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" |
5183411b | 67 | |
d5ea093e AG |
68 | #include <linux/suspend.h> |
69 | ||
e2a75f88 | 70 | MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin"); |
3f76dced | 71 | MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin"); |
2d2e5e7e | 72 | MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin"); |
ad5a67a7 | 73 | MODULE_FIRMWARE("amdgpu/picasso_gpu_info.bin"); |
54c4d17e | 74 | MODULE_FIRMWARE("amdgpu/raven2_gpu_info.bin"); |
65e60f6e | 75 | MODULE_FIRMWARE("amdgpu/arcturus_gpu_info.bin"); |
b51a26a0 | 76 | MODULE_FIRMWARE("amdgpu/renoir_gpu_info.bin"); |
23c6268e | 77 | MODULE_FIRMWARE("amdgpu/navi10_gpu_info.bin"); |
ed42cfe1 | 78 | MODULE_FIRMWARE("amdgpu/navi14_gpu_info.bin"); |
42b325e5 | 79 | MODULE_FIRMWARE("amdgpu/navi12_gpu_info.bin"); |
e2a75f88 | 80 | |
2dc80b00 S |
81 | #define AMDGPU_RESUME_MS 2000 |
82 | ||
050091ab | 83 | const char *amdgpu_asic_name[] = { |
da69c161 KW |
84 | "TAHITI", |
85 | "PITCAIRN", | |
86 | "VERDE", | |
87 | "OLAND", | |
88 | "HAINAN", | |
d38ceaf9 AD |
89 | "BONAIRE", |
90 | "KAVERI", | |
91 | "KABINI", | |
92 | "HAWAII", | |
93 | "MULLINS", | |
94 | "TOPAZ", | |
95 | "TONGA", | |
48299f95 | 96 | "FIJI", |
d38ceaf9 | 97 | "CARRIZO", |
139f4917 | 98 | "STONEY", |
2cc0c0b5 FC |
99 | "POLARIS10", |
100 | "POLARIS11", | |
c4642a47 | 101 | "POLARIS12", |
48ff108d | 102 | "VEGAM", |
d4196f01 | 103 | "VEGA10", |
8fab806a | 104 | "VEGA12", |
956fcddc | 105 | "VEGA20", |
2ca8a5d2 | 106 | "RAVEN", |
d6c3b24e | 107 | "ARCTURUS", |
1eee4228 | 108 | "RENOIR", |
852a6626 | 109 | "NAVI10", |
87dbad02 | 110 | "NAVI14", |
9802f5d7 | 111 | "NAVI12", |
d38ceaf9 AD |
112 | "LAST", |
113 | }; | |
114 | ||
dcea6e65 KR |
115 | /** |
116 | * DOC: pcie_replay_count | |
117 | * | |
118 | * The amdgpu driver provides a sysfs API for reporting the total number | |
119 | * of PCIe replays (NAKs) | |
120 | * The file pcie_replay_count is used for this and returns the total | |
121 | * number of replays as a sum of the NAKs generated and NAKs received | |
122 | */ | |
123 | ||
124 | static ssize_t amdgpu_device_get_pcie_replay_count(struct device *dev, | |
125 | struct device_attribute *attr, char *buf) | |
126 | { | |
127 | struct drm_device *ddev = dev_get_drvdata(dev); | |
128 | struct amdgpu_device *adev = ddev->dev_private; | |
129 | uint64_t cnt = amdgpu_asic_get_pcie_replay_count(adev); | |
130 | ||
131 | return snprintf(buf, PAGE_SIZE, "%llu\n", cnt); | |
132 | } | |
133 | ||
134 | static DEVICE_ATTR(pcie_replay_count, S_IRUGO, | |
135 | amdgpu_device_get_pcie_replay_count, NULL); | |
136 | ||
5494d864 AD |
137 | static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev); |
138 | ||
e3ecdffa | 139 | /** |
31af062a | 140 | * amdgpu_device_supports_boco - Is the device a dGPU with HG/PX power control |
e3ecdffa AD |
141 | * |
142 | * @dev: drm_device pointer | |
143 | * | |
144 | * Returns true if the device is a dGPU with HG/PX power control, | |
145 | * otherwise return false. | |
146 | */ | |
31af062a | 147 | bool amdgpu_device_supports_boco(struct drm_device *dev) |
d38ceaf9 AD |
148 | { |
149 | struct amdgpu_device *adev = dev->dev_private; | |
150 | ||
2f7d10b3 | 151 | if (adev->flags & AMD_IS_PX) |
d38ceaf9 AD |
152 | return true; |
153 | return false; | |
154 | } | |
155 | ||
a69cba42 AD |
156 | /** |
157 | * amdgpu_device_supports_baco - Does the device support BACO | |
158 | * | |
159 | * @dev: drm_device pointer | |
160 | * | |
161 | * Returns true if the device supporte BACO, | |
162 | * otherwise return false. | |
163 | */ | |
164 | bool amdgpu_device_supports_baco(struct drm_device *dev) | |
165 | { | |
166 | struct amdgpu_device *adev = dev->dev_private; | |
167 | ||
168 | return amdgpu_asic_supports_baco(adev); | |
169 | } | |
170 | ||
e35e2b11 TY |
171 | /** |
172 | * VRAM access helper functions. | |
173 | * | |
174 | * amdgpu_device_vram_access - read/write a buffer in vram | |
175 | * | |
176 | * @adev: amdgpu_device pointer | |
177 | * @pos: offset of the buffer in vram | |
178 | * @buf: virtual address of the buffer in system memory | |
179 | * @size: read/write size, sizeof(@buf) must > @size | |
180 | * @write: true - write to vram, otherwise - read from vram | |
181 | */ | |
182 | void amdgpu_device_vram_access(struct amdgpu_device *adev, loff_t pos, | |
183 | uint32_t *buf, size_t size, bool write) | |
184 | { | |
185 | uint64_t last; | |
186 | unsigned long flags; | |
187 | ||
188 | last = size - 4; | |
189 | for (last += pos; pos <= last; pos += 4) { | |
190 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
191 | WREG32_NO_KIQ(mmMM_INDEX, ((uint32_t)pos) | 0x80000000); | |
192 | WREG32_NO_KIQ(mmMM_INDEX_HI, pos >> 31); | |
193 | if (write) | |
194 | WREG32_NO_KIQ(mmMM_DATA, *buf++); | |
195 | else | |
196 | *buf++ = RREG32_NO_KIQ(mmMM_DATA); | |
197 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
198 | } | |
199 | } | |
200 | ||
d38ceaf9 AD |
201 | /* |
202 | * MMIO register access helper functions. | |
203 | */ | |
e3ecdffa AD |
204 | /** |
205 | * amdgpu_mm_rreg - read a memory mapped IO register | |
206 | * | |
207 | * @adev: amdgpu_device pointer | |
208 | * @reg: dword aligned register offset | |
209 | * @acc_flags: access flags which require special behavior | |
210 | * | |
211 | * Returns the 32 bit value from the offset specified. | |
212 | */ | |
d38ceaf9 | 213 | uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg, |
15d72fd7 | 214 | uint32_t acc_flags) |
d38ceaf9 | 215 | { |
f4b373f4 TSD |
216 | uint32_t ret; |
217 | ||
43ca8efa | 218 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) |
bc992ba5 | 219 | return amdgpu_virt_kiq_rreg(adev, reg); |
bc992ba5 | 220 | |
15d72fd7 | 221 | if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX)) |
f4b373f4 | 222 | ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); |
d38ceaf9 AD |
223 | else { |
224 | unsigned long flags; | |
d38ceaf9 AD |
225 | |
226 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
227 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
228 | ret = readl(((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
229 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
d38ceaf9 | 230 | } |
f4b373f4 TSD |
231 | trace_amdgpu_mm_rreg(adev->pdev->device, reg, ret); |
232 | return ret; | |
d38ceaf9 AD |
233 | } |
234 | ||
421a2a30 ML |
235 | /* |
236 | * MMIO register read with bytes helper functions | |
237 | * @offset:bytes offset from MMIO start | |
238 | * | |
239 | */ | |
240 | ||
e3ecdffa AD |
241 | /** |
242 | * amdgpu_mm_rreg8 - read a memory mapped IO register | |
243 | * | |
244 | * @adev: amdgpu_device pointer | |
245 | * @offset: byte aligned register offset | |
246 | * | |
247 | * Returns the 8 bit value from the offset specified. | |
248 | */ | |
421a2a30 ML |
249 | uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset) { |
250 | if (offset < adev->rmmio_size) | |
251 | return (readb(adev->rmmio + offset)); | |
252 | BUG(); | |
253 | } | |
254 | ||
255 | /* | |
256 | * MMIO register write with bytes helper functions | |
257 | * @offset:bytes offset from MMIO start | |
258 | * @value: the value want to be written to the register | |
259 | * | |
260 | */ | |
e3ecdffa AD |
261 | /** |
262 | * amdgpu_mm_wreg8 - read a memory mapped IO register | |
263 | * | |
264 | * @adev: amdgpu_device pointer | |
265 | * @offset: byte aligned register offset | |
266 | * @value: 8 bit value to write | |
267 | * | |
268 | * Writes the value specified to the offset specified. | |
269 | */ | |
421a2a30 ML |
270 | void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value) { |
271 | if (offset < adev->rmmio_size) | |
272 | writeb(value, adev->rmmio + offset); | |
273 | else | |
274 | BUG(); | |
275 | } | |
276 | ||
e3ecdffa AD |
277 | /** |
278 | * amdgpu_mm_wreg - write to a memory mapped IO register | |
279 | * | |
280 | * @adev: amdgpu_device pointer | |
281 | * @reg: dword aligned register offset | |
282 | * @v: 32 bit value to write to the register | |
283 | * @acc_flags: access flags which require special behavior | |
284 | * | |
285 | * Writes the value specified to the offset specified. | |
286 | */ | |
d38ceaf9 | 287 | void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, |
15d72fd7 | 288 | uint32_t acc_flags) |
d38ceaf9 | 289 | { |
f4b373f4 | 290 | trace_amdgpu_mm_wreg(adev->pdev->device, reg, v); |
4e99a44e | 291 | |
47ed4e1c KW |
292 | if (adev->asic_type >= CHIP_VEGA10 && reg == 0) { |
293 | adev->last_mm_index = v; | |
294 | } | |
295 | ||
43ca8efa | 296 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) |
bc992ba5 | 297 | return amdgpu_virt_kiq_wreg(adev, reg, v); |
bc992ba5 | 298 | |
15d72fd7 | 299 | if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX)) |
d38ceaf9 AD |
300 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); |
301 | else { | |
302 | unsigned long flags; | |
303 | ||
304 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
305 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
306 | writel(v, ((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
307 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
308 | } | |
47ed4e1c KW |
309 | |
310 | if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) { | |
311 | udelay(500); | |
312 | } | |
d38ceaf9 AD |
313 | } |
314 | ||
e3ecdffa AD |
315 | /** |
316 | * amdgpu_io_rreg - read an IO register | |
317 | * | |
318 | * @adev: amdgpu_device pointer | |
319 | * @reg: dword aligned register offset | |
320 | * | |
321 | * Returns the 32 bit value from the offset specified. | |
322 | */ | |
d38ceaf9 AD |
323 | u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg) |
324 | { | |
325 | if ((reg * 4) < adev->rio_mem_size) | |
326 | return ioread32(adev->rio_mem + (reg * 4)); | |
327 | else { | |
328 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
329 | return ioread32(adev->rio_mem + (mmMM_DATA * 4)); | |
330 | } | |
331 | } | |
332 | ||
e3ecdffa AD |
333 | /** |
334 | * amdgpu_io_wreg - write to an IO register | |
335 | * | |
336 | * @adev: amdgpu_device pointer | |
337 | * @reg: dword aligned register offset | |
338 | * @v: 32 bit value to write to the register | |
339 | * | |
340 | * Writes the value specified to the offset specified. | |
341 | */ | |
d38ceaf9 AD |
342 | void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v) |
343 | { | |
47ed4e1c KW |
344 | if (adev->asic_type >= CHIP_VEGA10 && reg == 0) { |
345 | adev->last_mm_index = v; | |
346 | } | |
d38ceaf9 AD |
347 | |
348 | if ((reg * 4) < adev->rio_mem_size) | |
349 | iowrite32(v, adev->rio_mem + (reg * 4)); | |
350 | else { | |
351 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
352 | iowrite32(v, adev->rio_mem + (mmMM_DATA * 4)); | |
353 | } | |
47ed4e1c KW |
354 | |
355 | if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) { | |
356 | udelay(500); | |
357 | } | |
d38ceaf9 AD |
358 | } |
359 | ||
360 | /** | |
361 | * amdgpu_mm_rdoorbell - read a doorbell dword | |
362 | * | |
363 | * @adev: amdgpu_device pointer | |
364 | * @index: doorbell index | |
365 | * | |
366 | * Returns the value in the doorbell aperture at the | |
367 | * requested doorbell index (CIK). | |
368 | */ | |
369 | u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index) | |
370 | { | |
371 | if (index < adev->doorbell.num_doorbells) { | |
372 | return readl(adev->doorbell.ptr + index); | |
373 | } else { | |
374 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
375 | return 0; | |
376 | } | |
377 | } | |
378 | ||
379 | /** | |
380 | * amdgpu_mm_wdoorbell - write a doorbell dword | |
381 | * | |
382 | * @adev: amdgpu_device pointer | |
383 | * @index: doorbell index | |
384 | * @v: value to write | |
385 | * | |
386 | * Writes @v to the doorbell aperture at the | |
387 | * requested doorbell index (CIK). | |
388 | */ | |
389 | void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v) | |
390 | { | |
391 | if (index < adev->doorbell.num_doorbells) { | |
392 | writel(v, adev->doorbell.ptr + index); | |
393 | } else { | |
394 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
395 | } | |
396 | } | |
397 | ||
832be404 KW |
398 | /** |
399 | * amdgpu_mm_rdoorbell64 - read a doorbell Qword | |
400 | * | |
401 | * @adev: amdgpu_device pointer | |
402 | * @index: doorbell index | |
403 | * | |
404 | * Returns the value in the doorbell aperture at the | |
405 | * requested doorbell index (VEGA10+). | |
406 | */ | |
407 | u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index) | |
408 | { | |
409 | if (index < adev->doorbell.num_doorbells) { | |
410 | return atomic64_read((atomic64_t *)(adev->doorbell.ptr + index)); | |
411 | } else { | |
412 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
413 | return 0; | |
414 | } | |
415 | } | |
416 | ||
417 | /** | |
418 | * amdgpu_mm_wdoorbell64 - write a doorbell Qword | |
419 | * | |
420 | * @adev: amdgpu_device pointer | |
421 | * @index: doorbell index | |
422 | * @v: value to write | |
423 | * | |
424 | * Writes @v to the doorbell aperture at the | |
425 | * requested doorbell index (VEGA10+). | |
426 | */ | |
427 | void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v) | |
428 | { | |
429 | if (index < adev->doorbell.num_doorbells) { | |
430 | atomic64_set((atomic64_t *)(adev->doorbell.ptr + index), v); | |
431 | } else { | |
432 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
433 | } | |
434 | } | |
435 | ||
d38ceaf9 AD |
436 | /** |
437 | * amdgpu_invalid_rreg - dummy reg read function | |
438 | * | |
439 | * @adev: amdgpu device pointer | |
440 | * @reg: offset of register | |
441 | * | |
442 | * Dummy register read function. Used for register blocks | |
443 | * that certain asics don't have (all asics). | |
444 | * Returns the value in the register. | |
445 | */ | |
446 | static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg) | |
447 | { | |
448 | DRM_ERROR("Invalid callback to read register 0x%04X\n", reg); | |
449 | BUG(); | |
450 | return 0; | |
451 | } | |
452 | ||
453 | /** | |
454 | * amdgpu_invalid_wreg - dummy reg write function | |
455 | * | |
456 | * @adev: amdgpu device pointer | |
457 | * @reg: offset of register | |
458 | * @v: value to write to the register | |
459 | * | |
460 | * Dummy register read function. Used for register blocks | |
461 | * that certain asics don't have (all asics). | |
462 | */ | |
463 | static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) | |
464 | { | |
465 | DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n", | |
466 | reg, v); | |
467 | BUG(); | |
468 | } | |
469 | ||
4fa1c6a6 TZ |
470 | /** |
471 | * amdgpu_invalid_rreg64 - dummy 64 bit reg read function | |
472 | * | |
473 | * @adev: amdgpu device pointer | |
474 | * @reg: offset of register | |
475 | * | |
476 | * Dummy register read function. Used for register blocks | |
477 | * that certain asics don't have (all asics). | |
478 | * Returns the value in the register. | |
479 | */ | |
480 | static uint64_t amdgpu_invalid_rreg64(struct amdgpu_device *adev, uint32_t reg) | |
481 | { | |
482 | DRM_ERROR("Invalid callback to read 64 bit register 0x%04X\n", reg); | |
483 | BUG(); | |
484 | return 0; | |
485 | } | |
486 | ||
487 | /** | |
488 | * amdgpu_invalid_wreg64 - dummy reg write function | |
489 | * | |
490 | * @adev: amdgpu device pointer | |
491 | * @reg: offset of register | |
492 | * @v: value to write to the register | |
493 | * | |
494 | * Dummy register read function. Used for register blocks | |
495 | * that certain asics don't have (all asics). | |
496 | */ | |
497 | static void amdgpu_invalid_wreg64(struct amdgpu_device *adev, uint32_t reg, uint64_t v) | |
498 | { | |
499 | DRM_ERROR("Invalid callback to write 64 bit register 0x%04X with 0x%08llX\n", | |
500 | reg, v); | |
501 | BUG(); | |
502 | } | |
503 | ||
d38ceaf9 AD |
504 | /** |
505 | * amdgpu_block_invalid_rreg - dummy reg read function | |
506 | * | |
507 | * @adev: amdgpu device pointer | |
508 | * @block: offset of instance | |
509 | * @reg: offset of register | |
510 | * | |
511 | * Dummy register read function. Used for register blocks | |
512 | * that certain asics don't have (all asics). | |
513 | * Returns the value in the register. | |
514 | */ | |
515 | static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev, | |
516 | uint32_t block, uint32_t reg) | |
517 | { | |
518 | DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n", | |
519 | reg, block); | |
520 | BUG(); | |
521 | return 0; | |
522 | } | |
523 | ||
524 | /** | |
525 | * amdgpu_block_invalid_wreg - dummy reg write function | |
526 | * | |
527 | * @adev: amdgpu device pointer | |
528 | * @block: offset of instance | |
529 | * @reg: offset of register | |
530 | * @v: value to write to the register | |
531 | * | |
532 | * Dummy register read function. Used for register blocks | |
533 | * that certain asics don't have (all asics). | |
534 | */ | |
535 | static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev, | |
536 | uint32_t block, | |
537 | uint32_t reg, uint32_t v) | |
538 | { | |
539 | DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n", | |
540 | reg, block, v); | |
541 | BUG(); | |
542 | } | |
543 | ||
e3ecdffa AD |
544 | /** |
545 | * amdgpu_device_vram_scratch_init - allocate the VRAM scratch page | |
546 | * | |
547 | * @adev: amdgpu device pointer | |
548 | * | |
549 | * Allocates a scratch page of VRAM for use by various things in the | |
550 | * driver. | |
551 | */ | |
06ec9070 | 552 | static int amdgpu_device_vram_scratch_init(struct amdgpu_device *adev) |
d38ceaf9 | 553 | { |
a4a02777 CK |
554 | return amdgpu_bo_create_kernel(adev, AMDGPU_GPU_PAGE_SIZE, |
555 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, | |
556 | &adev->vram_scratch.robj, | |
557 | &adev->vram_scratch.gpu_addr, | |
558 | (void **)&adev->vram_scratch.ptr); | |
d38ceaf9 AD |
559 | } |
560 | ||
e3ecdffa AD |
561 | /** |
562 | * amdgpu_device_vram_scratch_fini - Free the VRAM scratch page | |
563 | * | |
564 | * @adev: amdgpu device pointer | |
565 | * | |
566 | * Frees the VRAM scratch page. | |
567 | */ | |
06ec9070 | 568 | static void amdgpu_device_vram_scratch_fini(struct amdgpu_device *adev) |
d38ceaf9 | 569 | { |
078af1a3 | 570 | amdgpu_bo_free_kernel(&adev->vram_scratch.robj, NULL, NULL); |
d38ceaf9 AD |
571 | } |
572 | ||
573 | /** | |
9c3f2b54 | 574 | * amdgpu_device_program_register_sequence - program an array of registers. |
d38ceaf9 AD |
575 | * |
576 | * @adev: amdgpu_device pointer | |
577 | * @registers: pointer to the register array | |
578 | * @array_size: size of the register array | |
579 | * | |
580 | * Programs an array or registers with and and or masks. | |
581 | * This is a helper for setting golden registers. | |
582 | */ | |
9c3f2b54 AD |
583 | void amdgpu_device_program_register_sequence(struct amdgpu_device *adev, |
584 | const u32 *registers, | |
585 | const u32 array_size) | |
d38ceaf9 AD |
586 | { |
587 | u32 tmp, reg, and_mask, or_mask; | |
588 | int i; | |
589 | ||
590 | if (array_size % 3) | |
591 | return; | |
592 | ||
593 | for (i = 0; i < array_size; i +=3) { | |
594 | reg = registers[i + 0]; | |
595 | and_mask = registers[i + 1]; | |
596 | or_mask = registers[i + 2]; | |
597 | ||
598 | if (and_mask == 0xffffffff) { | |
599 | tmp = or_mask; | |
600 | } else { | |
601 | tmp = RREG32(reg); | |
602 | tmp &= ~and_mask; | |
e0d07657 HZ |
603 | if (adev->family >= AMDGPU_FAMILY_AI) |
604 | tmp |= (or_mask & and_mask); | |
605 | else | |
606 | tmp |= or_mask; | |
d38ceaf9 AD |
607 | } |
608 | WREG32(reg, tmp); | |
609 | } | |
610 | } | |
611 | ||
e3ecdffa AD |
612 | /** |
613 | * amdgpu_device_pci_config_reset - reset the GPU | |
614 | * | |
615 | * @adev: amdgpu_device pointer | |
616 | * | |
617 | * Resets the GPU using the pci config reset sequence. | |
618 | * Only applicable to asics prior to vega10. | |
619 | */ | |
8111c387 | 620 | void amdgpu_device_pci_config_reset(struct amdgpu_device *adev) |
d38ceaf9 AD |
621 | { |
622 | pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA); | |
623 | } | |
624 | ||
625 | /* | |
626 | * GPU doorbell aperture helpers function. | |
627 | */ | |
628 | /** | |
06ec9070 | 629 | * amdgpu_device_doorbell_init - Init doorbell driver information. |
d38ceaf9 AD |
630 | * |
631 | * @adev: amdgpu_device pointer | |
632 | * | |
633 | * Init doorbell driver information (CIK) | |
634 | * Returns 0 on success, error on failure. | |
635 | */ | |
06ec9070 | 636 | static int amdgpu_device_doorbell_init(struct amdgpu_device *adev) |
d38ceaf9 | 637 | { |
6585661d | 638 | |
705e519e CK |
639 | /* No doorbell on SI hardware generation */ |
640 | if (adev->asic_type < CHIP_BONAIRE) { | |
641 | adev->doorbell.base = 0; | |
642 | adev->doorbell.size = 0; | |
643 | adev->doorbell.num_doorbells = 0; | |
644 | adev->doorbell.ptr = NULL; | |
645 | return 0; | |
646 | } | |
647 | ||
d6895ad3 CK |
648 | if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET) |
649 | return -EINVAL; | |
650 | ||
22357775 AD |
651 | amdgpu_asic_init_doorbell_index(adev); |
652 | ||
d38ceaf9 AD |
653 | /* doorbell bar mapping */ |
654 | adev->doorbell.base = pci_resource_start(adev->pdev, 2); | |
655 | adev->doorbell.size = pci_resource_len(adev->pdev, 2); | |
656 | ||
edf600da | 657 | adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32), |
9564f192 | 658 | adev->doorbell_index.max_assignment+1); |
d38ceaf9 AD |
659 | if (adev->doorbell.num_doorbells == 0) |
660 | return -EINVAL; | |
661 | ||
ec3db8a6 | 662 | /* For Vega, reserve and map two pages on doorbell BAR since SDMA |
88dc26e4 OZ |
663 | * paging queue doorbell use the second page. The |
