(amdgpu_sched_jobs)){ |
5b011235 CZ |
1110 | dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n", |
1111 | amdgpu_sched_jobs); | |
1112 | amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs); | |
1113 | } | |
d38ceaf9 | 1114 | |
83e74db6 | 1115 | if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) { |
f9321cc4 CK |
1116 | /* gart size must be greater or equal to 32M */ |
1117 | dev_warn(adev->dev, "gart size (%d) too small\n", | |
1118 | amdgpu_gart_size); | |
83e74db6 | 1119 | amdgpu_gart_size = -1; |
d38ceaf9 AD |
1120 | } |
1121 | ||
36d38372 | 1122 | if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) { |
c4e1a13a | 1123 | /* gtt size must be greater or equal to 32M */ |
36d38372 CK |
1124 | dev_warn(adev->dev, "gtt size (%d) too small\n", |
1125 | amdgpu_gtt_size); | |
1126 | amdgpu_gtt_size = -1; | |
d38ceaf9 AD |
1127 | } |
1128 | ||
d07f14be RH |
1129 | /* valid range is between 4 and 9 inclusive */ |
1130 | if (amdgpu_vm_fragment_size != -1 && | |
1131 | (amdgpu_vm_fragment_size > 9 || amdgpu_vm_fragment_size < 4)) { | |
1132 | dev_warn(adev->dev, "valid range is between 4 and 9\n"); | |
1133 | amdgpu_vm_fragment_size = -1; | |
1134 | } | |
1135 | ||
7951e376 RZ |
1136 | amdgpu_device_check_smu_prv_buffer_size(adev); |
1137 | ||
06ec9070 | 1138 | amdgpu_device_check_vm_size(adev); |
d38ceaf9 | 1139 | |
06ec9070 | 1140 | amdgpu_device_check_block_size(adev); |
6a7f76e7 | 1141 | |
19aede77 | 1142 | adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type); |
912dfc84 | 1143 | |
c6252390 | 1144 | amdgpu_gmc_tmz_set(adev); |
01a8dcec | 1145 | |
e3c00faa | 1146 | return 0; |
d38ceaf9 AD |
1147 | } |
1148 | ||
1149 | /** | |
1150 | * amdgpu_switcheroo_set_state - set switcheroo state | |
1151 | * | |
1152 | * @pdev: pci dev pointer | |
1694467b | 1153 | * @state: vga_switcheroo state |
d38ceaf9 AD |
1154 | * |
1155 | * Callback for the switcheroo driver. Suspends or resumes the | |
1156 | * the asics before or after it is powered up using ACPI methods. | |
1157 | */ | |
1158 | static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_state state) | |
1159 | { | |
1160 | struct drm_device *dev = pci_get_drvdata(pdev); | |
de185019 | 1161 | int r; |
d38ceaf9 | 1162 | |
31af062a | 1163 | if (amdgpu_device_supports_boco(dev) && state == VGA_SWITCHEROO_OFF) |
d38ceaf9 AD |
1164 | return; |
1165 | ||
1166 | if (state == VGA_SWITCHEROO_ON) { | |
dd4fa6c1 | 1167 | pr_info("switched on\n"); |
d38ceaf9 AD |
1168 | /* don't suspend or resume card normally */ |
1169 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
1170 | ||
de185019 AD |
1171 | pci_set_power_state(dev->pdev, PCI_D0); |
1172 | pci_restore_state(dev->pdev); | |
1173 | r = pci_enable_device(dev->pdev); | |
1174 | if (r) | |
1175 | DRM_WARN("pci_enable_device failed (%d)\n", r); | |
1176 | amdgpu_device_resume(dev, true); | |
d38ceaf9 | 1177 | |
d38ceaf9 AD |
1178 | dev->switch_power_state = DRM_SWITCH_POWER_ON; |
1179 | drm_kms_helper_poll_enable(dev); | |
1180 | } else { | |
dd4fa6c1 | 1181 | pr_info("switched off\n"); |
d38ceaf9 AD |
1182 | drm_kms_helper_poll_disable(dev); |
1183 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
de185019 AD |
1184 | amdgpu_device_suspend(dev, true); |
1185 | pci_save_state(dev->pdev); | |
1186 | /* Shut down the device */ | |
1187 | pci_disable_device(dev->pdev); | |
1188 | pci_set_power_state(dev->pdev, PCI_D3cold); | |
d38ceaf9 AD |
1189 | dev->switch_power_state = DRM_SWITCH_POWER_OFF; |
1190 | } | |
1191 | } | |
1192 | ||
1193 | /** | |
1194 | * amdgpu_switcheroo_can_switch - see if switcheroo state can change | |
1195 | * | |
1196 | * @pdev: pci dev pointer | |
1197 | * | |
1198 | * Callback for the switcheroo driver. Check of the switcheroo | |
1199 | * state can be changed. | |
1200 | * Returns true if the state can be changed, false if not. | |
1201 | */ | |
1202 | static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev) | |
1203 | { | |
1204 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1205 | ||
1206 | /* | |
1207 | * FIXME: open_count is protected by drm_global_mutex but that would lead to | |
1208 | * locking inversion with the driver load path. And the access here is | |
1209 | * completely racy anyway. So don't bother with locking for now. | |
1210 | */ | |
7e13ad89 | 1211 | return atomic_read(&dev->open_count) == 0; |
d38ceaf9 AD |
1212 | } |
1213 | ||
1214 | static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = { | |
1215 | .set_gpu_state = amdgpu_switcheroo_set_state, | |
1216 | .reprobe = NULL, | |
1217 | .can_switch = amdgpu_switcheroo_can_switch, | |
1218 | }; | |
1219 | ||
e3ecdffa AD |
1220 | /** |
1221 | * amdgpu_device_ip_set_clockgating_state - set the CG state | |
1222 | * | |
87e3f136 | 1223 | * @dev: amdgpu_device pointer |
e3ecdffa AD |
1224 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) |
1225 | * @state: clockgating state (gate or ungate) | |
1226 | * | |
1227 | * Sets the requested clockgating state for all instances of | |
1228 | * the hardware IP specified. | |
1229 | * Returns the error code from the last instance. | |
1230 | */ | |
43fa561f | 1231 | int amdgpu_device_ip_set_clockgating_state(void *dev, |
2990a1fc AD |
1232 | enum amd_ip_block_type block_type, |
1233 | enum amd_clockgating_state state) | |
d38ceaf9 | 1234 | { |
43fa561f | 1235 | struct amdgpu_device *adev = dev; |
d38ceaf9 AD |
1236 | int i, r = 0; |
1237 | ||
1238 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1239 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1240 | continue; |
c722865a RZ |
1241 | if (adev->ip_blocks[i].version->type != block_type) |
1242 | continue; | |
1243 | if (!adev->ip_blocks[i].version->funcs->set_clockgating_state) | |
1244 | continue; | |
1245 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state( | |
1246 | (void *)adev, state); | |
1247 | if (r) | |
1248 | DRM_ERROR("set_clockgating_state of IP block <%s> failed %d\n", | |
1249 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1250 | } |
1251 | return r; | |
1252 | } | |
1253 | ||
e3ecdffa AD |
1254 | /** |
1255 | * amdgpu_device_ip_set_powergating_state - set the PG state | |
1256 | * | |
87e3f136 | 1257 | * @dev: amdgpu_device pointer |
e3ecdffa AD |
1258 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) |
1259 | * @state: powergating state (gate or ungate) | |
1260 | * | |
1261 | * Sets the requested powergating state for all instances of | |
1262 | * the hardware IP specified. | |
1263 | * Returns the error code from the last instance. | |
1264 | */ | |
43fa561f | 1265 | int amdgpu_device_ip_set_powergating_state(void *dev, |
2990a1fc AD |
1266 | enum amd_ip_block_type block_type, |
1267 | enum amd_powergating_state state) | |
d38ceaf9 | 1268 | { |
43fa561f | 1269 | struct amdgpu_device *adev = dev; |
d38ceaf9 AD |
1270 | int i, r = 0; |
1271 | ||
1272 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1273 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1274 | continue; |
c722865a RZ |
1275 | if (adev->ip_blocks[i].version->type != block_type) |
1276 | continue; | |
1277 | if (!adev->ip_blocks[i].version->funcs->set_powergating_state) | |
1278 | continue; | |
1279 | r = adev->ip_blocks[i].version->funcs->set_powergating_state( | |
1280 | (void *)adev, state); | |
1281 | if (r) | |
1282 | DRM_ERROR("set_powergating_state of IP block <%s> failed %d\n", | |
1283 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1284 | } |
1285 | return r; | |
1286 | } | |
1287 | ||
e3ecdffa AD |
1288 | /** |
1289 | * amdgpu_device_ip_get_clockgating_state - get the CG state | |
1290 | * | |
1291 | * @adev: amdgpu_device pointer | |
1292 | * @flags: clockgating feature flags | |
1293 | * | |
1294 | * Walks the list of IPs on the device and updates the clockgating | |
1295 | * flags for each IP. | |
1296 | * Updates @flags with the feature flags for each hardware IP where | |
1297 | * clockgating is enabled. | |
1298 | */ | |
2990a1fc AD |
1299 | void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev, |
1300 | u32 *flags) | |
6cb2d4e4 HR |
1301 | { |
1302 | int i; | |
1303 | ||
1304 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1305 | if (!adev->ip_blocks[i].status.valid) | |
1306 | continue; | |
1307 | if (adev->ip_blocks[i].version->funcs->get_clockgating_state) | |
1308 | adev->ip_blocks[i].version->funcs->get_clockgating_state((void *)adev, flags); | |
1309 | } | |
1310 | } | |
1311 | ||
e3ecdffa AD |
1312 | /** |
1313 | * amdgpu_device_ip_wait_for_idle - wait for idle | |
1314 | * | |
1315 | * @adev: amdgpu_device pointer | |
1316 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) | |
1317 | * | |
1318 | * Waits for the request hardware IP to be idle. | |
1319 | * Returns 0 for success or a negative error code on failure. | |
1320 | */ | |
2990a1fc AD |
1321 | int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev, |
1322 | enum amd_ip_block_type block_type) | |
5dbbb60b AD |
1323 | { |
1324 | int i, r; | |
1325 | ||
1326 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1327 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1328 | continue; |
a1255107 AD |
1329 | if (adev->ip_blocks[i].version->type == block_type) { |
1330 | r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev); | |
5dbbb60b AD |
1331 | if (r) |
1332 | return r; | |
1333 | break; | |
1334 | } | |
1335 | } | |
1336 | return 0; | |
1337 | ||
1338 | } | |
1339 | ||
e3ecdffa AD |
1340 | /** |
1341 | * amdgpu_device_ip_is_idle - is the hardware IP idle | |
1342 | * | |
1343 | * @adev: amdgpu_device pointer | |
1344 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) | |
1345 | * | |
1346 | * Check if the hardware IP is idle or not. | |
1347 | * Returns true if it the IP is idle, false if not. | |
1348 | */ | |
2990a1fc AD |
1349 | bool amdgpu_device_ip_is_idle(struct amdgpu_device *adev, |
1350 | enum amd_ip_block_type block_type) | |
5dbbb60b AD |
1351 | { |
1352 | int i; | |
1353 | ||
1354 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1355 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1356 | continue; |
a1255107 AD |
1357 | if (adev->ip_blocks[i].version->type == block_type) |
1358 | return adev->ip_blocks[i].version->funcs->is_idle((void *)adev); | |
5dbbb60b AD |
1359 | } |
1360 | return true; | |
1361 | ||
1362 | } | |
1363 | ||
e3ecdffa AD |
1364 | /** |
1365 | * amdgpu_device_ip_get_ip_block - get a hw IP pointer | |
1366 | * | |
1367 | * @adev: amdgpu_device pointer | |
87e3f136 | 1368 | * @type: Type of hardware IP (SMU, GFX, UVD, etc.) |
e3ecdffa AD |
1369 | * |
1370 | * Returns a pointer to the hardware IP block structure | |
1371 | * if it exists for the asic, otherwise NULL. | |
1372 | */ | |
2990a1fc AD |
1373 | struct amdgpu_ip_block * |
1374 | amdgpu_device_ip_get_ip_block(struct amdgpu_device *adev, | |
1375 | enum amd_ip_block_type type) | |
d38ceaf9 AD |
1376 | { |
1377 | int i; | |
1378 | ||
1379 | for (i = 0; i < adev->num_ip_blocks; i++) | |
a1255107 | 1380 | if (adev->ip_blocks[i].version->type == type) |
d38ceaf9 AD |
1381 | return &adev->ip_blocks[i]; |
1382 | ||
1383 | return NULL; | |
1384 | } | |
1385 | ||
1386 | /** | |
2990a1fc | 1387 | * amdgpu_device_ip_block_version_cmp |
d38ceaf9 AD |
1388 | * |
1389 | * @adev: amdgpu_device pointer | |
5fc3aeeb | 1390 | * @type: enum amd_ip_block_type |
d38ceaf9 AD |
1391 | * @major: major version |
1392 | * @minor: minor version | |
1393 | * | |
1394 | * return 0 if equal or greater | |
1395 | * return 1 if smaller or the ip_block doesn't exist | |
1396 | */ | |
2990a1fc AD |
1397 | int amdgpu_device_ip_block_version_cmp(struct amdgpu_device *adev, |
1398 | enum amd_ip_block_type type, | |
1399 | u32 major, u32 minor) | |
d38ceaf9 | 1400 | { |
2990a1fc | 1401 | struct amdgpu_ip_block *ip_block = amdgpu_device_ip_get_ip_block(adev, type); |
d38ceaf9 | 1402 | |
a1255107 AD |
1403 | if (ip_block && ((ip_block->version->major > major) || |
1404 | ((ip_block->version->major == major) && | |
1405 | (ip_block->version->minor >= minor)))) | |
d38ceaf9 AD |
1406 | return 0; |
1407 | ||
1408 | return 1; | |
1409 | } | |
1410 | ||
a1255107 | 1411 | /** |
2990a1fc | 1412 | * amdgpu_device_ip_block_add |
a1255107 AD |
1413 | * |
1414 | * @adev: amdgpu_device pointer | |
1415 | * @ip_block_version: pointer to the IP to add | |
1416 | * | |
1417 | * Adds the IP block driver information to the collection of IPs | |
1418 | * on the asic. | |
1419 | */ | |
2990a1fc AD |
1420 | int amdgpu_device_ip_block_add(struct amdgpu_device *adev, |
1421 | const struct amdgpu_ip_block_version *ip_block_version) | |
a1255107 AD |
1422 | { |
1423 | if (!ip_block_version) | |
1424 | return -EINVAL; | |
1425 | ||
e966a725 | 1426 | DRM_INFO("add ip block number %d <%s>\n", adev->num_ip_blocks, |
a0bae357 HR |
1427 | ip_block_version->funcs->name); |
1428 | ||
a1255107 AD |
1429 | adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version; |
1430 | ||
1431 | return 0; | |
1432 | } | |
1433 | ||
e3ecdffa AD |
1434 | /** |
1435 | * amdgpu_device_enable_virtual_display - enable virtual display feature | |
1436 | * | |
1437 | * @adev: amdgpu_device pointer | |
1438 | * | |
1439 | * Enabled the virtual display feature if the user has enabled it via | |
1440 | * the module parameter virtual_display. This feature provides a virtual | |
1441 | * display hardware on headless boards or in virtualized environments. | |
1442 | * This function parses and validates the configuration string specified by | |
1443 | * the user and configues the virtual display configuration (number of | |
1444 | * virtual connectors, crtcs, etc.) specified. | |
1445 | */ | |
483ef985 | 1446 | static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev) |
9accf2fd ED |
1447 | { |
1448 | adev->enable_virtual_display = false; | |
1449 | ||
1450 | if (amdgpu_virtual_display) { | |
1451 | struct drm_device *ddev = adev->ddev; | |
1452 | const char *pci_address_name = pci_name(ddev->pdev); | |
0f66356d | 1453 | char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname; |
9accf2fd ED |
1454 | |
1455 | pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL); | |
1456 | pciaddstr_tmp = pciaddstr; | |
0f66356d ED |
1457 | while ((pciaddname_tmp = strsep(&pciaddstr_tmp, ";"))) { |
1458 | pciaddname = strsep(&pciaddname_tmp, ","); | |
967de2a9 YT |
1459 | if (!strcmp("all", pciaddname) |
1460 | || !strcmp(pci_address_name, pciaddname)) { | |
0f66356d ED |
1461 | long num_crtc; |
1462 | int res = -1; | |
1463 | ||
9accf2fd | 1464 | adev->enable_virtual_display = true; |
0f66356d ED |
1465 | |
1466 | if (pciaddname_tmp) | |
1467 | res = kstrtol(pciaddname_tmp, 10, | |
1468 | &num_crtc); | |
1469 | ||
1470 | if (!res) { | |
1471 | if (num_crtc < 1) | |
1472 | num_crtc = 1; | |
1473 | if (num_crtc > 6) | |
1474 | num_crtc = 6; | |
1475 | adev->mode_info.num_crtc = num_crtc; | |
1476 | } else { | |
1477 | adev->mode_info.num_crtc = 1; | |
1478 | } | |
9accf2fd ED |
1479 | break; |
1480 | } | |
1481 | } | |
1482 | ||
0f66356d ED |
1483 | DRM_INFO("virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n", |
1484 | amdgpu_virtual_display, pci_address_name, | |
1485 | adev->enable_virtual_display, adev->mode_info.num_crtc); | |
9accf2fd ED |
1486 | |
1487 | kfree(pciaddstr); | |
1488 | } | |
1489 | } | |
1490 | ||
e3ecdffa AD |
1491 | /** |
1492 | * amdgpu_device_parse_gpu_info_fw - parse gpu info firmware | |
1493 | * | |
1494 | * @adev: amdgpu_device pointer | |
1495 | * | |
1496 | * Parses the asic configuration parameters specified in the gpu info | |
1497 | * firmware and makes them availale to the driver for use in configuring | |
1498 | * the asic. | |
1499 | * Returns 0 on success, -EINVAL on failure. | |
1500 | */ | |
e2a75f88 AD |
1501 | static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) |
1502 | { | |
e2a75f88 AD |
1503 | const char *chip_name; |
1504 | char fw_name[30]; | |
1505 | int err; | |
1506 | const struct gpu_info_firmware_header_v1_0 *hdr; | |
1507 | ||
ab4fe3e1 HR |
1508 | adev->firmware.gpu_info_fw = NULL; |
1509 | ||
e2a75f88 AD |
1510 | switch (adev->asic_type) { |
1511 | case CHIP_TOPAZ: | |
1512 | case CHIP_TONGA: | |
1513 | case CHIP_FIJI: | |
e2a75f88 | 1514 | case CHIP_POLARIS10: |
cc07f18d | 1515 | case CHIP_POLARIS11: |
e2a75f88 | 1516 | case CHIP_POLARIS12: |
cc07f18d | 1517 | case CHIP_VEGAM: |
e2a75f88 AD |
1518 | case CHIP_CARRIZO: |
1519 | case CHIP_STONEY: | |
1520 | #ifdef CONFIG_DRM_AMDGPU_SI | |
1521 | case CHIP_VERDE: | |
1522 | case CHIP_TAHITI: | |
1523 | case CHIP_PITCAIRN: | |
1524 | case CHIP_OLAND: | |
1525 | case CHIP_HAINAN: | |
1526 | #endif | |
1527 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
1528 | case CHIP_BONAIRE: | |
1529 | case CHIP_HAWAII: | |
1530 | case CHIP_KAVERI: | |
1531 | case CHIP_KABINI: | |
1532 | case CHIP_MULLINS: | |
1533 | #endif | |
27c0bc71 | 1534 | case CHIP_VEGA20: |
e2a75f88 AD |
1535 | default: |
1536 | return 0; | |
1537 | case CHIP_VEGA10: | |
1538 | chip_name = "vega10"; | |
1539 | break; | |
3f76dced AD |
1540 | case CHIP_VEGA12: |
1541 | chip_name = "vega12"; | |
1542 | break; | |
2d2e5e7e | 1543 | case CHIP_RAVEN: |
54c4d17e FX |
1544 | if (adev->rev_id >= 8) |
1545 | chip_name = "raven2"; | |
741deade AD |
1546 | else if (adev->pdev->device == 0x15d8) |
1547 | chip_name = "picasso"; | |
54c4d17e FX |
1548 | else |
1549 | chip_name = "raven"; | |
2d2e5e7e | 1550 | break; |
