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 | */ | |
0875dc9e | 28 | #include <linux/kthread.h> |
d38ceaf9 AD |
29 | #include <linux/console.h> |
30 | #include <linux/slab.h> | |
d38ceaf9 AD |
31 | #include <drm/drmP.h> |
32 | #include <drm/drm_crtc_helper.h> | |
4562236b | 33 | #include <drm/drm_atomic_helper.h> |
d38ceaf9 AD |
34 | #include <drm/amdgpu_drm.h> |
35 | #include <linux/vgaarb.h> | |
36 | #include <linux/vga_switcheroo.h> | |
37 | #include <linux/efi.h> | |
38 | #include "amdgpu.h" | |
f4b373f4 | 39 | #include "amdgpu_trace.h" |
d38ceaf9 AD |
40 | #include "amdgpu_i2c.h" |
41 | #include "atom.h" | |
42 | #include "amdgpu_atombios.h" | |
a5bde2f9 | 43 | #include "amdgpu_atomfirmware.h" |
d0dd7f0c | 44 | #include "amd_pcie.h" |
33f34802 KW |
45 | #ifdef CONFIG_DRM_AMDGPU_SI |
46 | #include "si.h" | |
47 | #endif | |
a2e73f56 AD |
48 | #ifdef CONFIG_DRM_AMDGPU_CIK |
49 | #include "cik.h" | |
50 | #endif | |
aaa36a97 | 51 | #include "vi.h" |
460826e6 | 52 | #include "soc15.h" |
d38ceaf9 | 53 | #include "bif/bif_4_1_d.h" |
9accf2fd | 54 | #include <linux/pci.h> |
bec86378 | 55 | #include <linux/firmware.h> |
89041940 | 56 | #include "amdgpu_vf_error.h" |
d38ceaf9 | 57 | |
ba997709 | 58 | #include "amdgpu_amdkfd.h" |
d2f52ac8 | 59 | #include "amdgpu_pm.h" |
d38ceaf9 | 60 | |
e2a75f88 | 61 | MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin"); |
2d2e5e7e | 62 | MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin"); |
e2a75f88 | 63 | |
2dc80b00 S |
64 | #define AMDGPU_RESUME_MS 2000 |
65 | ||
d38ceaf9 | 66 | static const char *amdgpu_asic_name[] = { |
da69c161 KW |
67 | "TAHITI", |
68 | "PITCAIRN", | |
69 | "VERDE", | |
70 | "OLAND", | |
71 | "HAINAN", | |
d38ceaf9 AD |
72 | "BONAIRE", |
73 | "KAVERI", | |
74 | "KABINI", | |
75 | "HAWAII", | |
76 | "MULLINS", | |
77 | "TOPAZ", | |
78 | "TONGA", | |
48299f95 | 79 | "FIJI", |
d38ceaf9 | 80 | "CARRIZO", |
139f4917 | 81 | "STONEY", |
2cc0c0b5 FC |
82 | "POLARIS10", |
83 | "POLARIS11", | |
c4642a47 | 84 | "POLARIS12", |
d4196f01 | 85 | "VEGA10", |
2ca8a5d2 | 86 | "RAVEN", |
d38ceaf9 AD |
87 | "LAST", |
88 | }; | |
89 | ||
90 | bool amdgpu_device_is_px(struct drm_device *dev) | |
91 | { | |
92 | struct amdgpu_device *adev = dev->dev_private; | |
93 | ||
2f7d10b3 | 94 | if (adev->flags & AMD_IS_PX) |
d38ceaf9 AD |
95 | return true; |
96 | return false; | |
97 | } | |
98 | ||
99 | /* | |
100 | * MMIO register access helper functions. | |
101 | */ | |
102 | uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg, | |
15d72fd7 | 103 | uint32_t acc_flags) |
d38ceaf9 | 104 | { |
f4b373f4 TSD |
105 | uint32_t ret; |
106 | ||
43ca8efa | 107 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) |
bc992ba5 | 108 | return amdgpu_virt_kiq_rreg(adev, reg); |
bc992ba5 | 109 | |
15d72fd7 | 110 | if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX)) |
f4b373f4 | 111 | ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); |
d38ceaf9 AD |
112 | else { |
113 | unsigned long flags; | |
d38ceaf9 AD |
114 | |
115 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
116 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
117 | ret = readl(((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
118 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
d38ceaf9 | 119 | } |
f4b373f4 TSD |
120 | trace_amdgpu_mm_rreg(adev->pdev->device, reg, ret); |
121 | return ret; | |
d38ceaf9 AD |
122 | } |
123 | ||
124 | void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, | |
15d72fd7 | 125 | uint32_t acc_flags) |
d38ceaf9 | 126 | { |
f4b373f4 | 127 | trace_amdgpu_mm_wreg(adev->pdev->device, reg, v); |
4e99a44e | 128 | |
47ed4e1c KW |
129 | if (adev->asic_type >= CHIP_VEGA10 && reg == 0) { |
130 | adev->last_mm_index = v; | |
131 | } | |
132 | ||
43ca8efa | 133 | if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) |
bc992ba5 | 134 | return amdgpu_virt_kiq_wreg(adev, reg, v); |
bc992ba5 | 135 | |
15d72fd7 | 136 | if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX)) |
d38ceaf9 AD |
137 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); |
138 | else { | |
139 | unsigned long flags; | |
140 | ||
141 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
142 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
143 | writel(v, ((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
144 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
145 | } | |
47ed4e1c KW |
146 | |
147 | if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) { | |
148 | udelay(500); | |
149 | } | |
d38ceaf9 AD |
150 | } |
151 | ||
152 | u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg) | |
153 | { | |
154 | if ((reg * 4) < adev->rio_mem_size) | |
155 | return ioread32(adev->rio_mem + (reg * 4)); | |
156 | else { | |
157 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
158 | return ioread32(adev->rio_mem + (mmMM_DATA * 4)); | |
159 | } | |
160 | } | |
161 | ||
162 | void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v) | |
163 | { | |
47ed4e1c KW |
164 | if (adev->asic_type >= CHIP_VEGA10 && reg == 0) { |
165 | adev->last_mm_index = v; | |
166 | } | |
d38ceaf9 AD |
167 | |
168 | if ((reg * 4) < adev->rio_mem_size) | |
169 | iowrite32(v, adev->rio_mem + (reg * 4)); | |
170 | else { | |
171 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
172 | iowrite32(v, adev->rio_mem + (mmMM_DATA * 4)); | |
173 | } | |
47ed4e1c KW |
174 | |
175 | if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) { | |
176 | udelay(500); | |
177 | } | |
d38ceaf9 AD |
178 | } |
179 | ||
180 | /** | |
181 | * amdgpu_mm_rdoorbell - read a doorbell dword | |
182 | * | |
183 | * @adev: amdgpu_device pointer | |
184 | * @index: doorbell index | |
185 | * | |
186 | * Returns the value in the doorbell aperture at the | |
187 | * requested doorbell index (CIK). | |
188 | */ | |
189 | u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index) | |
190 | { | |
191 | if (index < adev->doorbell.num_doorbells) { | |
192 | return readl(adev->doorbell.ptr + index); | |
193 | } else { | |
194 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
195 | return 0; | |
196 | } | |
197 | } | |
198 | ||
199 | /** | |