664 | * AMDGPU_DOORBELL64_MAX_ASSIGNMENT definition assumes all the | |
665 | * doorbells are in the first page. So with paging queue enabled, | |
666 | * the max num_doorbells should + 1 page (0x400 in dword) | |
ec3db8a6 PY |
667 | */ |
668 | if (adev->asic_type >= CHIP_VEGA10) | |
88dc26e4 | 669 | adev->doorbell.num_doorbells += 0x400; |
ec3db8a6 | 670 | |
8972e5d2 CK |
671 | adev->doorbell.ptr = ioremap(adev->doorbell.base, |
672 | adev->doorbell.num_doorbells * | |
673 | sizeof(u32)); | |
674 | if (adev->doorbell.ptr == NULL) | |
d38ceaf9 | 675 | return -ENOMEM; |
d38ceaf9 AD |
676 | |
677 | return 0; | |
678 | } | |
679 | ||
680 | /** | |
06ec9070 | 681 | * amdgpu_device_doorbell_fini - Tear down doorbell driver information. |
d38ceaf9 AD |
682 | * |
683 | * @adev: amdgpu_device pointer | |
684 | * | |
685 | * Tear down doorbell driver information (CIK) | |
686 | */ | |
06ec9070 | 687 | static void amdgpu_device_doorbell_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
688 | { |
689 | iounmap(adev->doorbell.ptr); | |
690 | adev->doorbell.ptr = NULL; | |
691 | } | |
692 | ||
22cb0164 | 693 | |
d38ceaf9 AD |
694 | |
695 | /* | |
06ec9070 | 696 | * amdgpu_device_wb_*() |
455a7bc2 | 697 | * Writeback is the method by which the GPU updates special pages in memory |
ea81a173 | 698 | * with the status of certain GPU events (fences, ring pointers,etc.). |
d38ceaf9 AD |
699 | */ |
700 | ||
701 | /** | |
06ec9070 | 702 | * amdgpu_device_wb_fini - Disable Writeback and free memory |
d38ceaf9 AD |
703 | * |
704 | * @adev: amdgpu_device pointer | |
705 | * | |
706 | * Disables Writeback and frees the Writeback memory (all asics). | |
707 | * Used at driver shutdown. | |
708 | */ | |
06ec9070 | 709 | static void amdgpu_device_wb_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
710 | { |
711 | if (adev->wb.wb_obj) { | |
a76ed485 AD |
712 | amdgpu_bo_free_kernel(&adev->wb.wb_obj, |
713 | &adev->wb.gpu_addr, | |
714 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
715 | adev->wb.wb_obj = NULL; |
716 | } | |
717 | } | |
718 | ||
719 | /** | |
06ec9070 | 720 | * amdgpu_device_wb_init- Init Writeback driver info and allocate memory |
d38ceaf9 AD |
721 | * |
722 | * @adev: amdgpu_device pointer | |
723 | * | |
455a7bc2 | 724 | * Initializes writeback and allocates writeback memory (all asics). |
d38ceaf9 AD |
725 | * Used at driver startup. |
726 | * Returns 0 on success or an -error on failure. | |
727 | */ | |
06ec9070 | 728 | static int amdgpu_device_wb_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
729 | { |
730 | int r; | |
731 | ||
732 | if (adev->wb.wb_obj == NULL) { | |
97407b63 AD |
733 | /* AMDGPU_MAX_WB * sizeof(uint32_t) * 8 = AMDGPU_MAX_WB 256bit slots */ |
734 | r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t) * 8, | |
a76ed485 AD |
735 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, |
736 | &adev->wb.wb_obj, &adev->wb.gpu_addr, | |
737 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
738 | if (r) { |
739 | dev_warn(adev->dev, "(%d) create WB bo failed\n", r); | |
740 | return r; | |
741 | } | |
d38ceaf9 AD |
742 | |
743 | adev->wb.num_wb = AMDGPU_MAX_WB; | |
744 | memset(&adev->wb.used, 0, sizeof(adev->wb.used)); | |
745 | ||
746 | /* clear wb memory */ | |
73469585 | 747 | memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t) * 8); |
d38ceaf9 AD |
748 | } |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
753 | /** | |
131b4b36 | 754 | * amdgpu_device_wb_get - Allocate a wb entry |
d38ceaf9 AD |
755 | * |
756 | * @adev: amdgpu_device pointer | |
757 | * @wb: wb index | |
758 | * | |
759 | * Allocate a wb slot for use by the driver (all asics). | |
760 | * Returns 0 on success or -EINVAL on failure. | |
761 | */ | |
131b4b36 | 762 | int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb) |
d38ceaf9 AD |
763 | { |
764 | unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb); | |
d38ceaf9 | 765 | |
97407b63 | 766 | if (offset < adev->wb.num_wb) { |
7014285a | 767 | __set_bit(offset, adev->wb.used); |
63ae07ca | 768 | *wb = offset << 3; /* convert to dw offset */ |
0915fdbc ML |
769 | return 0; |
770 | } else { | |
771 | return -EINVAL; | |
772 | } | |
773 | } | |
774 | ||
d38ceaf9 | 775 | /** |
131b4b36 | 776 | * amdgpu_device_wb_free - Free a wb entry |
d38ceaf9 AD |
777 | * |
778 | * @adev: amdgpu_device pointer | |
779 | * @wb: wb index | |
780 | * | |
781 | * Free a wb slot allocated for use by the driver (all asics) | |
782 | */ | |
131b4b36 | 783 | void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb) |
d38ceaf9 | 784 | { |
73469585 | 785 | wb >>= 3; |
d38ceaf9 | 786 | if (wb < adev->wb.num_wb) |
73469585 | 787 | __clear_bit(wb, adev->wb.used); |
d38ceaf9 AD |
788 | } |
789 | ||
d6895ad3 CK |
790 | /** |
791 | * amdgpu_device_resize_fb_bar - try to resize FB BAR | |
792 | * | |
793 | * @adev: amdgpu_device pointer | |
794 | * | |
795 | * Try to resize FB BAR to make all VRAM CPU accessible. We try very hard not | |
796 | * to fail, but if any of the BARs is not accessible after the size we abort | |
797 | * driver loading by returning -ENODEV. | |
798 | */ | |
799 | int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev) | |
800 | { | |
770d13b1 | 801 | u64 space_needed = roundup_pow_of_two(adev->gmc.real_vram_size); |
d6895ad3 | 802 | u32 rbar_size = order_base_2(((space_needed >> 20) | 1)) - 1; |
31b8adab CK |
803 | struct pci_bus *root; |
804 | struct resource *res; | |
805 | unsigned i; | |
d6895ad3 CK |
806 | u16 cmd; |
807 | int r; | |
808 | ||
0c03b912 | 809 | /* Bypass for VF */ |
810 | if (amdgpu_sriov_vf(adev)) | |
811 | return 0; | |
812 | ||
31b8adab CK |
813 | /* Check if the root BUS has 64bit memory resources */ |
814 | root = adev->pdev->bus; | |
815 | while (root->parent) | |
816 | root = root->parent; | |
817 | ||
818 | pci_bus_for_each_resource(root, res, i) { | |
0ebb7c54 | 819 | if (res && res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) && |
31b8adab CK |
820 | res->start > 0x100000000ull) |
821 | break; | |
822 | } | |
823 | ||
824 | /* Trying to resize is pointless without a root hub window above 4GB */ | |
825 | if (!res) | |
826 | return 0; | |
827 | ||
d6895ad3 CK |
828 | /* Disable memory decoding while we change the BAR addresses and size */ |
829 | pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd); | |
830 | pci_write_config_word(adev->pdev, PCI_COMMAND, | |
831 | cmd & ~PCI_COMMAND_MEMORY); | |
832 | ||
833 | /* Free the VRAM and doorbell BAR, we most likely need to move both. */ | |
06ec9070 | 834 | amdgpu_device_doorbell_fini(adev); |
d6895ad3 CK |
835 | if (adev->asic_type >= CHIP_BONAIRE) |
836 | pci_release_resource(adev->pdev, 2); | |
837 | ||
838 | pci_release_resource(adev->pdev, 0); | |
839 | ||
840 | r = pci_resize_resource(adev->pdev, 0, rbar_size); | |
841 | if (r == -ENOSPC) | |
842 | DRM_INFO("Not enough PCI address space for a large BAR."); | |
843 | else if (r && r != -ENOTSUPP) | |
844 | DRM_ERROR("Problem resizing BAR0 (%d).", r); | |
845 | ||
846 | pci_assign_unassigned_bus_resources(adev->pdev->bus); | |
847 | ||
848 | /* When the doorbell or fb BAR isn't available we have no chance of | |
849 | * using the device. | |
850 | */ | |
06ec9070 | 851 | r = amdgpu_device_doorbell_init(adev); |
d6895ad3 CK |
852 | if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET)) |
853 | return -ENODEV; | |
854 | ||
855 | pci_write_config_word(adev->pdev, PCI_COMMAND, cmd); | |
856 | ||
857 | return 0; | |
858 | } | |
a05502e5 | 859 | |
d38ceaf9 AD |
860 | /* |
861 | * GPU helpers function. | |
862 | */ | |
863 | /** | |
39c640c0 | 864 | * amdgpu_device_need_post - check if the hw need post or not |
d38ceaf9 AD |
865 | * |
866 | * @adev: amdgpu_device pointer | |
867 | * | |
c836fec5 JQ |
868 | * Check if the asic has been initialized (all asics) at driver startup |
869 | * or post is needed if hw reset is performed. | |
870 | * Returns true if need or false if not. | |
d38ceaf9 | 871 | */ |
39c640c0 | 872 | bool amdgpu_device_need_post(struct amdgpu_device *adev) |
d38ceaf9 AD |
873 | { |
874 | uint32_t reg; | |
875 | ||
bec86378 ML |
876 | if (amdgpu_sriov_vf(adev)) |
877 | return false; | |
878 | ||
879 | if (amdgpu_passthrough(adev)) { | |
1da2c326 ML |
880 | /* for FIJI: In whole GPU pass-through virtualization case, after VM reboot |
881 | * some old smc fw still need driver do vPost otherwise gpu hang, while | |
882 | * those smc fw version above 22.15 doesn't have this flaw, so we force | |
883 | * vpost executed for smc version below 22.15 | |
bec86378 ML |
884 | */ |
885 | if (adev->asic_type == CHIP_FIJI) { | |
886 | int err; | |
887 | uint32_t fw_ver; | |
888 | err = request_firmware(&adev->pm.fw, "amdgpu/fiji_smc.bin", adev->dev); | |
889 | /* force vPost if error occured */ | |
890 | if (err) | |
891 | return true; | |
892 | ||
893 | fw_ver = *((uint32_t *)adev->pm.fw->data + 69); | |
1da2c326 ML |
894 | if (fw_ver < 0x00160e00) |
895 | return true; | |
bec86378 | 896 | } |
bec86378 | 897 | } |
91fe77eb | 898 | |
899 | if (adev->has_hw_reset) { | |
900 | adev->has_hw_reset = false; | |
901 | return true; | |
902 | } | |
903 | ||
904 | /* bios scratch used on CIK+ */ | |
905 | if (adev->asic_type >= CHIP_BONAIRE) | |
906 | return amdgpu_atombios_scratch_need_asic_init(adev); | |
907 | ||
908 | /* check MEM_SIZE for older asics */ | |
909 | reg = amdgpu_asic_get_config_memsize(adev); | |
910 | ||
911 | if ((reg != 0) && (reg != 0xffffffff)) | |
912 | return false; | |
913 | ||
914 | return true; | |
bec86378 ML |
915 | } |
916 | ||
d38ceaf9 AD |
917 | /* if we get transitioned to only one device, take VGA back */ |
918 | /** | |
06ec9070 | 919 | * amdgpu_device_vga_set_decode - enable/disable vga decode |
d38ceaf9 AD |
920 | * |
921 | * @cookie: amdgpu_device pointer | |
922 | * @state: enable/disable vga decode | |
923 | * | |
924 | * Enable/disable vga decode (all asics). | |
925 | * Returns VGA resource flags. | |
926 | */ | |
06ec9070 | 927 | static unsigned int amdgpu_device_vga_set_decode(void *cookie, bool state) |
d38ceaf9 AD |
928 | { |
929 | struct amdgpu_device *adev = cookie; | |
930 | amdgpu_asic_set_vga_state(adev, state); | |
931 | if (state) | |
932 | return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM | | |
933 | VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
934 | else | |
935 | return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
936 | } | |
937 | ||
e3ecdffa AD |
938 | /** |
939 | * amdgpu_device_check_block_size - validate the vm block size | |
940 | * | |
941 | * @adev: amdgpu_device pointer | |
942 | * | |
943 | * Validates the vm block size specified via module parameter. | |
944 | * The vm block size defines number of bits in page table versus page directory, | |
945 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
946 | * page table and the remaining bits are in the page directory. | |
947 | */ | |
06ec9070 | 948 | static void amdgpu_device_check_block_size(struct amdgpu_device *adev) |
a1adf8be CZ |
949 | { |
950 | /* defines number of bits in page table versus page directory, | |
951 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
952 | * page table and the remaining bits are in the page directory */ | |
bab4fee7 JZ |
953 | if (amdgpu_vm_block_size == -1) |
954 | return; | |
a1adf8be | 955 | |
bab4fee7 | 956 | if (amdgpu_vm_block_size < 9) { |
a1adf8be CZ |
957 | dev_warn(adev->dev, "VM page table size (%d) too small\n", |
958 | amdgpu_vm_block_size); | |
97489129 | 959 | amdgpu_vm_block_size = -1; |
a1adf8be | 960 | } |
a1adf8be CZ |
961 | } |
962 | ||
e3ecdffa AD |
963 | /** |
964 | * amdgpu_device_check_vm_size - validate the vm size | |
965 | * | |
966 | * @adev: amdgpu_device pointer | |
967 | * | |
968 | * Validates the vm size in GB specified via module parameter. | |
969 | * The VM size is the size of the GPU virtual memory space in GB. | |
970 | */ | |
06ec9070 | 971 | static void amdgpu_device_check_vm_size(struct amdgpu_device *adev) |
83ca145d | 972 | { |
64dab074 AD |
973 | /* no need to check the default value */ |
974 | if (amdgpu_vm_size == -1) | |
975 | return; | |
976 | ||
83ca145d ZJ |
977 | if (amdgpu_vm_size < 1) { |
978 | dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n", | |
979 | amdgpu_vm_size); | |
f3368128 | 980 | amdgpu_vm_size = -1; |
83ca145d | 981 | } |
83ca145d ZJ |
982 | } |
983 | ||
7951e376 RZ |
984 | static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev) |
985 | { | |
986 | struct sysinfo si; | |
987 | bool is_os_64 = (sizeof(void *) == 8) ? true : false; | |
988 | uint64_t total_memory; | |
989 | uint64_t dram_size_seven_GB = 0x1B8000000; | |
990 | uint64_t dram_size_three_GB = 0xB8000000; | |
991 | ||
992 | if (amdgpu_smu_memory_pool_size == 0) | |
993 | return; | |
994 | ||
995 | if (!is_os_64) { | |
996 | DRM_WARN("Not 64-bit OS, feature not supported\n"); | |
997 | goto def_value; | |
998 | } | |
999 | si_meminfo(&si); | |
1000 | total_memory = (uint64_t)si.totalram * si.mem_unit; | |
1001 | ||
1002 | if ((amdgpu_smu_memory_pool_size == 1) || | |
1003 | (amdgpu_smu_memory_pool_size == 2)) { | |
1004 | if (total_memory < dram_size_three_GB) | |
1005 | goto def_value1; | |
1006 | } else if ((amdgpu_smu_memory_pool_size == 4) || | |
1007 | (amdgpu_smu_memory_pool_size == 8)) { | |
1008 | if (total_memory < dram_size_seven_GB) | |
1009 | goto def_value1; | |
1010 | } else { | |
1011 | DRM_WARN("Smu memory pool size not supported\n"); | |
1012 | goto def_value; | |
1013 | } | |
1014 | adev->pm.smu_prv_buffer_size = amdgpu_smu_memory_pool_size << 28; | |
1015 | ||
1016 | return; | |
1017 | ||
1018 | def_value1: | |
1019 | DRM_WARN("No enough system memory\n"); | |
1020 | def_value: | |
1021 | adev->pm.smu_prv_buffer_size = 0; | |
1022 | } | |
1023 | ||
d38ceaf9 | 1024 | /** |
06ec9070 | 1025 | * amdgpu_device_check_arguments - validate module params |
d38ceaf9 AD |
1026 | * |
1027 | * @adev: amdgpu_device pointer | |
1028 | * | |
1029 | * Validates certain module parameters and updates | |
1030 | * the associated values used by the driver (all asics). | |
1031 | */ | |
912dfc84 | 1032 | static int amdgpu_device_check_arguments(struct amdgpu_device *adev) |
d38ceaf9 | 1033 | { |
912dfc84 EQ |
1034 | int ret = 0; |
1035 | ||
5b011235 CZ |
1036 | if (amdgpu_sched_jobs < 4) { |
1037 | dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n", | |
1038 | amdgpu_sched_jobs); | |
1039 | amdgpu_sched_jobs = 4; | |
76117507 | 1040 | } else if (!is_power_of_2(amdgpu_sched_jobs)){ |
5b011235 CZ |
1041 | dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n", |
1042 | amdgpu_sched_jobs); | |
1043 | amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs); | |
1044 | } | |
d38ceaf9 | 1045 | |
83e74db6 | 1046 | if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) { |
f9321cc4 CK |
1047 | /* gart size must be greater or equal to 32M */ |
1048 | dev_warn(adev->dev, "gart size (%d) too small\n", | |
1049 | amdgpu_gart_size); | |
83e74db6 | 1050 | amdgpu_gart_size = -1; |
d38ceaf9 AD |
1051 | } |
1052 | ||
36d38372 | 1053 | if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) { |
c4e1a13a | 1054 | /* gtt size must be greater or equal to 32M */ |
36d38372 CK |
1055 | dev_warn(adev->dev, "gtt size (%d) too small\n", |
1056 | amdgpu_gtt_size); | |
1057 | amdgpu_gtt_size = -1; | |
d38ceaf9 AD |
1058 | } |
1059 | ||
d07f14be RH |
1060 | /* valid range is between 4 and 9 inclusive */ |
1061 | if (amdgpu_vm_fragment_size != -1 && | |
1062 | (amdgpu_vm_fragment_size > 9 || amdgpu_vm_fragment_size < 4)) { | |
1063 | dev_warn(adev->dev, "valid range is between 4 and 9\n"); | |
1064 | amdgpu_vm_fragment_size = -1; | |
1065 | } | |
1066 | ||
7951e376 RZ |
1067 | amdgpu_device_check_smu_prv_buffer_size(adev); |
1068 | ||
06ec9070 | 1069 | amdgpu_device_check_vm_size(adev); |
d38ceaf9 | 1070 | |
06ec9070 | 1071 | amdgpu_device_check_block_size(adev); |
6a7f76e7 | 1072 | |
19aede77 | 1073 | adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type); |
912dfc84 EQ |
1074 | |
1075 | return ret; | |
d38ceaf9 AD |
1076 | } |
1077 | ||
1078 | /** | |
1079 | * amdgpu_switcheroo_set_state - set switcheroo state | |
1080 | * | |
1081 | * @pdev: pci dev pointer | |
1694467b | 1082 | * @state: vga_switcheroo state |
d38ceaf9 AD |
1083 | * |
1084 | * Callback for the switcheroo driver. Suspends or resumes the | |
1085 | * the asics before or after it is powered up using ACPI methods. | |
1086 | */ | |
1087 | static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_state state) | |
1088 | { | |
1089 | struct drm_device *dev = pci_get_drvdata(pdev); | |
de185019 | 1090 | int r; |
d38ceaf9 | 1091 | |
31af062a | 1092 | if (amdgpu_device_supports_boco(dev) && state == VGA_SWITCHEROO_OFF) |
d38ceaf9 AD |
1093 | return; |
1094 | ||
1095 | if (state == VGA_SWITCHEROO_ON) { | |
7ca85295 | 1096 | pr_info("amdgpu: switched on\n"); |
d38ceaf9 AD |
1097 | /* don't suspend or resume card normally */ |
1098 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
1099 | ||
de185019 AD |
1100 | pci_set_power_state(dev->pdev, PCI_D0); |
1101 | pci_restore_state(dev->pdev); | |
1102 | r = pci_enable_device(dev->pdev); | |
1103 | if (r) | |
1104 | DRM_WARN("pci_enable_device failed (%d)\n", r); | |
1105 | amdgpu_device_resume(dev, true); | |
d38ceaf9 | 1106 | |
d38ceaf9 AD |
1107 | dev->switch_power_state = DRM_SWITCH_POWER_ON; |
1108 | drm_kms_helper_poll_enable(dev); | |
1109 | } else { | |
7ca85295 | 1110 | pr_info("amdgpu: switched off\n"); |
d38ceaf9 AD |
1111 | drm_kms_helper_poll_disable(dev); |
1112 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
de185019 AD |
1113 | amdgpu_device_suspend(dev, true); |
1114 | pci_save_state(dev->pdev); | |
1115 | /* Shut down the device */ | |
1116 | pci_disable_device(dev->pdev); | |
1117 | pci_set_power_state(dev->pdev, PCI_D3cold); | |
d38ceaf9 AD |
1118 | dev->switch_power_state = DRM_SWITCH_POWER_OFF; |
1119 | } | |
1120 | } | |
1121 | ||
1122 | /** | |
1123 | * amdgpu_switcheroo_can_switch - see if switcheroo state can change | |
1124 | * | |
1125 | * @pdev: pci dev pointer | |
1126 | * | |
1127 | * Callback for the switcheroo driver. Check of the switcheroo | |
1128 | * state can be changed. | |
1129 | * Returns true if the state can be changed, false if not. | |
1130 | */ | |
1131 | static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev) | |
1132 | { | |
1133 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1134 | ||
1135 | /* | |
1136 | * FIXME: open_count is protected by drm_global_mutex but that would lead to | |
1137 | * locking inversion with the driver load path. And the access here is | |
1138 | * completely racy anyway. So don't bother with locking for now. | |
1139 | */ | |
1140 | return dev->open_count == 0; | |
1141 | } | |
1142 | ||
1143 | static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = { | |
1144 | .set_gpu_state = amdgpu_switcheroo_set_state, | |
1145 | .reprobe = NULL, | |
1146 | .can_switch = amdgpu_switcheroo_can_switch, | |
1147 | }; | |
1148 | ||
e3ecdffa AD |
1149 | /** |
1150 | * amdgpu_device_ip_set_clockgating_state - set the CG state | |
1151 | * | |
87e3f136 | 1152 | * @dev: amdgpu_device pointer |
e3ecdffa AD |
1153 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) |
1154 | * @state: clockgating state (gate or ungate) | |
1155 | * | |
1156 | * Sets the requested clockgating state for all instances of | |
1157 | * the hardware IP specified. | |
1158 | * Returns the error code from the last instance. | |
1159 | */ | |
43fa561f | 1160 | int amdgpu_device_ip_set_clockgating_state(void *dev, |
2990a1fc AD |
1161 | enum amd_ip_block_type block_type, |
1162 | enum amd_clockgating_state state) | |
d38ceaf9 | 1163 | { |
43fa561f | 1164 | struct amdgpu_device *adev = dev; |
d38ceaf9 AD |
1165 | int i, r = 0; |
1166 | ||
1167 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1168 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1169 | continue; |
c722865a RZ |
1170 | if (adev->ip_blocks[i].version->type != block_type) |
1171 | continue; | |
1172 | if (!adev->ip_blocks[i].version->funcs->set_clockgating_state) | |
1173 | continue; | |
1174 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state( | |
1175 | (void *)adev, state); | |
1176 | if (r) | |
1177 | DRM_ERROR("set_clockgating_state of IP block <%s> failed %d\n", | |
1178 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1179 | } |
1180 | return r; | |
1181 | } | |
1182 | ||
e3ecdffa AD |
1183 | /** |
1184 | * amdgpu_device_ip_set_powergating_state - set the PG state | |
1185 | * | |
87e3f136 | 1186 | * @dev: amdgpu_device pointer |
e3ecdffa AD |
1187 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) |
1188 | * @state: powergating state (gate or ungate) | |
1189 | * | |
1190 | * Sets the requested powergating state for all instances of | |
1191 | * the hardware IP specified. | |
1192 | * Returns the error code from the last instance. | |
1193 | */ | |
43fa561f | 1194 | int amdgpu_device_ip_set_powergating_state(void *dev, |
2990a1fc AD |
1195 | enum amd_ip_block_type block_type, |
1196 | enum amd_powergating_state state) | |
d38ceaf9 | 1197 | { |
43fa561f | 1198 | struct amdgpu_device *adev = dev; |
d38ceaf9 AD |
1199 | int i, r = 0; |
1200 | ||
1201 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1202 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1203 | continue; |
c722865a RZ |
1204 | if (adev->ip_blocks[i].version->type != block_type) |
1205 | continue; | |
1206 | if (!adev->ip_blocks[i].version->funcs->set_powergating_state) | |