65e60f6e LM |
1551 | case CHIP_ARCTURUS: |
1552 | chip_name = "arcturus"; | |
1553 | break; | |
b51a26a0 HR |
1554 | case CHIP_RENOIR: |
1555 | chip_name = "renoir"; | |
1556 | break; | |
23c6268e HR |
1557 | case CHIP_NAVI10: |
1558 | chip_name = "navi10"; | |
1559 | break; | |
ed42cfe1 XY |
1560 | case CHIP_NAVI14: |
1561 | chip_name = "navi14"; | |
1562 | break; | |
42b325e5 XY |
1563 | case CHIP_NAVI12: |
1564 | chip_name = "navi12"; | |
1565 | break; | |
e2a75f88 AD |
1566 | } |
1567 | ||
1568 | snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name); | |
ab4fe3e1 | 1569 | err = request_firmware(&adev->firmware.gpu_info_fw, fw_name, adev->dev); |
e2a75f88 AD |
1570 | if (err) { |
1571 | dev_err(adev->dev, | |
1572 | "Failed to load gpu_info firmware \"%s\"\n", | |
1573 | fw_name); | |
1574 | goto out; | |
1575 | } | |
ab4fe3e1 | 1576 | err = amdgpu_ucode_validate(adev->firmware.gpu_info_fw); |
e2a75f88 AD |
1577 | if (err) { |
1578 | dev_err(adev->dev, | |
1579 | "Failed to validate gpu_info firmware \"%s\"\n", | |
1580 | fw_name); | |
1581 | goto out; | |
1582 | } | |
1583 | ||
ab4fe3e1 | 1584 | hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data; |
e2a75f88 AD |
1585 | amdgpu_ucode_print_gpu_info_hdr(&hdr->header); |
1586 | ||
1587 | switch (hdr->version_major) { | |
1588 | case 1: | |
1589 | { | |
1590 | const struct gpu_info_firmware_v1_0 *gpu_info_fw = | |
ab4fe3e1 | 1591 | (const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data + |
e2a75f88 AD |
1592 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); |
1593 | ||
ec51d3fa XY |
1594 | if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10) |
1595 | goto parse_soc_bounding_box; | |
1596 | ||
b5ab16bf AD |
1597 | adev->gfx.config.max_shader_engines = le32_to_cpu(gpu_info_fw->gc_num_se); |
1598 | adev->gfx.config.max_cu_per_sh = le32_to_cpu(gpu_info_fw->gc_num_cu_per_sh); | |
1599 | adev->gfx.config.max_sh_per_se = le32_to_cpu(gpu_info_fw->gc_num_sh_per_se); | |
1600 | adev->gfx.config.max_backends_per_se = le32_to_cpu(gpu_info_fw->gc_num_rb_per_se); | |
e2a75f88 | 1601 | adev->gfx.config.max_texture_channel_caches = |
b5ab16bf AD |
1602 | le32_to_cpu(gpu_info_fw->gc_num_tccs); |
1603 | adev->gfx.config.max_gprs = le32_to_cpu(gpu_info_fw->gc_num_gprs); | |
1604 | adev->gfx.config.max_gs_threads = le32_to_cpu(gpu_info_fw->gc_num_max_gs_thds); | |
1605 | adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gpu_info_fw->gc_gs_table_depth); | |
1606 | adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gpu_info_fw->gc_gsprim_buff_depth); | |
e2a75f88 | 1607 | adev->gfx.config.double_offchip_lds_buf = |
b5ab16bf AD |
1608 | le32_to_cpu(gpu_info_fw->gc_double_offchip_lds_buffer); |
1609 | adev->gfx.cu_info.wave_front_size = le32_to_cpu(gpu_info_fw->gc_wave_size); | |
51fd0370 HZ |
1610 | adev->gfx.cu_info.max_waves_per_simd = |
1611 | le32_to_cpu(gpu_info_fw->gc_max_waves_per_simd); | |
1612 | adev->gfx.cu_info.max_scratch_slots_per_cu = | |
1613 | le32_to_cpu(gpu_info_fw->gc_max_scratch_slots_per_cu); | |
1614 | adev->gfx.cu_info.lds_size = le32_to_cpu(gpu_info_fw->gc_lds_size); | |
48321c3d | 1615 | if (hdr->version_minor >= 1) { |
35c2e910 HZ |
1616 | const struct gpu_info_firmware_v1_1 *gpu_info_fw = |
1617 | (const struct gpu_info_firmware_v1_1 *)(adev->firmware.gpu_info_fw->data + | |
1618 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); | |
1619 | adev->gfx.config.num_sc_per_sh = | |
1620 | le32_to_cpu(gpu_info_fw->num_sc_per_sh); | |
1621 | adev->gfx.config.num_packer_per_sc = | |
1622 | le32_to_cpu(gpu_info_fw->num_packer_per_sc); | |
1623 | } | |
ec51d3fa XY |
1624 | |
1625 | parse_soc_bounding_box: | |
ec51d3fa XY |
1626 | /* |
1627 | * soc bounding box info is not integrated in disocovery table, | |
1628 | * we always need to parse it from gpu info firmware. | |
1629 | */ | |
48321c3d HW |
1630 | if (hdr->version_minor == 2) { |
1631 | const struct gpu_info_firmware_v1_2 *gpu_info_fw = | |
1632 | (const struct gpu_info_firmware_v1_2 *)(adev->firmware.gpu_info_fw->data + | |
1633 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); | |
1634 | adev->dm.soc_bounding_box = &gpu_info_fw->soc_bounding_box; | |
1635 | } | |
e2a75f88 AD |
1636 | break; |
1637 | } | |
1638 | default: | |
1639 | dev_err(adev->dev, | |
1640 | "Unsupported gpu_info table %d\n", hdr->header.ucode_version); | |
1641 | err = -EINVAL; | |
1642 | goto out; | |
1643 | } | |
1644 | out: | |
e2a75f88 AD |
1645 | return err; |
1646 | } | |
1647 | ||
e3ecdffa AD |
1648 | /** |
1649 | * amdgpu_device_ip_early_init - run early init for hardware IPs | |
1650 | * | |
1651 | * @adev: amdgpu_device pointer | |
1652 | * | |
1653 | * Early initialization pass for hardware IPs. The hardware IPs that make | |
1654 | * up each asic are discovered each IP's early_init callback is run. This | |
1655 | * is the first stage in initializing the asic. | |
1656 | * Returns 0 on success, negative error code on failure. | |
1657 | */ | |
06ec9070 | 1658 | static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) |
d38ceaf9 | 1659 | { |
aaa36a97 | 1660 | int i, r; |
d38ceaf9 | 1661 | |
483ef985 | 1662 | amdgpu_device_enable_virtual_display(adev); |
a6be7570 | 1663 | |
d38ceaf9 | 1664 | switch (adev->asic_type) { |
aaa36a97 AD |
1665 | case CHIP_TOPAZ: |
1666 | case CHIP_TONGA: | |
48299f95 | 1667 | case CHIP_FIJI: |
2cc0c0b5 | 1668 | case CHIP_POLARIS10: |
32cc7e53 | 1669 | case CHIP_POLARIS11: |
c4642a47 | 1670 | case CHIP_POLARIS12: |
32cc7e53 | 1671 | case CHIP_VEGAM: |
aaa36a97 | 1672 | case CHIP_CARRIZO: |
39bb0c92 SL |
1673 | case CHIP_STONEY: |
1674 | if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) | |
aaa36a97 AD |
1675 | adev->family = AMDGPU_FAMILY_CZ; |
1676 | else | |
1677 | adev->family = AMDGPU_FAMILY_VI; | |
1678 | ||
1679 | r = vi_set_ip_blocks(adev); | |
1680 | if (r) | |
1681 | return r; | |
1682 | break; | |
33f34802 KW |
1683 | #ifdef CONFIG_DRM_AMDGPU_SI |
1684 | case CHIP_VERDE: | |
1685 | case CHIP_TAHITI: | |
1686 | case CHIP_PITCAIRN: | |
1687 | case CHIP_OLAND: | |
1688 | case CHIP_HAINAN: | |
295d0daf | 1689 | adev->family = AMDGPU_FAMILY_SI; |
33f34802 KW |
1690 | r = si_set_ip_blocks(adev); |
1691 | if (r) | |
1692 | return r; | |
1693 | break; | |
1694 | #endif | |
a2e73f56 AD |
1695 | #ifdef CONFIG_DRM_AMDGPU_CIK |
1696 | case CHIP_BONAIRE: | |
1697 | case CHIP_HAWAII: | |
1698 | case CHIP_KAVERI: | |
1699 | case CHIP_KABINI: | |
1700 | case CHIP_MULLINS: | |
1701 | if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII)) | |
1702 | adev->family = AMDGPU_FAMILY_CI; | |
1703 | else | |
1704 | adev->family = AMDGPU_FAMILY_KV; | |
1705 | ||
1706 | r = cik_set_ip_blocks(adev); | |
1707 | if (r) | |
1708 | return r; | |
1709 | break; | |
1710 | #endif | |
e48a3cd9 AD |
1711 | case CHIP_VEGA10: |
1712 | case CHIP_VEGA12: | |
e4bd8170 | 1713 | case CHIP_VEGA20: |
e48a3cd9 | 1714 | case CHIP_RAVEN: |
61cf44c1 | 1715 | case CHIP_ARCTURUS: |
b51a26a0 HR |
1716 | case CHIP_RENOIR: |
1717 | if (adev->asic_type == CHIP_RAVEN || | |
1718 | adev->asic_type == CHIP_RENOIR) | |
2ca8a5d2 CZ |
1719 | adev->family = AMDGPU_FAMILY_RV; |
1720 | else | |
1721 | adev->family = AMDGPU_FAMILY_AI; | |
460826e6 KW |
1722 | |
1723 | r = soc15_set_ip_blocks(adev); | |
1724 | if (r) | |
1725 | return r; | |
1726 | break; | |
0a5b8c7b | 1727 | case CHIP_NAVI10: |
7ecb5cd4 | 1728 | case CHIP_NAVI14: |
4808cf9c | 1729 | case CHIP_NAVI12: |
0a5b8c7b HR |
1730 | adev->family = AMDGPU_FAMILY_NV; |
1731 | ||
1732 | r = nv_set_ip_blocks(adev); | |
1733 | if (r) | |
1734 | return r; | |
1735 | break; | |
d38ceaf9 AD |
1736 | default: |
1737 | /* FIXME: not supported yet */ | |
1738 | return -EINVAL; | |
1739 | } | |
1740 | ||
e2a75f88 AD |
1741 | r = amdgpu_device_parse_gpu_info_fw(adev); |
1742 | if (r) | |
1743 | return r; | |
1744 | ||
ec51d3fa XY |
1745 | if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10) |
1746 | amdgpu_discovery_get_gfx_info(adev); | |
1747 | ||
1884734a | 1748 | amdgpu_amdkfd_device_probe(adev); |
1749 | ||
3149d9da | 1750 | if (amdgpu_sriov_vf(adev)) { |
122078de ML |
1751 | /* handle vbios stuff prior full access mode for new handshake */ |
1752 | if (adev->virt.req_init_data_ver == 1) { | |
1753 | if (!amdgpu_get_bios(adev)) { | |
1754 | DRM_ERROR("failed to get vbios\n"); | |
1755 | return -EINVAL; | |
1756 | } | |
1757 | ||
1758 | r = amdgpu_atombios_init(adev); | |
1759 | if (r) { | |
1760 | dev_err(adev->dev, "amdgpu_atombios_init failed\n"); | |
1761 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0); | |
1762 | return r; | |
1763 | } | |
1764 | } | |
2f294132 | 1765 | } |
122078de | 1766 | |
2f294132 ML |
1767 | /* we need to send REQ_GPU here for legacy handshaker otherwise the vbios |
1768 | * will not be prepared by host for this VF */ | |
1769 | if (amdgpu_sriov_vf(adev) && adev->virt.req_init_data_ver < 1) { | |
3149d9da XY |
1770 | r = amdgpu_virt_request_full_gpu(adev, true); |
1771 | if (r) | |
2f294132 | 1772 | return r; |
3149d9da XY |
1773 | } |
1774 | ||
3b94fb10 | 1775 | adev->pm.pp_feature = amdgpu_pp_feature_mask; |
a35ad98b | 1776 | if (amdgpu_sriov_vf(adev) || sched_policy == KFD_SCHED_POLICY_NO_HWS) |
00544006 | 1777 | adev->pm.pp_feature &= ~PP_GFXOFF_MASK; |
00f54b97 | 1778 | |
d38ceaf9 AD |
1779 | for (i = 0; i < adev->num_ip_blocks; i++) { |
1780 | if ((amdgpu_ip_block_mask & (1 << i)) == 0) { | |
ed8cf00c HR |
1781 | DRM_ERROR("disabled ip block: %d <%s>\n", |
1782 | i, adev->ip_blocks[i].version->funcs->name); | |
a1255107 | 1783 | adev->ip_blocks[i].status.valid = false; |
d38ceaf9 | 1784 | } else { |
a1255107 AD |
1785 | if (adev->ip_blocks[i].version->funcs->early_init) { |
1786 | r = adev->ip_blocks[i].version->funcs->early_init((void *)adev); | |
2c1a2784 | 1787 | if (r == -ENOENT) { |
a1255107 | 1788 | adev->ip_blocks[i].status.valid = false; |
2c1a2784 | 1789 | } else if (r) { |
a1255107 AD |
1790 | DRM_ERROR("early_init of IP block <%s> failed %d\n", |
1791 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 1792 | return r; |
2c1a2784 | 1793 | } else { |
a1255107 | 1794 | adev->ip_blocks[i].status.valid = true; |
2c1a2784 | 1795 | } |
974e6b64 | 1796 | } else { |
a1255107 | 1797 | adev->ip_blocks[i].status.valid = true; |
d38ceaf9 | 1798 | } |
d38ceaf9 | 1799 | } |
21a249ca AD |
1800 | /* get the vbios after the asic_funcs are set up */ |
1801 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON) { | |
122078de ML |
1802 | /* skip vbios handling for new handshake */ |
1803 | if (amdgpu_sriov_vf(adev) && adev->virt.req_init_data_ver == 1) | |
1804 | continue; | |
1805 | ||
21a249ca AD |
1806 | /* Read BIOS */ |
1807 | if (!amdgpu_get_bios(adev)) | |
1808 | return -EINVAL; | |
1809 | ||
1810 | r = amdgpu_atombios_init(adev); | |
1811 | if (r) { | |
1812 | dev_err(adev->dev, "amdgpu_atombios_init failed\n"); | |
1813 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0); | |
1814 | return r; | |
1815 | } | |
1816 | } | |
d38ceaf9 AD |
1817 | } |
1818 | ||
395d1fb9 NH |
1819 | adev->cg_flags &= amdgpu_cg_mask; |
1820 | adev->pg_flags &= amdgpu_pg_mask; | |
1821 | ||
d38ceaf9 AD |
1822 | return 0; |
1823 | } | |
1824 | ||
0a4f2520 RZ |
1825 | static int amdgpu_device_ip_hw_init_phase1(struct amdgpu_device *adev) |
1826 | { | |
1827 | int i, r; | |
1828 | ||
1829 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1830 | if (!adev->ip_blocks[i].status.sw) | |
1831 | continue; | |
1832 | if (adev->ip_blocks[i].status.hw) | |
1833 | continue; | |
1834 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || | |
2d11fd3f | 1835 | (amdgpu_sriov_vf(adev) && (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP)) || |
0a4f2520 RZ |
1836 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) { |
1837 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
1838 | if (r) { | |
1839 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
1840 | adev->ip_blocks[i].version->funcs->name, r); | |
1841 | return r; | |
1842 | } | |
1843 | adev->ip_blocks[i].status.hw = true; | |
1844 | } | |
1845 | } | |
1846 | ||
1847 | return 0; | |
1848 | } | |
1849 | ||
1850 | static int amdgpu_device_ip_hw_init_phase2(struct amdgpu_device *adev) | |
1851 | { | |
1852 | int i, r; | |
1853 | ||
1854 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1855 | if (!adev->ip_blocks[i].status.sw) | |
1856 | continue; | |
1857 | if (adev->ip_blocks[i].status.hw) | |
1858 | continue; | |
1859 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
1860 | if (r) { | |
1861 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
1862 | adev->ip_blocks[i].version->funcs->name, r); | |
1863 | return r; | |
1864 | } | |
1865 | adev->ip_blocks[i].status.hw = true; | |
1866 | } | |
1867 | ||
1868 | return 0; | |
1869 | } | |
1870 | ||
7a3e0bb2 RZ |
1871 | static int amdgpu_device_fw_loading(struct amdgpu_device *adev) |
1872 | { | |
1873 | int r = 0; | |
1874 | int i; | |
80f41f84 | 1875 | uint32_t smu_version; |
7a3e0bb2 RZ |
1876 | |
1877 | if (adev->asic_type >= CHIP_VEGA10) { | |
1878 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
482f0e53 ML |
1879 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_PSP) |
1880 | continue; | |
1881 | ||
1882 | /* no need to do the fw loading again if already done*/ | |
1883 | if (adev->ip_blocks[i].status.hw == true) | |
1884 | break; | |
1885 | ||
1886 | if (adev->in_gpu_reset || adev->in_suspend) { | |
1887 | r = adev->ip_blocks[i].version->funcs->resume(adev); | |
1888 | if (r) { | |
1889 | DRM_ERROR("resume of IP block <%s> failed %d\n", | |
7a3e0bb2 | 1890 | adev->ip_blocks[i].version->funcs->name, r); |
482f0e53 ML |
1891 | return r; |
1892 | } | |
1893 | } else { | |
1894 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
1895 | if (r) { | |
1896 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
1897 | adev->ip_blocks[i].version->funcs->name, r); | |
1898 | return r; | |
7a3e0bb2 | 1899 | } |
7a3e0bb2 | 1900 | } |
482f0e53 ML |
1901 | |
1902 | adev->ip_blocks[i].status.hw = true; | |
1903 | break; | |
7a3e0bb2 RZ |
1904 | } |
1905 | } | |
482f0e53 | 1906 | |
8973d9ec ED |
1907 | if (!amdgpu_sriov_vf(adev) || adev->asic_type == CHIP_TONGA) |
1908 | r = amdgpu_pm_load_smu_firmware(adev, &smu_version); | |
7a3e0bb2 | 1909 | |
80f41f84 | 1910 | return r; |
7a3e0bb2 RZ |
1911 | } |
1912 | ||
e3ecdffa AD |
1913 | /** |
1914 | * amdgpu_device_ip_init - run init for hardware IPs | |
1915 | * | |
1916 | * @adev: amdgpu_device pointer | |
1917 | * | |
1918 | * Main initialization pass for hardware IPs. The list of all the hardware | |
1919 | * IPs that make up the asic is walked and the sw_init and hw_init callbacks | |
1920 | * are run. sw_init initializes the software state associated with each IP | |
1921 | * and hw_init initializes the hardware associated with each IP. | |
1922 | * Returns 0 on success, negative error code on failure. | |
1923 | */ | |
06ec9070 | 1924 | static int amdgpu_device_ip_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
1925 | { |
1926 | int i, r; | |
1927 | ||
c030f2e4 | 1928 | r = amdgpu_ras_init(adev); |
1929 | if (r) | |
1930 | return r; | |
1931 | ||
2f294132 ML |
1932 | if (amdgpu_sriov_vf(adev) && adev->virt.req_init_data_ver > 0) { |
1933 | r = amdgpu_virt_request_full_gpu(adev, true); | |
1934 | if (r) | |
1935 | return -EAGAIN; | |
1936 | } | |
1937 | ||
d38ceaf9 | 1938 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 1939 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 1940 | continue; |
a1255107 | 1941 | r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev); |
2c1a2784 | 1942 | if (r) { |
a1255107 AD |
1943 | DRM_ERROR("sw_init of IP block <%s> failed %d\n", |
1944 | adev->ip_blocks[i].version->funcs->name, r); | |
72d3f592 | 1945 | goto init_failed; |
2c1a2784 | 1946 | } |
a1255107 | 1947 | adev->ip_blocks[i].status.sw = true; |
bfca0289 | 1948 | |
d38ceaf9 | 1949 | /* need to do gmc hw init early so we can allocate gpu mem */ |
a1255107 | 1950 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { |
06ec9070 | 1951 | r = amdgpu_device_vram_scratch_init(adev); |
2c1a2784 AD |
1952 | if (r) { |
1953 | DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r); | |
72d3f592 | 1954 | goto init_failed; |
2c1a2784 | 1955 | } |
a1255107 | 1956 | r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); |
2c1a2784 AD |
1957 | if (r) { |
1958 | DRM_ERROR("hw_init %d failed %d\n", i, r); | |
72d3f592 | 1959 | goto init_failed; |
2c1a2784 | 1960 | } |
06ec9070 | 1961 | r = amdgpu_device_wb_init(adev); |
2c1a2784 | 1962 | if (r) { |
06ec9070 | 1963 | DRM_ERROR("amdgpu_device_wb_init failed %d\n", r); |
72d3f592 | 1964 | goto init_failed; |
2c1a2784 | 1965 | } |
a1255107 | 1966 | adev->ip_blocks[i].status.hw = true; |
2493664f ML |
1967 | |
1968 | /* right after GMC hw init, we create CSA */ | |
f92d5c61 | 1969 | if (amdgpu_mcbp || amdgpu_sriov_vf(adev)) { |
1e256e27 RZ |
1970 | r = amdgpu_allocate_static_csa(adev, &adev->virt.csa_obj, |
1971 | AMDGPU_GEM_DOMAIN_VRAM, | |
1972 | AMDGPU_CSA_SIZE); | |
2493664f ML |
1973 | if (r) { |
1974 | DRM_ERROR("allocate CSA failed %d\n", r); | |
72d3f592 | 1975 | goto init_failed; |
2493664f ML |
1976 | } |
1977 | } | |
d38ceaf9 AD |
1978 | } |
1979 | } | |
1980 | ||
c9ffa427 YT |
1981 | if (amdgpu_sriov_vf(adev)) |
1982 | amdgpu_virt_init_data_exchange(adev); | |
1983 | ||
533aed27 AG |