200 | * amdgpu_mm_wdoorbell - write a doorbell dword | |
201 | * | |
202 | * @adev: amdgpu_device pointer | |
203 | * @index: doorbell index | |
204 | * @v: value to write | |
205 | * | |
206 | * Writes @v to the doorbell aperture at the | |
207 | * requested doorbell index (CIK). | |
208 | */ | |
209 | void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v) | |
210 | { | |
211 | if (index < adev->doorbell.num_doorbells) { | |
212 | writel(v, adev->doorbell.ptr + index); | |
213 | } else { | |
214 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
215 | } | |
216 | } | |
217 | ||
832be404 KW |
218 | /** |
219 | * amdgpu_mm_rdoorbell64 - read a doorbell Qword | |
220 | * | |
221 | * @adev: amdgpu_device pointer | |
222 | * @index: doorbell index | |
223 | * | |
224 | * Returns the value in the doorbell aperture at the | |
225 | * requested doorbell index (VEGA10+). | |
226 | */ | |
227 | u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index) | |
228 | { | |
229 | if (index < adev->doorbell.num_doorbells) { | |
230 | return atomic64_read((atomic64_t *)(adev->doorbell.ptr + index)); | |
231 | } else { | |
232 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
233 | return 0; | |
234 | } | |
235 | } | |
236 | ||
237 | /** | |
238 | * amdgpu_mm_wdoorbell64 - write a doorbell Qword | |
239 | * | |
240 | * @adev: amdgpu_device pointer | |
241 | * @index: doorbell index | |
242 | * @v: value to write | |
243 | * | |
244 | * Writes @v to the doorbell aperture at the | |
245 | * requested doorbell index (VEGA10+). | |
246 | */ | |
247 | void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v) | |
248 | { | |
249 | if (index < adev->doorbell.num_doorbells) { | |
250 | atomic64_set((atomic64_t *)(adev->doorbell.ptr + index), v); | |
251 | } else { | |
252 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
253 | } | |
254 | } | |
255 | ||
d38ceaf9 AD |
256 | /** |
257 | * amdgpu_invalid_rreg - dummy reg read function | |
258 | * | |
259 | * @adev: amdgpu device pointer | |
260 | * @reg: offset of register | |
261 | * | |
262 | * Dummy register read function. Used for register blocks | |
263 | * that certain asics don't have (all asics). | |
264 | * Returns the value in the register. | |
265 | */ | |
266 | static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg) | |
267 | { | |
268 | DRM_ERROR("Invalid callback to read register 0x%04X\n", reg); | |
269 | BUG(); | |
270 | return 0; | |
271 | } | |
272 | ||
273 | /** | |
274 | * amdgpu_invalid_wreg - dummy reg write function | |
275 | * | |
276 | * @adev: amdgpu device pointer | |
277 | * @reg: offset of register | |
278 | * @v: value to write to the register | |
279 | * | |
280 | * Dummy register read function. Used for register blocks | |
281 | * that certain asics don't have (all asics). | |
282 | */ | |
283 | static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) | |
284 | { | |
285 | DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n", | |
286 | reg, v); | |
287 | BUG(); | |
288 | } | |
289 | ||
290 | /** | |
291 | * amdgpu_block_invalid_rreg - dummy reg read function | |
292 | * | |
293 | * @adev: amdgpu device pointer | |
294 | * @block: offset of instance | |
295 | * @reg: offset of register | |
296 | * | |
297 | * Dummy register read function. Used for register blocks | |
298 | * that certain asics don't have (all asics). | |
299 | * Returns the value in the register. | |
300 | */ | |
301 | static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev, | |
302 | uint32_t block, uint32_t reg) | |
303 | { | |
304 | DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n", | |
305 | reg, block); | |
306 | BUG(); | |
307 | return 0; | |
308 | } | |
309 | ||
310 | /** | |
311 | * amdgpu_block_invalid_wreg - dummy reg write function | |
312 | * | |
313 | * @adev: amdgpu device pointer | |
314 | * @block: offset of instance | |
315 | * @reg: offset of register | |
316 | * @v: value to write to the register | |
317 | * | |
318 | * Dummy register read function. Used for register blocks | |
319 | * that certain asics don't have (all asics). | |
320 | */ | |
321 | static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev, | |
322 | uint32_t block, | |
323 | uint32_t reg, uint32_t v) | |
324 | { | |
325 | DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n", | |
326 | reg, block, v); | |
327 | BUG(); | |
328 | } | |
329 | ||
06ec9070 | 330 | static int amdgpu_device_vram_scratch_init(struct amdgpu_device *adev) |
d38ceaf9 | 331 | { |
a4a02777 CK |
332 | return amdgpu_bo_create_kernel(adev, AMDGPU_GPU_PAGE_SIZE, |
333 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, | |
334 | &adev->vram_scratch.robj, | |
335 | &adev->vram_scratch.gpu_addr, | |
336 | (void **)&adev->vram_scratch.ptr); | |
d38ceaf9 AD |
337 | } |
338 | ||
06ec9070 | 339 | static void amdgpu_device_vram_scratch_fini(struct amdgpu_device *adev) |
d38ceaf9 | 340 | { |
078af1a3 | 341 | amdgpu_bo_free_kernel(&adev->vram_scratch.robj, NULL, NULL); |
d38ceaf9 AD |
342 | } |
343 | ||
344 | /** | |
9c3f2b54 | 345 | * amdgpu_device_program_register_sequence - program an array of registers. |
d38ceaf9 AD |
346 | * |
347 | * @adev: amdgpu_device pointer | |
348 | * @registers: pointer to the register array | |
349 | * @array_size: size of the register array | |
350 | * | |
351 | * Programs an array or registers with and and or masks. | |
352 | * This is a helper for setting golden registers. | |
353 | */ | |
9c3f2b54 AD |
354 | void amdgpu_device_program_register_sequence(struct amdgpu_device *adev, |
355 | const u32 *registers, | |
356 | const u32 array_size) | |
d38ceaf9 AD |
357 | { |
358 | u32 tmp, reg, and_mask, or_mask; | |
359 | int i; | |
360 | ||
361 | if (array_size % 3) | |
362 | return; | |
363 | ||
364 | for (i = 0; i < array_size; i +=3) { | |
365 | reg = registers[i + 0]; | |
366 | and_mask = registers[i + 1]; | |
367 | or_mask = registers[i + 2]; | |
368 | ||
369 | if (and_mask == 0xffffffff) { | |
370 | tmp = or_mask; | |
371 | } else { | |
372 | tmp = RREG32(reg); | |
373 | tmp &= ~and_mask; | |
374 | tmp |= or_mask; | |
375 | } | |
376 | WREG32(reg, tmp); | |
377 | } | |