1207 | continue; | |
1208 | r = adev->ip_blocks[i].version->funcs->set_powergating_state( | |
1209 | (void *)adev, state); | |
1210 | if (r) | |
1211 | DRM_ERROR("set_powergating_state of IP block <%s> failed %d\n", | |
1212 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1213 | } |
1214 | return r; | |
1215 | } | |
1216 | ||
e3ecdffa AD |
1217 | /** |
1218 | * amdgpu_device_ip_get_clockgating_state - get the CG state | |
1219 | * | |
1220 | * @adev: amdgpu_device pointer | |
1221 | * @flags: clockgating feature flags | |
1222 | * | |
1223 | * Walks the list of IPs on the device and updates the clockgating | |
1224 | * flags for each IP. | |
1225 | * Updates @flags with the feature flags for each hardware IP where | |
1226 | * clockgating is enabled. | |
1227 | */ | |
2990a1fc AD |
1228 | void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev, |
1229 | u32 *flags) | |
6cb2d4e4 HR |
1230 | { |
1231 | int i; | |
1232 | ||
1233 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1234 | if (!adev->ip_blocks[i].status.valid) | |
1235 | continue; | |
1236 | if (adev->ip_blocks[i].version->funcs->get_clockgating_state) | |
1237 | adev->ip_blocks[i].version->funcs->get_clockgating_state((void *)adev, flags); | |
1238 | } | |
1239 | } | |
1240 | ||
e3ecdffa AD |
1241 | /** |
1242 | * amdgpu_device_ip_wait_for_idle - wait for idle | |
1243 | * | |
1244 | * @adev: amdgpu_device pointer | |
1245 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) | |
1246 | * | |
1247 | * Waits for the request hardware IP to be idle. | |
1248 | * Returns 0 for success or a negative error code on failure. | |
1249 | */ | |
2990a1fc AD |
1250 | int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev, |
1251 | enum amd_ip_block_type block_type) | |
5dbbb60b AD |
1252 | { |
1253 | int i, r; | |
1254 | ||
1255 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1256 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1257 | continue; |
a1255107 AD |
1258 | if (adev->ip_blocks[i].version->type == block_type) { |
1259 | r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev); | |
5dbbb60b AD |
1260 | if (r) |
1261 | return r; | |
1262 | break; | |
1263 | } | |
1264 | } | |
1265 | return 0; | |
1266 | ||
1267 | } | |
1268 | ||
e3ecdffa AD |
1269 | /** |
1270 | * amdgpu_device_ip_is_idle - is the hardware IP idle | |
1271 | * | |
1272 | * @adev: amdgpu_device pointer | |
1273 | * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.) | |
1274 | * | |
1275 | * Check if the hardware IP is idle or not. | |
1276 | * Returns true if it the IP is idle, false if not. | |
1277 | */ | |
2990a1fc AD |
1278 | bool amdgpu_device_ip_is_idle(struct amdgpu_device *adev, |
1279 | enum amd_ip_block_type block_type) | |
5dbbb60b AD |
1280 | { |
1281 | int i; | |
1282 | ||
1283 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1284 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1285 | continue; |
a1255107 AD |
1286 | if (adev->ip_blocks[i].version->type == block_type) |
1287 | return adev->ip_blocks[i].version->funcs->is_idle((void *)adev); | |
5dbbb60b AD |
1288 | } |
1289 | return true; | |
1290 | ||
1291 | } | |
1292 | ||
e3ecdffa AD |
1293 | /** |
1294 | * amdgpu_device_ip_get_ip_block - get a hw IP pointer | |
1295 | * | |
1296 | * @adev: amdgpu_device pointer | |
87e3f136 | 1297 | * @type: Type of hardware IP (SMU, GFX, UVD, etc.) |
e3ecdffa AD |
1298 | * |
1299 | * Returns a pointer to the hardware IP block structure | |
1300 | * if it exists for the asic, otherwise NULL. | |
1301 | */ | |
2990a1fc AD |
1302 | struct amdgpu_ip_block * |
1303 | amdgpu_device_ip_get_ip_block(struct amdgpu_device *adev, | |
1304 | enum amd_ip_block_type type) | |
d38ceaf9 AD |
1305 | { |
1306 | int i; | |
1307 | ||
1308 | for (i = 0; i < adev->num_ip_blocks; i++) | |
a1255107 | 1309 | if (adev->ip_blocks[i].version->type == type) |
d38ceaf9 AD |
1310 | return &adev->ip_blocks[i]; |
1311 | ||
1312 | return NULL; | |
1313 | } | |
1314 | ||
1315 | /** | |
2990a1fc | 1316 | * amdgpu_device_ip_block_version_cmp |
d38ceaf9 AD |
1317 | * |
1318 | * @adev: amdgpu_device pointer | |
5fc3aeeb | 1319 | * @type: enum amd_ip_block_type |
d38ceaf9 AD |
1320 | * @major: major version |
1321 | * @minor: minor version | |
1322 | * | |
1323 | * return 0 if equal or greater | |
1324 | * return 1 if smaller or the ip_block doesn't exist | |
1325 | */ | |
2990a1fc AD |
1326 | int amdgpu_device_ip_block_version_cmp(struct amdgpu_device *adev, |
1327 | enum amd_ip_block_type type, | |
1328 | u32 major, u32 minor) | |
d38ceaf9 | 1329 | { |
2990a1fc | 1330 | struct amdgpu_ip_block *ip_block = amdgpu_device_ip_get_ip_block(adev, type); |
d38ceaf9 | 1331 | |
a1255107 AD |
1332 | if (ip_block && ((ip_block->version->major > major) || |
1333 | ((ip_block->version->major == major) && | |
1334 | (ip_block->version->minor >= minor)))) | |
d38ceaf9 AD |
1335 | return 0; |
1336 | ||
1337 | return 1; | |
1338 | } | |
1339 | ||
a1255107 | 1340 | /** |
2990a1fc | 1341 | * amdgpu_device_ip_block_add |
a1255107 AD |
1342 | * |
1343 | * @adev: amdgpu_device pointer | |
1344 | * @ip_block_version: pointer to the IP to add | |
1345 | * | |
1346 | * Adds the IP block driver information to the collection of IPs | |
1347 | * on the asic. | |
1348 | */ | |
2990a1fc AD |
1349 | int amdgpu_device_ip_block_add(struct amdgpu_device *adev, |
1350 | const struct amdgpu_ip_block_version *ip_block_version) | |
a1255107 AD |
1351 | { |
1352 | if (!ip_block_version) | |
1353 | return -EINVAL; | |
1354 | ||
e966a725 | 1355 | DRM_INFO("add ip block number %d <%s>\n", adev->num_ip_blocks, |
a0bae357 HR |
1356 | ip_block_version->funcs->name); |
1357 | ||
a1255107 AD |
1358 | adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version; |
1359 | ||
1360 | return 0; | |
1361 | } | |
1362 | ||
e3ecdffa AD |
1363 | /** |
1364 | * amdgpu_device_enable_virtual_display - enable virtual display feature | |
1365 | * | |
1366 | * @adev: amdgpu_device pointer | |
1367 | * | |
1368 | * Enabled the virtual display feature if the user has enabled it via | |
1369 | * the module parameter virtual_display. This feature provides a virtual | |
1370 | * display hardware on headless boards or in virtualized environments. | |
1371 | * This function parses and validates the configuration string specified by | |
1372 | * the user and configues the virtual display configuration (number of | |
1373 | * virtual connectors, crtcs, etc.) specified. | |
1374 | */ | |
483ef985 | 1375 | static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev) |
9accf2fd ED |
1376 | { |
1377 | adev->enable_virtual_display = false; | |
1378 | ||
1379 | if (amdgpu_virtual_display) { | |
1380 | struct drm_device *ddev = adev->ddev; | |
1381 | const char *pci_address_name = pci_name(ddev->pdev); | |
0f66356d | 1382 | char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname; |
9accf2fd ED |
1383 | |
1384 | pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL); | |
1385 | pciaddstr_tmp = pciaddstr; | |
0f66356d ED |
1386 | while ((pciaddname_tmp = strsep(&pciaddstr_tmp, ";"))) { |
1387 | pciaddname = strsep(&pciaddname_tmp, ","); | |
967de2a9 YT |
1388 | if (!strcmp("all", pciaddname) |
1389 | || !strcmp(pci_address_name, pciaddname)) { | |
0f66356d ED |
1390 | long num_crtc; |
1391 | int res = -1; | |
1392 | ||
9accf2fd | 1393 | adev->enable_virtual_display = true; |
0f66356d ED |
1394 | |
1395 | if (pciaddname_tmp) | |
1396 | res = kstrtol(pciaddname_tmp, 10, | |
1397 | &num_crtc); | |
1398 | ||
1399 | if (!res) { | |
1400 | if (num_crtc < 1) | |
1401 | num_crtc = 1; | |
1402 | if (num_crtc > 6) | |
1403 | num_crtc = 6; | |
1404 | adev->mode_info.num_crtc = num_crtc; | |
1405 | } else { | |
1406 | adev->mode_info.num_crtc = 1; | |
1407 | } | |
9accf2fd ED |
1408 | break; |
1409 | } | |
1410 | } | |
1411 | ||
0f66356d ED |
1412 | DRM_INFO("virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n", |
1413 | amdgpu_virtual_display, pci_address_name, | |
1414 | adev->enable_virtual_display, adev->mode_info.num_crtc); | |
9accf2fd ED |
1415 | |
1416 | kfree(pciaddstr); | |
1417 | } | |
1418 | } | |
1419 | ||
e3ecdffa AD |
1420 | /** |
1421 | * amdgpu_device_parse_gpu_info_fw - parse gpu info firmware | |
1422 | * | |
1423 | * @adev: amdgpu_device pointer | |
1424 | * | |
1425 | * Parses the asic configuration parameters specified in the gpu info | |
1426 | * firmware and makes them availale to the driver for use in configuring | |
1427 | * the asic. | |
1428 | * Returns 0 on success, -EINVAL on failure. | |
1429 | */ | |
e2a75f88 AD |
1430 | static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) |
1431 | { | |
e2a75f88 AD |
1432 | const char *chip_name; |
1433 | char fw_name[30]; | |
1434 | int err; | |
1435 | const struct gpu_info_firmware_header_v1_0 *hdr; | |
1436 | ||
ab4fe3e1 HR |
1437 | adev->firmware.gpu_info_fw = NULL; |
1438 | ||
e2a75f88 AD |
1439 | switch (adev->asic_type) { |
1440 | case CHIP_TOPAZ: | |
1441 | case CHIP_TONGA: | |
1442 | case CHIP_FIJI: | |
e2a75f88 | 1443 | case CHIP_POLARIS10: |
cc07f18d | 1444 | case CHIP_POLARIS11: |
e2a75f88 | 1445 | case CHIP_POLARIS12: |
cc07f18d | 1446 | case CHIP_VEGAM: |
e2a75f88 AD |
1447 | case CHIP_CARRIZO: |
1448 | case CHIP_STONEY: | |
1449 | #ifdef CONFIG_DRM_AMDGPU_SI | |
1450 | case CHIP_VERDE: | |
1451 | case CHIP_TAHITI: | |
1452 | case CHIP_PITCAIRN: | |
1453 | case CHIP_OLAND: | |
1454 | case CHIP_HAINAN: | |
1455 | #endif | |
1456 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
1457 | case CHIP_BONAIRE: | |
1458 | case CHIP_HAWAII: | |
1459 | case CHIP_KAVERI: | |
1460 | case CHIP_KABINI: | |
1461 | case CHIP_MULLINS: | |
1462 | #endif | |
27c0bc71 | 1463 | case CHIP_VEGA20: |
e2a75f88 AD |
1464 | default: |
1465 | return 0; | |
1466 | case CHIP_VEGA10: | |
1467 | chip_name = "vega10"; | |
1468 | break; | |
3f76dced AD |
1469 | case CHIP_VEGA12: |
1470 | chip_name = "vega12"; | |
1471 | break; | |
2d2e5e7e | 1472 | case CHIP_RAVEN: |
54c4d17e FX |
1473 | if (adev->rev_id >= 8) |
1474 | chip_name = "raven2"; | |
741deade AD |
1475 | else if (adev->pdev->device == 0x15d8) |
1476 | chip_name = "picasso"; | |
54c4d17e FX |
1477 | else |
1478 | chip_name = "raven"; | |
2d2e5e7e | 1479 | break; |
65e60f6e LM |
1480 | case CHIP_ARCTURUS: |
1481 | chip_name = "arcturus"; | |
1482 | break; | |
b51a26a0 HR |
1483 | case CHIP_RENOIR: |
1484 | chip_name = "renoir"; | |
1485 | break; | |
23c6268e HR |
1486 | case CHIP_NAVI10: |
1487 | chip_name = "navi10"; | |
1488 | break; | |
ed42cfe1 XY |
1489 | case CHIP_NAVI14: |
1490 | chip_name = "navi14"; | |
1491 | break; | |
42b325e5 XY |
1492 | case CHIP_NAVI12: |
1493 | chip_name = "navi12"; | |
1494 | break; | |
e2a75f88 AD |
1495 | } |
1496 | ||
1497 | snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name); | |
ab4fe3e1 | 1498 | err = request_firmware(&adev->firmware.gpu_info_fw, fw_name, adev->dev); |
e2a75f88 AD |
1499 | if (err) { |
1500 | dev_err(adev->dev, | |
1501 | "Failed to load gpu_info firmware \"%s\"\n", | |
1502 | fw_name); | |
1503 | goto out; | |
1504 | } | |
ab4fe3e1 | 1505 | err = amdgpu_ucode_validate(adev->firmware.gpu_info_fw); |
e2a75f88 AD |
1506 | if (err) { |
1507 | dev_err(adev->dev, | |
1508 | "Failed to validate gpu_info firmware \"%s\"\n", | |
1509 | fw_name); | |
1510 | goto out; | |
1511 | } | |
1512 | ||
ab4fe3e1 | 1513 | hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data; |
e2a75f88 AD |
1514 | amdgpu_ucode_print_gpu_info_hdr(&hdr->header); |
1515 | ||
1516 | switch (hdr->version_major) { | |
1517 | case 1: | |
1518 | { | |
1519 | const struct gpu_info_firmware_v1_0 *gpu_info_fw = | |
ab4fe3e1 | 1520 | (const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data + |
e2a75f88 AD |
1521 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); |
1522 | ||
ec51d3fa XY |
1523 | if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10) |
1524 | goto parse_soc_bounding_box; | |
1525 | ||
b5ab16bf AD |
1526 | adev->gfx.config.max_shader_engines = le32_to_cpu(gpu_info_fw->gc_num_se); |
1527 | adev->gfx.config.max_cu_per_sh = le32_to_cpu(gpu_info_fw->gc_num_cu_per_sh); | |
1528 | adev->gfx.config.max_sh_per_se = le32_to_cpu(gpu_info_fw->gc_num_sh_per_se); | |
1529 | adev->gfx.config.max_backends_per_se = le32_to_cpu(gpu_info_fw->gc_num_rb_per_se); | |
e2a75f88 | 1530 | adev->gfx.config.max_texture_channel_caches = |
b5ab16bf AD |
1531 | le32_to_cpu(gpu_info_fw->gc_num_tccs); |
1532 | adev->gfx.config.max_gprs = le32_to_cpu(gpu_info_fw->gc_num_gprs); | |
1533 | adev->gfx.config.max_gs_threads = le32_to_cpu(gpu_info_fw->gc_num_max_gs_thds); | |
1534 | adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gpu_info_fw->gc_gs_table_depth); | |
1535 | adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gpu_info_fw->gc_gsprim_buff_depth); | |
e2a75f88 | 1536 | adev->gfx.config.double_offchip_lds_buf = |
b5ab16bf AD |
1537 | le32_to_cpu(gpu_info_fw->gc_double_offchip_lds_buffer); |
1538 | adev->gfx.cu_info.wave_front_size = le32_to_cpu(gpu_info_fw->gc_wave_size); | |
51fd0370 HZ |
1539 | adev->gfx.cu_info.max_waves_per_simd = |
1540 | le32_to_cpu(gpu_info_fw->gc_max_waves_per_simd); | |
1541 | adev->gfx.cu_info.max_scratch_slots_per_cu = | |
1542 | le32_to_cpu(gpu_info_fw->gc_max_scratch_slots_per_cu); | |
1543 | adev->gfx.cu_info.lds_size = le32_to_cpu(gpu_info_fw->gc_lds_size); | |
48321c3d | 1544 | if (hdr->version_minor >= 1) { |
35c2e910 HZ |
1545 | const struct gpu_info_firmware_v1_1 *gpu_info_fw = |
1546 | (const struct gpu_info_firmware_v1_1 *)(adev->firmware.gpu_info_fw->data + | |
1547 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); | |
1548 | adev->gfx.config.num_sc_per_sh = | |
1549 | le32_to_cpu(gpu_info_fw->num_sc_per_sh); | |
1550 | adev->gfx.config.num_packer_per_sc = | |
1551 | le32_to_cpu(gpu_info_fw->num_packer_per_sc); | |
1552 | } | |
ec51d3fa XY |
1553 | |
1554 | parse_soc_bounding_box: | |
ec51d3fa XY |
1555 | /* |
1556 | * soc bounding box info is not integrated in disocovery table, | |
1557 | * we always need to parse it from gpu info firmware. | |
1558 | */ | |
48321c3d HW |
1559 | if (hdr->version_minor == 2) { |
1560 | const struct gpu_info_firmware_v1_2 *gpu_info_fw = | |
1561 | (const struct gpu_info_firmware_v1_2 *)(adev->firmware.gpu_info_fw->data + | |
1562 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); | |
1563 | adev->dm.soc_bounding_box = &gpu_info_fw->soc_bounding_box; | |
1564 | } | |
e2a75f88 AD |
1565 | break; |
1566 | } | |
1567 | default: | |
1568 | dev_err(adev->dev, | |
1569 | "Unsupported gpu_info table %d\n", hdr->header.ucode_version); | |
1570 | err = -EINVAL; | |
1571 | goto out; | |
1572 | } | |
1573 | out: | |
e2a75f88 AD |
1574 | return err; |
1575 | } | |
1576 | ||
e3ecdffa AD |
1577 | /** |
1578 | * amdgpu_device_ip_early_init - run early init for hardware IPs | |
1579 | * | |
1580 | * @adev: amdgpu_device pointer | |
1581 | * | |
1582 | * Early initialization pass for hardware IPs. The hardware IPs that make | |
1583 | * up each asic are discovered each IP's early_init callback is run. This | |
1584 | * is the first stage in initializing the asic. | |
1585 | * Returns 0 on success, negative error code on failure. | |
1586 | */ | |
06ec9070 | 1587 | static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) |
d38ceaf9 | 1588 | { |
aaa36a97 | 1589 | int i, r; |
d38ceaf9 | 1590 | |
483ef985 | 1591 | amdgpu_device_enable_virtual_display(adev); |
a6be7570 | 1592 | |
d38ceaf9 | 1593 | switch (adev->asic_type) { |
aaa36a97 AD |
1594 | case CHIP_TOPAZ: |
1595 | case CHIP_TONGA: | |
48299f95 | 1596 | case CHIP_FIJI: |
2cc0c0b5 | 1597 | case CHIP_POLARIS10: |
32cc7e53 | 1598 | case CHIP_POLARIS11: |
c4642a47 | 1599 | case CHIP_POLARIS12: |
32cc7e53 | 1600 | case CHIP_VEGAM: |
aaa36a97 | 1601 | case CHIP_CARRIZO: |
39bb0c92 SL |
1602 | case CHIP_STONEY: |
1603 | if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) | |
aaa36a97 AD |
1604 | adev->family = AMDGPU_FAMILY_CZ; |
1605 | else | |
1606 | adev->family = AMDGPU_FAMILY_VI; | |
1607 | ||
1608 | r = vi_set_ip_blocks(adev); | |
1609 | if (r) | |
1610 | return r; | |
1611 | break; | |
33f34802 KW |
1612 | #ifdef CONFIG_DRM_AMDGPU_SI |
1613 | case CHIP_VERDE: | |
1614 | case CHIP_TAHITI: | |
1615 | case CHIP_PITCAIRN: | |
1616 | case CHIP_OLAND: | |
1617 | case CHIP_HAINAN: | |
295d0daf | 1618 | adev->family = AMDGPU_FAMILY_SI; |
33f34802 KW |
1619 | r = si_set_ip_blocks(adev); |
1620 | if (r) | |
1621 | return r; | |
1622 | break; | |
1623 | #endif | |
a2e73f56 AD |
1624 | #ifdef CONFIG_DRM_AMDGPU_CIK |
1625 | case CHIP_BONAIRE: | |
1626 | case CHIP_HAWAII: | |
1627 | case CHIP_KAVERI: | |
1628 | case CHIP_KABINI: | |
1629 | case CHIP_MULLINS: | |
1630 | if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII)) | |
1631 | adev->family = AMDGPU_FAMILY_CI; | |
1632 | else | |
1633 | adev->family = AMDGPU_FAMILY_KV; | |
1634 | ||
1635 | r = cik_set_ip_blocks(adev); | |
1636 | if (r) | |
1637 | return r; | |
1638 | break; | |
1639 | #endif | |
e48a3cd9 AD |
1640 | case CHIP_VEGA10: |
1641 | case CHIP_VEGA12: | |
e4bd8170 | 1642 | case CHIP_VEGA20: |
e48a3cd9 | 1643 | case CHIP_RAVEN: |
61cf44c1 | 1644 | case CHIP_ARCTURUS: |
b51a26a0 HR |
1645 | case CHIP_RENOIR: |
1646 | if (adev->asic_type == CHIP_RAVEN || | |
1647 | adev->asic_type == CHIP_RENOIR) | |
2ca8a5d2 CZ |
1648 | adev->family = AMDGPU_FAMILY_RV; |
1649 | else | |
1650 | adev->family = AMDGPU_FAMILY_AI; | |
460826e6 KW |
1651 | |
1652 | r = soc15_set_ip_blocks(adev); | |
1653 | if (r) | |
1654 | return r; | |
1655 | break; | |
0a5b8c7b | 1656 | case CHIP_NAVI10: |
7ecb5cd4 | 1657 | case CHIP_NAVI14: |
4808cf9c | 1658 | case CHIP_NAVI12: |
0a5b8c7b HR |
1659 | adev->family = AMDGPU_FAMILY_NV; |
1660 | ||
1661 | r = nv_set_ip_blocks(adev); | |
1662 | if (r) | |
1663 | return r; | |
1664 | break; | |
d38ceaf9 AD |
1665 | default: |
1666 | /* FIXME: not supported yet */ | |
1667 | return -EINVAL; | |
1668 | } | |
1669 | ||
e2a75f88 AD |
1670 | r = amdgpu_device_parse_gpu_info_fw(adev); |
1671 | if (r) | |
1672 | return r; | |
1673 | ||
ec51d3fa XY |
1674 | if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10) |
1675 | amdgpu_discovery_get_gfx_info(adev); | |
1676 | ||
1884734a | 1677 | amdgpu_amdkfd_device_probe(adev); |
1678 | ||
3149d9da XY |
1679 | if (amdgpu_sriov_vf(adev)) { |
1680 | r = amdgpu_virt_request_full_gpu(adev, true); | |
1681 | if (r) | |
5ffa61c1 | 1682 | return -EAGAIN; |
3149d9da XY |
1683 | } |
1684 | ||
3b94fb10 | 1685 | adev->pm.pp_feature = amdgpu_pp_feature_mask; |
a35ad98b | 1686 | if (amdgpu_sriov_vf(adev) || sched_policy == KFD_SCHED_POLICY_NO_HWS) |
00544006 | 1687 | adev->pm.pp_feature &= ~PP_GFXOFF_MASK; |
00f54b97 | 1688 | |
d38ceaf9 AD |
1689 | for (i = 0; i < adev->num_ip_blocks; i++) { |
1690 | if ((amdgpu_ip_block_mask & (1 << i)) == 0) { | |
ed8cf00c HR |
1691 | DRM_ERROR("disabled ip block: %d <%s>\n", |
1692 | i, adev->ip_blocks[i].version->funcs->name); | |
a1255107 | 1693 | adev->ip_blocks[i].status.valid = false; |
d38ceaf9 | 1694 | } else { |
a1255107 AD |
1695 | if (adev->ip_blocks[i].version->funcs->early_init) { |
1696 | r = adev->ip_blocks[i].version->funcs->early_init((void *)adev); | |
2c1a2784 | 1697 | if (r == -ENOENT) { |
a1255107 | 1698 | adev->ip_blocks[i].status.valid = false; |
2c1a2784 | 1699 | } else if (r) { |
a1255107 AD |
1700 | DRM_ERROR("early_init of IP block <%s> failed %d\n", |
1701 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 1702 | return r; |
2c1a2784 | 1703 | } else { |
a1255107 | 1704 | adev->ip_blocks[i].status.valid = true; |
2c1a2784 | 1705 | } |
974e6b64 | 1706 | } else { |
a1255107 | 1707 | adev->ip_blocks[i].status.valid = true; |
d38ceaf9 | 1708 | } |
d38ceaf9 | 1709 | } |
21a249ca AD |
1710 | /* get the vbios after the asic_funcs are set up */ |
1711 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON) { | |
1712 | /* Read BIOS */ | |
1713 | if (!amdgpu_get_bios(adev)) | |
1714 | return -EINVAL; | |
1715 | ||
1716 | r = amdgpu_atombios_init(adev); | |
1717 | if (r) { | |
1718 | dev_err(adev->dev, "amdgpu_atombios_init failed\n"); | |
1719 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0); | |
1720 | return r; | |
1721 | } | |
1722 | } | |
d38ceaf9 AD |
1723 | } |
1724 | ||
395d1fb9 NH |
1725 | adev->cg_flags &= amdgpu_cg_mask; |
1726 | adev->pg_flags &= amdgpu_pg_mask; | |
1727 | ||
d38ceaf9 AD |
1728 | return 0; |
1729 | } | |
1730 | ||
0a4f2520 RZ |
1731 | static int amdgpu_device_ip_hw_init_phase1(struct amdgpu_device *adev) |
1732 | { | |
1733 | int i, r; | |
1734 | ||
1735 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1736 | if (!adev->ip_blocks[i].status.sw) | |
1737 | continue; | |
1738 | if (adev->ip_blocks[i].status.hw) | |
1739 | continue; | |
1740 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || | |
2d11fd3f | 1741 | (amdgpu_sriov_vf(adev) && (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP)) || |
0a4f2520 RZ |
1742 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) { |
1743 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
1744 | if (r) { | |
1745 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
1746 | adev->ip_blocks[i].version->funcs->name, r); | |
1747 | return r; | |
1748 | } | |
1749 | adev->ip_blocks[i].status.hw = true; | |
1750 | } | |
1751 | } | |
1752 | ||
1753 | return 0; | |
1754 | } | |
1755 | ||
1756 | static int amdgpu_device_ip_hw_init_phase2(struct amdgpu_device *adev) | |
1757 | { | |
1758 | int i, r; | |
1759 | ||
1760 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1761 | if (!adev->ip_blocks[i].status.sw) | |