1984 | r = amdgpu_ib_pool_init(adev); |
1985 | if (r) { | |
1986 | dev_err(adev->dev, "IB initialization failed (%d).\n", r); | |
1987 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r); | |
1988 | goto init_failed; | |
1989 | } | |
1990 | ||
c8963ea4 RZ |
1991 | r = amdgpu_ucode_create_bo(adev); /* create ucode bo when sw_init complete*/ |
1992 | if (r) | |
72d3f592 | 1993 | goto init_failed; |
0a4f2520 RZ |
1994 | |
1995 | r = amdgpu_device_ip_hw_init_phase1(adev); | |
1996 | if (r) | |
72d3f592 | 1997 | goto init_failed; |
0a4f2520 | 1998 | |
7a3e0bb2 RZ |
1999 | r = amdgpu_device_fw_loading(adev); |
2000 | if (r) | |
72d3f592 | 2001 | goto init_failed; |
7a3e0bb2 | 2002 | |
0a4f2520 RZ |
2003 | r = amdgpu_device_ip_hw_init_phase2(adev); |
2004 | if (r) | |
72d3f592 | 2005 | goto init_failed; |
d38ceaf9 | 2006 | |
121a2bc6 AG |
2007 | /* |
2008 | * retired pages will be loaded from eeprom and reserved here, | |
2009 | * it should be called after amdgpu_device_ip_hw_init_phase2 since | |
2010 | * for some ASICs the RAS EEPROM code relies on SMU fully functioning | |
2011 | * for I2C communication which only true at this point. | |
2012 | * recovery_init may fail, but it can free all resources allocated by | |
2013 | * itself and its failure should not stop amdgpu init process. | |
2014 | * | |
2015 | * Note: theoretically, this should be called before all vram allocations | |
2016 | * to protect retired page from abusing | |
2017 | */ | |
2018 | amdgpu_ras_recovery_init(adev); | |
2019 | ||
3e2e2ab5 HZ |
2020 | if (adev->gmc.xgmi.num_physical_nodes > 1) |
2021 | amdgpu_xgmi_add_device(adev); | |
1884734a | 2022 | amdgpu_amdkfd_device_init(adev); |
c6332b97 | 2023 | |
bd607166 KR |
2024 | amdgpu_fru_get_product_info(adev); |
2025 | ||
72d3f592 | 2026 | init_failed: |
c9ffa427 | 2027 | if (amdgpu_sriov_vf(adev)) |
c6332b97 | 2028 | amdgpu_virt_release_full_gpu(adev, true); |
2029 | ||
72d3f592 | 2030 | return r; |
d38ceaf9 AD |
2031 | } |
2032 | ||
e3ecdffa AD |
2033 | /** |
2034 | * amdgpu_device_fill_reset_magic - writes reset magic to gart pointer | |
2035 | * | |
2036 | * @adev: amdgpu_device pointer | |
2037 | * | |
2038 | * Writes a reset magic value to the gart pointer in VRAM. The driver calls | |
2039 | * this function before a GPU reset. If the value is retained after a | |
2040 | * GPU reset, VRAM has not been lost. Some GPU resets may destry VRAM contents. | |
2041 | */ | |
06ec9070 | 2042 | static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev) |
0c49e0b8 CZ |
2043 | { |
2044 | memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM); | |
2045 | } | |
2046 | ||
e3ecdffa AD |
2047 | /** |
2048 | * amdgpu_device_check_vram_lost - check if vram is valid | |
2049 | * | |
2050 | * @adev: amdgpu_device pointer | |
2051 | * | |
2052 | * Checks the reset magic value written to the gart pointer in VRAM. | |
2053 | * The driver calls this after a GPU reset to see if the contents of | |
2054 | * VRAM is lost or now. | |
2055 | * returns true if vram is lost, false if not. | |
2056 | */ | |
06ec9070 | 2057 | static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev) |
0c49e0b8 | 2058 | { |
dadce777 EQ |
2059 | if (memcmp(adev->gart.ptr, adev->reset_magic, |
2060 | AMDGPU_RESET_MAGIC_NUM)) | |
2061 | return true; | |
2062 | ||
2063 | if (!adev->in_gpu_reset) | |
2064 | return false; | |
2065 | ||
2066 | /* | |
2067 | * For all ASICs with baco/mode1 reset, the VRAM is | |
2068 | * always assumed to be lost. | |
2069 | */ | |
2070 | switch (amdgpu_asic_reset_method(adev)) { | |
2071 | case AMD_RESET_METHOD_BACO: | |
2072 | case AMD_RESET_METHOD_MODE1: | |
2073 | return true; | |
2074 | default: | |
2075 | return false; | |
2076 | } | |
0c49e0b8 CZ |
2077 | } |
2078 | ||
e3ecdffa | 2079 | /** |
1112a46b | 2080 | * amdgpu_device_set_cg_state - set clockgating for amdgpu device |
e3ecdffa AD |
2081 | * |
2082 | * @adev: amdgpu_device pointer | |
b8b72130 | 2083 | * @state: clockgating state (gate or ungate) |
e3ecdffa | 2084 | * |
e3ecdffa | 2085 | * The list of all the hardware IPs that make up the asic is walked and the |
1112a46b RZ |
2086 | * set_clockgating_state callbacks are run. |
2087 | * Late initialization pass enabling clockgating for hardware IPs. | |
2088 | * Fini or suspend, pass disabling clockgating for hardware IPs. | |
e3ecdffa AD |
2089 | * Returns 0 on success, negative error code on failure. |
2090 | */ | |
fdd34271 | 2091 | |
1112a46b RZ |
2092 | static int amdgpu_device_set_cg_state(struct amdgpu_device *adev, |
2093 | enum amd_clockgating_state state) | |
d38ceaf9 | 2094 | { |
1112a46b | 2095 | int i, j, r; |
d38ceaf9 | 2096 | |
4a2ba394 SL |
2097 | if (amdgpu_emu_mode == 1) |
2098 | return 0; | |
2099 | ||
1112a46b RZ |
2100 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2101 | i = state == AMD_CG_STATE_GATE ? j : adev->num_ip_blocks - j - 1; | |
a2d31dc3 | 2102 | if (!adev->ip_blocks[i].status.late_initialized) |
d38ceaf9 | 2103 | continue; |
4a446d55 | 2104 | /* skip CG for VCE/UVD, it's handled specially */ |
a1255107 | 2105 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && |
57716327 | 2106 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE && |
34319b32 | 2107 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN && |
52f2e779 | 2108 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG && |
57716327 | 2109 | adev->ip_blocks[i].version->funcs->set_clockgating_state) { |
4a446d55 | 2110 | /* enable clockgating to save power */ |
a1255107 | 2111 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev, |
1112a46b | 2112 | state); |
4a446d55 AD |
2113 | if (r) { |
2114 | DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n", | |
a1255107 | 2115 | adev->ip_blocks[i].version->funcs->name, r); |
4a446d55 AD |
2116 | return r; |
2117 | } | |
b0b00ff1 | 2118 | } |
d38ceaf9 | 2119 | } |
06b18f61 | 2120 | |
c9f96fd5 RZ |
2121 | return 0; |
2122 | } | |
2123 | ||
1112a46b | 2124 | static int amdgpu_device_set_pg_state(struct amdgpu_device *adev, enum amd_powergating_state state) |
c9f96fd5 | 2125 | { |
1112a46b | 2126 | int i, j, r; |
06b18f61 | 2127 | |
c9f96fd5 RZ |
2128 | if (amdgpu_emu_mode == 1) |
2129 | return 0; | |
2130 | ||
1112a46b RZ |
2131 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2132 | i = state == AMD_PG_STATE_GATE ? j : adev->num_ip_blocks - j - 1; | |
a2d31dc3 | 2133 | if (!adev->ip_blocks[i].status.late_initialized) |
c9f96fd5 RZ |
2134 | continue; |
2135 | /* skip CG for VCE/UVD, it's handled specially */ | |
2136 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && | |
2137 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE && | |
2138 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN && | |
52f2e779 | 2139 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG && |
c9f96fd5 RZ |
2140 | adev->ip_blocks[i].version->funcs->set_powergating_state) { |
2141 | /* enable powergating to save power */ | |
2142 | r = adev->ip_blocks[i].version->funcs->set_powergating_state((void *)adev, | |
1112a46b | 2143 | state); |
c9f96fd5 RZ |
2144 | if (r) { |
2145 | DRM_ERROR("set_powergating_state(gate) of IP block <%s> failed %d\n", | |
2146 | adev->ip_blocks[i].version->funcs->name, r); | |
2147 | return r; | |
2148 | } | |
2149 | } | |
2150 | } | |
2dc80b00 S |
2151 | return 0; |
2152 | } | |
2153 | ||
beff74bc AD |
2154 | static int amdgpu_device_enable_mgpu_fan_boost(void) |
2155 | { | |
2156 | struct amdgpu_gpu_instance *gpu_ins; | |
2157 | struct amdgpu_device *adev; | |
2158 | int i, ret = 0; | |
2159 | ||
2160 | mutex_lock(&mgpu_info.mutex); | |
2161 | ||
2162 | /* | |
2163 | * MGPU fan boost feature should be enabled | |
2164 | * only when there are two or more dGPUs in | |
2165 | * the system | |
2166 | */ | |
2167 | if (mgpu_info.num_dgpu < 2) | |
2168 | goto out; | |
2169 | ||
2170 | for (i = 0; i < mgpu_info.num_dgpu; i++) { | |
2171 | gpu_ins = &(mgpu_info.gpu_ins[i]); | |
2172 | adev = gpu_ins->adev; | |
2173 | if (!(adev->flags & AMD_IS_APU) && | |
2174 | !gpu_ins->mgpu_fan_enabled && | |
2175 | adev->powerplay.pp_funcs && | |
2176 | adev->powerplay.pp_funcs->enable_mgpu_fan_boost) { | |
2177 | ret = amdgpu_dpm_enable_mgpu_fan_boost(adev); | |
2178 | if (ret) | |
2179 | break; | |
2180 | ||
2181 | gpu_ins->mgpu_fan_enabled = 1; | |
2182 | } | |
2183 | } | |
2184 | ||
2185 | out: | |
2186 | mutex_unlock(&mgpu_info.mutex); | |
2187 | ||
2188 | return ret; | |
2189 | } | |
2190 | ||
e3ecdffa AD |
2191 | /** |
2192 | * amdgpu_device_ip_late_init - run late init for hardware IPs | |
2193 | * | |
2194 | * @adev: amdgpu_device pointer | |
2195 | * | |
2196 | * Late initialization pass for hardware IPs. The list of all the hardware | |
2197 | * IPs that make up the asic is walked and the late_init callbacks are run. | |
2198 | * late_init covers any special initialization that an IP requires | |
2199 | * after all of the have been initialized or something that needs to happen | |
2200 | * late in the init process. | |
2201 | * Returns 0 on success, negative error code on failure. | |
2202 | */ | |
06ec9070 | 2203 | static int amdgpu_device_ip_late_init(struct amdgpu_device *adev) |
2dc80b00 | 2204 | { |
60599a03 | 2205 | struct amdgpu_gpu_instance *gpu_instance; |
2dc80b00 S |
2206 | int i = 0, r; |
2207 | ||
2208 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
73f847db | 2209 | if (!adev->ip_blocks[i].status.hw) |
2dc80b00 S |
2210 | continue; |
2211 | if (adev->ip_blocks[i].version->funcs->late_init) { | |
2212 | r = adev->ip_blocks[i].version->funcs->late_init((void *)adev); | |
2213 | if (r) { | |
2214 | DRM_ERROR("late_init of IP block <%s> failed %d\n", | |
2215 | adev->ip_blocks[i].version->funcs->name, r); | |
2216 | return r; | |
2217 | } | |
2dc80b00 | 2218 | } |
73f847db | 2219 | adev->ip_blocks[i].status.late_initialized = true; |
2dc80b00 S |
2220 | } |
2221 | ||
a891d239 DL |
2222 | amdgpu_ras_set_error_query_ready(adev, true); |
2223 | ||
1112a46b RZ |
2224 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE); |
2225 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE); | |
916ac57f | 2226 | |
06ec9070 | 2227 | amdgpu_device_fill_reset_magic(adev); |
d38ceaf9 | 2228 | |
beff74bc AD |
2229 | r = amdgpu_device_enable_mgpu_fan_boost(); |
2230 | if (r) | |
2231 | DRM_ERROR("enable mgpu fan boost failed (%d).\n", r); | |
2232 | ||
60599a03 EQ |
2233 | |
2234 | if (adev->gmc.xgmi.num_physical_nodes > 1) { | |
2235 | mutex_lock(&mgpu_info.mutex); | |
2236 | ||
2237 | /* | |
2238 | * Reset device p-state to low as this was booted with high. | |
2239 | * | |
2240 | * This should be performed only after all devices from the same | |
2241 | * hive get initialized. | |
2242 | * | |
2243 | * However, it's unknown how many device in the hive in advance. | |
2244 | * As this is counted one by one during devices initializations. | |
2245 | * | |
2246 | * So, we wait for all XGMI interlinked devices initialized. | |
2247 | * This may bring some delays as those devices may come from | |
2248 | * different hives. But that should be OK. | |
2249 | */ | |
2250 | if (mgpu_info.num_dgpu == adev->gmc.xgmi.num_physical_nodes) { | |
2251 | for (i = 0; i < mgpu_info.num_gpu; i++) { | |
2252 | gpu_instance = &(mgpu_info.gpu_ins[i]); | |
2253 | if (gpu_instance->adev->flags & AMD_IS_APU) | |
2254 | continue; | |
2255 | ||
d84a430d JK |
2256 | r = amdgpu_xgmi_set_pstate(gpu_instance->adev, |
2257 | AMDGPU_XGMI_PSTATE_MIN); | |
60599a03 EQ |
2258 | if (r) { |
2259 | DRM_ERROR("pstate setting failed (%d).\n", r); | |
2260 | break; | |
2261 | } | |
2262 | } | |
2263 | } | |
2264 | ||
2265 | mutex_unlock(&mgpu_info.mutex); | |
2266 | } | |
2267 | ||
d38ceaf9 AD |
2268 | return 0; |
2269 | } | |
2270 | ||
e3ecdffa AD |
2271 | /** |
2272 | * amdgpu_device_ip_fini - run fini for hardware IPs | |
2273 | * | |
2274 | * @adev: amdgpu_device pointer | |
2275 | * | |
2276 | * Main teardown pass for hardware IPs. The list of all the hardware | |
2277 | * IPs that make up the asic is walked and the hw_fini and sw_fini callbacks | |
2278 | * are run. hw_fini tears down the hardware associated with each IP | |
2279 | * and sw_fini tears down any software state associated with each IP. | |
2280 | * Returns 0 on success, negative error code on failure. | |
2281 | */ | |
06ec9070 | 2282 | static int amdgpu_device_ip_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
2283 | { |
2284 | int i, r; | |
2285 | ||
c030f2e4 | 2286 | amdgpu_ras_pre_fini(adev); |
2287 | ||
a82400b5 AG |
2288 | if (adev->gmc.xgmi.num_physical_nodes > 1) |
2289 | amdgpu_xgmi_remove_device(adev); | |
2290 | ||
1884734a | 2291 | amdgpu_amdkfd_device_fini(adev); |
05df1f01 RZ |
2292 | |
2293 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); | |
fdd34271 RZ |
2294 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); |
2295 | ||
3e96dbfd AD |
2296 | /* need to disable SMC first */ |
2297 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 2298 | if (!adev->ip_blocks[i].status.hw) |
3e96dbfd | 2299 | continue; |
fdd34271 | 2300 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { |
a1255107 | 2301 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); |
3e96dbfd AD |
2302 | /* XXX handle errors */ |
2303 | if (r) { | |
2304 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", | |
a1255107 | 2305 | adev->ip_blocks[i].version->funcs->name, r); |
3e96dbfd | 2306 | } |
a1255107 | 2307 | adev->ip_blocks[i].status.hw = false; |
3e96dbfd AD |
2308 | break; |
2309 | } | |
2310 | } | |
2311 | ||
d38ceaf9 | 2312 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 2313 | if (!adev->ip_blocks[i].status.hw) |
d38ceaf9 | 2314 | continue; |
8201a67a | 2315 | |
a1255107 | 2316 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); |
d38ceaf9 | 2317 | /* XXX handle errors */ |
2c1a2784 | 2318 | if (r) { |
a1255107 AD |
2319 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", |
2320 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 2321 | } |
8201a67a | 2322 | |
a1255107 | 2323 | adev->ip_blocks[i].status.hw = false; |
d38ceaf9 AD |
2324 | } |
2325 | ||
9950cda2 | 2326 | |
d38ceaf9 | 2327 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 2328 | if (!adev->ip_blocks[i].status.sw) |
d38ceaf9 | 2329 | continue; |
c12aba3a ML |
2330 | |
2331 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { | |
c8963ea4 | 2332 | amdgpu_ucode_free_bo(adev); |
1e256e27 | 2333 | amdgpu_free_static_csa(&adev->virt.csa_obj); |
c12aba3a ML |
2334 | amdgpu_device_wb_fini(adev); |
2335 | amdgpu_device_vram_scratch_fini(adev); | |
533aed27 | 2336 | amdgpu_ib_pool_fini(adev); |
c12aba3a ML |
2337 | } |
2338 | ||
a1255107 | 2339 | r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev); |
d38ceaf9 | 2340 | /* XXX handle errors */ |
2c1a2784 | 2341 | if (r) { |
a1255107 AD |
2342 | DRM_DEBUG("sw_fini of IP block <%s> failed %d\n", |
2343 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 2344 | } |
a1255107 AD |
2345 | adev->ip_blocks[i].status.sw = false; |
2346 | adev->ip_blocks[i].status.valid = false; | |
d38ceaf9 AD |
2347 | } |
2348 | ||
a6dcfd9c | 2349 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 2350 | if (!adev->ip_blocks[i].status.late_initialized) |
8a2eef1d | 2351 | continue; |
a1255107 AD |
2352 | if (adev->ip_blocks[i].version->funcs->late_fini) |
2353 | adev->ip_blocks[i].version->funcs->late_fini((void *)adev); | |
2354 | adev->ip_blocks[i].status.late_initialized = false; | |
a6dcfd9c ML |
2355 | } |
2356 | ||
c030f2e4 | 2357 | amdgpu_ras_fini(adev); |
2358 | ||
030308fc | 2359 | if (amdgpu_sriov_vf(adev)) |
24136135 ML |
2360 | if (amdgpu_virt_release_full_gpu(adev, false)) |
2361 | DRM_ERROR("failed to release exclusive mode on fini\n"); | |
2493664f | 2362 | |
d38ceaf9 AD |
2363 | return 0; |
2364 | } | |
2365 | ||
e3ecdffa | 2366 | /** |
beff74bc | 2367 | * amdgpu_device_delayed_init_work_handler - work handler for IB tests |
e3ecdffa | 2368 | * |
1112a46b | 2369 | * @work: work_struct. |
e3ecdffa | 2370 | */ |
beff74bc | 2371 | static void amdgpu_device_delayed_init_work_handler(struct work_struct *work) |
2dc80b00 S |
2372 | { |
2373 | struct amdgpu_device *adev = | |
beff74bc | 2374 | container_of(work, struct amdgpu_device, delayed_init_work.work); |
916ac57f RZ |
2375 | int r; |
2376 | ||
2377 | r = amdgpu_ib_ring_tests(adev); | |
2378 | if (r) | |
2379 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
2dc80b00 S |
2380 | } |
2381 | ||
1e317b99 RZ |
2382 | static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work) |
2383 | { | |
2384 | struct amdgpu_device *adev = | |
2385 | container_of(work, struct amdgpu_device, gfx.gfx_off_delay_work.work); | |
2386 | ||
2387 | mutex_lock(&adev->gfx.gfx_off_mutex); | |
2388 | if (!adev->gfx.gfx_off_state && !adev->gfx.gfx_off_req_count) { | |
2389 | if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true)) | |
2390 | adev->gfx.gfx_off_state = true; | |
2391 | } | |
2392 | mutex_unlock(&adev->gfx.gfx_off_mutex); | |
2393 | } | |
2394 | ||
e3ecdffa | 2395 | /** |
e7854a03 | 2396 | * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1) |
e3ecdffa AD |
2397 | * |
2398 | * @adev: amdgpu_device pointer | |
2399 | * | |
2400 | * Main suspend function for hardware IPs. The list of all the hardware | |
2401 | * IPs that make up the asic is walked, clockgating is disabled and the | |