378 | } | |
379 | ||
380 | void amdgpu_pci_config_reset(struct amdgpu_device *adev) | |
381 | { | |
382 | pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA); | |
383 | } | |
384 | ||
385 | /* | |
386 | * GPU doorbell aperture helpers function. | |
387 | */ | |
388 | /** | |
06ec9070 | 389 | * amdgpu_device_doorbell_init - Init doorbell driver information. |
d38ceaf9 AD |
390 | * |
391 | * @adev: amdgpu_device pointer | |
392 | * | |
393 | * Init doorbell driver information (CIK) | |
394 | * Returns 0 on success, error on failure. | |
395 | */ | |
06ec9070 | 396 | static int amdgpu_device_doorbell_init(struct amdgpu_device *adev) |
d38ceaf9 | 397 | { |
705e519e CK |
398 | /* No doorbell on SI hardware generation */ |
399 | if (adev->asic_type < CHIP_BONAIRE) { | |
400 | adev->doorbell.base = 0; | |
401 | adev->doorbell.size = 0; | |
402 | adev->doorbell.num_doorbells = 0; | |
403 | adev->doorbell.ptr = NULL; | |
404 | return 0; | |
405 | } | |
406 | ||
d6895ad3 CK |
407 | if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET) |
408 | return -EINVAL; | |
409 | ||
d38ceaf9 AD |
410 | /* doorbell bar mapping */ |
411 | adev->doorbell.base = pci_resource_start(adev->pdev, 2); | |
412 | adev->doorbell.size = pci_resource_len(adev->pdev, 2); | |
413 | ||
edf600da | 414 | adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32), |
d38ceaf9 AD |
415 | AMDGPU_DOORBELL_MAX_ASSIGNMENT+1); |
416 | if (adev->doorbell.num_doorbells == 0) | |
417 | return -EINVAL; | |
418 | ||
8972e5d2 CK |
419 | adev->doorbell.ptr = ioremap(adev->doorbell.base, |
420 | adev->doorbell.num_doorbells * | |
421 | sizeof(u32)); | |
422 | if (adev->doorbell.ptr == NULL) | |
d38ceaf9 | 423 | return -ENOMEM; |
d38ceaf9 AD |
424 | |
425 | return 0; | |
426 | } | |
427 | ||
428 | /** | |
06ec9070 | 429 | * amdgpu_device_doorbell_fini - Tear down doorbell driver information. |
d38ceaf9 AD |
430 | * |
431 | * @adev: amdgpu_device pointer | |
432 | * | |
433 | * Tear down doorbell driver information (CIK) | |
434 | */ | |
06ec9070 | 435 | static void amdgpu_device_doorbell_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
436 | { |
437 | iounmap(adev->doorbell.ptr); | |
438 | adev->doorbell.ptr = NULL; | |
439 | } | |
440 | ||
441 | /** | |
442 | * amdgpu_doorbell_get_kfd_info - Report doorbell configuration required to | |
443 | * setup amdkfd | |
444 | * | |
445 | * @adev: amdgpu_device pointer | |
446 | * @aperture_base: output returning doorbell aperture base physical address | |
447 | * @aperture_size: output returning doorbell aperture size in bytes | |
448 | * @start_offset: output returning # of doorbell bytes reserved for amdgpu. | |
449 | * | |
450 | * amdgpu and amdkfd share the doorbell aperture. amdgpu sets it up, | |
451 | * takes doorbells required for its own rings and reports the setup to amdkfd. | |
452 | * amdgpu reserved doorbells are at the start of the doorbell aperture. | |
453 | */ | |
454 | void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev, | |
455 | phys_addr_t *aperture_base, | |
456 | size_t *aperture_size, | |
457 | size_t *start_offset) | |
458 | { | |
459 | /* | |
460 | * The first num_doorbells are used by amdgpu. | |
461 | * amdkfd takes whatever's left in the aperture. | |
462 | */ | |
463 | if (adev->doorbell.size > adev->doorbell.num_doorbells * sizeof(u32)) { | |
464 | *aperture_base = adev->doorbell.base; | |
465 | *aperture_size = adev->doorbell.size; | |
466 | *start_offset = adev->doorbell.num_doorbells * sizeof(u32); | |
467 | } else { | |
468 | *aperture_base = 0; | |
469 | *aperture_size = 0; | |
470 | *start_offset = 0; | |
471 | } | |
472 | } | |
473 | ||
474 | /* | |
06ec9070 | 475 | * amdgpu_device_wb_*() |
455a7bc2 | 476 | * Writeback is the method by which the GPU updates special pages in memory |
ea81a173 | 477 | * with the status of certain GPU events (fences, ring pointers,etc.). |
d38ceaf9 AD |
478 | */ |
479 | ||
480 | /** | |
06ec9070 | 481 | * amdgpu_device_wb_fini - Disable Writeback and free memory |
d38ceaf9 AD |
482 | * |
483 | * @adev: amdgpu_device pointer | |
484 | * | |
485 | * Disables Writeback and frees the Writeback memory (all asics). | |
486 | * Used at driver shutdown. | |
487 | */ | |
06ec9070 | 488 | static void amdgpu_device_wb_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
489 | { |
490 | if (adev->wb.wb_obj) { | |
a76ed485 AD |
491 | amdgpu_bo_free_kernel(&adev->wb.wb_obj, |
492 | &adev->wb.gpu_addr, | |
493 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
494 | adev->wb.wb_obj = NULL; |
495 | } | |
496 | } | |
497 | ||
498 | /** | |
06ec9070 | 499 | * amdgpu_device_wb_init- Init Writeback driver info and allocate memory |
d38ceaf9 AD |
500 | * |
501 | * @adev: amdgpu_device pointer | |
502 | * | |
455a7bc2 | 503 | * Initializes writeback and allocates writeback memory (all asics). |
d38ceaf9 AD |
504 | * Used at driver startup. |
505 | * Returns 0 on success or an -error on failure. | |
506 | */ | |
06ec9070 | 507 | static int amdgpu_device_wb_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
508 | { |
509 | int r; | |
510 | ||
511 | if (adev->wb.wb_obj == NULL) { | |
97407b63 AD |
512 | /* AMDGPU_MAX_WB * sizeof(uint32_t) * 8 = AMDGPU_MAX_WB 256bit slots */ |
513 | r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t) * 8, | |
a76ed485 AD |
514 | PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, |
515 | &adev->wb.wb_obj, &adev->wb.gpu_addr, | |
516 | (void **)&adev->wb.wb); | |
d38ceaf9 AD |
517 | if (r) { |
518 | dev_warn(adev->dev, "(%d) create WB bo failed\n", r); | |
519 | return r; | |
520 | } | |
d38ceaf9 AD |
521 | |
522 | adev->wb.num_wb = AMDGPU_MAX_WB; | |
523 | memset(&adev->wb.used, 0, sizeof(adev->wb.used)); | |
524 | ||
525 | /* clear wb memory */ | |
60a970a6 | 526 | memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t)); |
d38ceaf9 AD |
527 | } |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
532 | /** | |
131b4b36 | 533 | * amdgpu_device_wb_get - Allocate a wb entry |
d38ceaf9 AD |
534 | * |
535 | * @adev: amdgpu_device pointer | |
536 | * @wb: wb index | |
537 | * | |
538 | * Allocate a wb slot for use by the driver (all asics). | |