1762 | continue; | |
1763 | if (adev->ip_blocks[i].status.hw) | |
1764 | continue; | |
1765 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
1766 | if (r) { | |
1767 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
1768 | adev->ip_blocks[i].version->funcs->name, r); | |
1769 | return r; | |
1770 | } | |
1771 | adev->ip_blocks[i].status.hw = true; | |
1772 | } | |
1773 | ||
1774 | return 0; | |
1775 | } | |
1776 | ||
7a3e0bb2 RZ |
1777 | static int amdgpu_device_fw_loading(struct amdgpu_device *adev) |
1778 | { | |
1779 | int r = 0; | |
1780 | int i; | |
80f41f84 | 1781 | uint32_t smu_version; |
7a3e0bb2 RZ |
1782 | |
1783 | if (adev->asic_type >= CHIP_VEGA10) { | |
1784 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
482f0e53 ML |
1785 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_PSP) |
1786 | continue; | |
1787 | ||
1788 | /* no need to do the fw loading again if already done*/ | |
1789 | if (adev->ip_blocks[i].status.hw == true) | |
1790 | break; | |
1791 | ||
1792 | if (adev->in_gpu_reset || adev->in_suspend) { | |
1793 | r = adev->ip_blocks[i].version->funcs->resume(adev); | |
1794 | if (r) { | |
1795 | DRM_ERROR("resume of IP block <%s> failed %d\n", | |
7a3e0bb2 | 1796 | adev->ip_blocks[i].version->funcs->name, r); |
482f0e53 ML |
1797 | return r; |
1798 | } | |
1799 | } else { | |
1800 | r = adev->ip_blocks[i].version->funcs->hw_init(adev); | |
1801 | if (r) { | |
1802 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", | |
1803 | adev->ip_blocks[i].version->funcs->name, r); | |
1804 | return r; | |
7a3e0bb2 | 1805 | } |
7a3e0bb2 | 1806 | } |
482f0e53 ML |
1807 | |
1808 | adev->ip_blocks[i].status.hw = true; | |
1809 | break; | |
7a3e0bb2 RZ |
1810 | } |
1811 | } | |
482f0e53 | 1812 | |
8973d9ec ED |
1813 | if (!amdgpu_sriov_vf(adev) || adev->asic_type == CHIP_TONGA) |
1814 | r = amdgpu_pm_load_smu_firmware(adev, &smu_version); | |
7a3e0bb2 | 1815 | |
80f41f84 | 1816 | return r; |
7a3e0bb2 RZ |
1817 | } |
1818 | ||
e3ecdffa AD |
1819 | /** |
1820 | * amdgpu_device_ip_init - run init for hardware IPs | |
1821 | * | |
1822 | * @adev: amdgpu_device pointer | |
1823 | * | |
1824 | * Main initialization pass for hardware IPs. The list of all the hardware | |
1825 | * IPs that make up the asic is walked and the sw_init and hw_init callbacks | |
1826 | * are run. sw_init initializes the software state associated with each IP | |
1827 | * and hw_init initializes the hardware associated with each IP. | |
1828 | * Returns 0 on success, negative error code on failure. | |
1829 | */ | |
06ec9070 | 1830 | static int amdgpu_device_ip_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
1831 | { |
1832 | int i, r; | |
1833 | ||
c030f2e4 | 1834 | r = amdgpu_ras_init(adev); |
1835 | if (r) | |
1836 | return r; | |
1837 | ||
d38ceaf9 | 1838 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 1839 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 1840 | continue; |
a1255107 | 1841 | r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev); |
2c1a2784 | 1842 | if (r) { |
a1255107 AD |
1843 | DRM_ERROR("sw_init of IP block <%s> failed %d\n", |
1844 | adev->ip_blocks[i].version->funcs->name, r); | |
72d3f592 | 1845 | goto init_failed; |
2c1a2784 | 1846 | } |
a1255107 | 1847 | adev->ip_blocks[i].status.sw = true; |
bfca0289 | 1848 | |
d38ceaf9 | 1849 | /* need to do gmc hw init early so we can allocate gpu mem */ |
a1255107 | 1850 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { |
06ec9070 | 1851 | r = amdgpu_device_vram_scratch_init(adev); |
2c1a2784 AD |
1852 | if (r) { |
1853 | DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r); | |
72d3f592 | 1854 | goto init_failed; |
2c1a2784 | 1855 | } |
a1255107 | 1856 | r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); |
2c1a2784 AD |
1857 | if (r) { |
1858 | DRM_ERROR("hw_init %d failed %d\n", i, r); | |
72d3f592 | 1859 | goto init_failed; |
2c1a2784 | 1860 | } |
06ec9070 | 1861 | r = amdgpu_device_wb_init(adev); |
2c1a2784 | 1862 | if (r) { |
06ec9070 | 1863 | DRM_ERROR("amdgpu_device_wb_init failed %d\n", r); |
72d3f592 | 1864 | goto init_failed; |
2c1a2784 | 1865 | } |
a1255107 | 1866 | adev->ip_blocks[i].status.hw = true; |
2493664f ML |
1867 | |
1868 | /* right after GMC hw init, we create CSA */ | |
f92d5c61 | 1869 | if (amdgpu_mcbp || amdgpu_sriov_vf(adev)) { |
1e256e27 RZ |
1870 | r = amdgpu_allocate_static_csa(adev, &adev->virt.csa_obj, |
1871 | AMDGPU_GEM_DOMAIN_VRAM, | |
1872 | AMDGPU_CSA_SIZE); | |
2493664f ML |
1873 | if (r) { |
1874 | DRM_ERROR("allocate CSA failed %d\n", r); | |
72d3f592 | 1875 | goto init_failed; |
2493664f ML |
1876 | } |
1877 | } | |
d38ceaf9 AD |
1878 | } |
1879 | } | |
1880 | ||
c9ffa427 YT |
1881 | if (amdgpu_sriov_vf(adev)) |
1882 | amdgpu_virt_init_data_exchange(adev); | |
1883 | ||
533aed27 AG |
1884 | r = amdgpu_ib_pool_init(adev); |
1885 | if (r) { | |
1886 | dev_err(adev->dev, "IB initialization failed (%d).\n", r); | |
1887 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r); | |
1888 | goto init_failed; | |
1889 | } | |
1890 | ||
c8963ea4 RZ |
1891 | r = amdgpu_ucode_create_bo(adev); /* create ucode bo when sw_init complete*/ |
1892 | if (r) | |
72d3f592 | 1893 | goto init_failed; |
0a4f2520 RZ |
1894 | |
1895 | r = amdgpu_device_ip_hw_init_phase1(adev); | |
1896 | if (r) | |
72d3f592 | 1897 | goto init_failed; |
0a4f2520 | 1898 | |
7a3e0bb2 RZ |
1899 | r = amdgpu_device_fw_loading(adev); |
1900 | if (r) | |
72d3f592 | 1901 | goto init_failed; |
7a3e0bb2 | 1902 | |
0a4f2520 RZ |
1903 | r = amdgpu_device_ip_hw_init_phase2(adev); |
1904 | if (r) | |
72d3f592 | 1905 | goto init_failed; |
d38ceaf9 | 1906 | |
121a2bc6 AG |
1907 | /* |
1908 | * retired pages will be loaded from eeprom and reserved here, | |
1909 | * it should be called after amdgpu_device_ip_hw_init_phase2 since | |
1910 | * for some ASICs the RAS EEPROM code relies on SMU fully functioning | |
1911 | * for I2C communication which only true at this point. | |
1912 | * recovery_init may fail, but it can free all resources allocated by | |
1913 | * itself and its failure should not stop amdgpu init process. | |
1914 | * | |
1915 | * Note: theoretically, this should be called before all vram allocations | |
1916 | * to protect retired page from abusing | |
1917 | */ | |
1918 | amdgpu_ras_recovery_init(adev); | |
1919 | ||
3e2e2ab5 HZ |
1920 | if (adev->gmc.xgmi.num_physical_nodes > 1) |
1921 | amdgpu_xgmi_add_device(adev); | |
1884734a | 1922 | amdgpu_amdkfd_device_init(adev); |
c6332b97 | 1923 | |
72d3f592 | 1924 | init_failed: |
c9ffa427 | 1925 | if (amdgpu_sriov_vf(adev)) |
c6332b97 | 1926 | amdgpu_virt_release_full_gpu(adev, true); |
1927 | ||
72d3f592 | 1928 | return r; |
d38ceaf9 AD |
1929 | } |
1930 | ||
e3ecdffa AD |
1931 | /** |
1932 | * amdgpu_device_fill_reset_magic - writes reset magic to gart pointer | |
1933 | * | |
1934 | * @adev: amdgpu_device pointer | |
1935 | * | |
1936 | * Writes a reset magic value to the gart pointer in VRAM. The driver calls | |
1937 | * this function before a GPU reset. If the value is retained after a | |
1938 | * GPU reset, VRAM has not been lost. Some GPU resets may destry VRAM contents. | |
1939 | */ | |
06ec9070 | 1940 | static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev) |
0c49e0b8 CZ |
1941 | { |
1942 | memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM); | |
1943 | } | |
1944 | ||
e3ecdffa AD |
1945 | /** |
1946 | * amdgpu_device_check_vram_lost - check if vram is valid | |
1947 | * | |
1948 | * @adev: amdgpu_device pointer | |
1949 | * | |
1950 | * Checks the reset magic value written to the gart pointer in VRAM. | |
1951 | * The driver calls this after a GPU reset to see if the contents of | |
1952 | * VRAM is lost or now. | |
1953 | * returns true if vram is lost, false if not. | |
1954 | */ | |
06ec9070 | 1955 | static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev) |
0c49e0b8 CZ |
1956 | { |
1957 | return !!memcmp(adev->gart.ptr, adev->reset_magic, | |
1958 | AMDGPU_RESET_MAGIC_NUM); | |
1959 | } | |
1960 | ||
e3ecdffa | 1961 | /** |
1112a46b | 1962 | * amdgpu_device_set_cg_state - set clockgating for amdgpu device |
e3ecdffa AD |
1963 | * |
1964 | * @adev: amdgpu_device pointer | |
b8b72130 | 1965 | * @state: clockgating state (gate or ungate) |
e3ecdffa | 1966 | * |
e3ecdffa | 1967 | * The list of all the hardware IPs that make up the asic is walked and the |
1112a46b RZ |
1968 | * set_clockgating_state callbacks are run. |
1969 | * Late initialization pass enabling clockgating for hardware IPs. | |
1970 | * Fini or suspend, pass disabling clockgating for hardware IPs. | |
e3ecdffa AD |
1971 | * Returns 0 on success, negative error code on failure. |
1972 | */ | |
fdd34271 | 1973 | |
1112a46b RZ |
1974 | static int amdgpu_device_set_cg_state(struct amdgpu_device *adev, |
1975 | enum amd_clockgating_state state) | |
d38ceaf9 | 1976 | { |
1112a46b | 1977 | int i, j, r; |
d38ceaf9 | 1978 | |
4a2ba394 SL |
1979 | if (amdgpu_emu_mode == 1) |
1980 | return 0; | |
1981 | ||
1112a46b RZ |
1982 | for (j = 0; j < adev->num_ip_blocks; j++) { |
1983 | i = state == AMD_CG_STATE_GATE ? j : adev->num_ip_blocks - j - 1; | |
a2d31dc3 | 1984 | if (!adev->ip_blocks[i].status.late_initialized) |
d38ceaf9 | 1985 | continue; |
4a446d55 | 1986 | /* skip CG for VCE/UVD, it's handled specially */ |
a1255107 | 1987 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && |
57716327 | 1988 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE && |
34319b32 | 1989 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN && |
52f2e779 | 1990 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG && |
57716327 | 1991 | adev->ip_blocks[i].version->funcs->set_clockgating_state) { |
4a446d55 | 1992 | /* enable clockgating to save power */ |
a1255107 | 1993 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev, |
1112a46b | 1994 | state); |
4a446d55 AD |
1995 | if (r) { |
1996 | DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n", | |
a1255107 | 1997 | adev->ip_blocks[i].version->funcs->name, r); |
4a446d55 AD |
1998 | return r; |
1999 | } | |
b0b00ff1 | 2000 | } |
d38ceaf9 | 2001 | } |
06b18f61 | 2002 | |
c9f96fd5 RZ |
2003 | return 0; |
2004 | } | |
2005 | ||
1112a46b | 2006 | static int amdgpu_device_set_pg_state(struct amdgpu_device *adev, enum amd_powergating_state state) |
c9f96fd5 | 2007 | { |
1112a46b | 2008 | int i, j, r; |
06b18f61 | 2009 | |
c9f96fd5 RZ |
2010 | if (amdgpu_emu_mode == 1) |
2011 | return 0; | |
2012 | ||
1112a46b RZ |
2013 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2014 | i = state == AMD_PG_STATE_GATE ? j : adev->num_ip_blocks - j - 1; | |
a2d31dc3 | 2015 | if (!adev->ip_blocks[i].status.late_initialized) |
c9f96fd5 RZ |
2016 | continue; |
2017 | /* skip CG for VCE/UVD, it's handled specially */ | |
2018 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && | |
2019 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE && | |
2020 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN && | |
52f2e779 | 2021 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_JPEG && |
c9f96fd5 RZ |
2022 | adev->ip_blocks[i].version->funcs->set_powergating_state) { |
2023 | /* enable powergating to save power */ | |
2024 | r = adev->ip_blocks[i].version->funcs->set_powergating_state((void *)adev, | |
1112a46b | 2025 | state); |
c9f96fd5 RZ |
2026 | if (r) { |
2027 | DRM_ERROR("set_powergating_state(gate) of IP block <%s> failed %d\n", | |
2028 | adev->ip_blocks[i].version->funcs->name, r); | |
2029 | return r; | |
2030 | } | |
2031 | } | |
2032 | } | |
2dc80b00 S |
2033 | return 0; |
2034 | } | |
2035 | ||
beff74bc AD |
2036 | static int amdgpu_device_enable_mgpu_fan_boost(void) |
2037 | { | |
2038 | struct amdgpu_gpu_instance *gpu_ins; | |
2039 | struct amdgpu_device *adev; | |
2040 | int i, ret = 0; | |
2041 | ||
2042 | mutex_lock(&mgpu_info.mutex); | |
2043 | ||
2044 | /* | |
2045 | * MGPU fan boost feature should be enabled | |
2046 | * only when there are two or more dGPUs in | |
2047 | * the system | |
2048 | */ | |
2049 | if (mgpu_info.num_dgpu < 2) | |
2050 | goto out; | |
2051 | ||
2052 | for (i = 0; i < mgpu_info.num_dgpu; i++) { | |
2053 | gpu_ins = &(mgpu_info.gpu_ins[i]); | |
2054 | adev = gpu_ins->adev; | |
2055 | if (!(adev->flags & AMD_IS_APU) && | |
2056 | !gpu_ins->mgpu_fan_enabled && | |
2057 | adev->powerplay.pp_funcs && | |
2058 | adev->powerplay.pp_funcs->enable_mgpu_fan_boost) { | |
2059 | ret = amdgpu_dpm_enable_mgpu_fan_boost(adev); | |
2060 | if (ret) | |
2061 | break; | |
2062 | ||
2063 | gpu_ins->mgpu_fan_enabled = 1; | |
2064 | } | |
2065 | } | |
2066 | ||
2067 | out: | |
2068 | mutex_unlock(&mgpu_info.mutex); | |
2069 | ||
2070 | return ret; | |
2071 | } | |
2072 | ||
e3ecdffa AD |
2073 | /** |
2074 | * amdgpu_device_ip_late_init - run late init for hardware IPs | |
2075 | * | |
2076 | * @adev: amdgpu_device pointer | |
2077 | * | |
2078 | * Late initialization pass for hardware IPs. The list of all the hardware | |
2079 | * IPs that make up the asic is walked and the late_init callbacks are run. | |
2080 | * late_init covers any special initialization that an IP requires | |
2081 | * after all of the have been initialized or something that needs to happen | |
2082 | * late in the init process. | |
2083 | * Returns 0 on success, negative error code on failure. | |
2084 | */ | |
06ec9070 | 2085 | static int amdgpu_device_ip_late_init(struct amdgpu_device *adev) |
2dc80b00 | 2086 | { |
60599a03 | 2087 | struct amdgpu_gpu_instance *gpu_instance; |
2dc80b00 S |
2088 | int i = 0, r; |
2089 | ||
2090 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
73f847db | 2091 | if (!adev->ip_blocks[i].status.hw) |
2dc80b00 S |
2092 | continue; |
2093 | if (adev->ip_blocks[i].version->funcs->late_init) { | |
2094 | r = adev->ip_blocks[i].version->funcs->late_init((void *)adev); | |
2095 | if (r) { | |
2096 | DRM_ERROR("late_init of IP block <%s> failed %d\n", | |
2097 | adev->ip_blocks[i].version->funcs->name, r); | |
2098 | return r; | |
2099 | } | |
2dc80b00 | 2100 | } |
73f847db | 2101 | adev->ip_blocks[i].status.late_initialized = true; |
2dc80b00 S |
2102 | } |
2103 | ||
1112a46b RZ |
2104 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE); |
2105 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE); | |
916ac57f | 2106 | |
06ec9070 | 2107 | amdgpu_device_fill_reset_magic(adev); |
d38ceaf9 | 2108 | |
beff74bc AD |
2109 | r = amdgpu_device_enable_mgpu_fan_boost(); |
2110 | if (r) | |
2111 | DRM_ERROR("enable mgpu fan boost failed (%d).\n", r); | |
2112 | ||
60599a03 EQ |
2113 | |
2114 | if (adev->gmc.xgmi.num_physical_nodes > 1) { | |
2115 | mutex_lock(&mgpu_info.mutex); | |
2116 | ||
2117 | /* | |
2118 | * Reset device p-state to low as this was booted with high. | |
2119 | * | |
2120 | * This should be performed only after all devices from the same | |
2121 | * hive get initialized. | |
2122 | * | |
2123 | * However, it's unknown how many device in the hive in advance. | |
2124 | * As this is counted one by one during devices initializations. | |
2125 | * | |
2126 | * So, we wait for all XGMI interlinked devices initialized. | |
2127 | * This may bring some delays as those devices may come from | |
2128 | * different hives. But that should be OK. | |
2129 | */ | |
2130 | if (mgpu_info.num_dgpu == adev->gmc.xgmi.num_physical_nodes) { | |
2131 | for (i = 0; i < mgpu_info.num_gpu; i++) { | |
2132 | gpu_instance = &(mgpu_info.gpu_ins[i]); | |
2133 | if (gpu_instance->adev->flags & AMD_IS_APU) | |
2134 | continue; | |
2135 | ||
2136 | r = amdgpu_xgmi_set_pstate(gpu_instance->adev, 0); | |
2137 | if (r) { | |
2138 | DRM_ERROR("pstate setting failed (%d).\n", r); | |
2139 | break; | |
2140 | } | |
2141 | } | |
2142 | } | |
2143 | ||
2144 | mutex_unlock(&mgpu_info.mutex); | |
2145 | } | |
2146 | ||
d38ceaf9 AD |
2147 | return 0; |
2148 | } | |
2149 | ||
e3ecdffa AD |
2150 | /** |
2151 | * amdgpu_device_ip_fini - run fini for hardware IPs | |
2152 | * | |
2153 | * @adev: amdgpu_device pointer | |
2154 | * | |
2155 | * Main teardown pass for hardware IPs. The list of all the hardware | |
2156 | * IPs that make up the asic is walked and the hw_fini and sw_fini callbacks | |
2157 | * are run. hw_fini tears down the hardware associated with each IP | |
2158 | * and sw_fini tears down any software state associated with each IP. | |
2159 | * Returns 0 on success, negative error code on failure. | |
2160 | */ | |
06ec9070 | 2161 | static int amdgpu_device_ip_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
2162 | { |
2163 | int i, r; | |
2164 | ||
c030f2e4 | 2165 | amdgpu_ras_pre_fini(adev); |
2166 | ||
a82400b5 AG |
2167 | if (adev->gmc.xgmi.num_physical_nodes > 1) |
2168 | amdgpu_xgmi_remove_device(adev); | |
2169 | ||
1884734a | 2170 | amdgpu_amdkfd_device_fini(adev); |
05df1f01 RZ |
2171 | |
2172 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); | |
fdd34271 RZ |
2173 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); |
2174 | ||
3e96dbfd AD |
2175 | /* need to disable SMC first */ |
2176 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 2177 | if (!adev->ip_blocks[i].status.hw) |
3e96dbfd | 2178 | continue; |
fdd34271 | 2179 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { |
a1255107 | 2180 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); |
3e96dbfd AD |
2181 | /* XXX handle errors */ |
2182 | if (r) { | |
2183 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", | |
a1255107 | 2184 | adev->ip_blocks[i].version->funcs->name, r); |
3e96dbfd | 2185 | } |
a1255107 | 2186 | adev->ip_blocks[i].status.hw = false; |
3e96dbfd AD |
2187 | break; |
2188 | } | |
2189 | } | |
2190 | ||
d38ceaf9 | 2191 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 2192 | if (!adev->ip_blocks[i].status.hw) |
d38ceaf9 | 2193 | continue; |
8201a67a | 2194 | |
a1255107 | 2195 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); |
d38ceaf9 | 2196 | /* XXX handle errors */ |
2c1a2784 | 2197 | if (r) { |
a1255107 AD |
2198 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", |
2199 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 2200 | } |
8201a67a | 2201 | |
a1255107 | 2202 | adev->ip_blocks[i].status.hw = false; |
d38ceaf9 AD |
2203 | } |
2204 | ||
9950cda2 | 2205 | |
d38ceaf9 | 2206 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 2207 | if (!adev->ip_blocks[i].status.sw) |
d38ceaf9 | 2208 | continue; |
c12aba3a ML |
2209 | |
2210 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { | |
c8963ea4 | 2211 | amdgpu_ucode_free_bo(adev); |
1e256e27 | 2212 | amdgpu_free_static_csa(&adev->virt.csa_obj); |
c12aba3a ML |
2213 | amdgpu_device_wb_fini(adev); |
2214 | amdgpu_device_vram_scratch_fini(adev); | |
533aed27 | 2215 | amdgpu_ib_pool_fini(adev); |
c12aba3a ML |
2216 | } |
2217 | ||
a1255107 | 2218 | r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev); |
d38ceaf9 | 2219 | /* XXX handle errors */ |
2c1a2784 | 2220 | if (r) { |
a1255107 AD |
2221 | DRM_DEBUG("sw_fini of IP block <%s> failed %d\n", |
2222 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 2223 | } |
a1255107 AD |
2224 | adev->ip_blocks[i].status.sw = false; |
2225 | adev->ip_blocks[i].status.valid = false; | |
d38ceaf9 AD |
2226 | } |
2227 | ||
a6dcfd9c | 2228 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 2229 | if (!adev->ip_blocks[i].status.late_initialized) |
8a2eef1d | 2230 | continue; |
a1255107 AD |
2231 | if (adev->ip_blocks[i].version->funcs->late_fini) |
2232 | adev->ip_blocks[i].version->funcs->late_fini((void *)adev); | |
2233 | adev->ip_blocks[i].status.late_initialized = false; | |
a6dcfd9c ML |
2234 | } |
2235 | ||
c030f2e4 | 2236 | amdgpu_ras_fini(adev); |
2237 | ||
030308fc | 2238 | if (amdgpu_sriov_vf(adev)) |
24136135 ML |
2239 | if (amdgpu_virt_release_full_gpu(adev, false)) |
2240 | DRM_ERROR("failed to release exclusive mode on fini\n"); | |
2493664f | 2241 | |
d38ceaf9 AD |
2242 | return 0; |
2243 | } | |
2244 | ||
e3ecdffa | 2245 | /** |
beff74bc | 2246 | * amdgpu_device_delayed_init_work_handler - work handler for IB tests |
e3ecdffa | 2247 | * |
1112a46b | 2248 | * @work: work_struct. |
e3ecdffa | 2249 | */ |
beff74bc | 2250 | static void amdgpu_device_delayed_init_work_handler(struct work_struct *work) |
2dc80b00 S |
2251 | { |
2252 | struct amdgpu_device *adev = | |
beff74bc | 2253 | container_of(work, struct amdgpu_device, delayed_init_work.work); |
916ac57f RZ |
2254 | int r; |
2255 | ||
2256 | r = amdgpu_ib_ring_tests(adev); | |
2257 | if (r) | |
2258 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
2dc80b00 S |
2259 | } |
2260 | ||
1e317b99 RZ |
2261 | static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work) |
2262 | { | |
2263 | struct amdgpu_device *adev = | |
2264 | container_of(work, struct amdgpu_device, gfx.gfx_off_delay_work.work); | |
2265 | ||
2266 | mutex_lock(&adev->gfx.gfx_off_mutex); | |
2267 | if (!adev->gfx.gfx_off_state && !adev->gfx.gfx_off_req_count) { | |
2268 | if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true)) | |
2269 | adev->gfx.gfx_off_state = true; | |
2270 | } | |
2271 | mutex_unlock(&adev->gfx.gfx_off_mutex); | |
2272 | } | |
2273 | ||
e3ecdffa | 2274 | /** |
e7854a03 | 2275 | * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1) |
e3ecdffa AD |
2276 | * |
2277 | * @adev: amdgpu_device pointer | |
2278 | * | |
2279 | * Main suspend function for hardware IPs. The list of all the hardware | |
2280 | * IPs that make up the asic is walked, clockgating is disabled and the | |
2281 | * suspend callbacks are run. suspend puts the hardware and software state | |