2402 | * suspend callbacks are run. suspend puts the hardware and software state | |
2403 | * in each IP into a state suitable for suspend. | |
2404 | * Returns 0 on success, negative error code on failure. | |
2405 | */ | |
e7854a03 AD |
2406 | static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev) |
2407 | { | |
2408 | int i, r; | |
2409 | ||
ced1ba97 PL |
2410 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); |
2411 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); | |
05df1f01 | 2412 | |
e7854a03 AD |
2413 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
2414 | if (!adev->ip_blocks[i].status.valid) | |
2415 | continue; | |
2416 | /* displays are handled separately */ | |
2417 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) { | |
e7854a03 AD |
2418 | /* XXX handle errors */ |
2419 | r = adev->ip_blocks[i].version->funcs->suspend(adev); | |
2420 | /* XXX handle errors */ | |
2421 | if (r) { | |
2422 | DRM_ERROR("suspend of IP block <%s> failed %d\n", | |
2423 | adev->ip_blocks[i].version->funcs->name, r); | |
482f0e53 | 2424 | return r; |
e7854a03 | 2425 | } |
482f0e53 | 2426 | adev->ip_blocks[i].status.hw = false; |
e7854a03 AD |
2427 | } |
2428 | } | |
2429 | ||
e7854a03 AD |
2430 | return 0; |
2431 | } | |
2432 | ||
2433 | /** | |
2434 | * amdgpu_device_ip_suspend_phase2 - run suspend for hardware IPs (phase 2) | |
2435 | * | |
2436 | * @adev: amdgpu_device pointer | |
2437 | * | |
2438 | * Main suspend function for hardware IPs. The list of all the hardware | |
2439 | * IPs that make up the asic is walked, clockgating is disabled and the | |
2440 | * suspend callbacks are run. suspend puts the hardware and software state | |
2441 | * in each IP into a state suitable for suspend. | |
2442 | * Returns 0 on success, negative error code on failure. | |
2443 | */ | |
2444 | static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev) | |
d38ceaf9 AD |
2445 | { |
2446 | int i, r; | |
2447 | ||
2448 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
a1255107 | 2449 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 2450 | continue; |
e7854a03 AD |
2451 | /* displays are handled in phase1 */ |
2452 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) | |
2453 | continue; | |
bff77e86 LM |
2454 | /* PSP lost connection when err_event_athub occurs */ |
2455 | if (amdgpu_ras_intr_triggered() && | |
2456 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) { | |
2457 | adev->ip_blocks[i].status.hw = false; | |
2458 | continue; | |
2459 | } | |
d38ceaf9 | 2460 | /* XXX handle errors */ |
a1255107 | 2461 | r = adev->ip_blocks[i].version->funcs->suspend(adev); |
d38ceaf9 | 2462 | /* XXX handle errors */ |
2c1a2784 | 2463 | if (r) { |
a1255107 AD |
2464 | DRM_ERROR("suspend of IP block <%s> failed %d\n", |
2465 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 2466 | } |
876923fb | 2467 | adev->ip_blocks[i].status.hw = false; |
a3a09142 | 2468 | /* handle putting the SMC in the appropriate state */ |
86b93fd6 JZ |
2469 | if(!amdgpu_sriov_vf(adev)){ |
2470 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { | |
2471 | r = amdgpu_dpm_set_mp1_state(adev, adev->mp1_state); | |
2472 | if (r) { | |
2473 | DRM_ERROR("SMC failed to set mp1 state %d, %d\n", | |
2474 | adev->mp1_state, r); | |
2475 | return r; | |
2476 | } | |
a3a09142 AD |
2477 | } |
2478 | } | |
b5507c7e | 2479 | adev->ip_blocks[i].status.hw = false; |
d38ceaf9 AD |
2480 | } |
2481 | ||
2482 | return 0; | |
2483 | } | |
2484 | ||
e7854a03 AD |
2485 | /** |
2486 | * amdgpu_device_ip_suspend - run suspend for hardware IPs | |
2487 | * | |
2488 | * @adev: amdgpu_device pointer | |
2489 | * | |
2490 | * Main suspend function for hardware IPs. The list of all the hardware | |
2491 | * IPs that make up the asic is walked, clockgating is disabled and the | |
2492 | * suspend callbacks are run. suspend puts the hardware and software state | |
2493 | * in each IP into a state suitable for suspend. | |
2494 | * Returns 0 on success, negative error code on failure. | |
2495 | */ | |
2496 | int amdgpu_device_ip_suspend(struct amdgpu_device *adev) | |
2497 | { | |
2498 | int r; | |
2499 | ||
e7819644 YT |
2500 | if (amdgpu_sriov_vf(adev)) |
2501 | amdgpu_virt_request_full_gpu(adev, false); | |
2502 | ||
e7854a03 AD |
2503 | r = amdgpu_device_ip_suspend_phase1(adev); |
2504 | if (r) | |
2505 | return r; | |
2506 | r = amdgpu_device_ip_suspend_phase2(adev); | |
2507 | ||
e7819644 YT |
2508 | if (amdgpu_sriov_vf(adev)) |
2509 | amdgpu_virt_release_full_gpu(adev, false); | |
2510 | ||
e7854a03 AD |
2511 | return r; |
2512 | } | |
2513 | ||
06ec9070 | 2514 | static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
2515 | { |
2516 | int i, r; | |
2517 | ||
2cb681b6 ML |
2518 | static enum amd_ip_block_type ip_order[] = { |
2519 | AMD_IP_BLOCK_TYPE_GMC, | |
2520 | AMD_IP_BLOCK_TYPE_COMMON, | |
39186aef | 2521 | AMD_IP_BLOCK_TYPE_PSP, |
2cb681b6 ML |
2522 | AMD_IP_BLOCK_TYPE_IH, |
2523 | }; | |
a90ad3c2 | 2524 | |
2cb681b6 ML |
2525 | for (i = 0; i < ARRAY_SIZE(ip_order); i++) { |
2526 | int j; | |
2527 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 2528 | |
2cb681b6 ML |
2529 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2530 | block = &adev->ip_blocks[j]; | |
2531 | ||
482f0e53 | 2532 | block->status.hw = false; |
2cb681b6 ML |
2533 | if (block->version->type != ip_order[i] || |
2534 | !block->status.valid) | |
2535 | continue; | |
2536 | ||
2537 | r = block->version->funcs->hw_init(adev); | |
0aaeefcc | 2538 | DRM_INFO("RE-INIT-early: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); |
c41d1cf6 ML |
2539 | if (r) |
2540 | return r; | |
482f0e53 | 2541 | block->status.hw = true; |
a90ad3c2 ML |
2542 | } |
2543 | } | |
2544 | ||
2545 | return 0; | |
2546 | } | |
2547 | ||
06ec9070 | 2548 | static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
2549 | { |
2550 | int i, r; | |
2551 | ||
2cb681b6 ML |
2552 | static enum amd_ip_block_type ip_order[] = { |
2553 | AMD_IP_BLOCK_TYPE_SMC, | |
2554 | AMD_IP_BLOCK_TYPE_DCE, | |
2555 | AMD_IP_BLOCK_TYPE_GFX, | |
2556 | AMD_IP_BLOCK_TYPE_SDMA, | |
257deb8c | 2557 | AMD_IP_BLOCK_TYPE_UVD, |
d83c7a07 JJ |
2558 | AMD_IP_BLOCK_TYPE_VCE, |
2559 | AMD_IP_BLOCK_TYPE_VCN | |
2cb681b6 | 2560 | }; |
a90ad3c2 | 2561 | |
2cb681b6 ML |
2562 | for (i = 0; i < ARRAY_SIZE(ip_order); i++) { |
2563 | int j; | |
2564 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 2565 | |
2cb681b6 ML |
2566 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2567 | block = &adev->ip_blocks[j]; | |
2568 | ||
2569 | if (block->version->type != ip_order[i] || | |
482f0e53 ML |
2570 | !block->status.valid || |
2571 | block->status.hw) | |
2cb681b6 ML |
2572 | continue; |
2573 | ||
895bd048 JZ |
2574 | if (block->version->type == AMD_IP_BLOCK_TYPE_SMC) |
2575 | r = block->version->funcs->resume(adev); | |
2576 | else | |
2577 | r = block->version->funcs->hw_init(adev); | |
2578 | ||
0aaeefcc | 2579 | DRM_INFO("RE-INIT-late: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); |
c41d1cf6 ML |
2580 | if (r) |
2581 | return r; | |
482f0e53 | 2582 | block->status.hw = true; |
a90ad3c2 ML |
2583 | } |
2584 | } | |
2585 | ||
2586 | return 0; | |
2587 | } | |
2588 | ||
e3ecdffa AD |
2589 | /** |
2590 | * amdgpu_device_ip_resume_phase1 - run resume for hardware IPs | |
2591 | * | |
2592 | * @adev: amdgpu_device pointer | |
2593 | * | |
2594 | * First resume function for hardware IPs. The list of all the hardware | |
2595 | * IPs that make up the asic is walked and the resume callbacks are run for | |
2596 | * COMMON, GMC, and IH. resume puts the hardware into a functional state | |
2597 | * after a suspend and updates the software state as necessary. This | |
2598 | * function is also used for restoring the GPU after a GPU reset. | |
2599 | * Returns 0 on success, negative error code on failure. | |
2600 | */ | |
06ec9070 | 2601 | static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev) |
d38ceaf9 AD |
2602 | { |
2603 | int i, r; | |
2604 | ||
a90ad3c2 | 2605 | for (i = 0; i < adev->num_ip_blocks; i++) { |
482f0e53 | 2606 | if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw) |
a90ad3c2 | 2607 | continue; |
a90ad3c2 | 2608 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
e3ecdffa AD |
2609 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || |
2610 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) { | |
482f0e53 | 2611 | |
fcf0649f CZ |
2612 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
2613 | if (r) { | |
2614 | DRM_ERROR("resume of IP block <%s> failed %d\n", | |
2615 | adev->ip_blocks[i].version->funcs->name, r); | |
2616 | return r; | |
2617 | } | |
482f0e53 | 2618 | adev->ip_blocks[i].status.hw = true; |
a90ad3c2 ML |
2619 | } |
2620 | } | |
2621 | ||
2622 | return 0; | |
2623 | } | |
2624 | ||
e3ecdffa AD |
2625 | /** |
2626 | * amdgpu_device_ip_resume_phase2 - run resume for hardware IPs | |
2627 | * | |
2628 | * @adev: amdgpu_device pointer | |
2629 | * | |
2630 | * First resume function for hardware IPs. The list of all the hardware | |
2631 | * IPs that make up the asic is walked and the resume callbacks are run for | |
2632 | * all blocks except COMMON, GMC, and IH. resume puts the hardware into a | |
2633 | * functional state after a suspend and updates the software state as | |
2634 | * necessary. This function is also used for restoring the GPU after a GPU | |
2635 | * reset. | |
2636 | * Returns 0 on success, negative error code on failure. | |
2637 | */ | |
06ec9070 | 2638 | static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev) |
d38ceaf9 AD |
2639 | { |
2640 | int i, r; | |
2641 | ||
2642 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
482f0e53 | 2643 | if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw) |
d38ceaf9 | 2644 | continue; |
fcf0649f | 2645 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
e3ecdffa | 2646 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || |
7a3e0bb2 RZ |
2647 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH || |
2648 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) | |
fcf0649f | 2649 | continue; |
a1255107 | 2650 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
2c1a2784 | 2651 | if (r) { |
a1255107 AD |
2652 | DRM_ERROR("resume of IP block <%s> failed %d\n", |
2653 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 2654 | return r; |
2c1a2784 | 2655 | } |
482f0e53 | 2656 | adev->ip_blocks[i].status.hw = true; |
d38ceaf9 AD |
2657 | } |
2658 | ||
2659 | return 0; | |
2660 | } | |
2661 | ||
e3ecdffa AD |
2662 | /** |
2663 | * amdgpu_device_ip_resume - run resume for hardware IPs | |
2664 | * | |
2665 | * @adev: amdgpu_device pointer | |
2666 | * | |
2667 | * Main resume function for hardware IPs. The hardware IPs | |
2668 | * are split into two resume functions because they are | |
2669 | * are also used in in recovering from a GPU reset and some additional | |
2670 | * steps need to be take between them. In this case (S3/S4) they are | |
2671 | * run sequentially. | |
2672 | * Returns 0 on success, negative error code on failure. | |
2673 | */ | |
06ec9070 | 2674 | static int amdgpu_device_ip_resume(struct amdgpu_device *adev) |
fcf0649f CZ |
2675 | { |
2676 | int r; | |
2677 | ||
06ec9070 | 2678 | r = amdgpu_device_ip_resume_phase1(adev); |
fcf0649f CZ |
2679 | if (r) |
2680 | return r; | |
7a3e0bb2 RZ |
2681 | |
2682 | r = amdgpu_device_fw_loading(adev); | |
2683 | if (r) | |
2684 | return r; | |
2685 | ||
06ec9070 | 2686 | r = amdgpu_device_ip_resume_phase2(adev); |
fcf0649f CZ |
2687 | |
2688 | return r; | |
2689 | } | |
2690 | ||
e3ecdffa AD |
2691 | /** |
2692 | * amdgpu_device_detect_sriov_bios - determine if the board supports SR-IOV | |
2693 | * | |
2694 | * @adev: amdgpu_device pointer | |
2695 | * | |
2696 | * Query the VBIOS data tables to determine if the board supports SR-IOV. | |
2697 | */ | |
4e99a44e | 2698 | static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev) |
048765ad | 2699 | { |
6867e1b5 ML |
2700 | if (amdgpu_sriov_vf(adev)) { |
2701 | if (adev->is_atom_fw) { | |
2702 | if (amdgpu_atomfirmware_gpu_supports_virtualization(adev)) | |
2703 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; | |
2704 | } else { | |
2705 | if (amdgpu_atombios_has_gpu_virtualization_table(adev)) | |
2706 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; | |
2707 | } | |
2708 | ||
2709 | if (!(adev->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS)) | |
2710 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0); | |
a5bde2f9 | 2711 | } |
048765ad AR |
2712 | } |
2713 | ||
e3ecdffa AD |
2714 | /** |
2715 | * amdgpu_device_asic_has_dc_support - determine if DC supports the asic | |
2716 | * | |
2717 | * @asic_type: AMD asic type | |
2718 | * | |
2719 | * Check if there is DC (new modesetting infrastructre) support for an asic. | |
2720 | * returns true if DC has support, false if not. | |
2721 | */ | |
4562236b HW |
2722 | bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type) |
2723 | { | |
2724 | switch (asic_type) { | |
2725 | #if defined(CONFIG_DRM_AMD_DC) | |
2726 | case CHIP_BONAIRE: | |
0d6fbccb | 2727 | case CHIP_KAVERI: |
367e6687 AD |
2728 | case CHIP_KABINI: |
2729 | case CHIP_MULLINS: | |
d9fda248 HW |
2730 | /* |
2731 | * We have systems in the wild with these ASICs that require | |
2732 | * LVDS and VGA support which is not supported with DC. | |
2733 | * | |
2734 | * Fallback to the non-DC driver here by default so as not to | |
2735 | * cause regressions. | |
2736 | */ | |
2737 | return amdgpu_dc > 0; | |
2738 | case CHIP_HAWAII: | |
4562236b HW |
2739 | case CHIP_CARRIZO: |
2740 | case CHIP_STONEY: | |
4562236b | 2741 | case CHIP_POLARIS10: |
675fd32b | 2742 | case CHIP_POLARIS11: |
2c8ad2d5 | 2743 | case CHIP_POLARIS12: |
675fd32b | 2744 | case CHIP_VEGAM: |
4562236b HW |
2745 | case CHIP_TONGA: |
2746 | case CHIP_FIJI: | |
42f8ffa1 | 2747 | case CHIP_VEGA10: |
dca7b401 | 2748 | case CHIP_VEGA12: |
c6034aa2 | 2749 | case CHIP_VEGA20: |
b86a1aa3 | 2750 | #if defined(CONFIG_DRM_AMD_DC_DCN) |
fd187853 | 2751 | case CHIP_RAVEN: |
b4f199c7 | 2752 | case CHIP_NAVI10: |
8fceceb6 | 2753 | case CHIP_NAVI14: |
078655d9 | 2754 | case CHIP_NAVI12: |
e1c14c43 | 2755 | case CHIP_RENOIR: |
42f8ffa1 | 2756 | #endif |
fd187853 | 2757 | return amdgpu_dc != 0; |
4562236b HW |
2758 | #endif |
2759 | default: | |
93b09a9a SS |
2760 | if (amdgpu_dc > 0) |
2761 | DRM_INFO("Display Core has been requested via kernel parameter " | |
2762 | "but isn't supported by ASIC, ignoring\n"); | |
4562236b HW |
2763 | return false; |
2764 | } | |
2765 | } | |
2766 | ||
2767 | /** | |
2768 | * amdgpu_device_has_dc_support - check if dc is supported | |
2769 | * | |
2770 | * @adev: amdgpu_device_pointer | |
2771 | * | |
2772 | * Returns true for supported, false for not supported | |
2773 | */ | |
2774 | bool amdgpu_device_has_dc_support(struct amdgpu_device *adev) | |
2775 | { | |
2555039d XY |
2776 | if (amdgpu_sriov_vf(adev)) |
2777 | return false; | |
2778 | ||
4562236b HW |
2779 | return amdgpu_device_asic_has_dc_support(adev->asic_type); |
2780 | } | |
2781 | ||
d4535e2c AG |
2782 | |
2783 | static void amdgpu_device_xgmi_reset_func(struct work_struct *__work) | |
2784 | { | |
2785 | struct amdgpu_device *adev = | |
2786 | container_of(__work, struct amdgpu_device, xgmi_reset_work); | |
c6a6e2db | 2787 | struct amdgpu_hive_info *hive = amdgpu_get_xgmi_hive(adev, 0); |
d4535e2c | 2788 | |
c6a6e2db AG |
2789 | /* It's a bug to not have a hive within this function */ |
2790 | if (WARN_ON(!hive)) | |
2791 | return; | |
2792 | ||
2793 | /* | |
2794 | * Use task barrier to synchronize all xgmi reset works across the | |
2795 | * hive. task_barrier_enter and task_barrier_exit will block | |
2796 | * until all the threads running the xgmi reset works reach | |
2797 | * those points. task_barrier_full will do both blocks. | |
2798 | */ | |
2799 | if (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) { | |
2800 | ||
2801 | task_barrier_enter(&hive->tb); | |
2802 | adev->asic_reset_res = amdgpu_device_baco_enter(adev->ddev); | |
2803 | ||
2804 | if (adev->asic_reset_res) | |
2805 | goto fail; | |
2806 | ||
2807 | task_barrier_exit(&hive->tb); | |
2808 | adev->asic_reset_res = amdgpu_device_baco_exit(adev->ddev); | |
2809 | ||
2810 | if (adev->asic_reset_res) | |
2811 | goto fail; | |
43c4d576 JC |
2812 | |
2813 | if (adev->mmhub.funcs && adev->mmhub.funcs->reset_ras_error_count) | |
2814 | adev->mmhub.funcs->reset_ras_error_count(adev); | |
c6a6e2db AG |
2815 | } else { |
2816 | ||
2817 | task_barrier_full(&hive->tb); | |
2818 | adev->asic_reset_res = amdgpu_asic_reset(adev); | |
2819 | } | |
ce316fa5 | 2820 | |
c6a6e2db | 2821 | fail: |
d4535e2c | 2822 | if (adev->asic_reset_res) |
fed184e9 | 2823 | DRM_WARN("ASIC reset failed with error, %d for drm dev, %s", |
d4535e2c AG |
2824 | adev->asic_reset_res, adev->ddev->unique); |
2825 | } | |
2826 | ||
71f98027 AD |
2827 | static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev) |
2828 | { | |
2829 | char *input = amdgpu_lockup_timeout; | |
2830 | char *timeout_setting = NULL; | |
2831 | int index = 0; | |
2832 | long timeout; | |
2833 | int ret = 0; | |
2834 | ||
2835 | /* | |
2836 | * By default timeout for non compute jobs is 10000. | |
2837 | * And there is no timeout enforced on compute jobs. | |
2838 | * In SR-IOV or passthrough mode, timeout for compute | |