539 | * Returns 0 on success or -EINVAL on failure. | |
540 | */ | |
131b4b36 | 541 | int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb) |
d38ceaf9 AD |
542 | { |
543 | unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb); | |
d38ceaf9 | 544 | |
97407b63 | 545 | if (offset < adev->wb.num_wb) { |
7014285a | 546 | __set_bit(offset, adev->wb.used); |
63ae07ca | 547 | *wb = offset << 3; /* convert to dw offset */ |
0915fdbc ML |
548 | return 0; |
549 | } else { | |
550 | return -EINVAL; | |
551 | } | |
552 | } | |
553 | ||
d38ceaf9 | 554 | /** |
131b4b36 | 555 | * amdgpu_device_wb_free - Free a wb entry |
d38ceaf9 AD |
556 | * |
557 | * @adev: amdgpu_device pointer | |
558 | * @wb: wb index | |
559 | * | |
560 | * Free a wb slot allocated for use by the driver (all asics) | |
561 | */ | |
131b4b36 | 562 | void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb) |
d38ceaf9 AD |
563 | { |
564 | if (wb < adev->wb.num_wb) | |
63ae07ca | 565 | __clear_bit(wb >> 3, adev->wb.used); |
d38ceaf9 AD |
566 | } |
567 | ||
568 | /** | |
569 | * amdgpu_vram_location - try to find VRAM location | |
570 | * @adev: amdgpu device structure holding all necessary informations | |
571 | * @mc: memory controller structure holding memory informations | |
572 | * @base: base address at which to put VRAM | |
573 | * | |
455a7bc2 | 574 | * Function will try to place VRAM at base address provided |
3d647c8f | 575 | * as parameter. |
d38ceaf9 AD |
576 | */ |
577 | void amdgpu_vram_location(struct amdgpu_device *adev, struct amdgpu_mc *mc, u64 base) | |
578 | { | |
579 | uint64_t limit = (uint64_t)amdgpu_vram_limit << 20; | |
580 | ||
581 | mc->vram_start = base; | |
d38ceaf9 AD |
582 | mc->vram_end = mc->vram_start + mc->mc_vram_size - 1; |
583 | if (limit && limit < mc->real_vram_size) | |
584 | mc->real_vram_size = limit; | |
585 | dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n", | |
586 | mc->mc_vram_size >> 20, mc->vram_start, | |
587 | mc->vram_end, mc->real_vram_size >> 20); | |
588 | } | |
589 | ||
590 | /** | |
6f02a696 | 591 | * amdgpu_gart_location - try to find GTT location |
d38ceaf9 AD |
592 | * @adev: amdgpu device structure holding all necessary informations |
593 | * @mc: memory controller structure holding memory informations | |
594 | * | |
595 | * Function will place try to place GTT before or after VRAM. | |
596 | * | |
597 | * If GTT size is bigger than space left then we ajust GTT size. | |
598 | * Thus function will never fails. | |
599 | * | |
600 | * FIXME: when reducing GTT size align new size on power of 2. | |
601 | */ | |
6f02a696 | 602 | void amdgpu_gart_location(struct amdgpu_device *adev, struct amdgpu_mc *mc) |
d38ceaf9 AD |
603 | { |
604 | u64 size_af, size_bf; | |
605 | ||
ed21c047 CK |
606 | size_af = adev->mc.mc_mask - mc->vram_end; |
607 | size_bf = mc->vram_start; | |
d38ceaf9 | 608 | if (size_bf > size_af) { |
6f02a696 | 609 | if (mc->gart_size > size_bf) { |
d38ceaf9 | 610 | dev_warn(adev->dev, "limiting GTT\n"); |
6f02a696 | 611 | mc->gart_size = size_bf; |
d38ceaf9 | 612 | } |
6f02a696 | 613 | mc->gart_start = 0; |
d38ceaf9 | 614 | } else { |
6f02a696 | 615 | if (mc->gart_size > size_af) { |
d38ceaf9 | 616 | dev_warn(adev->dev, "limiting GTT\n"); |
6f02a696 | 617 | mc->gart_size = size_af; |
d38ceaf9 | 618 | } |
b98f1b9e CK |
619 | /* VCE doesn't like it when BOs cross a 4GB segment, so align |
620 | * the GART base on a 4GB boundary as well. | |
621 | */ | |
622 | mc->gart_start = ALIGN(mc->vram_end + 1, 0x100000000ULL); | |
d38ceaf9 | 623 | } |
6f02a696 | 624 | mc->gart_end = mc->gart_start + mc->gart_size - 1; |
d38ceaf9 | 625 | dev_info(adev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n", |
6f02a696 | 626 | mc->gart_size >> 20, mc->gart_start, mc->gart_end); |
d38ceaf9 AD |
627 | } |
628 | ||
a05502e5 HC |
629 | /* |
630 | * Firmware Reservation functions | |
631 | */ | |
632 | /** | |
633 | * amdgpu_fw_reserve_vram_fini - free fw reserved vram | |
634 | * | |
635 | * @adev: amdgpu_device pointer | |
636 | * | |
637 | * free fw reserved vram if it has been reserved. | |
638 | */ | |
639 | void amdgpu_fw_reserve_vram_fini(struct amdgpu_device *adev) | |
640 | { | |
641 | amdgpu_bo_free_kernel(&adev->fw_vram_usage.reserved_bo, | |
642 | NULL, &adev->fw_vram_usage.va); | |
643 | } | |
644 | ||
645 | /** | |
646 | * amdgpu_fw_reserve_vram_init - create bo vram reservation from fw | |
647 | * | |
648 | * @adev: amdgpu_device pointer | |
649 | * | |
650 | * create bo vram reservation from fw. | |
651 | */ | |
652 | int amdgpu_fw_reserve_vram_init(struct amdgpu_device *adev) | |
653 | { | |
c13c55d6 | 654 | struct ttm_operation_ctx ctx = { false, false }; |
a05502e5 | 655 | int r = 0; |
3c738893 | 656 | int i; |
a05502e5 | 657 | u64 vram_size = adev->mc.visible_vram_size; |
3c738893 HC |
658 | u64 offset = adev->fw_vram_usage.start_offset; |
659 | u64 size = adev->fw_vram_usage.size; | |
660 | struct amdgpu_bo *bo; | |
a05502e5 HC |
661 | |
662 | adev->fw_vram_usage.va = NULL; | |
663 | adev->fw_vram_usage.reserved_bo = NULL; | |
664 | ||
665 | if (adev->fw_vram_usage.size > 0 && | |
666 | adev->fw_vram_usage.size <= vram_size) { | |
667 | ||
668 | r = amdgpu_bo_create(adev, adev->fw_vram_usage.size, | |
3c738893 | 669 | PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, |
a05502e5 HC |
670 | AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED | |
671 | AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS, NULL, NULL, 0, | |
672 | &adev->fw_vram_usage.reserved_bo); | |
673 | if (r) | |
674 | goto error_create; | |
675 | ||
676 | r = amdgpu_bo_reserve(adev->fw_vram_usage.reserved_bo, false); | |
677 | if (r) | |
678 | goto error_reserve; | |
3c738893 HC |
679 | |
680 | /* remove the original mem node and create a new one at the | |
681 | * request position | |
682 | */ | |
683 | bo = adev->fw_vram_usage.reserved_bo; | |
684 | offset = ALIGN(offset, PAGE_SIZE); | |
685 | for (i = 0; i < bo->placement.num_placement; ++i) { | |
686 | bo->placements[i].fpfn = offset >> PAGE_SHIFT; | |
687 | bo->placements[i].lpfn = (offset + size) >> PAGE_SHIFT; | |