2282 | * in each IP into a state suitable for suspend. | |
2283 | * Returns 0 on success, negative error code on failure. | |
2284 | */ | |
e7854a03 AD |
2285 | static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev) |
2286 | { | |
2287 | int i, r; | |
2288 | ||
05df1f01 | 2289 | amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); |
fdd34271 | 2290 | amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); |
05df1f01 | 2291 | |
e7854a03 AD |
2292 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
2293 | if (!adev->ip_blocks[i].status.valid) | |
2294 | continue; | |
2295 | /* displays are handled separately */ | |
2296 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) { | |
e7854a03 AD |
2297 | /* XXX handle errors */ |
2298 | r = adev->ip_blocks[i].version->funcs->suspend(adev); | |
2299 | /* XXX handle errors */ | |
2300 | if (r) { | |
2301 | DRM_ERROR("suspend of IP block <%s> failed %d\n", | |
2302 | adev->ip_blocks[i].version->funcs->name, r); | |
482f0e53 | 2303 | return r; |
e7854a03 | 2304 | } |
482f0e53 | 2305 | adev->ip_blocks[i].status.hw = false; |
e7854a03 AD |
2306 | } |
2307 | } | |
2308 | ||
e7854a03 AD |
2309 | return 0; |
2310 | } | |
2311 | ||
2312 | /** | |
2313 | * amdgpu_device_ip_suspend_phase2 - run suspend for hardware IPs (phase 2) | |
2314 | * | |
2315 | * @adev: amdgpu_device pointer | |
2316 | * | |
2317 | * Main suspend function for hardware IPs. The list of all the hardware | |
2318 | * IPs that make up the asic is walked, clockgating is disabled and the | |
2319 | * suspend callbacks are run. suspend puts the hardware and software state | |
2320 | * in each IP into a state suitable for suspend. | |
2321 | * Returns 0 on success, negative error code on failure. | |
2322 | */ | |
2323 | static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev) | |
d38ceaf9 AD |
2324 | { |
2325 | int i, r; | |
2326 | ||
2327 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
a1255107 | 2328 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 2329 | continue; |
e7854a03 AD |
2330 | /* displays are handled in phase1 */ |
2331 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) | |
2332 | continue; | |
bff77e86 LM |
2333 | /* PSP lost connection when err_event_athub occurs */ |
2334 | if (amdgpu_ras_intr_triggered() && | |
2335 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) { | |
2336 | adev->ip_blocks[i].status.hw = false; | |
2337 | continue; | |
2338 | } | |
d38ceaf9 | 2339 | /* XXX handle errors */ |
a1255107 | 2340 | r = adev->ip_blocks[i].version->funcs->suspend(adev); |
d38ceaf9 | 2341 | /* XXX handle errors */ |
2c1a2784 | 2342 | if (r) { |
a1255107 AD |
2343 | DRM_ERROR("suspend of IP block <%s> failed %d\n", |
2344 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 2345 | } |
876923fb | 2346 | adev->ip_blocks[i].status.hw = false; |
a3a09142 AD |
2347 | /* handle putting the SMC in the appropriate state */ |
2348 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { | |
2349 | if (is_support_sw_smu(adev)) { | |
0e0b89c0 | 2350 | r = smu_set_mp1_state(&adev->smu, adev->mp1_state); |
a3a09142 | 2351 | } else if (adev->powerplay.pp_funcs && |
482f0e53 | 2352 | adev->powerplay.pp_funcs->set_mp1_state) { |
a3a09142 AD |
2353 | r = adev->powerplay.pp_funcs->set_mp1_state( |
2354 | adev->powerplay.pp_handle, | |
2355 | adev->mp1_state); | |
0e0b89c0 EQ |
2356 | } |
2357 | if (r) { | |
2358 | DRM_ERROR("SMC failed to set mp1 state %d, %d\n", | |
2359 | adev->mp1_state, r); | |
2360 | return r; | |
a3a09142 AD |
2361 | } |
2362 | } | |
b5507c7e AG |
2363 | |
2364 | adev->ip_blocks[i].status.hw = false; | |
d38ceaf9 AD |
2365 | } |
2366 | ||
2367 | return 0; | |
2368 | } | |
2369 | ||
e7854a03 AD |
2370 | /** |
2371 | * amdgpu_device_ip_suspend - run suspend for hardware IPs | |
2372 | * | |
2373 | * @adev: amdgpu_device pointer | |
2374 | * | |
2375 | * Main suspend function for hardware IPs. The list of all the hardware | |
2376 | * IPs that make up the asic is walked, clockgating is disabled and the | |
2377 | * suspend callbacks are run. suspend puts the hardware and software state | |
2378 | * in each IP into a state suitable for suspend. | |
2379 | * Returns 0 on success, negative error code on failure. | |
2380 | */ | |
2381 | int amdgpu_device_ip_suspend(struct amdgpu_device *adev) | |
2382 | { | |
2383 | int r; | |
2384 | ||
e7819644 YT |
2385 | if (amdgpu_sriov_vf(adev)) |
2386 | amdgpu_virt_request_full_gpu(adev, false); | |
2387 | ||
e7854a03 AD |
2388 | r = amdgpu_device_ip_suspend_phase1(adev); |
2389 | if (r) | |
2390 | return r; | |
2391 | r = amdgpu_device_ip_suspend_phase2(adev); | |
2392 | ||
e7819644 YT |
2393 | if (amdgpu_sriov_vf(adev)) |
2394 | amdgpu_virt_release_full_gpu(adev, false); | |
2395 | ||
e7854a03 AD |
2396 | return r; |
2397 | } | |
2398 | ||
06ec9070 | 2399 | static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
2400 | { |
2401 | int i, r; | |
2402 | ||
2cb681b6 ML |
2403 | static enum amd_ip_block_type ip_order[] = { |
2404 | AMD_IP_BLOCK_TYPE_GMC, | |
2405 | AMD_IP_BLOCK_TYPE_COMMON, | |
39186aef | 2406 | AMD_IP_BLOCK_TYPE_PSP, |
2cb681b6 ML |
2407 | AMD_IP_BLOCK_TYPE_IH, |
2408 | }; | |
a90ad3c2 | 2409 | |
2cb681b6 ML |
2410 | for (i = 0; i < ARRAY_SIZE(ip_order); i++) { |
2411 | int j; | |
2412 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 2413 | |
2cb681b6 ML |
2414 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2415 | block = &adev->ip_blocks[j]; | |
2416 | ||
482f0e53 | 2417 | block->status.hw = false; |
2cb681b6 ML |
2418 | if (block->version->type != ip_order[i] || |
2419 | !block->status.valid) | |
2420 | continue; | |
2421 | ||
2422 | r = block->version->funcs->hw_init(adev); | |
0aaeefcc | 2423 | DRM_INFO("RE-INIT-early: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); |
c41d1cf6 ML |
2424 | if (r) |
2425 | return r; | |
482f0e53 | 2426 | block->status.hw = true; |
a90ad3c2 ML |
2427 | } |
2428 | } | |
2429 | ||
2430 | return 0; | |
2431 | } | |
2432 | ||
06ec9070 | 2433 | static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
2434 | { |
2435 | int i, r; | |
2436 | ||
2cb681b6 ML |
2437 | static enum amd_ip_block_type ip_order[] = { |
2438 | AMD_IP_BLOCK_TYPE_SMC, | |
2439 | AMD_IP_BLOCK_TYPE_DCE, | |
2440 | AMD_IP_BLOCK_TYPE_GFX, | |
2441 | AMD_IP_BLOCK_TYPE_SDMA, | |
257deb8c FM |
2442 | AMD_IP_BLOCK_TYPE_UVD, |
2443 | AMD_IP_BLOCK_TYPE_VCE | |
2cb681b6 | 2444 | }; |
a90ad3c2 | 2445 | |
2cb681b6 ML |
2446 | for (i = 0; i < ARRAY_SIZE(ip_order); i++) { |
2447 | int j; | |
2448 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 2449 | |
2cb681b6 ML |
2450 | for (j = 0; j < adev->num_ip_blocks; j++) { |
2451 | block = &adev->ip_blocks[j]; | |
2452 | ||
2453 | if (block->version->type != ip_order[i] || | |
482f0e53 ML |
2454 | !block->status.valid || |
2455 | block->status.hw) | |
2cb681b6 ML |
2456 | continue; |
2457 | ||
2458 | r = block->version->funcs->hw_init(adev); | |
0aaeefcc | 2459 | DRM_INFO("RE-INIT-late: %s %s\n", block->version->funcs->name, r?"failed":"succeeded"); |
c41d1cf6 ML |
2460 | if (r) |
2461 | return r; | |
482f0e53 | 2462 | block->status.hw = true; |
a90ad3c2 ML |
2463 | } |
2464 | } | |
2465 | ||
2466 | return 0; | |
2467 | } | |
2468 | ||
e3ecdffa AD |
2469 | /** |
2470 | * amdgpu_device_ip_resume_phase1 - run resume for hardware IPs | |
2471 | * | |
2472 | * @adev: amdgpu_device pointer | |
2473 | * | |
2474 | * First resume function for hardware IPs. The list of all the hardware | |
2475 | * IPs that make up the asic is walked and the resume callbacks are run for | |
2476 | * COMMON, GMC, and IH. resume puts the hardware into a functional state | |
2477 | * after a suspend and updates the software state as necessary. This | |
2478 | * function is also used for restoring the GPU after a GPU reset. | |
2479 | * Returns 0 on success, negative error code on failure. | |
2480 | */ | |
06ec9070 | 2481 | static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev) |
d38ceaf9 AD |
2482 | { |
2483 | int i, r; | |
2484 | ||
a90ad3c2 | 2485 | for (i = 0; i < adev->num_ip_blocks; i++) { |
482f0e53 | 2486 | if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw) |
a90ad3c2 | 2487 | continue; |
a90ad3c2 | 2488 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
e3ecdffa AD |
2489 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || |
2490 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) { | |
482f0e53 | 2491 | |
fcf0649f CZ |
2492 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
2493 | if (r) { | |
2494 | DRM_ERROR("resume of IP block <%s> failed %d\n", | |
2495 | adev->ip_blocks[i].version->funcs->name, r); | |
2496 | return r; | |
2497 | } | |
482f0e53 | 2498 | adev->ip_blocks[i].status.hw = true; |
a90ad3c2 ML |
2499 | } |
2500 | } | |
2501 | ||
2502 | return 0; | |
2503 | } | |
2504 | ||
e3ecdffa AD |
2505 | /** |
2506 | * amdgpu_device_ip_resume_phase2 - run resume for hardware IPs | |
2507 | * | |
2508 | * @adev: amdgpu_device pointer | |
2509 | * | |
2510 | * First resume function for hardware IPs. The list of all the hardware | |
2511 | * IPs that make up the asic is walked and the resume callbacks are run for | |
2512 | * all blocks except COMMON, GMC, and IH. resume puts the hardware into a | |
2513 | * functional state after a suspend and updates the software state as | |
2514 | * necessary. This function is also used for restoring the GPU after a GPU | |
2515 | * reset. | |
2516 | * Returns 0 on success, negative error code on failure. | |
2517 | */ | |
06ec9070 | 2518 | static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev) |
d38ceaf9 AD |
2519 | { |
2520 | int i, r; | |
2521 | ||
2522 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
482f0e53 | 2523 | if (!adev->ip_blocks[i].status.valid || adev->ip_blocks[i].status.hw) |
d38ceaf9 | 2524 | continue; |
fcf0649f | 2525 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
e3ecdffa | 2526 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || |
7a3e0bb2 RZ |
2527 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH || |
2528 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) | |
fcf0649f | 2529 | continue; |
a1255107 | 2530 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
2c1a2784 | 2531 | if (r) { |
a1255107 AD |
2532 | DRM_ERROR("resume of IP block <%s> failed %d\n", |
2533 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 2534 | return r; |
2c1a2784 | 2535 | } |
482f0e53 | 2536 | adev->ip_blocks[i].status.hw = true; |
d38ceaf9 AD |
2537 | } |
2538 | ||
2539 | return 0; | |
2540 | } | |
2541 | ||
e3ecdffa AD |
2542 | /** |
2543 | * amdgpu_device_ip_resume - run resume for hardware IPs | |
2544 | * | |
2545 | * @adev: amdgpu_device pointer | |
2546 | * | |
2547 | * Main resume function for hardware IPs. The hardware IPs | |
2548 | * are split into two resume functions because they are | |
2549 | * are also used in in recovering from a GPU reset and some additional | |
2550 | * steps need to be take between them. In this case (S3/S4) they are | |
2551 | * run sequentially. | |
2552 | * Returns 0 on success, negative error code on failure. | |
2553 | */ | |
06ec9070 | 2554 | static int amdgpu_device_ip_resume(struct amdgpu_device *adev) |
fcf0649f CZ |
2555 | { |
2556 | int r; | |
2557 | ||
06ec9070 | 2558 | r = amdgpu_device_ip_resume_phase1(adev); |
fcf0649f CZ |
2559 | if (r) |
2560 | return r; | |
7a3e0bb2 RZ |
2561 | |
2562 | r = amdgpu_device_fw_loading(adev); | |
2563 | if (r) | |
2564 | return r; | |
2565 | ||
06ec9070 | 2566 | r = amdgpu_device_ip_resume_phase2(adev); |
fcf0649f CZ |
2567 | |
2568 | return r; | |
2569 | } | |
2570 | ||
e3ecdffa AD |
2571 | /** |
2572 | * amdgpu_device_detect_sriov_bios - determine if the board supports SR-IOV | |
2573 | * | |
2574 | * @adev: amdgpu_device pointer | |
2575 | * | |
2576 | * Query the VBIOS data tables to determine if the board supports SR-IOV. | |
2577 | */ | |
4e99a44e | 2578 | static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev) |
048765ad | 2579 | { |
6867e1b5 ML |
2580 | if (amdgpu_sriov_vf(adev)) { |
2581 | if (adev->is_atom_fw) { | |
2582 | if (amdgpu_atomfirmware_gpu_supports_virtualization(adev)) | |
2583 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; | |
2584 | } else { | |
2585 | if (amdgpu_atombios_has_gpu_virtualization_table(adev)) | |
2586 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; | |
2587 | } | |
2588 | ||
2589 | if (!(adev->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS)) | |
2590 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0); | |
a5bde2f9 | 2591 | } |
048765ad AR |
2592 | } |
2593 | ||
e3ecdffa AD |
2594 | /** |
2595 | * amdgpu_device_asic_has_dc_support - determine if DC supports the asic | |
2596 | * | |
2597 | * @asic_type: AMD asic type | |
2598 | * | |
2599 | * Check if there is DC (new modesetting infrastructre) support for an asic. | |
2600 | * returns true if DC has support, false if not. | |
2601 | */ | |
4562236b HW |
2602 | bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type) |
2603 | { | |
2604 | switch (asic_type) { | |
2605 | #if defined(CONFIG_DRM_AMD_DC) | |
2606 | case CHIP_BONAIRE: | |
0d6fbccb | 2607 | case CHIP_KAVERI: |
367e6687 AD |
2608 | case CHIP_KABINI: |
2609 | case CHIP_MULLINS: | |
d9fda248 HW |
2610 | /* |
2611 | * We have systems in the wild with these ASICs that require | |
2612 | * LVDS and VGA support which is not supported with DC. | |
2613 | * | |
2614 | * Fallback to the non-DC driver here by default so as not to | |
2615 | * cause regressions. | |
2616 | */ | |
2617 | return amdgpu_dc > 0; | |
2618 | case CHIP_HAWAII: | |
4562236b HW |
2619 | case CHIP_CARRIZO: |
2620 | case CHIP_STONEY: | |
4562236b | 2621 | case CHIP_POLARIS10: |
675fd32b | 2622 | case CHIP_POLARIS11: |
2c8ad2d5 | 2623 | case CHIP_POLARIS12: |
675fd32b | 2624 | case CHIP_VEGAM: |
4562236b HW |
2625 | case CHIP_TONGA: |
2626 | case CHIP_FIJI: | |
42f8ffa1 | 2627 | case CHIP_VEGA10: |
dca7b401 | 2628 | case CHIP_VEGA12: |
c6034aa2 | 2629 | case CHIP_VEGA20: |
b86a1aa3 | 2630 | #if defined(CONFIG_DRM_AMD_DC_DCN) |
fd187853 | 2631 | case CHIP_RAVEN: |
b4f199c7 | 2632 | case CHIP_NAVI10: |
8fceceb6 | 2633 | case CHIP_NAVI14: |
078655d9 | 2634 | case CHIP_NAVI12: |
e1c14c43 | 2635 | case CHIP_RENOIR: |
42f8ffa1 | 2636 | #endif |
fd187853 | 2637 | return amdgpu_dc != 0; |
4562236b HW |
2638 | #endif |
2639 | default: | |
93b09a9a SS |
2640 | if (amdgpu_dc > 0) |
2641 | DRM_INFO("Display Core has been requested via kernel parameter " | |
2642 | "but isn't supported by ASIC, ignoring\n"); | |
4562236b HW |
2643 | return false; |
2644 | } | |
2645 | } | |
2646 | ||
2647 | /** | |
2648 | * amdgpu_device_has_dc_support - check if dc is supported | |
2649 | * | |
2650 | * @adev: amdgpu_device_pointer | |
2651 | * | |
2652 | * Returns true for supported, false for not supported | |
2653 | */ | |
2654 | bool amdgpu_device_has_dc_support(struct amdgpu_device *adev) | |
2655 | { | |
2555039d XY |
2656 | if (amdgpu_sriov_vf(adev)) |
2657 | return false; | |
2658 | ||
4562236b HW |
2659 | return amdgpu_device_asic_has_dc_support(adev->asic_type); |
2660 | } | |
2661 | ||
d4535e2c AG |
2662 | |
2663 | static void amdgpu_device_xgmi_reset_func(struct work_struct *__work) | |
2664 | { | |
2665 | struct amdgpu_device *adev = | |
2666 | container_of(__work, struct amdgpu_device, xgmi_reset_work); | |
2667 | ||
ce316fa5 LM |
2668 | if (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) |
2669 | adev->asic_reset_res = (adev->in_baco == false) ? | |
2670 | amdgpu_device_baco_enter(adev->ddev) : | |
2671 | amdgpu_device_baco_exit(adev->ddev); | |
2672 | else | |
2673 | adev->asic_reset_res = amdgpu_asic_reset(adev); | |
2674 | ||
d4535e2c | 2675 | if (adev->asic_reset_res) |
fed184e9 | 2676 | DRM_WARN("ASIC reset failed with error, %d for drm dev, %s", |
d4535e2c AG |
2677 | adev->asic_reset_res, adev->ddev->unique); |
2678 | } | |
2679 | ||
71f98027 AD |
2680 | static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev) |
2681 | { | |
2682 | char *input = amdgpu_lockup_timeout; | |
2683 | char *timeout_setting = NULL; | |
2684 | int index = 0; | |
2685 | long timeout; | |
2686 | int ret = 0; | |
2687 | ||
2688 | /* | |
2689 | * By default timeout for non compute jobs is 10000. | |
2690 | * And there is no timeout enforced on compute jobs. | |
2691 | * In SR-IOV or passthrough mode, timeout for compute | |
2692 | * jobs are 10000 by default. | |
2693 | */ | |
2694 | adev->gfx_timeout = msecs_to_jiffies(10000); | |
2695 | adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; | |
2696 | if (amdgpu_sriov_vf(adev) || amdgpu_passthrough(adev)) | |
2697 | adev->compute_timeout = adev->gfx_timeout; | |
2698 | else | |
2699 | adev->compute_timeout = MAX_SCHEDULE_TIMEOUT; | |
2700 | ||
f440ff44 | 2701 | if (strnlen(input, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH)) { |
71f98027 | 2702 | while ((timeout_setting = strsep(&input, ",")) && |
f440ff44 | 2703 | strnlen(timeout_setting, AMDGPU_MAX_TIMEOUT_PARAM_LENGTH)) { |
71f98027 AD |
2704 | ret = kstrtol(timeout_setting, 0, &timeout); |
2705 | if (ret) | |
2706 | return ret; | |
2707 | ||
2708 | if (timeout == 0) { | |
2709 | index++; | |
2710 | continue; | |
2711 | } else if (timeout < 0) { | |
2712 | timeout = MAX_SCHEDULE_TIMEOUT; | |
2713 | } else { | |
2714 | timeout = msecs_to_jiffies(timeout); | |
2715 | } | |
2716 | ||
2717 | switch (index++) { | |
2718 | case 0: | |
2719 | adev->gfx_timeout = timeout; | |
2720 | break; | |
2721 | case 1: | |
2722 | adev->compute_timeout = timeout; | |
2723 | break; | |
2724 | case 2: | |
2725 | adev->sdma_timeout = timeout; | |
2726 | break; | |
2727 | case 3: | |
2728 | adev->video_timeout = timeout; | |
2729 | break; | |
2730 | default: | |
2731 | break; | |
2732 | } | |
2733 | } | |
2734 | /* | |
2735 | * There is only one value specified and | |
2736 | * it should apply to all non-compute jobs. | |
2737 | */ | |
bcccee89 | 2738 | if (index == 1) { |
71f98027 | 2739 | adev->sdma_timeout = adev->video_timeout = adev->gfx_timeout; |
bcccee89 ED |
2740 | if (amdgpu_sriov_vf(adev) || amdgpu_passthrough(adev)) |
2741 | adev->compute_timeout = adev->gfx_timeout; | |
2742 | } | |
71f98027 AD |
2743 | } |
2744 | ||
2745 | return ret; | |
2746 | } | |
d4535e2c | 2747 | |
d38ceaf9 AD |
2748 | /** |
2749 | * amdgpu_device_init - initialize the driver | |
2750 | * | |
2751 | * @adev: amdgpu_device pointer | |
87e3f136 | 2752 | * @ddev: drm dev pointer |
d38ceaf9 AD |
2753 | * @pdev: pci dev pointer |
2754 | * @flags: driver flags | |
2755 | * | |
2756 | * Initializes the driver info and hw (all asics). | |
2757 | * Returns 0 for success or an error on failure. | |
2758 | * Called at driver startup. | |
2759 | */ | |
2760 | int amdgpu_device_init(struct amdgpu_device *adev, | |
2761 | struct drm_device *ddev, | |
2762 | struct pci_dev *pdev, | |
2763 | uint32_t flags) | |
2764 | { | |
2765 | int r, i; | |
3840c5bc | 2766 | bool boco = false; |
95844d20 | 2767 | u32 max_MBps; |
d38ceaf9 AD |
2768 | |
2769 | adev->shutdown = false; | |
2770 | adev->dev = &pdev->dev; | |
2771 | adev->ddev = ddev; | |
2772 | adev->pdev = pdev; | |
2773 | adev->flags = flags; | |
4e66d7d2 YZ |
2774 | |
2775 | if (amdgpu_force_asic_type >= 0 && amdgpu_force_asic_type < CHIP_LAST) | |
2776 | adev->asic_type = amdgpu_force_asic_type; | |
2777 | else | |
2778 | adev->asic_type = flags & AMD_ASIC_MASK; | |
2779 | ||
d38ceaf9 | 2780 | adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT; |
593aa2d2 SL |
2781 | if (amdgpu_emu_mode == 1) |
2782 | adev->usec_timeout *= 2; | |
770d13b1 | 2783 | adev->gmc.gart_size = 512 * 1024 * 1024; |
d38ceaf9 AD |
2784 | adev->accel_working = false; |
2785 | adev->num_rings = 0; | |
2786 | adev->mman.buffer_funcs = NULL; | |
2787 | adev->mman.buffer_funcs_ring = NULL; | |
2788 | adev->vm_manager.vm_pte_funcs = NULL; | |
0c88b430 | 2789 | adev->vm_manager.vm_pte_num_scheds = 0; |
132f34e4 | 2790 | adev->gmc.gmc_funcs = NULL; |
f54d1867 | 2791 | adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS); |
b8866c26 | 2792 | bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); |
d38ceaf9 AD |
2793 | |
2794 | adev->smc_rreg = &amdgpu_invalid_rreg; | |
2795 | adev->smc_wreg = &amdgpu_invalid_wreg; | |
2796 | adev->pcie_rreg = &amdgpu_invalid_rreg; | |
2797 | adev->pcie_wreg = &amdgpu_invalid_wreg; | |
36b9a952 HR |
2798 | adev->pciep_rreg = &amdgpu_invalid_rreg; |
2799 | adev->pciep_wreg = &amdgpu_invalid_wreg; | |
4fa1c6a6 TZ |
2800 | adev->pcie_rreg64 = &amdgpu_invalid_rreg64; |
2801 | adev->pcie_wreg64 = &amdgpu_invalid_wreg64; | |
d38ceaf9 AD |
2802 | adev->uvd_ctx_rreg = &amdgpu_invalid_rreg; |
2803 | adev->uvd_ctx_wreg = &amdgpu_invalid_wreg; | |
2804 | adev->didt_rreg = &amdgpu_invalid_rreg; | |
2805 | adev->didt_wreg = &amdgpu_invalid_wreg; | |
ccdbb20a RZ |
2806 | adev->gc_cac_rreg = &amdgpu_invalid_rreg; |
2807 | adev->gc_cac_wreg = &amdgpu_invalid_wreg; | |
d38ceaf9 AD |
2808 | adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg; |
2809 | adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg; | |
2810 | ||
3e39ab90 AD |
2811 | DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n", |
2812 | amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device, | |
2813 | pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision); | |
d38ceaf9 AD |
2814 | |
2815 | /* mutex initialization are all done here so we | |
2816 | * can recall function without having locking issues */ | |
d38ceaf9 | 2817 | atomic_set(&adev->irq.ih.lock, 0); |