b7b2a316 | 2839 | * jobs are 60000 by default. |
71f98027 AD |
2840 | */ |
2841 | adev->gfx_timeout = msecs_to_jiffies(10000); | |
2842 | adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; | |
2843 | if (amdgpu_sriov_vf(adev) || amdgpu_passthrough(adev)) | |
b7b2a316 | 2844 | adev->compute_timeout = msecs_to_jiffies(60000); |
71f98027 AD |
2845 | else |
2846 | adev->compute_timeout = MAX_SCHEDULE_TIMEOUT; | |
2847 | ||
f440ff44 | 2848 | if (strnlen(input, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH)) { |
71f98027 | 2849 | while ((timeout_setting = strsep(&input, ",")) && |
f440ff44 | 2850 | strnlen(timeout_setting, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH)) { |
71f98027 AD |
2851 | ret = kstrtol(timeout_setting, 0, &timeout); |
2852 | if (ret) | |
2853 | return ret; | |
2854 | ||
2855 | if (timeout == 0) { | |
2856 | index++; | |
2857 | continue; | |
2858 | } else if (timeout < 0) { | |
2859 | timeout = MAX_SCHEDULE_TIMEOUT; | |
2860 | } else { | |
2861 | timeout = msecs_to_jiffies(timeout); | |
2862 | } | |
2863 | ||
2864 | switch (index++) { | |
2865 | case 0: | |
2866 | adev->gfx_timeout = timeout; | |
2867 | break; | |
2868 | case 1: | |
2869 | adev->compute_timeout = timeout; | |
2870 | break; | |
2871 | case 2: | |
2872 | adev->sdma_timeout = timeout; | |
2873 | break; | |
2874 | case 3: | |
2875 | adev->video_timeout = timeout; | |
2876 | break; | |
2877 | default: | |
2878 | break; | |
2879 | } | |
2880 | } | |
2881 | /* | |
2882 | * There is only one value specified and | |
2883 | * it should apply to all non-compute jobs. | |
2884 | */ | |
bcccee89 | 2885 | if (index == 1) { |
71f98027 | 2886 | adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; |
bcccee89 ED |
2887 | if (amdgpu_sriov_vf(adev) || amdgpu_passthrough(adev)) |
2888 | adev->compute_timeout = adev->gfx_timeout; | |
2889 | } | |
71f98027 AD |
2890 | } |
2891 | ||
2892 | return ret; | |
2893 | } | |
d4535e2c | 2894 | |
d38ceaf9 AD |
2895 | /** |
2896 | * amdgpu_device_init - initialize the driver | |
2897 | * | |
2898 | * @adev: amdgpu_device pointer | |
87e3f136 | 2899 | * @ddev: drm dev pointer |
d38ceaf9 AD |
2900 | * @pdev: pci dev pointer |
2901 | * @flags: driver flags | |
2902 | * | |
2903 | * Initializes the driver info and hw (all asics). | |
2904 | * Returns 0 for success or an error on failure. | |
2905 | * Called at driver startup. | |
2906 | */ | |
2907 | int amdgpu_device_init(struct amdgpu_device *adev, | |
2908 | struct drm_device *ddev, | |
2909 | struct pci_dev *pdev, | |
2910 | uint32_t flags) | |
2911 | { | |
2912 | int r, i; | |
3840c5bc | 2913 | bool boco = false; |
95844d20 | 2914 | u32 max_MBps; |
d38ceaf9 AD |
2915 | |
2916 | adev->shutdown = false; | |
2917 | adev->dev = &pdev->dev; | |
2918 | adev->ddev = ddev; | |
2919 | adev->pdev = pdev; | |
2920 | adev->flags = flags; | |
4e66d7d2 YZ |
2921 | |
2922 | if (amdgpu_force_asic_type >= 0 && amdgpu_force_asic_type < CHIP_LAST) | |
2923 | adev->asic_type = amdgpu_force_asic_type; | |
2924 | else | |
2925 | adev->asic_type = flags & AMD_ASIC_MASK; | |
2926 | ||
d38ceaf9 | 2927 | adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT; |
593aa2d2 | 2928 | if (amdgpu_emu_mode == 1) |
8bdab6bb | 2929 | adev->usec_timeout *= 10; |
770d13b1 | 2930 | adev->gmc.gart_size = 512 * 1024 * 1024; |
d38ceaf9 AD |
2931 | adev->accel_working = false; |
2932 | adev->num_rings = 0; | |
2933 | adev->mman.buffer_funcs = NULL; | |
2934 | adev->mman.buffer_funcs_ring = NULL; | |
2935 | adev->vm_manager.vm_pte_funcs = NULL; | |
0c88b430 | 2936 | adev->vm_manager.vm_pte_num_scheds = 0; |
132f34e4 | 2937 | adev->gmc.gmc_funcs = NULL; |
f54d1867 | 2938 | adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS); |
b8866c26 | 2939 | bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); |
d38ceaf9 AD |
2940 | |
2941 | adev->smc_rreg = &amdgpu_invalid_rreg; | |
2942 | adev->smc_wreg = &amdgpu_invalid_wreg; | |
2943 | adev->pcie_rreg = &amdgpu_invalid_rreg; | |
2944 | adev->pcie_wreg = &amdgpu_invalid_wreg; | |
36b9a952 HR |
2945 | adev->pciep_rreg = &amdgpu_invalid_rreg; |
2946 | adev->pciep_wreg = &amdgpu_invalid_wreg; | |
4fa1c6a6 TZ |
2947 | adev->pcie_rreg64 = &amdgpu_invalid_rreg64; |
2948 | adev->pcie_wreg64 = &amdgpu_invalid_wreg64; | |
d38ceaf9 AD |
2949 | adev->uvd_ctx_rreg = &amdgpu_invalid_rreg; |
2950 | adev->uvd_ctx_wreg = &amdgpu_invalid_wreg; | |
2951 | adev->didt_rreg = &amdgpu_invalid_rreg; | |
2952 | adev->didt_wreg = &amdgpu_invalid_wreg; | |
ccdbb20a RZ |
2953 | adev->gc_cac_rreg = &amdgpu_invalid_rreg; |
2954 | adev->gc_cac_wreg = &amdgpu_invalid_wreg; | |
d38ceaf9 AD |
2955 | adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg; |
2956 | adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg; | |
2957 | ||
3e39ab90 AD |
2958 | DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n", |
2959 | amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device, | |
2960 | pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision); | |
d38ceaf9 AD |
2961 | |
2962 | /* mutex initialization are all done here so we | |
2963 | * can recall function without having locking issues */ | |
d38ceaf9 | 2964 | atomic_set(&adev->irq.ih.lock, 0); |
0e5ca0d1 | 2965 | mutex_init(&adev->firmware.mutex); |
d38ceaf9 AD |
2966 | mutex_init(&adev->pm.mutex); |
2967 | mutex_init(&adev->gfx.gpu_clock_mutex); | |
2968 | mutex_init(&adev->srbm_mutex); | |
b8866c26 | 2969 | mutex_init(&adev->gfx.pipe_reserve_mutex); |
d23ee13f | 2970 | mutex_init(&adev->gfx.gfx_off_mutex); |
d38ceaf9 | 2971 | mutex_init(&adev->grbm_idx_mutex); |
d38ceaf9 | 2972 | mutex_init(&adev->mn_lock); |
e23b74aa | 2973 | mutex_init(&adev->virt.vf_errors.lock); |
d38ceaf9 | 2974 | hash_init(adev->mn_hash); |
13a752e3 | 2975 | mutex_init(&adev->lock_reset); |
32eaeae0 | 2976 | mutex_init(&adev->psp.mutex); |
bd052211 | 2977 | mutex_init(&adev->notifier_lock); |
d38ceaf9 | 2978 | |
912dfc84 EQ |
2979 | r = amdgpu_device_check_arguments(adev); |
2980 | if (r) | |
2981 | return r; | |
d38ceaf9 | 2982 | |
d38ceaf9 AD |
2983 | spin_lock_init(&adev->mmio_idx_lock); |
2984 | spin_lock_init(&adev->smc_idx_lock); | |
2985 | spin_lock_init(&adev->pcie_idx_lock); | |
2986 | spin_lock_init(&adev->uvd_ctx_idx_lock); | |
2987 | spin_lock_init(&adev->didt_idx_lock); | |
ccdbb20a | 2988 | spin_lock_init(&adev->gc_cac_idx_lock); |
16abb5d2 | 2989 | spin_lock_init(&adev->se_cac_idx_lock); |
d38ceaf9 | 2990 | spin_lock_init(&adev->audio_endpt_idx_lock); |
95844d20 | 2991 | spin_lock_init(&adev->mm_stats.lock); |
d38ceaf9 | 2992 | |
0c4e7fa5 CZ |
2993 | INIT_LIST_HEAD(&adev->shadow_list); |
2994 | mutex_init(&adev->shadow_list_lock); | |
2995 | ||
beff74bc AD |
2996 | INIT_DELAYED_WORK(&adev->delayed_init_work, |
2997 | amdgpu_device_delayed_init_work_handler); | |
1e317b99 RZ |
2998 | INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work, |
2999 | amdgpu_device_delay_enable_gfx_off); | |
2dc80b00 | 3000 | |
d4535e2c AG |
3001 | INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func); |
3002 | ||
d23ee13f | 3003 | adev->gfx.gfx_off_req_count = 1; |
b6e79d9a | 3004 | adev->pm.ac_power = power_supply_is_system_supplied() > 0; |
b1ddf548 | 3005 | |
0fa49558 AX |
3006 | /* Registers mapping */ |
3007 | /* TODO: block userspace mapping of io register */ | |
da69c161 KW |
3008 | if (adev->asic_type >= CHIP_BONAIRE) { |
3009 | adev->rmmio_base = pci_resource_start(adev->pdev, 5); | |
3010 | adev->rmmio_size = pci_resource_len(adev->pdev, 5); | |
3011 | } else { | |
3012 | adev->rmmio_base = pci_resource_start(adev->pdev, 2); | |
3013 | adev->rmmio_size = pci_resource_len(adev->pdev, 2); | |
3014 | } | |
d38ceaf9 | 3015 | |
d38ceaf9 AD |
3016 | adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size); |
3017 | if (adev->rmmio == NULL) { | |
3018 | return -ENOMEM; | |
3019 | } | |
3020 | DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base); | |
3021 | DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size); | |
3022 | ||
d38ceaf9 AD |
3023 | /* io port mapping */ |
3024 | for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { | |
3025 | if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) { | |
3026 | adev->rio_mem_size = pci_resource_len(adev->pdev, i); | |
3027 | adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size); | |
3028 | break; | |
3029 | } | |
3030 | } | |
3031 | if (adev->rio_mem == NULL) | |
b64a18c5 | 3032 | DRM_INFO("PCI I/O BAR is not found.\n"); |
d38ceaf9 | 3033 | |
b2109d8e JX |
3034 | /* enable PCIE atomic ops */ |
3035 | r = pci_enable_atomic_ops_to_root(adev->pdev, | |
3036 | PCI_EXP_DEVCAP2_ATOMIC_COMP32 | | |
3037 | PCI_EXP_DEVCAP2_ATOMIC_COMP64); | |
3038 | if (r) { | |
3039 | adev->have_atomics_support = false; | |
3040 | DRM_INFO("PCIE atomic ops is not supported\n"); | |
3041 | } else { | |
3042 | adev->have_atomics_support = true; | |
3043 | } | |
3044 | ||
5494d864 AD |
3045 | amdgpu_device_get_pcie_info(adev); |
3046 | ||
b239c017 JX |
3047 | if (amdgpu_mcbp) |
3048 | DRM_INFO("MCBP is enabled\n"); | |
3049 | ||
5f84cc63 JX |
3050 | if (amdgpu_mes && adev->asic_type >= CHIP_NAVI10) |
3051 | adev->enable_mes = true; | |
3052 | ||
3aa0115d ML |
3053 | /* detect hw virtualization here */ |
3054 | amdgpu_detect_virtualization(adev); | |
3055 | ||
dffa11b4 ML |
3056 | r = amdgpu_device_get_job_timeout_settings(adev); |
3057 | if (r) { | |
3058 | dev_err(adev->dev, "invalid lockup_timeout parameter syntax\n"); | |
3059 | return r; | |
a190d1c7 XY |
3060 | } |
3061 | ||
d38ceaf9 | 3062 | /* early init functions */ |
06ec9070 | 3063 | r = amdgpu_device_ip_early_init(adev); |
d38ceaf9 AD |
3064 | if (r) |
3065 | return r; | |
3066 | ||
6585661d OZ |
3067 | /* doorbell bar mapping and doorbell index init*/ |
3068 | amdgpu_device_doorbell_init(adev); | |
3069 | ||
d38ceaf9 AD |
3070 | /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */ |
3071 | /* this will fail for cards that aren't VGA class devices, just | |
3072 | * ignore it */ | |
06ec9070 | 3073 | vga_client_register(adev->pdev, adev, NULL, amdgpu_device_vga_set_decode); |
d38ceaf9 | 3074 | |
31af062a | 3075 | if (amdgpu_device_supports_boco(ddev)) |
3840c5bc AD |
3076 | boco = true; |
3077 | if (amdgpu_has_atpx() && | |
3078 | (amdgpu_is_atpx_hybrid() || | |
3079 | amdgpu_has_atpx_dgpu_power_cntl()) && | |
3080 | !pci_is_thunderbolt_attached(adev->pdev)) | |
84c8b22e | 3081 | vga_switcheroo_register_client(adev->pdev, |
3840c5bc AD |
3082 | &amdgpu_switcheroo_ops, boco); |
3083 | if (boco) | |
d38ceaf9 AD |
3084 | vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain); |
3085 | ||
9475a943 SL |
3086 | if (amdgpu_emu_mode == 1) { |
3087 | /* post the asic on emulation mode */ | |
3088 | emu_soc_asic_init(adev); | |
bfca0289 | 3089 | goto fence_driver_init; |
9475a943 | 3090 | } |
bfca0289 | 3091 | |
4e99a44e ML |
3092 | /* detect if we are with an SRIOV vbios */ |
3093 | amdgpu_device_detect_sriov_bios(adev); | |
048765ad | 3094 | |
95e8e59e AD |
3095 | /* check if we need to reset the asic |
3096 | * E.g., driver was not cleanly unloaded previously, etc. | |
3097 | */ | |
f14899fd | 3098 | if (!amdgpu_sriov_vf(adev) && amdgpu_asic_need_reset_on_init(adev)) { |
95e8e59e AD |
3099 | r = amdgpu_asic_reset(adev); |
3100 | if (r) { | |
3101 | dev_err(adev->dev, "asic reset on init failed\n"); | |
3102 | goto failed; | |
3103 | } | |
3104 | } | |
3105 | ||
d38ceaf9 | 3106 | /* Post card if necessary */ |
39c640c0 | 3107 | if (amdgpu_device_need_post(adev)) { |
d38ceaf9 | 3108 | if (!adev->bios) { |
bec86378 | 3109 | dev_err(adev->dev, "no vBIOS found\n"); |
83ba126a AD |
3110 | r = -EINVAL; |
3111 | goto failed; | |
d38ceaf9 | 3112 | } |
bec86378 | 3113 | DRM_INFO("GPU posting now...\n"); |
4e99a44e ML |
3114 | r = amdgpu_atom_asic_init(adev->mode_info.atom_context); |
3115 | if (r) { | |
3116 | dev_err(adev->dev, "gpu post error!\n"); | |
3117 | goto failed; | |
3118 | } | |
d38ceaf9 AD |
3119 | } |
3120 | ||
88b64e95 AD |
3121 | if (adev->is_atom_fw) { |
3122 | /* Initialize clocks */ | |
3123 | r = amdgpu_atomfirmware_get_clock_info(adev); | |
3124 | if (r) { | |
3125 | dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n"); | |
e23b74aa | 3126 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); |
88b64e95 AD |
3127 | goto failed; |
3128 | } | |
3129 | } else { | |
a5bde2f9 AD |
3130 | /* Initialize clocks */ |
3131 | r = amdgpu_atombios_get_clock_info(adev); | |
3132 | if (r) { | |
3133 | dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n"); | |
e23b74aa | 3134 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); |
89041940 | 3135 | goto failed; |
a5bde2f9 AD |
3136 | } |
3137 | /* init i2c buses */ | |
4562236b HW |
3138 | if (!amdgpu_device_has_dc_support(adev)) |
3139 | amdgpu_atombios_i2c_init(adev); | |
2c1a2784 | 3140 | } |
d38ceaf9 | 3141 | |
bfca0289 | 3142 | fence_driver_init: |
d38ceaf9 AD |
3143 | /* Fence driver */ |
3144 | r = amdgpu_fence_driver_init(adev); | |
2c1a2784 AD |
3145 | if (r) { |
3146 | dev_err(adev->dev, "amdgpu_fence_driver_init failed\n"); | |
e23b74aa | 3147 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0); |
83ba126a | 3148 | goto failed; |
2c1a2784 | 3149 | } |
d38ceaf9 AD |
3150 | |
3151 | /* init the mode config */ | |
3152 | drm_mode_config_init(adev->ddev); | |
3153 | ||
06ec9070 | 3154 | r = amdgpu_device_ip_init(adev); |
d38ceaf9 | 3155 | if (r) { |
8840a387 | 3156 | /* failed in exclusive mode due to timeout */ |
3157 | if (amdgpu_sriov_vf(adev) && | |
3158 | !amdgpu_sriov_runtime(adev) && | |
3159 | amdgpu_virt_mmio_blocked(adev) && | |
3160 | !amdgpu_virt_wait_reset(adev)) { | |
3161 | dev_err(adev->dev, "VF exclusive mode timeout\n"); | |
1daee8b4 PD |
3162 | /* Don't send request since VF is inactive. */ |
3163 | adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME; | |
3164 | adev->virt.ops = NULL; | |
8840a387 | 3165 | r = -EAGAIN; |
3166 | goto failed; | |
3167 | } | |
06ec9070 | 3168 | dev_err(adev->dev, "amdgpu_device_ip_init failed\n"); |
e23b74aa | 3169 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0); |
83ba126a | 3170 | goto failed; |
d38ceaf9 AD |
3171 | } |
3172 | ||
d69b8971 YZ |
3173 | dev_info(adev->dev, |
3174 | "SE %d, SH per SE %d, CU per SH %d, active_cu_number %d\n", | |
d7f72fe4 YZ |
3175 | adev->gfx.config.max_shader_engines, |
3176 | adev->gfx.config.max_sh_per_se, | |
3177 | adev->gfx.config.max_cu_per_sh, | |
3178 | adev->gfx.cu_info.number); | |
3179 | ||
d38ceaf9 AD |
3180 | adev->accel_working = true; |
3181 | ||
e59c0205 AX |
3182 | amdgpu_vm_check_compute_bug(adev); |
3183 | ||
95844d20 MO |
3184 | /* Initialize the buffer migration limit. */ |
3185 | if (amdgpu_moverate >= 0) | |
3186 | max_MBps = amdgpu_moverate; | |
3187 | else | |
3188 | max_MBps = 8; /* Allow 8 MB/s. */ | |
3189 | /* Get a log2 for easy divisions. */ | |
3190 | adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps)); | |
3191 | ||
9bc92b9c ML |
3192 | amdgpu_fbdev_init(adev); |
3193 | ||
d2f52ac8 | 3194 | r = amdgpu_pm_sysfs_init(adev); |
7c868b59 YT |
3195 | if (r) { |
3196 | adev->pm_sysfs_en = false; | |
d2f52ac8 | 3197 | DRM_ERROR("registering pm debugfs failed (%d).\n", r); |
7c868b59 YT |
3198 | } else |
3199 | adev->pm_sysfs_en = true; | |
d2f52ac8 | 3200 | |
5bb23532 | 3201 | r = amdgpu_ucode_sysfs_init(adev); |
7c868b59 YT |
3202 | if (r) { |
3203 | adev->ucode_sysfs_en = false; | |
5bb23532 | 3204 | DRM_ERROR("Creating firmware sysfs failed (%d).\n", r); |
7c868b59 YT |
3205 | } else |
3206 | adev->ucode_sysfs_en = true; | |
5bb23532 | 3207 | |
d38ceaf9 AD |
3208 | if ((amdgpu_testing & 1)) { |
3209 | if (adev->accel_working) | |
3210 | amdgpu_test_moves(adev); | |
3211 | else | |
3212 | DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n"); | |
3213 | } | |
d38ceaf9 AD |
3214 | if (amdgpu_benchmarking) { |
3215 | if (adev->accel_working) | |
3216 | amdgpu_benchmark(adev, amdgpu_benchmarking); | |
3217 | else | |
3218 | DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n"); | |
3219 | } | |
3220 | ||
b0adca4d EQ |
3221 | /* |
3222 | * Register gpu instance before amdgpu_device_enable_mgpu_fan_boost. | |
3223 | * Otherwise the mgpu fan boost feature will be skipped due to the | |
3224 | * gpu instance is counted less. | |
3225 | */ | |
3226 | amdgpu_register_gpu_instance(adev); | |
3227 | ||
d38ceaf9 AD |
3228 | /* enable clockgating, etc. after ib tests, etc. since some blocks require |
3229 | * explicit gating rather than handling it automatically. | |
3230 | */ | |
06ec9070 | 3231 | r = amdgpu_device_ip_late_init(adev); |
2c1a2784 | 3232 | if (r) { |
06ec9070 | 3233 | dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n"); |
e23b74aa | 3234 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r); |
83ba126a | 3235 | goto failed; |
2c1a2784 | 3236 | } |
d38ceaf9 | 3237 | |
108c6a63 | 3238 | /* must succeed. */ |
511fdbc3 | 3239 | amdgpu_ras_resume(adev); |
108c6a63 | 3240 | |
beff74bc AD |
3241 | queue_delayed_work(system_wq, &adev->delayed_init_work, |
3242 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
3243 | ||
dcea6e65 KR |
3244 | r = device_create_file(adev->dev, &dev_attr_pcie_replay_count); |
3245 | if (r) { | |
3246 | dev_err(adev->dev, "Could not create pcie_replay_count"); | |
3247 | return r; | |
3248 | } | |
108c6a63 | 3249 | |
bd607166 KR |
3250 | r = device_create_file(adev->dev, &dev_attr_product_name); |