688 | } | |
689 | ||
690 | ttm_bo_mem_put(&bo->tbo, &bo->tbo.mem); | |
c13c55d6 CK |
691 | r = ttm_bo_mem_space(&bo->tbo, &bo->placement, |
692 | &bo->tbo.mem, &ctx); | |
3c738893 HC |
693 | if (r) |
694 | goto error_pin; | |
695 | ||
a05502e5 HC |
696 | r = amdgpu_bo_pin_restricted(adev->fw_vram_usage.reserved_bo, |
697 | AMDGPU_GEM_DOMAIN_VRAM, | |
698 | adev->fw_vram_usage.start_offset, | |
699 | (adev->fw_vram_usage.start_offset + | |
9921167d | 700 | adev->fw_vram_usage.size), NULL); |
a05502e5 HC |
701 | if (r) |
702 | goto error_pin; | |
703 | r = amdgpu_bo_kmap(adev->fw_vram_usage.reserved_bo, | |
704 | &adev->fw_vram_usage.va); | |
705 | if (r) | |
706 | goto error_kmap; | |
707 | ||
708 | amdgpu_bo_unreserve(adev->fw_vram_usage.reserved_bo); | |
709 | } | |
710 | return r; | |
711 | ||
712 | error_kmap: | |
713 | amdgpu_bo_unpin(adev->fw_vram_usage.reserved_bo); | |
714 | error_pin: | |
715 | amdgpu_bo_unreserve(adev->fw_vram_usage.reserved_bo); | |
716 | error_reserve: | |
717 | amdgpu_bo_unref(&adev->fw_vram_usage.reserved_bo); | |
718 | error_create: | |
719 | adev->fw_vram_usage.va = NULL; | |
720 | adev->fw_vram_usage.reserved_bo = NULL; | |
721 | return r; | |
722 | } | |
723 | ||
d6895ad3 CK |
724 | /** |
725 | * amdgpu_device_resize_fb_bar - try to resize FB BAR | |
726 | * | |
727 | * @adev: amdgpu_device pointer | |
728 | * | |
729 | * Try to resize FB BAR to make all VRAM CPU accessible. We try very hard not | |
730 | * to fail, but if any of the BARs is not accessible after the size we abort | |
731 | * driver loading by returning -ENODEV. | |
732 | */ | |
733 | int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev) | |
734 | { | |
735 | u64 space_needed = roundup_pow_of_two(adev->mc.real_vram_size); | |
736 | u32 rbar_size = order_base_2(((space_needed >> 20) | 1)) - 1; | |
31b8adab CK |
737 | struct pci_bus *root; |
738 | struct resource *res; | |
739 | unsigned i; | |
d6895ad3 CK |
740 | u16 cmd; |
741 | int r; | |
742 | ||
0c03b912 | 743 | /* Bypass for VF */ |
744 | if (amdgpu_sriov_vf(adev)) | |
745 | return 0; | |
746 | ||
31b8adab CK |
747 | /* Check if the root BUS has 64bit memory resources */ |
748 | root = adev->pdev->bus; | |
749 | while (root->parent) | |
750 | root = root->parent; | |
751 | ||
752 | pci_bus_for_each_resource(root, res, i) { | |
753 | if (res && res->flags & IORESOURCE_MEM_64 && | |
754 | res->start > 0x100000000ull) | |
755 | break; | |
756 | } | |
757 | ||
758 | /* Trying to resize is pointless without a root hub window above 4GB */ | |
759 | if (!res) | |
760 | return 0; | |
761 | ||
d6895ad3 CK |
762 | /* Disable memory decoding while we change the BAR addresses and size */ |
763 | pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd); | |
764 | pci_write_config_word(adev->pdev, PCI_COMMAND, | |
765 | cmd & ~PCI_COMMAND_MEMORY); | |
766 | ||
767 | /* Free the VRAM and doorbell BAR, we most likely need to move both. */ | |
06ec9070 | 768 | amdgpu_device_doorbell_fini(adev); |
d6895ad3 CK |
769 | if (adev->asic_type >= CHIP_BONAIRE) |
770 | pci_release_resource(adev->pdev, 2); | |
771 | ||
772 | pci_release_resource(adev->pdev, 0); | |
773 | ||
774 | r = pci_resize_resource(adev->pdev, 0, rbar_size); | |
775 | if (r == -ENOSPC) | |
776 | DRM_INFO("Not enough PCI address space for a large BAR."); | |
777 | else if (r && r != -ENOTSUPP) | |
778 | DRM_ERROR("Problem resizing BAR0 (%d).", r); | |
779 | ||
780 | pci_assign_unassigned_bus_resources(adev->pdev->bus); | |
781 | ||
782 | /* When the doorbell or fb BAR isn't available we have no chance of | |
783 | * using the device. | |
784 | */ | |
06ec9070 | 785 | r = amdgpu_device_doorbell_init(adev); |
d6895ad3 CK |
786 | if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET)) |
787 | return -ENODEV; | |
788 | ||
789 | pci_write_config_word(adev->pdev, PCI_COMMAND, cmd); | |
790 | ||
791 | return 0; | |
792 | } | |
a05502e5 | 793 | |
d38ceaf9 AD |
794 | /* |
795 | * GPU helpers function. | |
796 | */ | |
797 | /** | |
c836fec5 | 798 | * amdgpu_need_post - check if the hw need post or not |
d38ceaf9 AD |
799 | * |
800 | * @adev: amdgpu_device pointer | |
801 | * | |
c836fec5 JQ |
802 | * Check if the asic has been initialized (all asics) at driver startup |
803 | * or post is needed if hw reset is performed. | |
804 | * Returns true if need or false if not. | |
d38ceaf9 | 805 | */ |
c836fec5 | 806 | bool amdgpu_need_post(struct amdgpu_device *adev) |
d38ceaf9 AD |
807 | { |
808 | uint32_t reg; | |
809 | ||
bec86378 ML |
810 | if (amdgpu_sriov_vf(adev)) |
811 | return false; | |
812 | ||
813 | if (amdgpu_passthrough(adev)) { | |
1da2c326 ML |
814 | /* for FIJI: In whole GPU pass-through virtualization case, after VM reboot |
815 | * some old smc fw still need driver do vPost otherwise gpu hang, while | |
816 | * those smc fw version above 22.15 doesn't have this flaw, so we force | |
817 | * vpost executed for smc version below 22.15 | |
bec86378 ML |
818 | */ |
819 | if (adev->asic_type == CHIP_FIJI) { | |
820 | int err; | |
821 | uint32_t fw_ver; | |
822 | err = request_firmware(&adev->pm.fw, "amdgpu/fiji_smc.bin", adev->dev); | |
823 | /* force vPost if error occured */ | |
824 | if (err) | |
825 | return true; | |
826 | ||
827 | fw_ver = *((uint32_t *)adev->pm.fw->data + 69); | |
1da2c326 ML |
828 | if (fw_ver < 0x00160e00) |
829 | return true; | |
bec86378 | 830 | } |
bec86378 | 831 | } |
91fe77eb | 832 | |
833 | if (adev->has_hw_reset) { | |
834 | adev->has_hw_reset = false; | |
835 | return true; | |
836 | } | |
837 | ||
838 | /* bios scratch used on CIK+ */ | |
839 | if (adev->asic_type >= CHIP_BONAIRE) | |
840 | return amdgpu_atombios_scratch_need_asic_init(adev); | |
841 | ||
842 | /* check MEM_SIZE for older asics */ | |
843 | reg = amdgpu_asic_get_config_memsize(adev); | |
844 | ||
845 | if ((reg != 0) && (reg != 0xffffffff)) | |
846 | return false; | |
847 | ||
848 | return true; | |
bec86378 ML |
849 | } |
850 | ||
d38ceaf9 AD |
851 | /** |
852 | * amdgpu_dummy_page_init - init dummy page used by the driver | |
853 | * | |
854 | * @adev: amdgpu_device pointer | |
855 | * | |