0e5ca0d1 | 2818 | mutex_init(&adev->firmware.mutex); |
d38ceaf9 AD |
2819 | mutex_init(&adev->pm.mutex); |
2820 | mutex_init(&adev->gfx.gpu_clock_mutex); | |
2821 | mutex_init(&adev->srbm_mutex); | |
b8866c26 | 2822 | mutex_init(&adev->gfx.pipe_reserve_mutex); |
d23ee13f | 2823 | mutex_init(&adev->gfx.gfx_off_mutex); |
d38ceaf9 | 2824 | mutex_init(&adev->grbm_idx_mutex); |
d38ceaf9 | 2825 | mutex_init(&adev->mn_lock); |
e23b74aa | 2826 | mutex_init(&adev->virt.vf_errors.lock); |
d38ceaf9 | 2827 | hash_init(adev->mn_hash); |
13a752e3 | 2828 | mutex_init(&adev->lock_reset); |
32eaeae0 | 2829 | mutex_init(&adev->psp.mutex); |
d38ceaf9 | 2830 | |
912dfc84 EQ |
2831 | r = amdgpu_device_check_arguments(adev); |
2832 | if (r) | |
2833 | return r; | |
d38ceaf9 | 2834 | |
d38ceaf9 AD |
2835 | spin_lock_init(&adev->mmio_idx_lock); |
2836 | spin_lock_init(&adev->smc_idx_lock); | |
2837 | spin_lock_init(&adev->pcie_idx_lock); | |
2838 | spin_lock_init(&adev->uvd_ctx_idx_lock); | |
2839 | spin_lock_init(&adev->didt_idx_lock); | |
ccdbb20a | 2840 | spin_lock_init(&adev->gc_cac_idx_lock); |
16abb5d2 | 2841 | spin_lock_init(&adev->se_cac_idx_lock); |
d38ceaf9 | 2842 | spin_lock_init(&adev->audio_endpt_idx_lock); |
95844d20 | 2843 | spin_lock_init(&adev->mm_stats.lock); |
d38ceaf9 | 2844 | |
0c4e7fa5 CZ |
2845 | INIT_LIST_HEAD(&adev->shadow_list); |
2846 | mutex_init(&adev->shadow_list_lock); | |
2847 | ||
795f2813 AR |
2848 | INIT_LIST_HEAD(&adev->ring_lru_list); |
2849 | spin_lock_init(&adev->ring_lru_list_lock); | |
2850 | ||
beff74bc AD |
2851 | INIT_DELAYED_WORK(&adev->delayed_init_work, |
2852 | amdgpu_device_delayed_init_work_handler); | |
1e317b99 RZ |
2853 | INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work, |
2854 | amdgpu_device_delay_enable_gfx_off); | |
2dc80b00 | 2855 | |
d4535e2c AG |
2856 | INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func); |
2857 | ||
d23ee13f | 2858 | adev->gfx.gfx_off_req_count = 1; |
b1ddf548 RZ |
2859 | adev->pm.ac_power = power_supply_is_system_supplied() > 0 ? true : false; |
2860 | ||
0fa49558 AX |
2861 | /* Registers mapping */ |
2862 | /* TODO: block userspace mapping of io register */ | |
da69c161 KW |
2863 | if (adev->asic_type >= CHIP_BONAIRE) { |
2864 | adev->rmmio_base = pci_resource_start(adev->pdev, 5); | |
2865 | adev->rmmio_size = pci_resource_len(adev->pdev, 5); | |
2866 | } else { | |
2867 | adev->rmmio_base = pci_resource_start(adev->pdev, 2); | |
2868 | adev->rmmio_size = pci_resource_len(adev->pdev, 2); | |
2869 | } | |
d38ceaf9 | 2870 | |
d38ceaf9 AD |
2871 | adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size); |
2872 | if (adev->rmmio == NULL) { | |
2873 | return -ENOMEM; | |
2874 | } | |
2875 | DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base); | |
2876 | DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size); | |
2877 | ||
d38ceaf9 AD |
2878 | /* io port mapping */ |
2879 | for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { | |
2880 | if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) { | |
2881 | adev->rio_mem_size = pci_resource_len(adev->pdev, i); | |
2882 | adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size); | |
2883 | break; | |
2884 | } | |
2885 | } | |
2886 | if (adev->rio_mem == NULL) | |
b64a18c5 | 2887 | DRM_INFO("PCI I/O BAR is not found.\n"); |
d38ceaf9 | 2888 | |
b2109d8e JX |
2889 | /* enable PCIE atomic ops */ |
2890 | r = pci_enable_atomic_ops_to_root(adev->pdev, | |
2891 | PCI_EXP_DEVCAP2_ATOMIC_COMP32 | | |
2892 | PCI_EXP_DEVCAP2_ATOMIC_COMP64); | |
2893 | if (r) { | |
2894 | adev->have_atomics_support = false; | |
2895 | DRM_INFO("PCIE atomic ops is not supported\n"); | |
2896 | } else { | |
2897 | adev->have_atomics_support = true; | |
2898 | } | |
2899 | ||
5494d864 AD |
2900 | amdgpu_device_get_pcie_info(adev); |
2901 | ||
b239c017 JX |
2902 | if (amdgpu_mcbp) |
2903 | DRM_INFO("MCBP is enabled\n"); | |
2904 | ||
5f84cc63 JX |
2905 | if (amdgpu_mes && adev->asic_type >= CHIP_NAVI10) |
2906 | adev->enable_mes = true; | |
2907 | ||
f54eeab4 | 2908 | if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10) { |
a190d1c7 XY |
2909 | r = amdgpu_discovery_init(adev); |
2910 | if (r) { | |
2911 | dev_err(adev->dev, "amdgpu_discovery_init failed\n"); | |
2912 | return r; | |
2913 | } | |
2914 | } | |
2915 | ||
d38ceaf9 | 2916 | /* early init functions */ |
06ec9070 | 2917 | r = amdgpu_device_ip_early_init(adev); |
d38ceaf9 AD |
2918 | if (r) |
2919 | return r; | |
2920 | ||
df99ac0f JZ |
2921 | r = amdgpu_device_get_job_timeout_settings(adev); |
2922 | if (r) { | |
2923 | dev_err(adev->dev, "invalid lockup_timeout parameter syntax\n"); | |
2924 | return r; | |
2925 | } | |
2926 | ||
6585661d OZ |
2927 | /* doorbell bar mapping and doorbell index init*/ |
2928 | amdgpu_device_doorbell_init(adev); | |
2929 | ||
d38ceaf9 AD |
2930 | /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */ |
2931 | /* this will fail for cards that aren't VGA class devices, just | |
2932 | * ignore it */ | |
06ec9070 | 2933 | vga_client_register(adev->pdev, adev, NULL, amdgpu_device_vga_set_decode); |
d38ceaf9 | 2934 | |
31af062a | 2935 | if (amdgpu_device_supports_boco(ddev)) |
3840c5bc AD |
2936 | boco = true; |
2937 | if (amdgpu_has_atpx() && | |
2938 | (amdgpu_is_atpx_hybrid() || | |
2939 | amdgpu_has_atpx_dgpu_power_cntl()) && | |
2940 | !pci_is_thunderbolt_attached(adev->pdev)) | |
84c8b22e | 2941 | vga_switcheroo_register_client(adev->pdev, |
3840c5bc AD |
2942 | &amdgpu_switcheroo_ops, boco); |
2943 | if (boco) | |
d38ceaf9 AD |
2944 | vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain); |
2945 | ||
9475a943 SL |
2946 | if (amdgpu_emu_mode == 1) { |
2947 | /* post the asic on emulation mode */ | |
2948 | emu_soc_asic_init(adev); | |
bfca0289 | 2949 | goto fence_driver_init; |
9475a943 | 2950 | } |
bfca0289 | 2951 | |
4e99a44e ML |
2952 | /* detect if we are with an SRIOV vbios */ |
2953 | amdgpu_device_detect_sriov_bios(adev); | |
048765ad | 2954 | |
95e8e59e AD |
2955 | /* check if we need to reset the asic |
2956 | * E.g., driver was not cleanly unloaded previously, etc. | |
2957 | */ | |
f14899fd | 2958 | if (!amdgpu_sriov_vf(adev) && amdgpu_asic_need_reset_on_init(adev)) { |
95e8e59e AD |
2959 | r = amdgpu_asic_reset(adev); |
2960 | if (r) { | |
2961 | dev_err(adev->dev, "asic reset on init failed\n"); | |
2962 | goto failed; | |
2963 | } | |
2964 | } | |
2965 | ||
d38ceaf9 | 2966 | /* Post card if necessary */ |
39c640c0 | 2967 | if (amdgpu_device_need_post(adev)) { |
d38ceaf9 | 2968 | if (!adev->bios) { |
bec86378 | 2969 | dev_err(adev->dev, "no vBIOS found\n"); |
83ba126a AD |
2970 | r = -EINVAL; |
2971 | goto failed; | |
d38ceaf9 | 2972 | } |
bec86378 | 2973 | DRM_INFO("GPU posting now...\n"); |
4e99a44e ML |
2974 | r = amdgpu_atom_asic_init(adev->mode_info.atom_context); |
2975 | if (r) { | |
2976 | dev_err(adev->dev, "gpu post error!\n"); | |
2977 | goto failed; | |
2978 | } | |
d38ceaf9 AD |
2979 | } |
2980 | ||
88b64e95 AD |
2981 | if (adev->is_atom_fw) { |
2982 | /* Initialize clocks */ | |
2983 | r = amdgpu_atomfirmware_get_clock_info(adev); | |
2984 | if (r) { | |
2985 | dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n"); | |
e23b74aa | 2986 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); |
88b64e95 AD |
2987 | goto failed; |
2988 | } | |
2989 | } else { | |
a5bde2f9 AD |
2990 | /* Initialize clocks */ |
2991 | r = amdgpu_atombios_get_clock_info(adev); | |
2992 | if (r) { | |
2993 | dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n"); | |
e23b74aa | 2994 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); |
89041940 | 2995 | goto failed; |
a5bde2f9 AD |
2996 | } |
2997 | /* init i2c buses */ | |
4562236b HW |
2998 | if (!amdgpu_device_has_dc_support(adev)) |
2999 | amdgpu_atombios_i2c_init(adev); | |
2c1a2784 | 3000 | } |
d38ceaf9 | 3001 | |
bfca0289 | 3002 | fence_driver_init: |
d38ceaf9 AD |
3003 | /* Fence driver */ |
3004 | r = amdgpu_fence_driver_init(adev); | |
2c1a2784 AD |
3005 | if (r) { |
3006 | dev_err(adev->dev, "amdgpu_fence_driver_init failed\n"); | |
e23b74aa | 3007 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0); |
83ba126a | 3008 | goto failed; |
2c1a2784 | 3009 | } |
d38ceaf9 AD |
3010 | |
3011 | /* init the mode config */ | |
3012 | drm_mode_config_init(adev->ddev); | |
3013 | ||
06ec9070 | 3014 | r = amdgpu_device_ip_init(adev); |
d38ceaf9 | 3015 | if (r) { |
8840a387 | 3016 | /* failed in exclusive mode due to timeout */ |
3017 | if (amdgpu_sriov_vf(adev) && | |
3018 | !amdgpu_sriov_runtime(adev) && | |
3019 | amdgpu_virt_mmio_blocked(adev) && | |
3020 | !amdgpu_virt_wait_reset(adev)) { | |
3021 | dev_err(adev->dev, "VF exclusive mode timeout\n"); | |
1daee8b4 PD |
3022 | /* Don't send request since VF is inactive. */ |
3023 | adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME; | |
3024 | adev->virt.ops = NULL; | |
8840a387 | 3025 | r = -EAGAIN; |
3026 | goto failed; | |
3027 | } | |
06ec9070 | 3028 | dev_err(adev->dev, "amdgpu_device_ip_init failed\n"); |
e23b74aa | 3029 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0); |
83ba126a | 3030 | goto failed; |
d38ceaf9 AD |
3031 | } |
3032 | ||
d7f72fe4 YZ |
3033 | DRM_DEBUG("SE %d, SH per SE %d, CU per SH %d, active_cu_number %d\n", |
3034 | adev->gfx.config.max_shader_engines, | |
3035 | adev->gfx.config.max_sh_per_se, | |
3036 | adev->gfx.config.max_cu_per_sh, | |
3037 | adev->gfx.cu_info.number); | |
3038 | ||
f880799d ND |
3039 | amdgpu_ctx_init_sched(adev); |
3040 | ||
d38ceaf9 AD |
3041 | adev->accel_working = true; |
3042 | ||
e59c0205 AX |
3043 | amdgpu_vm_check_compute_bug(adev); |
3044 | ||
95844d20 MO |
3045 | /* Initialize the buffer migration limit. */ |
3046 | if (amdgpu_moverate >= 0) | |
3047 | max_MBps = amdgpu_moverate; | |
3048 | else | |
3049 | max_MBps = 8; /* Allow 8 MB/s. */ | |
3050 | /* Get a log2 for easy divisions. */ | |
3051 | adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps)); | |
3052 | ||
9bc92b9c ML |
3053 | amdgpu_fbdev_init(adev); |
3054 | ||
d2f52ac8 | 3055 | r = amdgpu_pm_sysfs_init(adev); |
7c868b59 YT |
3056 | if (r) { |
3057 | adev->pm_sysfs_en = false; | |
d2f52ac8 | 3058 | DRM_ERROR("registering pm debugfs failed (%d).\n", r); |
7c868b59 YT |
3059 | } else |
3060 | adev->pm_sysfs_en = true; | |
d2f52ac8 | 3061 | |
5bb23532 | 3062 | r = amdgpu_ucode_sysfs_init(adev); |
7c868b59 YT |
3063 | if (r) { |
3064 | adev->ucode_sysfs_en = false; | |
5bb23532 | 3065 | DRM_ERROR("Creating firmware sysfs failed (%d).\n", r); |
7c868b59 YT |
3066 | } else |
3067 | adev->ucode_sysfs_en = true; | |
5bb23532 | 3068 | |
75758255 | 3069 | r = amdgpu_debugfs_gem_init(adev); |
3f14e623 | 3070 | if (r) |
d38ceaf9 | 3071 | DRM_ERROR("registering gem debugfs failed (%d).\n", r); |
d38ceaf9 AD |
3072 | |
3073 | r = amdgpu_debugfs_regs_init(adev); | |
3f14e623 | 3074 | if (r) |
d38ceaf9 | 3075 | DRM_ERROR("registering register debugfs failed (%d).\n", r); |
d38ceaf9 | 3076 | |
50ab2533 | 3077 | r = amdgpu_debugfs_firmware_init(adev); |
3f14e623 | 3078 | if (r) |
50ab2533 | 3079 | DRM_ERROR("registering firmware debugfs failed (%d).\n", r); |
50ab2533 | 3080 | |
763efb6c | 3081 | r = amdgpu_debugfs_init(adev); |
db95e218 | 3082 | if (r) |
763efb6c | 3083 | DRM_ERROR("Creating debugfs files failed (%d).\n", r); |
db95e218 | 3084 | |
d38ceaf9 AD |
3085 | if ((amdgpu_testing & 1)) { |
3086 | if (adev->accel_working) | |
3087 | amdgpu_test_moves(adev); | |
3088 | else | |
3089 | DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n"); | |
3090 | } | |
d38ceaf9 AD |
3091 | if (amdgpu_benchmarking) { |
3092 | if (adev->accel_working) | |
3093 | amdgpu_benchmark(adev, amdgpu_benchmarking); | |
3094 | else | |
3095 | DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n"); | |
3096 | } | |
3097 | ||
b0adca4d EQ |
3098 | /* |
3099 | * Register gpu instance before amdgpu_device_enable_mgpu_fan_boost. | |
3100 | * Otherwise the mgpu fan boost feature will be skipped due to the | |
3101 | * gpu instance is counted less. | |
3102 | */ | |
3103 | amdgpu_register_gpu_instance(adev); | |
3104 | ||
d38ceaf9 AD |
3105 | /* enable clockgating, etc. after ib tests, etc. since some blocks require |
3106 | * explicit gating rather than handling it automatically. | |
3107 | */ | |
06ec9070 | 3108 | r = amdgpu_device_ip_late_init(adev); |
2c1a2784 | 3109 | if (r) { |
06ec9070 | 3110 | dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n"); |
e23b74aa | 3111 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r); |
83ba126a | 3112 | goto failed; |
2c1a2784 | 3113 | } |
d38ceaf9 | 3114 | |
108c6a63 | 3115 | /* must succeed. */ |
511fdbc3 | 3116 | amdgpu_ras_resume(adev); |
108c6a63 | 3117 | |
beff74bc AD |
3118 | queue_delayed_work(system_wq, &adev->delayed_init_work, |
3119 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
3120 | ||
dcea6e65 KR |
3121 | r = device_create_file(adev->dev, &dev_attr_pcie_replay_count); |
3122 | if (r) { | |
3123 | dev_err(adev->dev, "Could not create pcie_replay_count"); | |
3124 | return r; | |
3125 | } | |
108c6a63 | 3126 | |
d155bef0 AB |
3127 | if (IS_ENABLED(CONFIG_PERF_EVENTS)) |
3128 | r = amdgpu_pmu_init(adev); | |
9c7c85f7 JK |
3129 | if (r) |
3130 | dev_err(adev->dev, "amdgpu_pmu_init failed\n"); | |
3131 | ||
d38ceaf9 | 3132 | return 0; |
83ba126a AD |
3133 | |
3134 | failed: | |
89041940 | 3135 | amdgpu_vf_error_trans_all(adev); |
3840c5bc | 3136 | if (boco) |
83ba126a | 3137 | vga_switcheroo_fini_domain_pm_ops(adev->dev); |
8840a387 | 3138 | |
83ba126a | 3139 | return r; |
d38ceaf9 AD |
3140 | } |
3141 | ||
d38ceaf9 AD |
3142 | /** |
3143 | * amdgpu_device_fini - tear down the driver | |
3144 | * | |
3145 | * @adev: amdgpu_device pointer | |
3146 | * | |
3147 | * Tear down the driver info (all asics). | |
3148 | * Called at driver shutdown. | |
3149 | */ | |
3150 | void amdgpu_device_fini(struct amdgpu_device *adev) | |
3151 | { | |
3152 | int r; | |
3153 | ||
3154 | DRM_INFO("amdgpu: finishing device.\n"); | |
9f875167 | 3155 | flush_delayed_work(&adev->delayed_init_work); |
d0d13fe8 | 3156 | adev->shutdown = true; |
9f875167 | 3157 | |
e5b03032 ML |
3158 | /* disable all interrupts */ |
3159 | amdgpu_irq_disable_all(adev); | |
ff97cba8 ML |
3160 | if (adev->mode_info.mode_config_initialized){ |
3161 | if (!amdgpu_device_has_dc_support(adev)) | |
c2d88e06 | 3162 | drm_helper_force_disable_all(adev->ddev); |
ff97cba8 ML |
3163 | else |
3164 | drm_atomic_helper_shutdown(adev->ddev); | |
3165 | } | |
d38ceaf9 | 3166 | amdgpu_fence_driver_fini(adev); |
7c868b59 YT |
3167 | if (adev->pm_sysfs_en) |
3168 | amdgpu_pm_sysfs_fini(adev); | |
d38ceaf9 | 3169 | amdgpu_fbdev_fini(adev); |
06ec9070 | 3170 | r = amdgpu_device_ip_fini(adev); |
ab4fe3e1 HR |
3171 | if (adev->firmware.gpu_info_fw) { |
3172 | release_firmware(adev->firmware.gpu_info_fw); | |
3173 | adev->firmware.gpu_info_fw = NULL; | |
3174 | } | |
d38ceaf9 AD |
3175 | adev->accel_working = false; |
3176 | /* free i2c buses */ | |
4562236b HW |
3177 | if (!amdgpu_device_has_dc_support(adev)) |
3178 | amdgpu_i2c_fini(adev); | |
bfca0289 SL |
3179 | |
3180 | if (amdgpu_emu_mode != 1) | |
3181 | amdgpu_atombios_fini(adev); | |
3182 | ||
d38ceaf9 AD |
3183 | kfree(adev->bios); |
3184 | adev->bios = NULL; | |
3840c5bc AD |
3185 | if (amdgpu_has_atpx() && |
3186 | (amdgpu_is_atpx_hybrid() || | |
3187 | amdgpu_has_atpx_dgpu_power_cntl()) && | |
3188 | !pci_is_thunderbolt_attached(adev->pdev)) | |
84c8b22e | 3189 | vga_switcheroo_unregister_client(adev->pdev); |
3840c5bc | 3190 | if (amdgpu_device_supports_boco(adev->ddev)) |
83ba126a | 3191 | vga_switcheroo_fini_domain_pm_ops(adev->dev); |
d38ceaf9 AD |
3192 | vga_client_register(adev->pdev, NULL, NULL, NULL); |
3193 | if (adev->rio_mem) | |
3194 | pci_iounmap(adev->pdev, adev->rio_mem); | |
3195 | adev->rio_mem = NULL; | |
3196 | iounmap(adev->rmmio); | |
3197 | adev->rmmio = NULL; | |
06ec9070 | 3198 | amdgpu_device_doorbell_fini(adev); |
e9bc1bf7 | 3199 | |
d38ceaf9 | 3200 | amdgpu_debugfs_regs_cleanup(adev); |
dcea6e65 | 3201 | device_remove_file(adev->dev, &dev_attr_pcie_replay_count); |
7c868b59 YT |
3202 | if (adev->ucode_sysfs_en) |
3203 | amdgpu_ucode_sysfs_fini(adev); | |
d155bef0 AB |
3204 | if (IS_ENABLED(CONFIG_PERF_EVENTS)) |
3205 | amdgpu_pmu_fini(adev); | |
6698a3d0 | 3206 | amdgpu_debugfs_preempt_cleanup(adev); |
f54eeab4 | 3207 | if (amdgpu_discovery && adev->asic_type >= CHIP_NAVI10) |
a190d1c7 | 3208 | amdgpu_discovery_fini(adev); |
d38ceaf9 AD |
3209 | } |
3210 | ||
3211 | ||
3212 | /* | |
3213 | * Suspend & resume. | |
3214 | */ | |
3215 | /** | |
810ddc3a | 3216 | * amdgpu_device_suspend - initiate device suspend |
d38ceaf9 | 3217 | * |
87e3f136 DP |
3218 | * @dev: drm dev pointer |
3219 | * @suspend: suspend state | |
3220 | * @fbcon : notify the fbdev of suspend | |
d38ceaf9 AD |
3221 | * |
3222 | * Puts the hw in the suspend state (all asics). | |
3223 | * Returns 0 for success or an error on failure. | |
3224 | * Called at driver suspend. | |
3225 | */ | |
de185019 | 3226 | int amdgpu_device_suspend(struct drm_device *dev, bool fbcon) |
d38ceaf9 AD |
3227 | { |
3228 | struct amdgpu_device *adev; | |
3229 | struct drm_crtc *crtc; | |
3230 | struct drm_connector *connector; | |
f8d2d39e | 3231 | struct drm_connector_list_iter iter; |
5ceb54c6 | 3232 | int r; |
d38ceaf9 AD |
3233 | |
3234 | if (dev == NULL || dev->dev_private == NULL) { | |
3235 | return -ENODEV; | |
3236 | } | |
3237 | ||
3238 | adev = dev->dev_private; | |
3239 | ||
3240 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
3241 | return 0; | |
3242 | ||
44779b43 | 3243 | adev->in_suspend = true; |
d38ceaf9 AD |
3244 | drm_kms_helper_poll_disable(dev); |
3245 | ||
5f818173 S |
3246 | if (fbcon) |
3247 | amdgpu_fbdev_set_suspend(adev, 1); | |
3248 | ||
beff74bc | 3249 | cancel_delayed_work_sync(&adev->delayed_init_work); |
a5459475 | 3250 | |
4562236b HW |
3251 | if (!amdgpu_device_has_dc_support(adev)) { |
3252 | /* turn off display hw */ | |
3253 | drm_modeset_lock_all(dev); | |
f8d2d39e LP |
3254 | drm_connector_list_iter_begin(dev, &iter); |
3255 | drm_for_each_connector_iter(connector, &iter) | |
3256 | drm_helper_connector_dpms(connector, | |
3257 | DRM_MODE_DPMS_OFF); | |
3258 | drm_connector_list_iter_end(&iter); | |
4562236b | 3259 | drm_modeset_unlock_all(dev); |
fe1053b7 AD |
3260 | /* unpin the front buffers and cursors */ |
3261 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
3262 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); | |
3263 | struct drm_framebuffer *fb = crtc->primary->fb; | |
3264 | struct amdgpu_bo *robj; | |
3265 | ||
91334223 | 3266 | if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) { |
fe1053b7 AD |
3267 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); |
3268 | r = amdgpu_bo_reserve(aobj, true); | |
3269 | if (r == 0) { | |
3270 | amdgpu_bo_unpin(aobj); | |
3271 | amdgpu_bo_unreserve(aobj); | |
3272 | } | |
756e6880 | 3273 | } |
756e6880 | 3274 | |
fe1053b7 AD |
3275 | if (fb == NULL || fb->obj[0] == NULL) { |
3276 | continue; | |
3277 | } | |
3278 | robj = gem_to_amdgpu_bo(fb->obj[0]); | |
3279 | /* don't unpin kernel fb objects */ | |
3280 | if (!amdgpu_fbdev_robj_is_fb(adev, robj)) { | |
3281 | r = amdgpu_bo_reserve(robj, true); | |
3282 | if (r == 0) { | |
3283 | amdgpu_bo_unpin(robj); | |
3284 | amdgpu_bo_unreserve(robj); | |
3285 | } | |
d38ceaf9 AD |
3286 | } |
3287 | } | |
3288 | } | |
fe1053b7 AD |
3289 | |
3290 | amdgpu_amdkfd_suspend(adev); | |
3291 | ||
5e6932fe | 3292 | amdgpu_ras_suspend(adev); |
3293 | ||
fe1053b7 AD |
3294 | r = amdgpu_device_ip_suspend_phase1(adev); |
3295 | ||
d38ceaf9 AD |
3296 | /* evict vram memory */ |
3297 | amdgpu_bo_evict_vram(adev); | |
3298 | ||
5ceb54c6 | 3299 | amdgpu_fence_driver_suspend(adev); |
d38ceaf9 | 3300 | |
fe1053b7 | 3301 | r = amdgpu_device_ip_suspend_phase2(adev); |
d38ceaf9 | 3302 | |
a0a71e49 AD |
3303 | /* evict remaining vram memory |
3304 | * This second call to evict vram is to evict the gart page table | |
3305 | * using the CPU. | |
3306 | */ | |
d38ceaf9 AD |
3307 | amdgpu_bo_evict_vram(adev); |
3308 | ||
d38ceaf9 AD |
3309 | return 0; |
3310 | } | |
3311 | ||
3312 | /** | |
810ddc3a | 3313 | * amdgpu_device_resume - initiate device resume |
d38ceaf9 | 3314 | * |
87e3f136 DP |
3315 | * @dev: drm dev pointer |
3316 | * @resume: resume state | |
3317 | * @fbcon : notify the fbdev of resume | |
d38ceaf9 AD |
3318 | * |
3319 | * Bring the hw back to operating state (all asics). | |
3320 | * Returns 0 for success or an error on failure. | |
3321 | * Called at driver resume. | |
3322 | */ | |
de185019 | 3323 | int amdgpu_device_resume(struct drm_device *dev, bool fbcon) |
d38ceaf9 AD |
3324 | { |
3325 | struct drm_connector *connector; | |
f8d2d39e | 3326 | struct drm_connector_list_iter iter; |
d38ceaf9 | 3327 | struct amdgpu_device *adev = dev->dev_private; |
756e6880 | 3328 | struct drm_crtc *crtc; |
03161a6e | 3329 | int r = 0; |
d38ceaf9 AD |
3330 | |
3331 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
3332 | return 0; | |
3333 | ||
d38ceaf9 | 3334 | /* post card */ |
39c640c0 | 3335 | if (amdgpu_device_need_post(adev)) { |
74b0b157 | 3336 | r = amdgpu_atom_asic_init(adev->mode_info.atom_context); |
3337 | if (r) | |
3338 | DRM_ERROR("amdgpu asic init failed\n"); | |
3339 | } | |
d38ceaf9 | 3340 | |
06ec9070 | 3341 | r = amdgpu_device_ip_resume(adev); |
e6707218 | 3342 | if (r) { |
06ec9070 | 3343 | DRM_ERROR("amdgpu_device_ip_resume failed (%d).\n", r); |
4d3b9ae5 | 3344 | return r; |
e6707218 | 3345 | } |
5ceb54c6 AD |