3251 | if (r) { | |
3252 | dev_err(adev->dev, "Could not create product_name"); | |
3253 | return r; | |
3254 | } | |
3255 | ||
3256 | r = device_create_file(adev->dev, &dev_attr_product_number); | |
3257 | if (r) { | |
3258 | dev_err(adev->dev, "Could not create product_number"); | |
3259 | return r; | |
3260 | } | |
3261 | ||
3262 | r = device_create_file(adev->dev, &dev_attr_serial_number); | |
3263 | if (r) { | |
3264 | dev_err(adev->dev, "Could not create serial_number"); | |
3265 | return r; | |
3266 | } | |
3267 | ||
d155bef0 AB |
3268 | if (IS_ENABLED(CONFIG_PERF_EVENTS)) |
3269 | r = amdgpu_pmu_init(adev); | |
9c7c85f7 JK |
3270 | if (r) |
3271 | dev_err(adev->dev, "amdgpu_pmu_init failed\n"); | |
3272 | ||
d38ceaf9 | 3273 | return 0; |
83ba126a AD |
3274 | |
3275 | failed: | |
89041940 | 3276 | amdgpu_vf_error_trans_all(adev); |
3840c5bc | 3277 | if (boco) |
83ba126a | 3278 | vga_switcheroo_fini_domain_pm_ops(adev->dev); |
8840a387 | 3279 | |
83ba126a | 3280 | return r; |
d38ceaf9 AD |
3281 | } |
3282 | ||
d38ceaf9 AD |
3283 | /** |
3284 | * amdgpu_device_fini - tear down the driver | |
3285 | * | |
3286 | * @adev: amdgpu_device pointer | |
3287 | * | |
3288 | * Tear down the driver info (all asics). | |
3289 | * Called at driver shutdown. | |
3290 | */ | |
3291 | void amdgpu_device_fini(struct amdgpu_device *adev) | |
3292 | { | |
3293 | int r; | |
3294 | ||
3295 | DRM_INFO("amdgpu: finishing device.\n"); | |
9f875167 | 3296 | flush_delayed_work(&adev->delayed_init_work); |
d0d13fe8 | 3297 | adev->shutdown = true; |
9f875167 | 3298 | |
752c683d ML |
3299 | /* make sure IB test finished before entering exclusive mode |
3300 | * to avoid preemption on IB test | |
3301 | * */ | |
3302 | if (amdgpu_sriov_vf(adev)) | |
3303 | amdgpu_virt_request_full_gpu(adev, false); | |
3304 | ||
e5b03032 ML |
3305 | /* disable all interrupts */ |
3306 | amdgpu_irq_disable_all(adev); | |
ff97cba8 ML |
3307 | if (adev->mode_info.mode_config_initialized){ |
3308 | if (!amdgpu_device_has_dc_support(adev)) | |
c2d88e06 | 3309 | drm_helper_force_disable_all(adev->ddev); |
ff97cba8 ML |
3310 | else |
3311 | drm_atomic_helper_shutdown(adev->ddev); | |
3312 | } | |
d38ceaf9 | 3313 | amdgpu_fence_driver_fini(adev); |
7c868b59 YT |
3314 | if (adev->pm_sysfs_en) |
3315 | amdgpu_pm_sysfs_fini(adev); | |
d38ceaf9 | 3316 | amdgpu_fbdev_fini(adev); |
06ec9070 | 3317 | r = amdgpu_device_ip_fini(adev); |
ab4fe3e1 HR |
3318 | if (adev->firmware.gpu_info_fw) { |
3319 | release_firmware(adev->firmware.gpu_info_fw); | |
3320 | adev->firmware.gpu_info_fw = NULL; | |
3321 | } | |
d38ceaf9 AD |
3322 | adev->accel_working = false; |
3323 | /* free i2c buses */ | |
4562236b HW |
3324 | if (!amdgpu_device_has_dc_support(adev)) |
3325 | amdgpu_i2c_fini(adev); | |
bfca0289 SL |
3326 | |
3327 | if (amdgpu_emu_mode != 1) | |
3328 | amdgpu_atombios_fini(adev); | |
3329 | ||
d38ceaf9 AD |
3330 | kfree(adev->bios); |
3331 | adev->bios = NULL; | |
3840c5bc AD |
3332 | if (amdgpu_has_atpx() && |
3333 | (amdgpu_is_atpx_hybrid() || | |
3334 | amdgpu_has_atpx_dgpu_power_cntl()) && | |
3335 | !pci_is_thunderbolt_attached(adev->pdev)) | |
84c8b22e | 3336 | vga_switcheroo_unregister_client(adev->pdev); |
3840c5bc | 3337 | if (amdgpu_device_supports_boco(adev->ddev)) |
83ba126a | 3338 | vga_switcheroo_fini_domain_pm_ops(adev->dev); |
d38ceaf9 AD |
3339 | vga_client_register(adev->pdev, NULL, NULL, NULL); |
3340 | if (adev->rio_mem) | |
3341 | pci_iounmap(adev->pdev, adev->rio_mem); | |
3342 | adev->rio_mem = NULL; | |
3343 | iounmap(adev->rmmio); | |
3344 | adev->rmmio = NULL; | |
06ec9070 | 3345 | amdgpu_device_doorbell_fini(adev); |
e9bc1bf7 | 3346 | |
dcea6e65 | 3347 | device_remove_file(adev->dev, &dev_attr_pcie_replay_count); |
7c868b59 YT |
3348 | if (adev->ucode_sysfs_en) |
3349 | amdgpu_ucode_sysfs_fini(adev); | |
bd607166 KR |
3350 | device_remove_file(adev->dev, &dev_attr_product_name); |
3351 | device_remove_file(adev->dev, &dev_attr_product_number); | |
3352 | device_remove_file(adev->dev, &dev_attr_serial_number); | |
d155bef0 AB |
3353 | if (IS_ENABLED(CONFIG_PERF_EVENTS)) |
3354 | amdgpu_pmu_fini(adev); | |
f54eeab4 | 3355 | if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10) |
a190d1c7 | 3356 | amdgpu_discovery_fini(adev); |
d38ceaf9 AD |
3357 | } |
3358 | ||
3359 | ||
3360 | /* | |
3361 | * Suspend & resume. | |
3362 | */ | |
3363 | /** | |
810ddc3a | 3364 | * amdgpu_device_suspend - initiate device suspend |
d38ceaf9 | 3365 | * |
87e3f136 DP |
3366 | * @dev: drm dev pointer |
3367 | * @suspend: suspend state | |
3368 | * @fbcon : notify the fbdev of suspend | |
d38ceaf9 AD |
3369 | * |
3370 | * Puts the hw in the suspend state (all asics). | |
3371 | * Returns 0 for success or an error on failure. | |
3372 | * Called at driver suspend. | |
3373 | */ | |
de185019 | 3374 | int amdgpu_device_suspend(struct drm_device *dev, bool fbcon) |
d38ceaf9 AD |
3375 | { |
3376 | struct amdgpu_device *adev; | |
3377 | struct drm_crtc *crtc; | |
3378 | struct drm_connector *connector; | |
f8d2d39e | 3379 | struct drm_connector_list_iter iter; |
5ceb54c6 | 3380 | int r; |
d38ceaf9 AD |
3381 | |
3382 | if (dev == NULL || dev->dev_private == NULL) { | |
3383 | return -ENODEV; | |
3384 | } | |
3385 | ||
3386 | adev = dev->dev_private; | |
3387 | ||
3388 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
3389 | return 0; | |
3390 | ||
44779b43 | 3391 | adev->in_suspend = true; |
d38ceaf9 AD |
3392 | drm_kms_helper_poll_disable(dev); |
3393 | ||
5f818173 S |
3394 | if (fbcon) |
3395 | amdgpu_fbdev_set_suspend(adev, 1); | |
3396 | ||
beff74bc | 3397 | cancel_delayed_work_sync(&adev->delayed_init_work); |
a5459475 | 3398 | |
4562236b HW |
3399 | if (!amdgpu_device_has_dc_support(adev)) { |
3400 | /* turn off display hw */ | |
3401 | drm_modeset_lock_all(dev); | |
f8d2d39e LP |
3402 | drm_connector_list_iter_begin(dev, &iter); |
3403 | drm_for_each_connector_iter(connector, &iter) | |
3404 | drm_helper_connector_dpms(connector, | |
3405 | DRM_MODE_DPMS_OFF); | |
3406 | drm_connector_list_iter_end(&iter); | |
4562236b | 3407 | drm_modeset_unlock_all(dev); |
fe1053b7 AD |
3408 | /* unpin the front buffers and cursors */ |
3409 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
3410 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); | |
3411 | struct drm_framebuffer *fb = crtc->primary->fb; | |
3412 | struct amdgpu_bo *robj; | |
3413 | ||
91334223 | 3414 | if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) { |
fe1053b7 AD |
3415 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); |
3416 | r = amdgpu_bo_reserve(aobj, true); | |
3417 | if (r == 0) { | |
3418 | amdgpu_bo_unpin(aobj); | |
3419 | amdgpu_bo_unreserve(aobj); | |
3420 | } | |
756e6880 | 3421 | } |
756e6880 | 3422 | |
fe1053b7 AD |
3423 | if (fb == NULL || fb->obj[0] == NULL) { |
3424 | continue; | |
3425 | } | |
3426 | robj = gem_to_amdgpu_bo(fb->obj[0]); | |
3427 | /* don't unpin kernel fb objects */ | |
3428 | if (!amdgpu_fbdev_robj_is_fb(adev, robj)) { | |
3429 | r = amdgpu_bo_reserve(robj, true); | |
3430 | if (r == 0) { | |
3431 | amdgpu_bo_unpin(robj); | |
3432 | amdgpu_bo_unreserve(robj); | |
3433 | } | |
d38ceaf9 AD |
3434 | } |
3435 | } | |
3436 | } | |
fe1053b7 | 3437 | |
5e6932fe | 3438 | amdgpu_ras_suspend(adev); |
3439 | ||
fe1053b7 AD |
3440 | r = amdgpu_device_ip_suspend_phase1(adev); |
3441 | ||
94fa5660 EQ |
3442 | amdgpu_amdkfd_suspend(adev, !fbcon); |
3443 | ||
d38ceaf9 AD |
3444 | /* evict vram memory */ |
3445 | amdgpu_bo_evict_vram(adev); | |
3446 | ||
5ceb54c6 | 3447 | amdgpu_fence_driver_suspend(adev); |
d38ceaf9 | 3448 | |
fe1053b7 | 3449 | r = amdgpu_device_ip_suspend_phase2(adev); |
d38ceaf9 | 3450 | |
a0a71e49 AD |
3451 | /* evict remaining vram memory |
3452 | * This second call to evict vram is to evict the gart page table | |
3453 | * using the CPU. | |
3454 | */ | |
d38ceaf9 AD |
3455 | amdgpu_bo_evict_vram(adev); |
3456 | ||
d38ceaf9 AD |
3457 | return 0; |
3458 | } | |
3459 | ||
3460 | /** | |
810ddc3a | 3461 | * amdgpu_device_resume - initiate device resume |
d38ceaf9 | 3462 | * |
87e3f136 DP |
3463 | * @dev: drm dev pointer |
3464 | * @resume: resume state | |
3465 | * @fbcon : notify the fbdev of resume | |
d38ceaf9 AD |
3466 | * |
3467 | * Bring the hw back to operating state (all asics). | |
3468 | * Returns 0 for success or an error on failure. | |
3469 | * Called at driver resume. | |
3470 | */ | |
de185019 | 3471 | int amdgpu_device_resume(struct drm_device *dev, bool fbcon) |
d38ceaf9 AD |
3472 | { |
3473 | struct drm_connector *connector; | |
f8d2d39e | 3474 | struct drm_connector_list_iter iter; |
d38ceaf9 | 3475 | struct amdgpu_device *adev = dev->dev_private; |
756e6880 | 3476 | struct drm_crtc *crtc; |
03161a6e | 3477 | int r = 0; |
d38ceaf9 AD |
3478 | |
3479 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
3480 | return 0; | |
3481 | ||
d38ceaf9 | 3482 | /* post card */ |
39c640c0 | 3483 | if (amdgpu_device_need_post(adev)) { |
74b0b157 | 3484 | r = amdgpu_atom_asic_init(adev->mode_info.atom_context); |
3485 | if (r) | |
3486 | DRM_ERROR("amdgpu asic init failed\n"); | |
3487 | } | |
d38ceaf9 | 3488 | |
06ec9070 | 3489 | r = amdgpu_device_ip_resume(adev); |
e6707218 | 3490 | if (r) { |
06ec9070 | 3491 | DRM_ERROR("amdgpu_device_ip_resume failed (%d).\n", r); |
4d3b9ae5 | 3492 | return r; |
e6707218 | 3493 | } |
5ceb54c6 AD |
3494 | amdgpu_fence_driver_resume(adev); |
3495 | ||
d38ceaf9 | 3496 | |
06ec9070 | 3497 | r = amdgpu_device_ip_late_init(adev); |
03161a6e | 3498 | if (r) |
4d3b9ae5 | 3499 | return r; |
d38ceaf9 | 3500 | |
beff74bc AD |
3501 | queue_delayed_work(system_wq, &adev->delayed_init_work, |
3502 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
3503 | ||
fe1053b7 AD |
3504 | if (!amdgpu_device_has_dc_support(adev)) { |
3505 | /* pin cursors */ | |
3506 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
3507 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); | |
3508 | ||
91334223 | 3509 | if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) { |
fe1053b7 AD |
3510 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); |
3511 | r = amdgpu_bo_reserve(aobj, true); | |
3512 | if (r == 0) { | |
3513 | r = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM); | |
3514 | if (r != 0) | |
3515 | DRM_ERROR("Failed to pin cursor BO (%d)\n", r); | |
3516 | amdgpu_crtc->cursor_addr = amdgpu_bo_gpu_offset(aobj); | |
3517 | amdgpu_bo_unreserve(aobj); | |
3518 | } | |
756e6880 AD |
3519 | } |
3520 | } | |
3521 | } | |
9593f4d6 | 3522 | r = amdgpu_amdkfd_resume(adev, !fbcon); |
ba997709 YZ |
3523 | if (r) |
3524 | return r; | |
756e6880 | 3525 | |
96a5d8d4 | 3526 | /* Make sure IB tests flushed */ |
beff74bc | 3527 | flush_delayed_work(&adev->delayed_init_work); |
96a5d8d4 | 3528 | |
d38ceaf9 AD |
3529 | /* blat the mode back in */ |
3530 | if (fbcon) { | |
4562236b HW |
3531 | if (!amdgpu_device_has_dc_support(adev)) { |
3532 | /* pre DCE11 */ | |
3533 | drm_helper_resume_force_mode(dev); | |
3534 | ||
3535 | /* turn on display hw */ | |
3536 | drm_modeset_lock_all(dev); | |
f8d2d39e LP |
3537 | |
3538 | drm_connector_list_iter_begin(dev, &iter); | |
3539 | drm_for_each_connector_iter(connector, &iter) | |
3540 | drm_helper_connector_dpms(connector, | |
3541 | DRM_MODE_DPMS_ON); | |
3542 | drm_connector_list_iter_end(&iter); | |
3543 | ||
4562236b | 3544 | drm_modeset_unlock_all(dev); |
d38ceaf9 | 3545 | } |
4d3b9ae5 | 3546 | amdgpu_fbdev_set_suspend(adev, 0); |
d38ceaf9 AD |
3547 | } |
3548 | ||
3549 | drm_kms_helper_poll_enable(dev); | |
23a1a9e5 | 3550 | |
5e6932fe | 3551 | amdgpu_ras_resume(adev); |
3552 | ||
23a1a9e5 L |
3553 | /* |
3554 | * Most of the connector probing functions try to acquire runtime pm | |
3555 | * refs to ensure that the GPU is powered on when connector polling is | |
3556 | * performed. Since we're calling this from a runtime PM callback, | |
3557 | * trying to acquire rpm refs will cause us to deadlock. | |
3558 | * | |
3559 | * Since we're guaranteed to be holding the rpm lock, it's safe to | |
3560 | * temporarily disable the rpm helpers so this doesn't deadlock us. | |
3561 | */ | |
3562 | #ifdef CONFIG_PM | |
3563 | dev->dev->power.disable_depth++; | |
3564 | #endif | |
4562236b HW |
3565 | if (!amdgpu_device_has_dc_support(adev)) |
3566 | drm_helper_hpd_irq_event(dev); | |
3567 | else | |
3568 | drm_kms_helper_hotplug_event(dev); | |
23a1a9e5 L |
3569 | #ifdef CONFIG_PM |
3570 | dev->dev->power.disable_depth--; | |
3571 | #endif | |
44779b43 RZ |
3572 | adev->in_suspend = false; |
3573 | ||
4d3b9ae5 | 3574 | return 0; |
d38ceaf9 AD |
3575 | } |
3576 | ||
e3ecdffa AD |
3577 | /** |
3578 | * amdgpu_device_ip_check_soft_reset - did soft reset succeed | |
3579 | * | |
3580 | * @adev: amdgpu_device pointer | |
3581 | * | |
3582 | * The list of all the hardware IPs that make up the asic is walked and | |
3583 | * the check_soft_reset callbacks are run. check_soft_reset determines | |
3584 | * if the asic is still hung or not. | |
3585 | * Returns true if any of the IPs are still in a hung state, false if not. | |
3586 | */ | |
06ec9070 | 3587 | static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev) |
63fbf42f CZ |
3588 | { |
3589 | int i; | |
3590 | bool asic_hang = false; | |
3591 | ||
f993d628 ML |
3592 | if (amdgpu_sriov_vf(adev)) |
3593 | return true; | |
3594 | ||
8bc04c29 AD |
3595 | if (amdgpu_asic_need_full_reset(adev)) |
3596 | return true; | |
3597 | ||
63fbf42f | 3598 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 3599 | if (!adev->ip_blocks[i].status.valid) |
63fbf42f | 3600 | continue; |
a1255107 AD |
3601 | if (adev->ip_blocks[i].version->funcs->check_soft_reset) |
3602 | adev->ip_blocks[i].status.hang = | |
3603 | adev->ip_blocks[i].version->funcs->check_soft_reset(adev); | |
3604 | if (adev->ip_blocks[i].status.hang) { | |
3605 | DRM_INFO("IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name); | |
63fbf42f CZ |
3606 | asic_hang = true; |
3607 | } | |
3608 | } | |
3609 | return asic_hang; | |
3610 | } | |
3611 | ||
e3ecdffa AD |
3612 | /** |
3613 | * amdgpu_device_ip_pre_soft_reset - prepare for soft reset | |
3614 | * | |
3615 | * @adev: amdgpu_device pointer | |
3616 | * | |
3617 | * The list of all the hardware IPs that make up the asic is walked and the | |
3618 | * pre_soft_reset callbacks are run if the block is hung. pre_soft_reset | |
3619 | * handles any IP specific hardware or software state changes that are | |
3620 | * necessary for a soft reset to succeed. | |
3621 | * Returns 0 on success, negative error code on failure. | |
3622 | */ | |
06ec9070 | 3623 | static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev) |
d31a501e CZ |
3624 | { |
3625 | int i, r = 0; | |
3626 | ||
3627 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 3628 | if (!adev->ip_blocks[i].status.valid) |
d31a501e | 3629 | continue; |
a1255107 AD |
3630 | if (adev->ip_blocks[i].status.hang && |
3631 | adev->ip_blocks[i].version->funcs->pre_soft_reset) { | |
3632 | r = adev->ip_blocks[i].version->funcs->pre_soft_reset(adev); | |
d31a501e CZ |
3633 | if (r) |
3634 | return r; | |
3635 | } | |
3636 | } | |
3637 | ||
3638 | return 0; | |
3639 | } | |
3640 | ||
e3ecdffa AD |
3641 | /** |
3642 | * amdgpu_device_ip_need_full_reset - check if a full asic reset is needed | |
3643 | * | |
3644 | * @adev: amdgpu_device pointer | |
3645 | * | |
3646 | * Some hardware IPs cannot be soft reset. If they are hung, a full gpu | |
3647 | * reset is necessary to recover. | |
3648 | * Returns true if a full asic reset is required, false if not. | |
3649 | */ | |
06ec9070 | 3650 | static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev) |
35d782fe | 3651 | { |
da146d3b AD |
3652 | int i; |
3653 | ||
8bc04c29 AD |
3654 | if (amdgpu_asic_need_full_reset(adev)) |
3655 | return true; | |
3656 | ||
da146d3b | 3657 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 3658 | if (!adev->ip_blocks[i].status.valid) |
da146d3b | 3659 | continue; |
a1255107 AD |
3660 | if ((adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) || |
3661 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) || | |
3662 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_ACP) || | |
98512bb8 KW |
3663 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) || |
3664 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) { | |
a1255107 | 3665 | if (adev->ip_blocks[i].status.hang) { |
da146d3b AD |
3666 | DRM_INFO("Some block need full reset!\n"); |
3667 | return true; | |
3668 | } | |
3669 | } | |
35d782fe CZ |
3670 | } |
3671 | return false; | |
3672 | } | |
3673 | ||
e3ecdffa AD |
3674 | /** |
3675 | * amdgpu_device_ip_soft_reset - do a soft reset | |
3676 | * | |
3677 | * @adev: amdgpu_device pointer | |
3678 | * | |
3679 | * The list of all the hardware IPs that make up the asic is walked and the | |
3680 | * soft_reset callbacks are run if the block is hung. soft_reset handles any | |
3681 | * IP specific hardware or software state changes that are necessary to soft | |
3682 | * reset the IP. | |
3683 | * Returns 0 on success, negative error code on failure. | |
3684 | */ | |
06ec9070 | 3685 | static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
3686 | { |
3687 | int i, r = 0; | |
3688 | ||
3689 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 3690 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 3691 | continue; |