856 | * Allocate the dummy page used by the driver (all asics). | |
857 | * This dummy page is used by the driver as a filler for gart entries | |
858 | * when pages are taken out of the GART | |
859 | * Returns 0 on sucess, -ENOMEM on failure. | |
860 | */ | |
861 | int amdgpu_dummy_page_init(struct amdgpu_device *adev) | |
862 | { | |
863 | if (adev->dummy_page.page) | |
864 | return 0; | |
865 | adev->dummy_page.page = alloc_page(GFP_DMA32 | GFP_KERNEL | __GFP_ZERO); | |
866 | if (adev->dummy_page.page == NULL) | |
867 | return -ENOMEM; | |
868 | adev->dummy_page.addr = pci_map_page(adev->pdev, adev->dummy_page.page, | |
869 | 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
870 | if (pci_dma_mapping_error(adev->pdev, adev->dummy_page.addr)) { | |
871 | dev_err(&adev->pdev->dev, "Failed to DMA MAP the dummy page\n"); | |
872 | __free_page(adev->dummy_page.page); | |
873 | adev->dummy_page.page = NULL; | |
874 | return -ENOMEM; | |
875 | } | |
876 | return 0; | |
877 | } | |
878 | ||
879 | /** | |
880 | * amdgpu_dummy_page_fini - free dummy page used by the driver | |
881 | * | |
882 | * @adev: amdgpu_device pointer | |
883 | * | |
884 | * Frees the dummy page used by the driver (all asics). | |
885 | */ | |
886 | void amdgpu_dummy_page_fini(struct amdgpu_device *adev) | |
887 | { | |
888 | if (adev->dummy_page.page == NULL) | |
889 | return; | |
890 | pci_unmap_page(adev->pdev, adev->dummy_page.addr, | |
891 | PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
892 | __free_page(adev->dummy_page.page); | |
893 | adev->dummy_page.page = NULL; | |
894 | } | |
895 | ||
d38ceaf9 AD |
896 | /* if we get transitioned to only one device, take VGA back */ |
897 | /** | |
06ec9070 | 898 | * amdgpu_device_vga_set_decode - enable/disable vga decode |
d38ceaf9 AD |
899 | * |
900 | * @cookie: amdgpu_device pointer | |
901 | * @state: enable/disable vga decode | |
902 | * | |
903 | * Enable/disable vga decode (all asics). | |
904 | * Returns VGA resource flags. | |
905 | */ | |
06ec9070 | 906 | static unsigned int amdgpu_device_vga_set_decode(void *cookie, bool state) |
d38ceaf9 AD |
907 | { |
908 | struct amdgpu_device *adev = cookie; | |
909 | amdgpu_asic_set_vga_state(adev, state); | |
910 | if (state) | |
911 | return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM | | |
912 | VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
913 | else | |
914 | return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
915 | } | |
916 | ||
06ec9070 | 917 | static void amdgpu_device_check_block_size(struct amdgpu_device *adev) |
a1adf8be CZ |
918 | { |
919 | /* defines number of bits in page table versus page directory, | |
920 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
921 | * page table and the remaining bits are in the page directory */ | |
bab4fee7 JZ |
922 | if (amdgpu_vm_block_size == -1) |
923 | return; | |
a1adf8be | 924 | |
bab4fee7 | 925 | if (amdgpu_vm_block_size < 9) { |
a1adf8be CZ |
926 | dev_warn(adev->dev, "VM page table size (%d) too small\n", |
927 | amdgpu_vm_block_size); | |
97489129 | 928 | amdgpu_vm_block_size = -1; |
a1adf8be | 929 | } |
a1adf8be CZ |
930 | } |
931 | ||
06ec9070 | 932 | static void amdgpu_device_check_vm_size(struct amdgpu_device *adev) |
83ca145d | 933 | { |
64dab074 AD |
934 | /* no need to check the default value */ |
935 | if (amdgpu_vm_size == -1) | |
936 | return; | |
937 | ||
83ca145d ZJ |
938 | if (amdgpu_vm_size < 1) { |
939 | dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n", | |
940 | amdgpu_vm_size); | |
f3368128 | 941 | amdgpu_vm_size = -1; |
83ca145d | 942 | } |
83ca145d ZJ |
943 | } |
944 | ||
d38ceaf9 | 945 | /** |
06ec9070 | 946 | * amdgpu_device_check_arguments - validate module params |
d38ceaf9 AD |
947 | * |
948 | * @adev: amdgpu_device pointer | |
949 | * | |
950 | * Validates certain module parameters and updates | |
951 | * the associated values used by the driver (all asics). | |
952 | */ | |
06ec9070 | 953 | static void amdgpu_device_check_arguments(struct amdgpu_device *adev) |
d38ceaf9 | 954 | { |
5b011235 CZ |
955 | if (amdgpu_sched_jobs < 4) { |
956 | dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n", | |
957 | amdgpu_sched_jobs); | |
958 | amdgpu_sched_jobs = 4; | |
76117507 | 959 | } else if (!is_power_of_2(amdgpu_sched_jobs)){ |
5b011235 CZ |
960 | dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n", |
961 | amdgpu_sched_jobs); | |
962 | amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs); | |
963 | } | |
d38ceaf9 | 964 | |
83e74db6 | 965 | if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) { |
f9321cc4 CK |
966 | /* gart size must be greater or equal to 32M */ |
967 | dev_warn(adev->dev, "gart size (%d) too small\n", | |
968 | amdgpu_gart_size); | |
83e74db6 | 969 | amdgpu_gart_size = -1; |
d38ceaf9 AD |
970 | } |
971 | ||
36d38372 | 972 | if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) { |
c4e1a13a | 973 | /* gtt size must be greater or equal to 32M */ |
36d38372 CK |
974 | dev_warn(adev->dev, "gtt size (%d) too small\n", |
975 | amdgpu_gtt_size); | |
976 | amdgpu_gtt_size = -1; | |
d38ceaf9 AD |
977 | } |
978 | ||
d07f14be RH |
979 | /* valid range is between 4 and 9 inclusive */ |
980 | if (amdgpu_vm_fragment_size != -1 && | |
981 | (amdgpu_vm_fragment_size > 9 || amdgpu_vm_fragment_size < 4)) { | |
982 | dev_warn(adev->dev, "valid range is between 4 and 9\n"); | |
983 | amdgpu_vm_fragment_size = -1; | |
984 | } | |
985 | ||
06ec9070 | 986 | amdgpu_device_check_vm_size(adev); |
d38ceaf9 | 987 | |
06ec9070 | 988 | amdgpu_device_check_block_size(adev); |
6a7f76e7 | 989 | |
526bae37 | 990 | if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 || |
76117507 | 991 | !is_power_of_2(amdgpu_vram_page_split))) { |
6a7f76e7 CK |
992 | dev_warn(adev->dev, "invalid VRAM page split (%d)\n", |
993 | amdgpu_vram_page_split); | |
994 | amdgpu_vram_page_split = 1024; | |
995 | } | |
8854695a AG |
996 | |
997 | if (amdgpu_lockup_timeout == 0) { | |
998 | dev_warn(adev->dev, "lockup_timeout msut be > 0, adjusting to 10000\n"); | |
999 | amdgpu_lockup_timeout = 10000; | |
1000 | } | |
d38ceaf9 AD |
1001 | } |
1002 | ||
1003 | /** | |