3346 | amdgpu_fence_driver_resume(adev); |
3347 | ||
d38ceaf9 | 3348 | |
06ec9070 | 3349 | r = amdgpu_device_ip_late_init(adev); |
03161a6e | 3350 | if (r) |
4d3b9ae5 | 3351 | return r; |
d38ceaf9 | 3352 | |
beff74bc AD |
3353 | queue_delayed_work(system_wq, &adev->delayed_init_work, |
3354 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
3355 | ||
fe1053b7 AD |
3356 | if (!amdgpu_device_has_dc_support(adev)) { |
3357 | /* pin cursors */ | |
3358 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
3359 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); | |
3360 | ||
91334223 | 3361 | if (amdgpu_crtc->cursor_bo && !adev->enable_virtual_display) { |
fe1053b7 AD |
3362 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); |
3363 | r = amdgpu_bo_reserve(aobj, true); | |
3364 | if (r == 0) { | |
3365 | r = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM); | |
3366 | if (r != 0) | |
3367 | DRM_ERROR("Failed to pin cursor BO (%d)\n", r); | |
3368 | amdgpu_crtc->cursor_addr = amdgpu_bo_gpu_offset(aobj); | |
3369 | amdgpu_bo_unreserve(aobj); | |
3370 | } | |
756e6880 AD |
3371 | } |
3372 | } | |
3373 | } | |
ba997709 YZ |
3374 | r = amdgpu_amdkfd_resume(adev); |
3375 | if (r) | |
3376 | return r; | |
756e6880 | 3377 | |
96a5d8d4 | 3378 | /* Make sure IB tests flushed */ |
beff74bc | 3379 | flush_delayed_work(&adev->delayed_init_work); |
96a5d8d4 | 3380 | |
d38ceaf9 AD |
3381 | /* blat the mode back in */ |
3382 | if (fbcon) { | |
4562236b HW |
3383 | if (!amdgpu_device_has_dc_support(adev)) { |
3384 | /* pre DCE11 */ | |
3385 | drm_helper_resume_force_mode(dev); | |
3386 | ||
3387 | /* turn on display hw */ | |
3388 | drm_modeset_lock_all(dev); | |
f8d2d39e LP |
3389 | |
3390 | drm_connector_list_iter_begin(dev, &iter); | |
3391 | drm_for_each_connector_iter(connector, &iter) | |
3392 | drm_helper_connector_dpms(connector, | |
3393 | DRM_MODE_DPMS_ON); | |
3394 | drm_connector_list_iter_end(&iter); | |
3395 | ||
4562236b | 3396 | drm_modeset_unlock_all(dev); |
d38ceaf9 | 3397 | } |
4d3b9ae5 | 3398 | amdgpu_fbdev_set_suspend(adev, 0); |
d38ceaf9 AD |
3399 | } |
3400 | ||
3401 | drm_kms_helper_poll_enable(dev); | |
23a1a9e5 | 3402 | |
5e6932fe | 3403 | amdgpu_ras_resume(adev); |
3404 | ||
23a1a9e5 L |
3405 | /* |
3406 | * Most of the connector probing functions try to acquire runtime pm | |
3407 | * refs to ensure that the GPU is powered on when connector polling is | |
3408 | * performed. Since we're calling this from a runtime PM callback, | |
3409 | * trying to acquire rpm refs will cause us to deadlock. | |
3410 | * | |
3411 | * Since we're guaranteed to be holding the rpm lock, it's safe to | |
3412 | * temporarily disable the rpm helpers so this doesn't deadlock us. | |
3413 | */ | |
3414 | #ifdef CONFIG_PM | |
3415 | dev->dev->power.disable_depth++; | |
3416 | #endif | |
4562236b HW |
3417 | if (!amdgpu_device_has_dc_support(adev)) |
3418 | drm_helper_hpd_irq_event(dev); | |
3419 | else | |
3420 | drm_kms_helper_hotplug_event(dev); | |
23a1a9e5 L |
3421 | #ifdef CONFIG_PM |
3422 | dev->dev->power.disable_depth--; | |
3423 | #endif | |
44779b43 RZ |
3424 | adev->in_suspend = false; |
3425 | ||
4d3b9ae5 | 3426 | return 0; |
d38ceaf9 AD |
3427 | } |
3428 | ||
e3ecdffa AD |
3429 | /** |
3430 | * amdgpu_device_ip_check_soft_reset - did soft reset succeed | |
3431 | * | |
3432 | * @adev: amdgpu_device pointer | |
3433 | * | |
3434 | * The list of all the hardware IPs that make up the asic is walked and | |
3435 | * the check_soft_reset callbacks are run. check_soft_reset determines | |
3436 | * if the asic is still hung or not. | |
3437 | * Returns true if any of the IPs are still in a hung state, false if not. | |
3438 | */ | |
06ec9070 | 3439 | static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev) |
63fbf42f CZ |
3440 | { |
3441 | int i; | |
3442 | bool asic_hang = false; | |
3443 | ||
f993d628 ML |
3444 | if (amdgpu_sriov_vf(adev)) |
3445 | return true; | |
3446 | ||
8bc04c29 AD |
3447 | if (amdgpu_asic_need_full_reset(adev)) |
3448 | return true; | |
3449 | ||
63fbf42f | 3450 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 3451 | if (!adev->ip_blocks[i].status.valid) |
63fbf42f | 3452 | continue; |
a1255107 AD |
3453 | if (adev->ip_blocks[i].version->funcs->check_soft_reset) |
3454 | adev->ip_blocks[i].status.hang = | |
3455 | adev->ip_blocks[i].version->funcs->check_soft_reset(adev); | |
3456 | if (adev->ip_blocks[i].status.hang) { | |
3457 | DRM_INFO("IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name); | |
63fbf42f CZ |
3458 | asic_hang = true; |
3459 | } | |
3460 | } | |
3461 | return asic_hang; | |
3462 | } | |
3463 | ||
e3ecdffa AD |
3464 | /** |
3465 | * amdgpu_device_ip_pre_soft_reset - prepare for soft reset | |
3466 | * | |
3467 | * @adev: amdgpu_device pointer | |
3468 | * | |
3469 | * The list of all the hardware IPs that make up the asic is walked and the | |
3470 | * pre_soft_reset callbacks are run if the block is hung. pre_soft_reset | |
3471 | * handles any IP specific hardware or software state changes that are | |
3472 | * necessary for a soft reset to succeed. | |
3473 | * Returns 0 on success, negative error code on failure. | |
3474 | */ | |
06ec9070 | 3475 | static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev) |
d31a501e CZ |
3476 | { |
3477 | int i, r = 0; | |
3478 | ||
3479 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 3480 | if (!adev->ip_blocks[i].status.valid) |
d31a501e | 3481 | continue; |
a1255107 AD |
3482 | if (adev->ip_blocks[i].status.hang && |
3483 | adev->ip_blocks[i].version->funcs->pre_soft_reset) { | |
3484 | r = adev->ip_blocks[i].version->funcs->pre_soft_reset(adev); | |
d31a501e CZ |
3485 | if (r) |
3486 | return r; | |
3487 | } | |
3488 | } | |
3489 | ||
3490 | return 0; | |
3491 | } | |
3492 | ||
e3ecdffa AD |
3493 | /** |
3494 | * amdgpu_device_ip_need_full_reset - check if a full asic reset is needed | |
3495 | * | |
3496 | * @adev: amdgpu_device pointer | |
3497 | * | |
3498 | * Some hardware IPs cannot be soft reset. If they are hung, a full gpu | |
3499 | * reset is necessary to recover. | |
3500 | * Returns true if a full asic reset is required, false if not. | |
3501 | */ | |
06ec9070 | 3502 | static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev) |
35d782fe | 3503 | { |
da146d3b AD |
3504 | int i; |
3505 | ||
8bc04c29 AD |
3506 | if (amdgpu_asic_need_full_reset(adev)) |
3507 | return true; | |
3508 | ||
da146d3b | 3509 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 3510 | if (!adev->ip_blocks[i].status.valid) |
da146d3b | 3511 | continue; |
a1255107 AD |
3512 | if ((adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) || |
3513 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) || | |
3514 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_ACP) || | |
98512bb8 KW |
3515 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) || |
3516 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) { | |
a1255107 | 3517 | if (adev->ip_blocks[i].status.hang) { |
da146d3b AD |
3518 | DRM_INFO("Some block need full reset!\n"); |
3519 | return true; | |
3520 | } | |
3521 | } | |
35d782fe CZ |
3522 | } |
3523 | return false; | |
3524 | } | |
3525 | ||
e3ecdffa AD |
3526 | /** |
3527 | * amdgpu_device_ip_soft_reset - do a soft reset | |
3528 | * | |
3529 | * @adev: amdgpu_device pointer | |
3530 | * | |
3531 | * The list of all the hardware IPs that make up the asic is walked and the | |
3532 | * soft_reset callbacks are run if the block is hung. soft_reset handles any | |
3533 | * IP specific hardware or software state changes that are necessary to soft | |
3534 | * reset the IP. | |
3535 | * Returns 0 on success, negative error code on failure. | |
3536 | */ | |
06ec9070 | 3537 | static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
3538 | { |
3539 | int i, r = 0; | |
3540 | ||
3541 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 3542 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 3543 | continue; |
a1255107 AD |
3544 | if (adev->ip_blocks[i].status.hang && |
3545 | adev->ip_blocks[i].version->funcs->soft_reset) { | |
3546 | r = adev->ip_blocks[i].version->funcs->soft_reset(adev); | |
35d782fe CZ |
3547 | if (r) |
3548 | return r; | |
3549 | } | |
3550 | } | |
3551 | ||
3552 | return 0; | |
3553 | } | |
3554 | ||
e3ecdffa AD |
3555 | /** |
3556 | * amdgpu_device_ip_post_soft_reset - clean up from soft reset | |
3557 | * | |
3558 | * @adev: amdgpu_device pointer | |
3559 | * | |
3560 | * The list of all the hardware IPs that make up the asic is walked and the | |
3561 | * post_soft_reset callbacks are run if the asic was hung. post_soft_reset | |
3562 | * handles any IP specific hardware or software state changes that are | |
3563 | * necessary after the IP has been soft reset. | |
3564 | * Returns 0 on success, negative error code on failure. | |
3565 | */ | |
06ec9070 | 3566 | static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
3567 | { |
3568 | int i, r = 0; | |
3569 | ||
3570 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 3571 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 3572 | continue; |
a1255107 AD |
3573 | if (adev->ip_blocks[i].status.hang && |
3574 | adev->ip_blocks[i].version->funcs->post_soft_reset) | |
3575 | r = adev->ip_blocks[i].version->funcs->post_soft_reset(adev); | |
35d782fe CZ |
3576 | if (r) |
3577 | return r; | |
3578 | } | |
3579 | ||
3580 | return 0; | |
3581 | } | |
3582 | ||
e3ecdffa | 3583 | /** |
c33adbc7 | 3584 | * amdgpu_device_recover_vram - Recover some VRAM contents |
e3ecdffa AD |
3585 | * |
3586 | * @adev: amdgpu_device pointer | |
3587 | * | |
3588 | * Restores the contents of VRAM buffers from the shadows in GTT. Used to | |
3589 | * restore things like GPUVM page tables after a GPU reset where | |
3590 | * the contents of VRAM might be lost. | |
403009bf CK |
3591 | * |
3592 | * Returns: | |
3593 | * 0 on success, negative error code on failure. | |
e3ecdffa | 3594 | */ |
c33adbc7 | 3595 | static int amdgpu_device_recover_vram(struct amdgpu_device *adev) |
c41d1cf6 | 3596 | { |
c41d1cf6 | 3597 | struct dma_fence *fence = NULL, *next = NULL; |
403009bf CK |
3598 | struct amdgpu_bo *shadow; |
3599 | long r = 1, tmo; | |
c41d1cf6 ML |
3600 | |
3601 | if (amdgpu_sriov_runtime(adev)) | |
b045d3af | 3602 | tmo = msecs_to_jiffies(8000); |
c41d1cf6 ML |
3603 | else |
3604 | tmo = msecs_to_jiffies(100); | |
3605 | ||
3606 | DRM_INFO("recover vram bo from shadow start\n"); | |
3607 | mutex_lock(&adev->shadow_list_lock); | |
403009bf CK |
3608 | list_for_each_entry(shadow, &adev->shadow_list, shadow_list) { |
3609 | ||
3610 | /* No need to recover an evicted BO */ | |
3611 | if (shadow->tbo.mem.mem_type != TTM_PL_TT || | |
b575f10d | 3612 | shadow->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET || |
403009bf CK |
3613 | shadow->parent->tbo.mem.mem_type != TTM_PL_VRAM) |
3614 | continue; | |
3615 | ||
3616 | r = amdgpu_bo_restore_shadow(shadow, &next); | |
3617 | if (r) | |
3618 | break; | |
3619 | ||
c41d1cf6 | 3620 | if (fence) { |
1712fb1a | 3621 | tmo = dma_fence_wait_timeout(fence, false, tmo); |
403009bf CK |
3622 | dma_fence_put(fence); |
3623 | fence = next; | |
1712fb1a | 3624 | if (tmo == 0) { |
3625 | r = -ETIMEDOUT; | |
c41d1cf6 | 3626 | break; |
1712fb1a | 3627 | } else if (tmo < 0) { |
3628 | r = tmo; | |
3629 | break; | |
3630 | } | |
403009bf CK |
3631 | } else { |
3632 | fence = next; | |
c41d1cf6 | 3633 | } |
c41d1cf6 ML |
3634 | } |
3635 | mutex_unlock(&adev->shadow_list_lock); | |
3636 | ||
403009bf CK |
3637 | if (fence) |
3638 | tmo = dma_fence_wait_timeout(fence, false, tmo); | |
c41d1cf6 ML |
3639 | dma_fence_put(fence); |
3640 | ||
1712fb1a | 3641 | if (r < 0 || tmo <= 0) { |
3642 | DRM_ERROR("recover vram bo from shadow failed, r is %ld, tmo is %ld\n", r, tmo); | |
403009bf CK |
3643 | return -EIO; |
3644 | } | |
c41d1cf6 | 3645 | |
403009bf CK |
3646 | DRM_INFO("recover vram bo from shadow done\n"); |
3647 | return 0; | |
c41d1cf6 ML |
3648 | } |
3649 | ||
a90ad3c2 | 3650 | |
e3ecdffa | 3651 | /** |
06ec9070 | 3652 | * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf |
5740682e ML |
3653 | * |
3654 | * @adev: amdgpu device pointer | |
87e3f136 | 3655 | * @from_hypervisor: request from hypervisor |
5740682e ML |
3656 | * |
3657 | * do VF FLR and reinitialize Asic | |
3f48c681 | 3658 | * return 0 means succeeded otherwise failed |
e3ecdffa AD |
3659 | */ |
3660 | static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, | |
3661 | bool from_hypervisor) | |
5740682e ML |
3662 | { |
3663 | int r; | |
3664 | ||
3665 | if (from_hypervisor) | |
3666 | r = amdgpu_virt_request_full_gpu(adev, true); | |
3667 | else | |
3668 | r = amdgpu_virt_reset_gpu(adev); | |
3669 | if (r) | |
3670 | return r; | |
a90ad3c2 ML |
3671 | |
3672 | /* Resume IP prior to SMC */ | |
06ec9070 | 3673 | r = amdgpu_device_ip_reinit_early_sriov(adev); |
5740682e ML |
3674 | if (r) |
3675 | goto error; | |
a90ad3c2 | 3676 | |
c9ffa427 | 3677 | amdgpu_virt_init_data_exchange(adev); |
a90ad3c2 | 3678 | /* we need recover gart prior to run SMC/CP/SDMA resume */ |
c1c7ce8f | 3679 | amdgpu_gtt_mgr_recover(&adev->mman.bdev.man[TTM_PL_TT]); |
a90ad3c2 | 3680 | |
7a3e0bb2 RZ |
3681 | r = amdgpu_device_fw_loading(adev); |
3682 | if (r) | |
3683 | return r; | |
3684 | ||
a90ad3c2 | 3685 | /* now we are okay to resume SMC/CP/SDMA */ |
06ec9070 | 3686 | r = amdgpu_device_ip_reinit_late_sriov(adev); |
5740682e ML |
3687 | if (r) |
3688 | goto error; | |
a90ad3c2 ML |
3689 | |
3690 | amdgpu_irq_gpu_reset_resume_helper(adev); | |
5740682e | 3691 | r = amdgpu_ib_ring_tests(adev); |
f81e8d53 | 3692 | amdgpu_amdkfd_post_reset(adev); |
a90ad3c2 | 3693 | |
abc34253 ED |
3694 | error: |
3695 | amdgpu_virt_release_full_gpu(adev, true); | |
c41d1cf6 | 3696 | if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) { |
e3526257 | 3697 | amdgpu_inc_vram_lost(adev); |
c33adbc7 | 3698 | r = amdgpu_device_recover_vram(adev); |
a90ad3c2 ML |
3699 | } |
3700 | ||
3701 | return r; | |
3702 | } | |
3703 | ||
12938fad CK |
3704 | /** |
3705 | * amdgpu_device_should_recover_gpu - check if we should try GPU recovery | |
3706 | * | |
3707 | * @adev: amdgpu device pointer | |
3708 | * | |
3709 | * Check amdgpu_gpu_recovery and SRIOV status to see if we should try to recover | |
3710 | * a hung GPU. | |
3711 | */ | |
3712 | bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev) | |
3713 | { | |
3714 | if (!amdgpu_device_ip_check_soft_reset(adev)) { | |
3715 | DRM_INFO("Timeout, but no hardware hang detected.\n"); | |
3716 | return false; | |
3717 | } | |
3718 | ||
3ba7b418 AG |
3719 | if (amdgpu_gpu_recovery == 0) |
3720 | goto disabled; | |
3721 | ||
3722 | if (amdgpu_sriov_vf(adev)) | |
3723 | return true; | |
3724 | ||
3725 | if (amdgpu_gpu_recovery == -1) { | |
3726 | switch (adev->asic_type) { | |
fc42d47c AG |
3727 | case CHIP_BONAIRE: |
3728 | case CHIP_HAWAII: | |
3ba7b418 AG |
3729 | case CHIP_TOPAZ: |
3730 | case CHIP_TONGA: | |
3731 | case CHIP_FIJI: | |
3732 | case CHIP_POLARIS10: | |
3733 | case CHIP_POLARIS11: | |
3734 | case CHIP_POLARIS12: | |
3735 | case CHIP_VEGAM: | |
3736 | case CHIP_VEGA20: | |
3737 | case CHIP_VEGA10: | |
3738 | case CHIP_VEGA12: | |
c43b849f | 3739 | case CHIP_RAVEN: |
3ba7b418 AG |
3740 | break; |
3741 | default: | |
3742 | goto disabled; | |
3743 | } | |
12938fad CK |
3744 | } |
3745 | ||
3746 | return true; | |
3ba7b418 AG |
3747 | |
3748 | disabled: | |
3749 | DRM_INFO("GPU recovery disabled.\n"); | |
3750 | return false; | |
12938fad CK |
3751 | } |
3752 | ||
5c6dd71e | 3753 | |
26bc5340 AG |
3754 | static int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, |
3755 | struct amdgpu_job *job, | |
3756 | bool *need_full_reset_arg) | |
3757 | { | |
3758 | int i, r = 0; | |
3759 | bool need_full_reset = *need_full_reset_arg; | |
71182665 | 3760 | |
71182665 | 3761 | /* block all schedulers and reset given job's ring */ |
0875dc9e CZ |
3762 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
3763 | struct amdgpu_ring *ring = adev->rings[i]; | |
3764 | ||
51687759 | 3765 | if (!ring || !ring->sched.thread) |
0875dc9e | 3766 | continue; |
5740682e | 3767 | |
2f9d4084 ML |
3768 | /* after all hw jobs are reset, hw fence is meaningless, so force_completion */ |
3769 | amdgpu_fence_driver_force_completion(ring); | |
0875dc9e | 3770 | } |
d38ceaf9 | 3771 | |
222b5f04 AG |
3772 | if(job) |
3773 | drm_sched_increase_karma(&job->base); | |
3774 | ||
1d721ed6 | 3775 | /* Don't suspend on bare metal if we are not going to HW reset the ASIC */ |
26bc5340 AG |
3776 | if (!amdgpu_sriov_vf(adev)) { |
3777 | ||
3778 | if (!need_full_reset) | |
3779 | need_full_reset = amdgpu_device_ip_need_full_reset(adev); | |
3780 | ||
3781 | if (!need_full_reset) { | |
3782 | amdgpu_device_ip_pre_soft_reset(adev); | |
3783 | r = amdgpu_device_ip_soft_reset(adev); | |
3784 | amdgpu_device_ip_post_soft_reset(adev); | |
3785 | if (r || amdgpu_device_ip_check_soft_reset(adev)) { | |
3786 | DRM_INFO("soft reset failed, will fallback to full reset!\n"); | |
3787 | need_full_reset = true; | |
3788 | } | |
3789 | } | |
3790 | ||
3791 | if (need_full_reset) | |
3792 | r = amdgpu_device_ip_suspend(adev); | |
3793 | ||
3794 | *need_full_reset_arg = need_full_reset; | |
3795 | } | |
3796 | ||
3797 | return r; | |
3798 | } | |
3799 | ||
ce316fa5 LM |
3800 | static int amdgpu_do_asic_reset(struct amdgpu_device *adev, |
3801 | struct amdgpu_hive_info *hive, | |
26bc5340 AG |
3802 | struct list_head *device_list_handle, |
3803 | bool *need_full_reset_arg) | |
3804 | { | |
3805 | struct amdgpu_device *tmp_adev = NULL; | |
3806 | bool need_full_reset = *need_full_reset_arg, vram_lost = false; | |
3807 | int r = 0; | |
ce316fa5 LM |
3808 | int cpu = smp_processor_id(); |
3809 | bool use_baco = | |
3810 | (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) ? | |
3811 | true : false; | |
26bc5340 AG |
3812 | |
3813 | /* | |
3814 | * ASIC reset has to be done on all HGMI hive nodes ASAP | |
3815 | * to allow proper links negotiation in FW (within 1 sec) | |
3816 | */ | |
3817 | if (need_full_reset) { | |
3818 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
ce316fa5 LM |
3819 | /* |
3820 | * For XGMI run all resets in parallel to speed up the | |
3821 | * process by scheduling the highpri wq on different | |
3822 | * cpus. For XGMI with baco reset, all nodes must enter | |
3823 | * baco within close proximity before anyone exit. | |
3824 | */ | |
d4535e2c | 3825 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { |
ce316fa5 LM |
3826 | if (!queue_work_on(cpu, system_highpri_wq, |
3827 | &tmp_adev->xgmi_reset_work)) | |
d4535e2c | 3828 | r = -EALREADY; |
ce316fa5 | 3829 | cpu = cpumask_next(cpu, cpu_online_mask); |
d4535e2c AG |
3830 | } else |
3831 | r = amdgpu_asic_reset(tmp_adev); | |
ce316fa5 | 3832 | if (r) |
d4535e2c | 3833 | break; |
d4535e2c AG |
3834 | } |
3835 | ||
ce316fa5 | 3836 | /* For XGMI wait for all work to complete before proceed */ |
d4535e2c AG |
3837 | if (!r) { |
3838 | list_for_each_entry(tmp_adev, device_list_handle, | |
3839 | gmc.xgmi.head) { | |
3840 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { | |
3841 | flush_work(&tmp_adev->xgmi_reset_work); | |
3842 | r = tmp_adev->asic_reset_res; | |
3843 | if (r) | |
3844 | break; | |
ce316fa5 LM |
3845 | if (use_baco) |
3846 | tmp_adev->in_baco = true; | |
d4535e2c AG |
3847 | } |
3848 | } | |
26bc5340 | 3849 | } |
26bc5340 | 3850 | |
ce316fa5 LM |
3851 | /* |
3852 | * For XGMI with baco reset, need exit baco phase by scheduling | |
3853 | * xgmi_reset_work one more time. PSP reset and sGPU skips this | |
3854 | * phase. Not assume the situation that PSP reset and baco reset | |
3855 | * coexist within an XGMI hive. | |
3856 | */ | |
3857 | ||
3858 | if (!r && use_baco) { | |
3859 | cpu = smp_processor_id(); | |
3860 | list_for_each_entry(tmp_adev, device_list_handle, | |
3861 | gmc.xgmi.head) { | |
3862 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { | |
3863 | if (!queue_work_on(cpu, | |
3864 | system_highpri_wq, | |
3865 | &tmp_adev->xgmi_reset_work)) | |
3866 | r = -EALREADY; | |
3867 | if (r) | |
3868 | break; | |
3869 | cpu = cpumask_next(cpu, cpu_online_mask); | |
3870 | } | |
3871 | } | |
3872 | } | |
3873 | ||
3874 | if (!r && use_baco) { | |
3875 | list_for_each_entry(tmp_adev, device_list_handle, | |
3876 | gmc.xgmi.head) { | |
3877 | if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { | |
3878 | flush_work(&tmp_adev->xgmi_reset_work); | |
3879 | r = tmp_adev->asic_reset_res; | |
3880 | if (r) | |
3881 | break; | |
3882 | tmp_adev->in_baco = false; | |
3883 | } | |
3884 | } | |
3885 | } | |
3886 | ||
3887 | if (r) { | |
3888 | DRM_ERROR("ASIC reset failed with error, %d for drm dev, %s", | |
3889 | r, tmp_adev->ddev->unique); | |
3890 | goto end; | |
3891 | } | |
3892 | } | |
26bc5340 | 3893 | |
00eaa571 LM |
3894 | if (!r && amdgpu_ras_intr_triggered()) |
3895 | amdgpu_ras_intr_cleared(); | |
3896 | ||
26bc5340 AG |
3897 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { |
3898 | if (need_full_reset) { | |
3899 | /* post card */ | |
3900 | if (amdgpu_atom_asic_init(tmp_adev->mode_info.atom_context)) | |
3901 | DRM_WARN("asic atom init failed!"); | |
3902 | ||
3903 | if (!r) { | |
3904 | dev_info(tmp_adev->dev, "GPU reset succeeded, trying to resume\n"); | |
3905 | r = amdgpu_device_ip_resume_phase1(tmp_adev); | |
3906 | if (r) | |
3907 | goto out; | |
3908 | ||
3909 | vram_lost = amdgpu_device_check_vram_lost(tmp_adev); | |
3910 | if (vram_lost) { | |