a1255107 AD |
3692 | if (adev->ip_blocks[i].status.hang && |
3693 | adev->ip_blocks[i].version->funcs->soft_reset) { | |
3694 | r = adev->ip_blocks[i].version->funcs->soft_reset(adev); | |
35d782fe CZ |
3695 | if (r) |
3696 | return r; | |
3697 | } | |
3698 | } | |
3699 | ||
3700 | return 0; | |
3701 | } | |
3702 | ||
e3ecdffa AD |
3703 | /** |
3704 | * amdgpu_device_ip_post_soft_reset - clean up from soft reset | |
3705 | * | |
3706 | * @adev: amdgpu_device pointer | |
3707 | * | |
3708 | * The list of all the hardware IPs that make up the asic is walked and the | |
3709 | * post_soft_reset callbacks are run if the asic was hung. post_soft_reset | |
3710 | * handles any IP specific hardware or software state changes that are | |
3711 | * necessary after the IP has been soft reset. | |
3712 | * Returns 0 on success, negative error code on failure. | |
3713 | */ | |
06ec9070 | 3714 | static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
3715 | { |
3716 | int i, r = 0; | |
3717 | ||
3718 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 3719 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 3720 | continue; |
a1255107 AD |
3721 | if (adev->ip_blocks[i].status.hang && |
3722 | adev->ip_blocks[i].version->funcs->post_soft_reset) | |
3723 | r = adev->ip_blocks[i].version->funcs->post_soft_reset(adev); | |
35d782fe CZ |
3724 | if (r) |
3725 | return r; | |
3726 | } | |
3727 | ||
3728 | return 0; | |
3729 | } | |
3730 | ||
e3ecdffa | 3731 | /** |
c33adbc7 | 3732 | * amdgpu_device_recover_vram - Recover some VRAM contents |
e3ecdffa AD |
3733 | * |
3734 | * @adev: amdgpu_device pointer | |
3735 | * | |
3736 | * Restores the contents of VRAM buffers from the shadows in GTT. Used to | |
3737 | * restore things like GPUVM page tables after a GPU reset where | |
3738 | * the contents of VRAM might be lost. | |
403009bf CK |
3739 | * |
3740 | * Returns: | |
3741 | * 0 on success, negative error code on failure. | |
e3ecdffa | 3742 | */ |
c33adbc7 | 3743 | static int amdgpu_device_recover_vram(struct amdgpu_device *adev) |
c41d1cf6 | 3744 | { |
c41d1cf6 | 3745 | struct dma_fence *fence = NULL, *next = NULL; |
403009bf CK |
3746 | struct amdgpu_bo *shadow; |
3747 | long r = 1, tmo; | |
c41d1cf6 ML |
3748 | |
3749 | if (amdgpu_sriov_runtime(adev)) | |
b045d3af | 3750 | tmo = msecs_to_jiffies(8000); |
c41d1cf6 ML |
3751 | else |
3752 | tmo = msecs_to_jiffies(100); | |
3753 | ||
3754 | DRM_INFO("recover vram bo from shadow start\n"); | |
3755 | mutex_lock(&adev->shadow_list_lock); | |
403009bf CK |
3756 | list_for_each_entry(shadow, &adev->shadow_list, shadow_list) { |
3757 | ||
3758 | /* No need to recover an evicted BO */ | |
3759 | if (shadow->tbo.mem.mem_type != TTM_PL_TT || | |
b575f10d | 3760 | shadow->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET || |
403009bf CK |
3761 | shadow->parent->tbo.mem.mem_type != TTM_PL_VRAM) |
3762 | continue; | |
3763 | ||
3764 | r = amdgpu_bo_restore_shadow(shadow, &next); | |
3765 | if (r) | |
3766 | break; | |
3767 | ||
c41d1cf6 | 3768 | if (fence) { |
1712fb1a | 3769 | tmo = dma_fence_wait_timeout(fence, false, tmo); |
403009bf CK |
3770 | dma_fence_put(fence); |
3771 | fence = next; | |
1712fb1a | 3772 | if (tmo == 0) { |
3773 | r = -ETIMEDOUT; | |
c41d1cf6 | 3774 | break; |
1712fb1a | 3775 | } else if (tmo < 0) { |
3776 | r = tmo; | |
3777 | break; | |
3778 | } | |
403009bf CK |
3779 | } else { |
3780 | fence = next; | |
c41d1cf6 | 3781 | } |
c41d1cf6 ML |
3782 | } |
3783 | mutex_unlock(&adev->shadow_list_lock); | |
3784 | ||
403009bf CK |
3785 | if (fence) |
3786 | tmo = dma_fence_wait_timeout(fence, false, tmo); | |
c41d1cf6 ML |
3787 | dma_fence_put(fence); |
3788 | ||
1712fb1a | 3789 | if (r < 0 || tmo <= 0) { |
3790 | DRM_ERROR("recover vram bo from shadow failed, r is %ld, tmo is %ld\n", r, tmo); | |
403009bf CK |
3791 | return -EIO; |
3792 | } | |
c41d1cf6 | 3793 | |
403009bf CK |
3794 | DRM_INFO("recover vram bo from shadow done\n"); |
3795 | return 0; | |
c41d1cf6 ML |
3796 | } |
3797 | ||
a90ad3c2 | 3798 | |
e3ecdffa | 3799 | /** |
06ec9070 | 3800 | * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf |
5740682e ML |
3801 | * |
3802 | * @adev: amdgpu device pointer | |
87e3f136 | 3803 | * @from_hypervisor: request from hypervisor |
5740682e ML |
3804 | * |
3805 | * do VF FLR and reinitialize Asic | |
3f48c681 | 3806 | * return 0 means succeeded otherwise failed |
e3ecdffa AD |
3807 | */ |
3808 | static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, | |
3809 | bool from_hypervisor) | |
5740682e ML |
3810 | { |
3811 | int r; | |
3812 | ||
3813 | if (from_hypervisor) | |
3814 | r = amdgpu_virt_request_full_gpu(adev, true); | |
3815 | else | |
3816 | r = amdgpu_virt_reset_gpu(adev); | |
3817 | if (r) | |
3818 | return r; | |
a90ad3c2 | 3819 | |
b639c22c JZ |
3820 | amdgpu_amdkfd_pre_reset(adev); |
3821 | ||
a90ad3c2 | 3822 | /* Resume IP prior to SMC */ |
06ec9070 | 3823 | r = amdgpu_device_ip_reinit_early_sriov(adev); |
5740682e ML |
3824 | if (r) |
3825 | goto error; | |
a90ad3c2 | 3826 | |
c9ffa427 | 3827 | amdgpu_virt_init_data_exchange(adev); |
a90ad3c2 | 3828 | /* we need recover gart prior to run SMC/CP/SDMA resume */ |
c1c7ce8f | 3829 | amdgpu_gtt_mgr_recover(&adev->mman.bdev.man[TTM_PL_TT]); |
a90ad3c2 | 3830 | |
7a3e0bb2 RZ |
3831 | r = amdgpu_device_fw_loading(adev); |
3832 | if (r) | |
3833 | return r; | |
3834 | ||
a90ad3c2 | 3835 | /* now we are okay to resume SMC/CP/SDMA */ |
06ec9070 | 3836 | r = amdgpu_device_ip_reinit_late_sriov(adev); |
5740682e ML |
3837 | if (r) |
3838 | goto error; | |
a90ad3c2 ML |
3839 | |
3840 | amdgpu_irq_gpu_reset_resume_helper(adev); | |
5740682e | 3841 | r = amdgpu_ib_ring_tests(adev); |
f81e8d53 | 3842 | amdgpu_amdkfd_post_reset(adev); |
a90ad3c2 | 3843 | |
abc34253 ED |
3844 | error: |
3845 | amdgpu_virt_release_full_gpu(adev, true); | |
c41d1cf6 | 3846 | if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) { |
e3526257 | 3847 | amdgpu_inc_vram_lost(adev); |
c33adbc7 | 3848 | r = amdgpu_device_recover_vram(adev); |
a90ad3c2 ML |
3849 | } |
3850 | ||
3851 | return r; | |
3852 | } | |
3853 | ||
12938fad CK |
3854 | /** |
3855 | * amdgpu_device_should_recover_gpu - check if we should try GPU recovery | |
3856 | * | |
3857 | * @adev: amdgpu device pointer | |
3858 | * | |
3859 | * Check amdgpu_gpu_recovery and SRIOV status to see if we should try to recover | |
3860 | * a hung GPU. | |
3861 | */ | |
3862 | bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev) | |
3863 | { | |
3864 | if (!amdgpu_device_ip_check_soft_reset(adev)) { | |
3865 | DRM_INFO("Timeout, but no hardware hang detected.\n"); | |
3866 | return false; | |
3867 | } | |
3868 | ||
3ba7b418 AG |
3869 | if (amdgpu_gpu_recovery == 0) |
3870 | goto disabled; | |
3871 | ||
3872 | if (amdgpu_sriov_vf(adev)) | |
3873 | return true; | |
3874 | ||
3875 | if (amdgpu_gpu_recovery == -1) { | |
3876 | switch (adev->asic_type) { | |
fc42d47c AG |
3877 | case CHIP_BONAIRE: |
3878 | case CHIP_HAWAII: | |
3ba7b418 AG |
3879 | case CHIP_TOPAZ: |
3880 | case CHIP_TONGA: | |
3881 | case CHIP_FIJI: | |
3882 | case CHIP_POLARIS10: | |
3883 | case CHIP_POLARIS11: | |
3884 | case CHIP_POLARIS12: | |
3885 | case CHIP_VEGAM: | |
3886 | case CHIP_VEGA20: | |
3887 | case CHIP_VEGA10: | |
3888 | case CHIP_VEGA12: | |
c43b849f | 3889 | case CHIP_RAVEN: |
e9d4cf91 | 3890 | case CHIP_ARCTURUS: |
2cb44fb0 | 3891 | case CHIP_RENOIR: |
658c6639 AD |
3892 | case CHIP_NAVI10: |
3893 | case CHIP_NAVI14: | |
3894 | case CHIP_NAVI12: | |
3ba7b418 AG |
3895 | break; |
3896 | default: | |
3897 | goto disabled; | |
3898 | } | |
12938fad CK |
3899 | } |
3900 | ||
3901 | return true; | |
3ba7b418 AG |
3902 | |
3903 | disabled: | |
3904 | DRM_INFO("GPU recovery disabled.\n"); | |
3905 | return false; | |
12938fad CK |
3906 | } |
3907 | ||
5c6dd71e | 3908 | |
26bc5340 AG |
3909 | static int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, |
3910 | struct amdgpu_job *job, | |
3911 | bool *need_full_reset_arg) | |
3912 | { | |
3913 | int i, r = 0; | |
3914 | bool need_full_reset = *need_full_reset_arg; | |
71182665 | 3915 | |
71182665 | 3916 | /* block all schedulers and reset given job's ring */ |
0875dc9e CZ |
3917 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
3918 | struct amdgpu_ring *ring = adev->rings[i]; | |
3919 | ||
51687759 | 3920 | if (!ring || !ring->sched.thread) |
0875dc9e | 3921 | continue; |
5740682e | 3922 | |
2f9d4084 ML |
3923 | /* after all hw jobs are reset, hw fence is meaningless, so force_completion */ |
3924 | amdgpu_fence_driver_force_completion(ring); | |
0875dc9e | 3925 | } |
d38ceaf9 | 3926 | |
222b5f04 AG |
3927 | if(job) |
3928 | drm_sched_increase_karma(&job->base); | |
3929 | ||
1d721ed6 | 3930 | /* Don't suspend on bare metal if we are not going to HW reset the ASIC */ |
26bc5340 AG |
3931 | if (!amdgpu_sriov_vf(adev)) { |
3932 | ||
3933 | if (!need_full_reset) | |
3934 | need_full_reset = amdgpu_device_ip_need_full_reset(adev); | |
3935 | ||
3936 | if (!need_full_reset) { | |
3937 | amdgpu_device_ip_pre_soft_reset(adev); | |
3938 | r = amdgpu_device_ip_soft_reset(adev); | |
3939 | amdgpu_device_ip_post_soft_reset(adev); | |
3940 | if (r || amdgpu_device_ip_check_soft_reset(adev)) { | |
3941 | DRM_INFO("soft reset failed, will fallback to full reset!\n"); | |
3942 | need_full_reset = true; | |
3943 | } | |
3944 | } | |
3945 | ||
3946 | if (need_full_reset) | |
3947 | r = amdgpu_device_ip_suspend(adev); | |
3948 | ||
3949 | *need_full_reset_arg = need_full_reset; | |
3950 | } | |
3951 | ||
3952 | return r; | |
3953 | } | |
3954 | ||
041a62bc | 3955 | static int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, |
26bc5340 AG |
3956 | struct list_head *device_list_handle, |
3957 | bool *need_full_reset_arg) | |
3958 | { | |
3959 | struct amdgpu_device *tmp_adev = NULL; | |
3960 | bool need_full_reset = *need_full_reset_arg, vram_lost = false; | |
3961 | int r = 0; | |
3962 | ||
3963 | /* | |
3964 | * ASIC reset has to be done on all HGMI hive nodes ASAP | |
3965 | * to allow proper links negotiation in FW (within 1 sec) | |
3966 | */ | |
3967 | if (need_full_reset) { | |
3968 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
041a62bc | 3969 | /* For XGMI run all resets in parallel to speed up the process */ |
d4535e2c | 3970 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { |
c96cf282 | 3971 | if (!queue_work(system_unbound_wq, &tmp_adev->xgmi_reset_work)) |
d4535e2c AG |
3972 | r = -EALREADY; |
3973 | } else | |
3974 | r = amdgpu_asic_reset(tmp_adev); | |
d4535e2c | 3975 | |
041a62bc AG |
3976 | if (r) { |
3977 | DRM_ERROR("ASIC reset failed with error, %d for drm dev, %s", | |
3978 | r, tmp_adev->ddev->unique); | |
3979 | break; | |
ce316fa5 LM |
3980 | } |
3981 | } | |
3982 | ||
041a62bc AG |
3983 | /* For XGMI wait for all resets to complete before proceed */ |
3984 | if (!r) { | |
ce316fa5 LM |
3985 | list_for_each_entry(tmp_adev, device_list_handle, |
3986 | gmc.xgmi.head) { | |
3987 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { | |
3988 | flush_work(&tmp_adev->xgmi_reset_work); | |
3989 | r = tmp_adev->asic_reset_res; | |
3990 | if (r) | |
3991 | break; | |
ce316fa5 LM |
3992 | } |
3993 | } | |
3994 | } | |
ce316fa5 | 3995 | } |
26bc5340 | 3996 | |
43c4d576 JC |
3997 | if (!r && amdgpu_ras_intr_triggered()) { |
3998 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
3999 | if (tmp_adev->mmhub.funcs && | |
4000 | tmp_adev->mmhub.funcs->reset_ras_error_count) | |
4001 | tmp_adev->mmhub.funcs->reset_ras_error_count(tmp_adev); | |
4002 | } | |
4003 | ||
00eaa571 | 4004 | amdgpu_ras_intr_cleared(); |
43c4d576 | 4005 | } |
00eaa571 | 4006 | |
26bc5340 AG |
4007 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { |
4008 | if (need_full_reset) { | |
4009 | /* post card */ | |
4010 | if (amdgpu_atom_asic_init(tmp_adev->mode_info.atom_context)) | |
4011 | DRM_WARN("asic atom init failed!"); | |
4012 | ||
4013 | if (!r) { | |
4014 | dev_info(tmp_adev->dev, "GPU reset succeeded, trying to resume\n"); | |
4015 | r = amdgpu_device_ip_resume_phase1(tmp_adev); | |
4016 | if (r) | |
4017 | goto out; | |
4018 | ||
4019 | vram_lost = amdgpu_device_check_vram_lost(tmp_adev); | |
4020 | if (vram_lost) { | |
77e7f829 | 4021 | DRM_INFO("VRAM is lost due to GPU reset!\n"); |
e3526257 | 4022 | amdgpu_inc_vram_lost(tmp_adev); |
26bc5340 AG |
4023 | } |
4024 | ||
4025 | r = amdgpu_gtt_mgr_recover( | |
4026 | &tmp_adev->mman.bdev.man[TTM_PL_TT]); | |
4027 | if (r) | |
4028 | goto out; | |
4029 | ||
4030 | r = amdgpu_device_fw_loading(tmp_adev); | |
4031 | if (r) | |
4032 | return r; | |
4033 | ||
4034 | r = amdgpu_device_ip_resume_phase2(tmp_adev); | |
4035 | if (r) | |
4036 | goto out; | |
4037 | ||
4038 | if (vram_lost) | |
4039 | amdgpu_device_fill_reset_magic(tmp_adev); | |
4040 | ||
fdafb359 EQ |
4041 | /* |
4042 | * Add this ASIC as tracked as reset was already | |
4043 | * complete successfully. | |
4044 | */ | |
4045 | amdgpu_register_gpu_instance(tmp_adev); | |
4046 | ||
7c04ca50 | 4047 | r = amdgpu_device_ip_late_init(tmp_adev); |
4048 | if (r) | |
4049 | goto out; | |
4050 | ||
565d1941 EQ |
4051 | amdgpu_fbdev_set_suspend(tmp_adev, 0); |
4052 | ||
e79a04d5 | 4053 | /* must succeed. */ |
511fdbc3 | 4054 | amdgpu_ras_resume(tmp_adev); |
e79a04d5 | 4055 | |
26bc5340 AG |
4056 | /* Update PSP FW topology after reset */ |
4057 | if (hive && tmp_adev->gmc.xgmi.num_physical_nodes > 1) | |
4058 | r = amdgpu_xgmi_update_topology(hive, tmp_adev); | |
4059 | } | |
4060 | } | |
4061 | ||
4062 | ||
4063 | out: | |
4064 | if (!r) { | |
4065 | amdgpu_irq_gpu_reset_resume_helper(tmp_adev); | |
4066 | r = amdgpu_ib_ring_tests(tmp_adev); | |
4067 | if (r) { | |
4068 | dev_err(tmp_adev->dev, "ib ring test failed (%d).\n", r); | |
4069 | r = amdgpu_device_ip_suspend(tmp_adev); | |
4070 | need_full_reset = true; | |
4071 | r = -EAGAIN; | |
4072 | goto end; | |
4073 | } | |
4074 | } | |
4075 | ||
4076 | if (!r) | |
4077 | r = amdgpu_device_recover_vram(tmp_adev); | |
4078 | else | |
4079 | tmp_adev->asic_reset_res = r; | |
4080 | } | |
4081 | ||
4082 | end: | |
4083 | *need_full_reset_arg = need_full_reset; | |
4084 | return r; | |
4085 | } | |
4086 | ||
1d721ed6 | 4087 | static bool amdgpu_device_lock_adev(struct amdgpu_device *adev, bool trylock) |
26bc5340 | 4088 | { |
1d721ed6 AG |
4089 | if (trylock) { |
4090 | if (!mutex_trylock(&adev->lock_reset)) | |
4091 | return false; | |
4092 | } else | |
4093 | mutex_lock(&adev->lock_reset); | |
5740682e | 4094 | |
26bc5340 | 4095 | atomic_inc(&adev->gpu_reset_counter); |
2a9b90ae | 4096 | adev->in_gpu_reset = true; |
a3a09142 AD |
4097 | switch (amdgpu_asic_reset_method(adev)) { |
4098 | case AMD_RESET_METHOD_MODE1: | |
4099 | adev->mp1_state = PP_MP1_STATE_SHUTDOWN; | |
4100 | break; | |
4101 | case AMD_RESET_METHOD_MODE2: | |
4102 | adev->mp1_state = PP_MP1_STATE_RESET; | |
4103 | break; | |
4104 | default: | |
4105 | adev->mp1_state = PP_MP1_STATE_NONE; | |
4106 | break; | |
4107 | } | |
1d721ed6 AG |
4108 | |
4109 | return true; | |
26bc5340 | 4110 | } |
d38ceaf9 | 4111 | |
26bc5340 AG |
4112 | static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) |
4113 | { | |
89041940 | 4114 | amdgpu_vf_error_trans_all(adev); |
a3a09142 | 4115 | adev->mp1_state = PP_MP1_STATE_NONE; |
2a9b90ae | 4116 | adev->in_gpu_reset = false; |
13a752e3 | 4117 | mutex_unlock(&adev->lock_reset); |
26bc5340 AG |
4118 | } |
4119 | ||
3f12acc8 EQ |
4120 | static void amdgpu_device_resume_display_audio(struct amdgpu_device *adev) |
4121 | { | |
4122 | struct pci_dev *p = NULL; | |
4123 | ||
4124 | p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus), | |
4125 | adev->pdev->bus->number, 1); | |
4126 | if (p) { | |
4127 | pm_runtime_enable(&(p->dev)); | |
4128 | pm_runtime_resume(&(p->dev)); | |
4129 | } | |
4130 | } | |
4131 | ||
4132 | static int amdgpu_device_suspend_display_audio(struct amdgpu_device *adev) | |
4133 | { | |
4134 | enum amd_reset_method reset_method; | |
4135 | struct pci_dev *p = NULL; | |
4136 | u64 expires; | |
4137 | ||
4138 | /* | |
4139 | * For now, only BACO and mode1 reset are confirmed | |
4140 | * to suffer the audio issue without proper suspended. | |
4141 | */ | |
4142 | reset_method = amdgpu_asic_reset_method(adev); | |
4143 | if ((reset_method != AMD_RESET_METHOD_BACO) && | |
4144 | (reset_method != AMD_RESET_METHOD_MODE1)) | |
4145 | return -EINVAL; | |
4146 | ||
4147 | p = pci_get_domain_bus_and_slot(pci_domain_nr(adev->pdev->bus), | |
4148 | adev->pdev->bus->number, 1); | |
4149 | if (!p) | |
4150 | return -ENODEV; | |
4151 | ||
4152 | expires = pm_runtime_autosuspend_expiration(&(p->dev)); | |
4153 | if (!expires) | |
4154 | /* | |
4155 | * If we cannot get the audio device autosuspend delay, | |
4156 | * a fixed 4S interval will be used. Considering 3S is | |
4157 | * the audio controller default autosuspend delay setting. | |
4158 | * 4S used here is guaranteed to cover that. | |
4159 | */ | |
54b7feb9 | 4160 | expires = ktime_get_mono_fast_ns() + NSEC_PER_SEC * 4ULL; |
3f12acc8 EQ |
4161 | |
4162 | while (!pm_runtime_status_suspended(&(p->dev))) { | |
4163 | if (!pm_runtime_suspend(&(p->dev))) | |
4164 | break; | |
4165 | ||
4166 | if (expires < ktime_get_mono_fast_ns()) { | |
4167 | dev_warn(adev->dev, "failed to suspend display audio\n"); | |