1004 | * amdgpu_switcheroo_set_state - set switcheroo state | |
1005 | * | |
1006 | * @pdev: pci dev pointer | |
1694467b | 1007 | * @state: vga_switcheroo state |
d38ceaf9 AD |
1008 | * |
1009 | * Callback for the switcheroo driver. Suspends or resumes the | |
1010 | * the asics before or after it is powered up using ACPI methods. | |
1011 | */ | |
1012 | static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_state state) | |
1013 | { | |
1014 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1015 | ||
1016 | if (amdgpu_device_is_px(dev) && state == VGA_SWITCHEROO_OFF) | |
1017 | return; | |
1018 | ||
1019 | if (state == VGA_SWITCHEROO_ON) { | |
7ca85295 | 1020 | pr_info("amdgpu: switched on\n"); |
d38ceaf9 AD |
1021 | /* don't suspend or resume card normally */ |
1022 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
1023 | ||
810ddc3a | 1024 | amdgpu_device_resume(dev, true, true); |
d38ceaf9 | 1025 | |
d38ceaf9 AD |
1026 | dev->switch_power_state = DRM_SWITCH_POWER_ON; |
1027 | drm_kms_helper_poll_enable(dev); | |
1028 | } else { | |
7ca85295 | 1029 | pr_info("amdgpu: switched off\n"); |
d38ceaf9 AD |
1030 | drm_kms_helper_poll_disable(dev); |
1031 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
810ddc3a | 1032 | amdgpu_device_suspend(dev, true, true); |
d38ceaf9 AD |
1033 | dev->switch_power_state = DRM_SWITCH_POWER_OFF; |
1034 | } | |
1035 | } | |
1036 | ||
1037 | /** | |
1038 | * amdgpu_switcheroo_can_switch - see if switcheroo state can change | |
1039 | * | |
1040 | * @pdev: pci dev pointer | |
1041 | * | |
1042 | * Callback for the switcheroo driver. Check of the switcheroo | |
1043 | * state can be changed. | |
1044 | * Returns true if the state can be changed, false if not. | |
1045 | */ | |
1046 | static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev) | |
1047 | { | |
1048 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1049 | ||
1050 | /* | |
1051 | * FIXME: open_count is protected by drm_global_mutex but that would lead to | |
1052 | * locking inversion with the driver load path. And the access here is | |
1053 | * completely racy anyway. So don't bother with locking for now. | |
1054 | */ | |
1055 | return dev->open_count == 0; | |
1056 | } | |
1057 | ||
1058 | static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = { | |
1059 | .set_gpu_state = amdgpu_switcheroo_set_state, | |
1060 | .reprobe = NULL, | |
1061 | .can_switch = amdgpu_switcheroo_can_switch, | |
1062 | }; | |
1063 | ||
1064 | int amdgpu_set_clockgating_state(struct amdgpu_device *adev, | |
5fc3aeeb | 1065 | enum amd_ip_block_type block_type, |
1066 | enum amd_clockgating_state state) | |
d38ceaf9 AD |
1067 | { |
1068 | int i, r = 0; | |
1069 | ||
1070 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1071 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1072 | continue; |
c722865a RZ |
1073 | if (adev->ip_blocks[i].version->type != block_type) |
1074 | continue; | |
1075 | if (!adev->ip_blocks[i].version->funcs->set_clockgating_state) | |
1076 | continue; | |
1077 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state( | |
1078 | (void *)adev, state); | |
1079 | if (r) | |
1080 | DRM_ERROR("set_clockgating_state of IP block <%s> failed %d\n", | |
1081 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1082 | } |
1083 | return r; | |
1084 | } | |
1085 | ||
1086 | int amdgpu_set_powergating_state(struct amdgpu_device *adev, | |
5fc3aeeb | 1087 | enum amd_ip_block_type block_type, |
1088 | enum amd_powergating_state state) | |
d38ceaf9 AD |
1089 | { |
1090 | int i, r = 0; | |
1091 | ||
1092 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1093 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1094 | continue; |
c722865a RZ |
1095 | if (adev->ip_blocks[i].version->type != block_type) |
1096 | continue; | |
1097 | if (!adev->ip_blocks[i].version->funcs->set_powergating_state) | |
1098 | continue; | |
1099 | r = adev->ip_blocks[i].version->funcs->set_powergating_state( | |
1100 | (void *)adev, state); | |
1101 | if (r) | |
1102 | DRM_ERROR("set_powergating_state of IP block <%s> failed %d\n", | |
1103 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 AD |
1104 | } |
1105 | return r; | |
1106 | } | |
1107 | ||
6cb2d4e4 HR |
1108 | void amdgpu_get_clockgating_state(struct amdgpu_device *adev, u32 *flags) |
1109 | { | |
1110 | int i; | |
1111 | ||
1112 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1113 | if (!adev->ip_blocks[i].status.valid) | |
1114 | continue; | |
1115 | if (adev->ip_blocks[i].version->funcs->get_clockgating_state) | |
1116 | adev->ip_blocks[i].version->funcs->get_clockgating_state((void *)adev, flags); | |
1117 | } | |
1118 | } | |
1119 | ||
5dbbb60b AD |
1120 | int amdgpu_wait_for_idle(struct amdgpu_device *adev, |
1121 | enum amd_ip_block_type block_type) | |
1122 | { | |
1123 | int i, r; | |
1124 | ||
1125 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1126 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1127 | continue; |
a1255107 AD |
1128 | if (adev->ip_blocks[i].version->type == block_type) { |
1129 | r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev); | |
5dbbb60b AD |
1130 | if (r) |
1131 | return r; | |
1132 | break; | |
1133 | } | |
1134 | } | |
1135 | return 0; | |
1136 | ||
1137 | } | |
1138 | ||
1139 | bool amdgpu_is_idle(struct amdgpu_device *adev, | |
1140 | enum amd_ip_block_type block_type) | |
1141 | { | |
1142 | int i; | |
1143 | ||
1144 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1145 | if (!adev->ip_blocks[i].status.valid) |
9ecbe7f5 | 1146 | continue; |
a1255107 AD |
1147 | if (adev->ip_blocks[i].version->type == block_type) |
1148 | return adev->ip_blocks[i].version->funcs->is_idle((void *)adev); | |
5dbbb60b AD |
1149 | } |
1150 | return true; | |
1151 | ||
1152 | } | |
1153 | ||
a1255107 AD |
1154 | struct amdgpu_ip_block * amdgpu_get_ip_block(struct amdgpu_device *adev, |
1155 | enum amd_ip_block_type type) | |
d38ceaf9 AD |
1156 | { |
1157 | int i; | |
1158 | ||
1159 | for (i = 0; i < adev->num_ip_blocks; i++) | |
a1255107 | 1160 | if (adev->ip_blocks[i].version->type == type) |
d38ceaf9 AD |
1161 | return &adev->ip_blocks[i]; |