77e7f829 | 3911 | DRM_INFO("VRAM is lost due to GPU reset!\n"); |
e3526257 | 3912 | amdgpu_inc_vram_lost(tmp_adev); |
26bc5340 AG |
3913 | } |
3914 | ||
3915 | r = amdgpu_gtt_mgr_recover( | |
3916 | &tmp_adev->mman.bdev.man[TTM_PL_TT]); | |
3917 | if (r) | |
3918 | goto out; | |
3919 | ||
3920 | r = amdgpu_device_fw_loading(tmp_adev); | |
3921 | if (r) | |
3922 | return r; | |
3923 | ||
3924 | r = amdgpu_device_ip_resume_phase2(tmp_adev); | |
3925 | if (r) | |
3926 | goto out; | |
3927 | ||
3928 | if (vram_lost) | |
3929 | amdgpu_device_fill_reset_magic(tmp_adev); | |
3930 | ||
fdafb359 EQ |
3931 | /* |
3932 | * Add this ASIC as tracked as reset was already | |
3933 | * complete successfully. | |
3934 | */ | |
3935 | amdgpu_register_gpu_instance(tmp_adev); | |
3936 | ||
7c04ca50 | 3937 | r = amdgpu_device_ip_late_init(tmp_adev); |
3938 | if (r) | |
3939 | goto out; | |
3940 | ||
e79a04d5 | 3941 | /* must succeed. */ |
511fdbc3 | 3942 | amdgpu_ras_resume(tmp_adev); |
e79a04d5 | 3943 | |
26bc5340 AG |
3944 | /* Update PSP FW topology after reset */ |
3945 | if (hive && tmp_adev->gmc.xgmi.num_physical_nodes > 1) | |
3946 | r = amdgpu_xgmi_update_topology(hive, tmp_adev); | |
3947 | } | |
3948 | } | |
3949 | ||
3950 | ||
3951 | out: | |
3952 | if (!r) { | |
3953 | amdgpu_irq_gpu_reset_resume_helper(tmp_adev); | |
3954 | r = amdgpu_ib_ring_tests(tmp_adev); | |
3955 | if (r) { | |
3956 | dev_err(tmp_adev->dev, "ib ring test failed (%d).\n", r); | |
3957 | r = amdgpu_device_ip_suspend(tmp_adev); | |
3958 | need_full_reset = true; | |
3959 | r = -EAGAIN; | |
3960 | goto end; | |
3961 | } | |
3962 | } | |
3963 | ||
3964 | if (!r) | |
3965 | r = amdgpu_device_recover_vram(tmp_adev); | |
3966 | else | |
3967 | tmp_adev->asic_reset_res = r; | |
3968 | } | |
3969 | ||
3970 | end: | |
3971 | *need_full_reset_arg = need_full_reset; | |
3972 | return r; | |
3973 | } | |
3974 | ||
1d721ed6 | 3975 | static bool amdgpu_device_lock_adev(struct amdgpu_device *adev, bool trylock) |
26bc5340 | 3976 | { |
1d721ed6 AG |
3977 | if (trylock) { |
3978 | if (!mutex_trylock(&adev->lock_reset)) | |
3979 | return false; | |
3980 | } else | |
3981 | mutex_lock(&adev->lock_reset); | |
5740682e | 3982 | |
26bc5340 AG |
3983 | atomic_inc(&adev->gpu_reset_counter); |
3984 | adev->in_gpu_reset = 1; | |
a3a09142 AD |
3985 | switch (amdgpu_asic_reset_method(adev)) { |
3986 | case AMD_RESET_METHOD_MODE1: | |
3987 | adev->mp1_state = PP_MP1_STATE_SHUTDOWN; | |
3988 | break; | |
3989 | case AMD_RESET_METHOD_MODE2: | |
3990 | adev->mp1_state = PP_MP1_STATE_RESET; | |
3991 | break; | |
3992 | default: | |
3993 | adev->mp1_state = PP_MP1_STATE_NONE; | |
3994 | break; | |
3995 | } | |
1d721ed6 AG |
3996 | |
3997 | return true; | |
26bc5340 | 3998 | } |
d38ceaf9 | 3999 | |
26bc5340 AG |
4000 | static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) |
4001 | { | |
89041940 | 4002 | amdgpu_vf_error_trans_all(adev); |
a3a09142 | 4003 | adev->mp1_state = PP_MP1_STATE_NONE; |
13a752e3 ML |
4004 | adev->in_gpu_reset = 0; |
4005 | mutex_unlock(&adev->lock_reset); | |
26bc5340 AG |
4006 | } |
4007 | ||
26bc5340 AG |
4008 | /** |
4009 | * amdgpu_device_gpu_recover - reset the asic and recover scheduler | |
4010 | * | |
4011 | * @adev: amdgpu device pointer | |
4012 | * @job: which job trigger hang | |
4013 | * | |
4014 | * Attempt to reset the GPU if it has hung (all asics). | |
4015 | * Attempt to do soft-reset or full-reset and reinitialize Asic | |
4016 | * Returns 0 for success or an error on failure. | |
4017 | */ | |
4018 | ||
4019 | int amdgpu_device_gpu_recover(struct amdgpu_device *adev, | |
4020 | struct amdgpu_job *job) | |
4021 | { | |
1d721ed6 AG |
4022 | struct list_head device_list, *device_list_handle = NULL; |
4023 | bool need_full_reset, job_signaled; | |
26bc5340 | 4024 | struct amdgpu_hive_info *hive = NULL; |
26bc5340 | 4025 | struct amdgpu_device *tmp_adev = NULL; |
1d721ed6 | 4026 | int i, r = 0; |
7c6e68c7 | 4027 | bool in_ras_intr = amdgpu_ras_intr_triggered(); |
b823821f LM |
4028 | bool use_baco = |
4029 | (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO) ? | |
4030 | true : false; | |
26bc5340 | 4031 | |
d5ea093e AG |
4032 | /* |
4033 | * Flush RAM to disk so that after reboot | |
4034 | * the user can read log and see why the system rebooted. | |
4035 | */ | |
b823821f | 4036 | if (in_ras_intr && !use_baco && amdgpu_ras_get_context(adev)->reboot) { |
d5ea093e AG |
4037 | |
4038 | DRM_WARN("Emergency reboot."); | |
4039 | ||
4040 | ksys_sync_helper(); | |
4041 | emergency_restart(); | |
4042 | } | |
4043 | ||
1d721ed6 | 4044 | need_full_reset = job_signaled = false; |
26bc5340 AG |
4045 | INIT_LIST_HEAD(&device_list); |
4046 | ||
b823821f LM |
4047 | dev_info(adev->dev, "GPU %s begin!\n", |
4048 | (in_ras_intr && !use_baco) ? "jobs stop":"reset"); | |
26bc5340 | 4049 | |
beff74bc | 4050 | cancel_delayed_work_sync(&adev->delayed_init_work); |
c53e4db7 | 4051 | |
1d721ed6 AG |
4052 | hive = amdgpu_get_xgmi_hive(adev, false); |
4053 | ||
26bc5340 | 4054 | /* |
1d721ed6 AG |
4055 | * Here we trylock to avoid chain of resets executing from |
4056 | * either trigger by jobs on different adevs in XGMI hive or jobs on | |
4057 | * different schedulers for same device while this TO handler is running. | |
4058 | * We always reset all schedulers for device and all devices for XGMI | |
4059 | * hive so that should take care of them too. | |
26bc5340 | 4060 | */ |
1d721ed6 AG |
4061 | |
4062 | if (hive && !mutex_trylock(&hive->reset_lock)) { | |
4063 | DRM_INFO("Bailing on TDR for s_job:%llx, hive: %llx as another already in progress", | |
0b2d2c2e | 4064 | job ? job->base.id : -1, hive->hive_id); |
26bc5340 | 4065 | return 0; |
1d721ed6 | 4066 | } |
26bc5340 AG |
4067 | |
4068 | /* Start with adev pre asic reset first for soft reset check.*/ | |
1d721ed6 AG |
4069 | if (!amdgpu_device_lock_adev(adev, !hive)) { |
4070 | DRM_INFO("Bailing on TDR for s_job:%llx, as another already in progress", | |
0b2d2c2e | 4071 | job ? job->base.id : -1); |
1d721ed6 | 4072 | return 0; |
26bc5340 AG |
4073 | } |
4074 | ||
7c6e68c7 AG |
4075 | /* Block kfd: SRIOV would do it separately */ |
4076 | if (!amdgpu_sriov_vf(adev)) | |
4077 | amdgpu_amdkfd_pre_reset(adev); | |
4078 | ||
26bc5340 | 4079 | /* Build list of devices to reset */ |
1d721ed6 | 4080 | if (adev->gmc.xgmi.num_physical_nodes > 1) { |
26bc5340 | 4081 | if (!hive) { |
7c6e68c7 AG |
4082 | /*unlock kfd: SRIOV would do it separately */ |
4083 | if (!amdgpu_sriov_vf(adev)) | |
4084 | amdgpu_amdkfd_post_reset(adev); | |
26bc5340 AG |
4085 | amdgpu_device_unlock_adev(adev); |
4086 | return -ENODEV; | |
4087 | } | |
4088 | ||
4089 | /* | |
4090 | * In case we are in XGMI hive mode device reset is done for all the | |
4091 | * nodes in the hive to retrain all XGMI links and hence the reset | |
4092 | * sequence is executed in loop on all nodes. | |
4093 | */ | |
4094 | device_list_handle = &hive->device_list; | |
4095 | } else { | |
4096 | list_add_tail(&adev->gmc.xgmi.head, &device_list); | |
4097 | device_list_handle = &device_list; | |
4098 | } | |
4099 | ||
1d721ed6 AG |
4100 | /* block all schedulers and reset given job's ring */ |
4101 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
7c6e68c7 | 4102 | if (tmp_adev != adev) { |
12ffa55d | 4103 | amdgpu_device_lock_adev(tmp_adev, false); |
7c6e68c7 AG |
4104 | if (!amdgpu_sriov_vf(tmp_adev)) |
4105 | amdgpu_amdkfd_pre_reset(tmp_adev); | |
4106 | } | |
4107 | ||
12ffa55d AG |
4108 | /* |
4109 | * Mark these ASICs to be reseted as untracked first | |
4110 | * And add them back after reset completed | |
4111 | */ | |
4112 | amdgpu_unregister_gpu_instance(tmp_adev); | |
4113 | ||
f1c1314b | 4114 | /* disable ras on ALL IPs */ |
b823821f LM |
4115 | if (!(in_ras_intr && !use_baco) && |
4116 | amdgpu_device_ip_need_full_reset(tmp_adev)) | |
f1c1314b | 4117 | amdgpu_ras_suspend(tmp_adev); |
4118 | ||
1d721ed6 AG |
4119 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
4120 | struct amdgpu_ring *ring = tmp_adev->rings[i]; | |
4121 | ||
4122 | if (!ring || !ring->sched.thread) | |
4123 | continue; | |
4124 | ||
0b2d2c2e | 4125 | drm_sched_stop(&ring->sched, job ? &job->base : NULL); |
7c6e68c7 | 4126 | |
b823821f | 4127 | if (in_ras_intr && !use_baco) |
7c6e68c7 | 4128 | amdgpu_job_stop_all_jobs_on_sched(&ring->sched); |
1d721ed6 AG |
4129 | } |
4130 | } | |
4131 | ||
4132 | ||
b823821f | 4133 | if (in_ras_intr && !use_baco) |
7c6e68c7 AG |
4134 | goto skip_sched_resume; |
4135 | ||
1d721ed6 AG |
4136 | /* |
4137 | * Must check guilty signal here since after this point all old | |
4138 | * HW fences are force signaled. | |
4139 | * | |
4140 | * job->base holds a reference to parent fence | |
4141 | */ | |
4142 | if (job && job->base.s_fence->parent && | |
4143 | dma_fence_is_signaled(job->base.s_fence->parent)) | |
4144 | job_signaled = true; | |
4145 | ||
1d721ed6 AG |
4146 | if (job_signaled) { |
4147 | dev_info(adev->dev, "Guilty job already signaled, skipping HW reset"); | |
4148 | goto skip_hw_reset; | |
4149 | } | |
4150 | ||
4151 | ||
4152 | /* Guilty job will be freed after this*/ | |
0b2d2c2e | 4153 | r = amdgpu_device_pre_asic_reset(adev, job, &need_full_reset); |
1d721ed6 AG |
4154 | if (r) { |
4155 | /*TODO Should we stop ?*/ | |
4156 | DRM_ERROR("GPU pre asic reset failed with err, %d for drm dev, %s ", | |
4157 | r, adev->ddev->unique); | |
4158 | adev->asic_reset_res = r; | |
4159 | } | |
4160 | ||
26bc5340 AG |
4161 | retry: /* Rest of adevs pre asic reset from XGMI hive. */ |
4162 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
4163 | ||
4164 | if (tmp_adev == adev) | |
4165 | continue; | |
4166 | ||
26bc5340 AG |
4167 | r = amdgpu_device_pre_asic_reset(tmp_adev, |
4168 | NULL, | |
4169 | &need_full_reset); | |
4170 | /*TODO Should we stop ?*/ | |
4171 | if (r) { | |
4172 | DRM_ERROR("GPU pre asic reset failed with err, %d for drm dev, %s ", | |
4173 | r, tmp_adev->ddev->unique); | |
4174 | tmp_adev->asic_reset_res = r; | |
4175 | } | |
4176 | } | |
4177 | ||
4178 | /* Actual ASIC resets if needed.*/ | |
4179 | /* TODO Implement XGMI hive reset logic for SRIOV */ | |
4180 | if (amdgpu_sriov_vf(adev)) { | |
4181 | r = amdgpu_device_reset_sriov(adev, job ? false : true); | |
4182 | if (r) | |
4183 | adev->asic_reset_res = r; | |
4184 | } else { | |
ce316fa5 LM |
4185 | r = amdgpu_do_asic_reset(adev, hive, device_list_handle, |
4186 | &need_full_reset); | |
26bc5340 AG |
4187 | if (r && r == -EAGAIN) |
4188 | goto retry; | |
4189 | } | |
4190 | ||
1d721ed6 AG |
4191 | skip_hw_reset: |
4192 | ||
26bc5340 AG |
4193 | /* Post ASIC reset for all devs .*/ |
4194 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
7c6e68c7 | 4195 | |
1d721ed6 AG |
4196 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
4197 | struct amdgpu_ring *ring = tmp_adev->rings[i]; | |
4198 | ||
4199 | if (!ring || !ring->sched.thread) | |
4200 | continue; | |
4201 | ||
4202 | /* No point to resubmit jobs if we didn't HW reset*/ | |
4203 | if (!tmp_adev->asic_reset_res && !job_signaled) | |
4204 | drm_sched_resubmit_jobs(&ring->sched); | |
4205 | ||
4206 | drm_sched_start(&ring->sched, !tmp_adev->asic_reset_res); | |
4207 | } | |
4208 | ||
4209 | if (!amdgpu_device_has_dc_support(tmp_adev) && !job_signaled) { | |
4210 | drm_helper_resume_force_mode(tmp_adev->ddev); | |
4211 | } | |
4212 | ||
4213 | tmp_adev->asic_reset_res = 0; | |
26bc5340 AG |
4214 | |
4215 | if (r) { | |
4216 | /* bad news, how to tell it to userspace ? */ | |
12ffa55d | 4217 | dev_info(tmp_adev->dev, "GPU reset(%d) failed\n", atomic_read(&tmp_adev->gpu_reset_counter)); |
26bc5340 AG |
4218 | amdgpu_vf_error_put(tmp_adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r); |
4219 | } else { | |
12ffa55d | 4220 | dev_info(tmp_adev->dev, "GPU reset(%d) succeeded!\n", atomic_read(&tmp_adev->gpu_reset_counter)); |
26bc5340 | 4221 | } |
7c6e68c7 | 4222 | } |
26bc5340 | 4223 | |
7c6e68c7 AG |
4224 | skip_sched_resume: |
4225 | list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) { | |
4226 | /*unlock kfd: SRIOV would do it separately */ | |
b823821f | 4227 | if (!(in_ras_intr && !use_baco) && !amdgpu_sriov_vf(tmp_adev)) |
7c6e68c7 | 4228 | amdgpu_amdkfd_post_reset(tmp_adev); |
26bc5340 AG |
4229 | amdgpu_device_unlock_adev(tmp_adev); |
4230 | } | |
4231 | ||
1d721ed6 | 4232 | if (hive) |
22d6575b | 4233 | mutex_unlock(&hive->reset_lock); |
26bc5340 AG |
4234 | |
4235 | if (r) | |
4236 | dev_info(adev->dev, "GPU reset end with ret = %d\n", r); | |
d38ceaf9 AD |
4237 | return r; |
4238 | } | |
4239 | ||
e3ecdffa AD |
4240 | /** |
4241 | * amdgpu_device_get_pcie_info - fence pcie info about the PCIE slot | |
4242 | * | |
4243 | * @adev: amdgpu_device pointer | |
4244 | * | |
4245 | * Fetchs and stores in the driver the PCIE capabilities (gen speed | |
4246 | * and lanes) of the slot the device is in. Handles APUs and | |
4247 | * virtualized environments where PCIE config space may not be available. | |
4248 | */ | |
5494d864 | 4249 | static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev) |
d0dd7f0c | 4250 | { |
5d9a6330 | 4251 | struct pci_dev *pdev; |
c5313457 HK |
4252 | enum pci_bus_speed speed_cap, platform_speed_cap; |
4253 | enum pcie_link_width platform_link_width; | |
d0dd7f0c | 4254 | |
cd474ba0 AD |
4255 | if (amdgpu_pcie_gen_cap) |
4256 | adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap; | |
d0dd7f0c | 4257 | |
cd474ba0 AD |
4258 | if (amdgpu_pcie_lane_cap) |
4259 | adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap; | |
d0dd7f0c | 4260 | |
cd474ba0 AD |
4261 | /* covers APUs as well */ |
4262 | if (pci_is_root_bus(adev->pdev->bus)) { | |
4263 | if (adev->pm.pcie_gen_mask == 0) | |
4264 | adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; | |
4265 | if (adev->pm.pcie_mlw_mask == 0) | |
4266 | adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; | |
d0dd7f0c | 4267 | return; |
cd474ba0 | 4268 | } |
d0dd7f0c | 4269 | |
c5313457 HK |
4270 | if (adev->pm.pcie_gen_mask && adev->pm.pcie_mlw_mask) |
4271 | return; | |
4272 | ||
dbaa922b AD |
4273 | pcie_bandwidth_available(adev->pdev, NULL, |
4274 | &platform_speed_cap, &platform_link_width); | |
c5313457 | 4275 | |
cd474ba0 | 4276 | if (adev->pm.pcie_gen_mask == 0) { |
5d9a6330 AD |
4277 | /* asic caps */ |
4278 | pdev = adev->pdev; | |
4279 | speed_cap = pcie_get_speed_cap(pdev); | |
4280 | if (speed_cap == PCI_SPEED_UNKNOWN) { | |
4281 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
cd474ba0 AD |
4282 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | |
4283 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
cd474ba0 | 4284 | } else { |
5d9a6330 AD |
4285 | if (speed_cap == PCIE_SPEED_16_0GT) |
4286 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
4287 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4288 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
4289 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4); | |
4290 | else if (speed_cap == PCIE_SPEED_8_0GT) | |
4291 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
4292 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4293 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
4294 | else if (speed_cap == PCIE_SPEED_5_0GT) | |
4295 | adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
4296 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
4297 | else | |
4298 | adev->pm.pcie_gen_mask |= CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
4299 | } | |
4300 | /* platform caps */ | |
c5313457 | 4301 | if (platform_speed_cap == PCI_SPEED_UNKNOWN) { |
5d9a6330 AD |
4302 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4303 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
4304 | } else { | |
c5313457 | 4305 | if (platform_speed_cap == PCIE_SPEED_16_0GT) |
5d9a6330 AD |
4306 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4307 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4308 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 | | |
4309 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4); | |
c5313457 | 4310 | else if (platform_speed_cap == PCIE_SPEED_8_0GT) |
5d9a6330 AD |
4311 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4312 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
4313 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
c5313457 | 4314 | else if (platform_speed_cap == PCIE_SPEED_5_0GT) |
5d9a6330 AD |
4315 | adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 | |
4316 | CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2); | |
4317 | else | |
4318 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
4319 | ||
cd474ba0 AD |
4320 | } |
4321 | } | |
4322 | if (adev->pm.pcie_mlw_mask == 0) { | |
c5313457 | 4323 | if (platform_link_width == PCIE_LNK_WIDTH_UNKNOWN) { |
5d9a6330 AD |
4324 | adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_PCIE_MLW_MASK; |
4325 | } else { | |
c5313457 | 4326 | switch (platform_link_width) { |
5d9a6330 | 4327 | case PCIE_LNK_X32: |
cd474ba0 AD |
4328 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 | |
4329 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | | |
4330 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
4331 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
4332 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4333 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4334 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4335 | break; | |
5d9a6330 | 4336 | case PCIE_LNK_X16: |
cd474ba0 AD |
4337 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | |
4338 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
4339 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
4340 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4341 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4342 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4343 | break; | |
5d9a6330 | 4344 | case PCIE_LNK_X12: |
cd474ba0 AD |
4345 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | |
4346 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
4347 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4348 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4349 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4350 | break; | |
5d9a6330 | 4351 | case PCIE_LNK_X8: |
cd474ba0 AD |
4352 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | |
4353 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
4354 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4355 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4356 | break; | |
5d9a6330 | 4357 | case PCIE_LNK_X4: |
cd474ba0 AD |
4358 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | |
4359 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
4360 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4361 | break; | |
5d9a6330 | 4362 | case PCIE_LNK_X2: |
cd474ba0 AD |
4363 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | |
4364 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
4365 | break; | |
5d9a6330 | 4366 | case PCIE_LNK_X1: |
cd474ba0 AD |
4367 | adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1; |
4368 | break; | |
4369 | default: | |
4370 | break; | |
4371 | } | |
d0dd7f0c AD |
4372 | } |
4373 | } | |
4374 | } | |
d38ceaf9 | 4375 | |
361dbd01 AD |
4376 | int amdgpu_device_baco_enter(struct drm_device *dev) |
4377 | { | |
4378 | struct amdgpu_device *adev = dev->dev_private; | |
7a22677b | 4379 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
361dbd01 AD |
4380 | |
4381 | if (!amdgpu_device_supports_baco(adev->ddev)) | |
4382 | return -ENOTSUPP; | |
4383 | ||
7a22677b LM |
4384 | if (ras && ras->supported) |
4385 | adev->nbio.funcs->enable_doorbell_interrupt(adev, false); | |
4386 | ||
361dbd01 AD |
4387 | if (is_support_sw_smu(adev)) { |
4388 | struct smu_context *smu = &adev->smu; | |
4389 | int ret; | |
4390 | ||
4391 | ret = smu_baco_enter(smu); | |
4392 | if (ret) | |
4393 | return ret; | |
361dbd01 AD |
4394 | } else { |
4395 | void *pp_handle = adev->powerplay.pp_handle; | |
4396 | const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs; | |
4397 | ||
4398 | if (!pp_funcs ||!pp_funcs->get_asic_baco_state ||!pp_funcs->set_asic_baco_state) | |
4399 | return -ENOENT; | |
4400 | ||
4401 | /* enter BACO state */ | |
4402 | if (pp_funcs->set_asic_baco_state(pp_handle, 1)) | |
4403 | return -EIO; | |
361dbd01 | 4404 | } |
7a22677b LM |
4405 | |
4406 | return 0; | |
361dbd01 AD |
4407 | } |
4408 | ||
4409 | int amdgpu_device_baco_exit(struct drm_device *dev) | |
4410 | { | |
4411 | struct amdgpu_device *adev = dev->dev_private; | |
7a22677b | 4412 | struct amdgpu_ras *ras = amdgpu_ras_get_context(adev); |
361dbd01 AD |
4413 | |
4414 | if (!amdgpu_device_supports_baco(adev->ddev)) | |
4415 | return -ENOTSUPP; | |
4416 | ||
4417 | if (is_support_sw_smu(adev)) { | |
4418 | struct smu_context *smu = &adev->smu; | |
4419 | int ret; | |
4420 | ||
4421 | ret = smu_baco_exit(smu); | |
4422 | if (ret) | |
4423 | return ret; | |
4424 | ||
361dbd01 AD |
4425 | } else { |
4426 | void *pp_handle = adev->powerplay.pp_handle; | |
4427 | const struct amd_pm_funcs *pp_funcs = adev->powerplay.pp_funcs; | |
4428 | ||
4429 | if (!pp_funcs ||!pp_funcs->get_asic_baco_state ||!pp_funcs->set_asic_baco_state) | |
4430 | return -ENOENT; | |
4431 | ||
4432 | /* exit BACO state */ | |
4433 | if (pp_funcs->set_asic_baco_state(pp_handle, 0)) | |
4434 | return -EIO; | |
361dbd01 | 4435 | } |
7a22677b LM |
4436 | |
4437 | if (ras && ras->supported) | |
4438 | adev->nbio.funcs->enable_doorbell_interrupt(adev, true); | |
4439 | ||
4440 | return 0; | |
361dbd01 | 4441 | } |