4168 | /* TODO: abort the succeeding gpu reset? */ | |
4169 | return -ETIMEDOUT; | |
4170 | } | |
4171 | } | |
4172 | ||
4173 | pm_runtime_disable(&(p->dev)); | |
4174 | ||
4175 | return 0; | |
4176 | } | |
4177 | ||
26bc5340 AG |
4178 | /** |
4179 | * amdgpu_device_gpu_recover - reset the asic and recover scheduler | |
4180 | * | |
4181 | * @adev: amdgpu device pointer | |
4182 | * @job: which job trigger hang | |
4183 | * | |
4184 | * Attempt to reset the GPU if it has hung (all asics). | |
4185 | * Attempt to do soft-reset or full-reset and reinitialize Asic | |
4186 | * Returns 0 for success or an error on failure. | |
4187 | */ | |
4188 | ||
4189 | int amdgpu_device_gpu_recover(struct amdgpu_device *adev, | |
4190 | struct amdgpu_job *job) | |
4191 | { | |
1d721ed6 | 4192 | struct list_head device_list, *device_list_handle = NULL; |
7dd8c205 EQ |
4193 | bool need_full_reset = false; |
4194 | bool job_signaled = false; | |
26bc5340 | 4195 | struct amdgpu_hive_info *hive = NULL; |
26bc5340 | 4196 | struct amdgpu_device *tmp_adev = NULL; |
1d721ed6 | 4197 | int i, r = 0; |
7c6e68c7 | 4198 | bool in_ras_intr = amdgpu_ras_intr_triggered(); |
b823821f LM |
4199 | bool use_baco = |
4200 | (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) ? | |
4201 | true : false; | |
3f12acc8 | 4202 | bool audio_suspended = false; |
26bc5340 | 4203 | |
d5ea093e AG |
4204 | /* |
4205 | * Flush RAM to disk so that after reboot | |
4206 | * the user can read log and see why the system rebooted. | |
4207 | */ | |
b823821f | 4208 | if (in_ras_intr && !use_baco && amdgpu_ras_get_context(adev)->reboot) { |
d5ea093e AG |
4209 | |
4210 | DRM_WARN("Emergency reboot."); | |
4211 | ||
4212 | ksys_sync_helper(); | |
4213 | emergency_restart(); | |
4214 | } | |
4215 | ||
b823821f LM |
4216 | dev_info(adev->dev, "GPU %s begin!\n", |
4217 | (in_ras_intr && !use_baco) ? "jobs stop":"reset"); | |
26bc5340 AG |
4218 | |
4219 | /* | |
1d721ed6 AG |
4220 | * Here we trylock to avoid chain of resets executing from |
4221 | * either trigger by jobs on different adevs in XGMI hive or jobs on | |
4222 | * different schedulers for same device while this TO handler is running. | |
4223 | * We always reset all schedulers for device and all devices for XGMI | |
4224 | * hive so that should take care of them too. | |
26bc5340 | 4225 | */ |
7dd8c205 | 4226 | hive = amdgpu_get_xgmi_hive(adev, true); |
1d721ed6 AG |
4227 | if (hive && !mutex_trylock(&hive->reset_lock)) { |
4228 | DRM_INFO("Bailing on TDR for s_job:%llx, hive: %llx as another already in progress", | |
0b2d2c2e | 4229 | job ? job->base.id : -1, hive->hive_id); |
9e94d22c | 4230 | mutex_unlock(&hive->hive_lock); |
26bc5340 | 4231 | return 0; |
1d721ed6 | 4232 | } |
26bc5340 | 4233 | |
9e94d22c EQ |
4234 | /* |
4235 | * Build list of devices to reset. | |
4236 | * In case we are in XGMI hive mode, resort the device list | |
4237 | * to put adev in the 1st position. | |
4238 | */ | |
4239 | INIT_LIST_HEAD(&device_list); | |
4240 | if (adev->gmc.xgmi.num_physical_nodes > 1) { | |
4241 | if (!hive) | |
26bc5340 | 4242 | return -ENODEV; |
9e94d22c EQ |
4243 | if (!list_is_first(&adev->gmc.xgmi.head, &hive->device_list)) |
4244 | list_rotate_to_front(&adev->gmc.xgmi.head, &hive->device_list); | |
26bc5340 AG |
4245 | device_list_handle = &hive->device_list; |
4246 | } else { | |
4247 | list_add_tail(&adev->gmc.xgmi.head, &device_list); | |
4248 | device_list_handle = &device_list; | |
4249 | } | |
4250 | ||
1d721ed6 AG |
4251 | /* block all schedulers and reset given job's ring */ |
4252 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
9e94d22c EQ |
4253 | if (!amdgpu_device_lock_adev(tmp_adev, !hive)) { |
4254 | DRM_INFO("Bailing on TDR for s_job:%llx, as another already in progress", | |
4255 | job ? job->base.id : -1); | |
4256 | mutex_unlock(&hive->hive_lock); | |
4257 | return 0; | |
7c6e68c7 AG |
4258 | } |
4259 | ||
3f12acc8 EQ |
4260 | /* |
4261 | * Try to put the audio codec into suspend state | |
4262 | * before gpu reset started. | |
4263 | * | |
4264 | * Due to the power domain of the graphics device | |
4265 | * is shared with AZ power domain. Without this, | |
4266 | * we may change the audio hardware from behind | |
4267 | * the audio driver's back. That will trigger | |
4268 | * some audio codec errors. | |
4269 | */ | |
4270 | if (!amdgpu_device_suspend_display_audio(tmp_adev)) | |
4271 | audio_suspended = true; | |
4272 | ||
9e94d22c EQ |
4273 | amdgpu_ras_set_error_query_ready(tmp_adev, false); |
4274 | ||
52fb44cf EQ |
4275 | cancel_delayed_work_sync(&tmp_adev->delayed_init_work); |
4276 | ||
9e94d22c EQ |
4277 | if (!amdgpu_sriov_vf(tmp_adev)) |
4278 | amdgpu_amdkfd_pre_reset(tmp_adev); | |
4279 | ||
12ffa55d AG |
4280 | /* |
4281 | * Mark these ASICs to be reseted as untracked first | |
4282 | * And add them back after reset completed | |
4283 | */ | |
4284 | amdgpu_unregister_gpu_instance(tmp_adev); | |
4285 | ||
a2f63ee8 | 4286 | amdgpu_fbdev_set_suspend(tmp_adev, 1); |
565d1941 | 4287 | |
f1c1314b | 4288 | /* disable ras on ALL IPs */ |
b823821f LM |
4289 | if (!(in_ras_intr && !use_baco) && |
4290 | amdgpu_device_ip_need_full_reset(tmp_adev)) | |
f1c1314b | 4291 | amdgpu_ras_suspend(tmp_adev); |
4292 | ||
1d721ed6 AG |
4293 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
4294 | struct amdgpu_ring *ring = tmp_adev->rings[i]; | |
4295 | ||
4296 | if (!ring || !ring->sched.thread) | |
4297 | continue; | |
4298 | ||
0b2d2c2e | 4299 | drm_sched_stop(&ring->sched, job ? &job->base : NULL); |
7c6e68c7 | 4300 | |
b823821f | 4301 | if (in_ras_intr && !use_baco) |
7c6e68c7 | 4302 | amdgpu_job_stop_all_jobs_on_sched(&ring->sched); |
1d721ed6 AG |
4303 | } |
4304 | } | |
4305 | ||
b823821f | 4306 | if (in_ras_intr && !use_baco) |
7c6e68c7 AG |
4307 | goto skip_sched_resume; |
4308 | ||
1d721ed6 AG |
4309 | /* |
4310 | * Must check guilty signal here since after this point all old | |
4311 | * HW fences are force signaled. | |
4312 | * | |
4313 | * job->base holds a reference to parent fence | |
4314 | */ | |
4315 | if (job && job->base.s_fence->parent && | |
7dd8c205 | 4316 | dma_fence_is_signaled(job->base.s_fence->parent)) { |
1d721ed6 | 4317 | job_signaled = true; |
1d721ed6 AG |
4318 | dev_info(adev->dev, "Guilty job already signaled, skipping HW reset"); |
4319 | goto skip_hw_reset; | |
4320 | } | |
4321 | ||
26bc5340 AG |
4322 | retry: /* Rest of adevs pre asic reset from XGMI hive. */ |
4323 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
26bc5340 AG |
4324 | r = amdgpu_device_pre_asic_reset(tmp_adev, |
4325 | NULL, | |
4326 | &need_full_reset); | |
4327 | /*TODO Should we stop ?*/ | |
4328 | if (r) { | |
4329 | DRM_ERROR("GPU pre asic reset failed with err, %d for drm dev, %s ", | |
4330 | r, tmp_adev->ddev->unique); | |
4331 | tmp_adev->asic_reset_res = r; | |
4332 | } | |
4333 | } | |
4334 | ||
4335 | /* Actual ASIC resets if needed.*/ | |
4336 | /* TODO Implement XGMI hive reset logic for SRIOV */ | |
4337 | if (amdgpu_sriov_vf(adev)) { | |
4338 | r = amdgpu_device_reset_sriov(adev, job ? false : true); | |
4339 | if (r) | |
4340 | adev->asic_reset_res = r; | |
4341 | } else { | |
041a62bc | 4342 | r = amdgpu_do_asic_reset(hive, device_list_handle, &need_full_reset); |
26bc5340 AG |
4343 | if (r && r == -EAGAIN) |
4344 | goto retry; | |
4345 | } | |
4346 | ||
1d721ed6 AG |
4347 | skip_hw_reset: |
4348 | ||
26bc5340 AG |
4349 | /* Post ASIC reset for all devs .*/ |
4350 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
7c6e68c7 | 4351 | |
1d721ed6 AG |
4352 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
4353 | struct amdgpu_ring *ring = tmp_adev->rings[i]; | |
4354 | ||
4355 | if (!ring || !ring->sched.thread) | |
4356 | continue; | |
4357 | ||
4358 | /* No point to resubmit jobs if we didn't HW reset*/ | |
4359 | if (!tmp_adev->asic_reset_res && !job_signaled) | |
4360 | drm_sched_resubmit_jobs(&ring->sched); | |
4361 | ||
4362 | drm_sched_start(&ring->sched, !tmp_adev->asic_reset_res); | |
4363 | } | |
4364 | ||
4365 | if (!amdgpu_device_has_dc_support(tmp_adev) && !job_signaled) { | |
4366 | drm_helper_resume_force_mode(tmp_adev->ddev); | |
4367 | } | |
4368 | ||
4369 | tmp_adev->asic_reset_res = 0; | |
26bc5340 AG |
4370 | |
4371 | if (r) { | |
4372 | /* bad news, how to tell it to userspace ? */ | |
12ffa55d | 4373 | dev_info(tmp_adev->dev, "GPU reset(%d) failed\n", atomic_read(&tmp_adev->gpu_reset_counter)); |
26bc5340 AG |
4374 | amdgpu_vf_error_put(tmp_adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r); |
4375 | } else { | |
12ffa55d | 4376 | dev_info(tmp_adev->dev, "GPU reset(%d) succeeded!\n", atomic_read(&tmp_adev->gpu_reset_counter)); |
26bc5340 | 4377 | } |
7c6e68c7 | 4378 | } |
26bc5340 | 4379 | |
7c6e68c7 AG |
4380 | skip_sched_resume: |
4381 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
4382 | /*unlock kfd: SRIOV would do it separately */ | |
b823821f | 4383 | if (!(in_ras_intr && !use_baco) && !amdgpu_sriov_vf(tmp_adev)) |
7c6e68c7 | 4384 | amdgpu_amdkfd_post_reset(tmp_adev); |
3f12acc8 EQ |
4385 | if (audio_suspended) |
4386 | amdgpu_device_resume_display_audio(tmp_adev); | |
26bc5340 AG |
4387 | amdgpu_device_unlock_adev(tmp_adev); |
4388 | } | |
4389 | ||
9e94d22c | 4390 | if (hive) { |
22d6575b | 4391 | mutex_unlock(&hive->reset_lock); |
9e94d22c EQ |
4392 | mutex_unlock(&hive->hive_lock); |
4393 | } | |
26bc5340 AG |
4394 | |
4395 | if (r) | |
4396 | dev_info(adev->dev, "GPU reset end with ret = %d\n", r); | |
d38ceaf9 AD |
4397 | return r; |
4398 | } | |
4399 | ||
e3ecdffa AD |
4400 | /** |
4401 | * amdgpu_device_get_pcie_info - fence pcie info about the PCIE slot | |
4402 | * | |
4403 | * @adev: amdgpu_device pointer | |
4404 | * | |
4405 | * Fetchs and stores in the driver the PCIE capabilities (gen speed | |
4406 | * and lanes) of the slot the device is in. Handles APUs and | |
4407 | * virtualized environments where PCIE config space may not be available. | |
4408 | */ | |
5494d864 | 4409 | static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) |
d0dd7f0c | 4410 | { |
5d9a6330 | 4411 | struct pci_dev *pdev; |
c5313457 HK |
4412 | enum pci_bus_speed speed_cap, platform_speed_cap; |
4413 | enum pcie_link_width platform_link_width; | |
d0dd7f0c | 4414 | |
cd474ba0 AD |
4415 | if (amdgpu_pcie_gen_cap) |
4416 | adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap; | |
d0dd7f0c | 4417 | |
cd474ba0 AD |
4418 | if (amdgpu_pcie_lane_cap) |
4419 | adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap; | |
d0dd7f0c | 4420 | |
cd474ba0 AD |
4421 | /* covers APUs as well */ |
4422 | if (pci_is_root_bus(adev->pdev->bus)) { | |
4423 | if (adev->pm.pcie_gen_mask == 0) | |
4424 | adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; | |
4425 | if (adev->pm.pcie_mlw_mask == 0) | |
4426 | adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; | |
d0dd7f0c | 4427 | return; |
cd474ba0 | 4428 | } |
d0dd7f0c | 4429 | |
c5313457 HK |
4430 | if (adev->pm.pcie_gen_mask && adev->pm.pcie_mlw_mask) |
4431 | return; | |
4432 | ||
dbaa922b AD |
4433 | pcie_bandwidth_available(adev->pdev, NULL, |
4434 | &platform_speed_cap, &platform_link_width); | |
c5313457 | 4435 | |
cd474ba0 | 4436 | if (adev->pm.pcie_gen_mask == 0) { |
5d9a6330 AD |
4437 | /* asic caps */ |
4438 | pdev = adev->pdev; | |
4439 | speed_cap = pcie_get_speed_cap(pdev); | |
4440 | if (speed_cap == PCI_SPEED_UNKNOWN) { | |
4441 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
cd474ba0 AD |
4442 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | |
4443 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
cd474ba0 | 4444 | } else { |
5d9a6330 AD |
4445 | if (speed_cap == PCIE_SPEED_16_0GT) |
4446 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
4447 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4448 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
4449 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4); | |
4450 | else if (speed_cap == PCIE_SPEED_8_0GT) | |
4451 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
4452 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4453 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
4454 | else if (speed_cap == PCIE_SPEED_5_0GT) | |
4455 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
4456 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
4457 | else | |
4458 | adev->pm.pcie_gen_mask |= CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
4459 | } | |
4460 | /* platform caps */ | |
c5313457 | 4461 | if (platform_speed_cap == PCI_SPEED_UNKNOWN) { |
5d9a6330 AD |
4462 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4463 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
4464 | } else { | |
c5313457 | 4465 | if (platform_speed_cap == PCIE_SPEED_16_0GT) |
5d9a6330 AD |
4466 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4467 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4468 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
4469 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4); | |
c5313457 | 4470 | else if (platform_speed_cap == PCIE_SPEED_8_0GT) |
5d9a6330 AD |
4471 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4472 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4473 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
c5313457 | 4474 | else if (platform_speed_cap == PCIE_SPEED_5_0GT) |
5d9a6330 AD |
4475 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4476 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
4477 | else | |
4478 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
4479 | ||
cd474ba0 AD |
4480 | } |
4481 | } | |
4482 | if (adev->pm.pcie_mlw_mask == 0) { | |
c5313457 | 4483 | if (platform_link_width == PCIE_LNK_WIDTH_UNKNOWN) { |
5d9a6330 AD |
4484 | adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_PCIE_MLW_MASK; |
4485 | } else { | |
c5313457 | 4486 | switch (platform_link_width) { |
5d9a6330 | 4487 | case PCIE_LNK_X32: |
cd474ba0 AD |
4488 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 | |
4489 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | | |
4490 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
4491 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
4492 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4493 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4494 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4495 | break; | |
5d9a6330 | 4496 | case PCIE_LNK_X16: |
cd474ba0 AD |
4497 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | |
4498 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
4499 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
4500 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4501 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4502 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4503 | break; | |
5d9a6330 | 4504 | case PCIE_LNK_X12: |
cd474ba0 AD |
4505 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | |
4506 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
4507 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4508 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4509 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4510 | break; | |
5d9a6330 | 4511 | case PCIE_LNK_X8: |
cd474ba0 AD |
4512 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | |
4513 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4514 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4515 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4516 | break; | |
5d9a6330 | 4517 | case PCIE_LNK_X4: |
cd474ba0 AD |
4518 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | |
4519 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4520 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4521 | break; | |
5d9a6330 | 4522 | case PCIE_LNK_X2: |
cd474ba0 AD |
4523 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | |
4524 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4525 | break; | |
5d9a6330 | 4526 | case PCIE_LNK_X1: |
cd474ba0 AD |
4527 | adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1; |
4528 | break; | |
4529 | default: | |
4530 | break; | |
4531 | } | |
d0dd7f0c AD |
4532 | } |
4533 | } | |
4534 | } | |
d38ceaf9 | 4535 | |
361dbd01 AD |
4536 | int amdgpu_device_baco_enter(struct drm_device *dev) |
4537 | { | |
4538 | struct amdgpu_device *adev = dev->dev_private; | |
7a22677b | 4539 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
361dbd01 AD |
4540 | |
4541 | if (!amdgpu_device_supports_baco(adev->ddev)) | |
4542 | return -ENOTSUPP; | |
4543 | ||
7a22677b LM |
4544 | if (ras && ras->supported) |
4545 | adev->nbio.funcs->enable_doorbell_interrupt(adev, false); | |
4546 | ||
9530273e | 4547 | return amdgpu_dpm_baco_enter(adev); |
361dbd01 AD |
4548 | } |
4549 | ||
4550 | int amdgpu_device_baco_exit(struct drm_device *dev) | |
4551 | { | |
4552 | struct amdgpu_device *adev = dev->dev_private; | |
7a22677b | 4553 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
9530273e | 4554 | int ret = 0; |
361dbd01 AD |
4555 | |
4556 | if (!amdgpu_device_supports_baco(adev->ddev)) | |
4557 | return -ENOTSUPP; | |
4558 | ||
9530273e EQ |
4559 | ret = amdgpu_dpm_baco_exit(adev); |
4560 | if (ret) | |
4561 | return ret; | |
7a22677b LM |
4562 | |
4563 | if (ras && ras->supported) | |
4564 | adev->nbio.funcs->enable_doorbell_interrupt(adev, true); | |
4565 | ||
4566 | return 0; | |
361dbd01 | 4567 | } |