1162 | ||
1163 | return NULL; | |
1164 | } | |
1165 | ||
1166 | /** | |
1167 | * amdgpu_ip_block_version_cmp | |
1168 | * | |
1169 | * @adev: amdgpu_device pointer | |
5fc3aeeb | 1170 | * @type: enum amd_ip_block_type |
d38ceaf9 AD |
1171 | * @major: major version |
1172 | * @minor: minor version | |
1173 | * | |
1174 | * return 0 if equal or greater | |
1175 | * return 1 if smaller or the ip_block doesn't exist | |
1176 | */ | |
1177 | int amdgpu_ip_block_version_cmp(struct amdgpu_device *adev, | |
5fc3aeeb | 1178 | enum amd_ip_block_type type, |
d38ceaf9 AD |
1179 | u32 major, u32 minor) |
1180 | { | |
a1255107 | 1181 | struct amdgpu_ip_block *ip_block = amdgpu_get_ip_block(adev, type); |
d38ceaf9 | 1182 | |
a1255107 AD |
1183 | if (ip_block && ((ip_block->version->major > major) || |
1184 | ((ip_block->version->major == major) && | |
1185 | (ip_block->version->minor >= minor)))) | |
d38ceaf9 AD |
1186 | return 0; |
1187 | ||
1188 | return 1; | |
1189 | } | |
1190 | ||
a1255107 AD |
1191 | /** |
1192 | * amdgpu_ip_block_add | |
1193 | * | |
1194 | * @adev: amdgpu_device pointer | |
1195 | * @ip_block_version: pointer to the IP to add | |
1196 | * | |
1197 | * Adds the IP block driver information to the collection of IPs | |
1198 | * on the asic. | |
1199 | */ | |
1200 | int amdgpu_ip_block_add(struct amdgpu_device *adev, | |
1201 | const struct amdgpu_ip_block_version *ip_block_version) | |
1202 | { | |
1203 | if (!ip_block_version) | |
1204 | return -EINVAL; | |
1205 | ||
a0bae357 HR |
1206 | DRM_DEBUG("add ip block number %d <%s>\n", adev->num_ip_blocks, |
1207 | ip_block_version->funcs->name); | |
1208 | ||
a1255107 AD |
1209 | adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version; |
1210 | ||
1211 | return 0; | |
1212 | } | |
1213 | ||
483ef985 | 1214 | static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev) |
9accf2fd ED |
1215 | { |
1216 | adev->enable_virtual_display = false; | |
1217 | ||
1218 | if (amdgpu_virtual_display) { | |
1219 | struct drm_device *ddev = adev->ddev; | |
1220 | const char *pci_address_name = pci_name(ddev->pdev); | |
0f66356d | 1221 | char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname; |
9accf2fd ED |
1222 | |
1223 | pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL); | |
1224 | pciaddstr_tmp = pciaddstr; | |
0f66356d ED |
1225 | while ((pciaddname_tmp = strsep(&pciaddstr_tmp, ";"))) { |
1226 | pciaddname = strsep(&pciaddname_tmp, ","); | |
967de2a9 YT |
1227 | if (!strcmp("all", pciaddname) |
1228 | || !strcmp(pci_address_name, pciaddname)) { | |
0f66356d ED |
1229 | long num_crtc; |
1230 | int res = -1; | |
1231 | ||
9accf2fd | 1232 | adev->enable_virtual_display = true; |
0f66356d ED |
1233 | |
1234 | if (pciaddname_tmp) | |
1235 | res = kstrtol(pciaddname_tmp, 10, | |
1236 | &num_crtc); | |
1237 | ||
1238 | if (!res) { | |
1239 | if (num_crtc < 1) | |
1240 | num_crtc = 1; | |
1241 | if (num_crtc > 6) | |
1242 | num_crtc = 6; | |
1243 | adev->mode_info.num_crtc = num_crtc; | |
1244 | } else { | |
1245 | adev->mode_info.num_crtc = 1; | |
1246 | } | |
9accf2fd ED |
1247 | break; |
1248 | } | |
1249 | } | |
1250 | ||
0f66356d ED |
1251 | DRM_INFO("virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n", |
1252 | amdgpu_virtual_display, pci_address_name, | |
1253 | adev->enable_virtual_display, adev->mode_info.num_crtc); | |
9accf2fd ED |
1254 | |
1255 | kfree(pciaddstr); | |
1256 | } | |
1257 | } | |
1258 | ||
e2a75f88 AD |
1259 | static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev) |
1260 | { | |
e2a75f88 AD |
1261 | const char *chip_name; |
1262 | char fw_name[30]; | |
1263 | int err; | |
1264 | const struct gpu_info_firmware_header_v1_0 *hdr; | |
1265 | ||
ab4fe3e1 HR |
1266 | adev->firmware.gpu_info_fw = NULL; |
1267 | ||
e2a75f88 AD |
1268 | switch (adev->asic_type) { |
1269 | case CHIP_TOPAZ: | |
1270 | case CHIP_TONGA: | |
1271 | case CHIP_FIJI: | |
1272 | case CHIP_POLARIS11: | |
1273 | case CHIP_POLARIS10: | |
1274 | case CHIP_POLARIS12: | |
1275 | case CHIP_CARRIZO: | |
1276 | case CHIP_STONEY: | |
1277 | #ifdef CONFIG_DRM_AMDGPU_SI | |
1278 | case CHIP_VERDE: | |
1279 | case CHIP_TAHITI: | |
1280 | case CHIP_PITCAIRN: | |
1281 | case CHIP_OLAND: | |
1282 | case CHIP_HAINAN: | |
1283 | #endif | |
1284 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
1285 | case CHIP_BONAIRE: | |
1286 | case CHIP_HAWAII: | |
1287 | case CHIP_KAVERI: | |
1288 | case CHIP_KABINI: | |
1289 | case CHIP_MULLINS: | |
1290 | #endif | |
1291 | default: | |
1292 | return 0; | |
1293 | case CHIP_VEGA10: | |
1294 | chip_name = "vega10"; | |
1295 | break; | |
2d2e5e7e AD |
1296 | case CHIP_RAVEN: |
1297 | chip_name = "raven"; | |
1298 | break; | |
e2a75f88 AD |
1299 | } |
1300 | ||
1301 | snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name); | |
ab4fe3e1 | 1302 | err = request_firmware(&adev->firmware.gpu_info_fw, fw_name, adev->dev); |
e2a75f88 AD |
1303 | if (err) { |
1304 | dev_err(adev->dev, | |
1305 | "Failed to load gpu_info firmware \"%s\"\n", | |
1306 | fw_name); | |
1307 | goto out; | |
1308 | } | |
ab4fe3e1 | 1309 | err = amdgpu_ucode_validate(adev->firmware.gpu_info_fw); |
e2a75f88 AD |
1310 | if (err) { |
1311 | dev_err(adev->dev, | |
1312 | "Failed to validate gpu_info firmware \"%s\"\n", | |
1313 | fw_name); | |
1314 | goto out; | |
1315 | } | |
1316 | ||
ab4fe3e1 | 1317 | hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data; |
e2a75f88 AD |
1318 | amdgpu_ucode_print_gpu_info_hdr(&hdr->header); |
1319 | ||
1320 | switch (hdr->version_major) { | |
1321 | case 1: | |
1322 | { | |
1323 | const struct gpu_info_firmware_v1_0 *gpu_info_fw = | |
ab4fe3e1 | 1324 | (const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data + |
e2a75f88 AD |
1325 | le32_to_cpu(hdr->header.ucode_array_offset_bytes)); |
1326 | ||
b5ab16bf AD |
1327 | adev->gfx.config.max_shader_engines = le32_to_cpu(gpu_info_fw->gc_num_se); |
1328 | adev->gfx.config.max_cu_per_sh = le32_to_cpu(gpu_info_fw->gc_num_cu_per_sh); | |
1329 | adev->gfx.config.max_sh_per_se = le32_to_cpu(gpu_info_fw->gc_num_sh_per_se); | |
1330 | adev->gfx.config.max_backends_per_se = le32_to_cpu(gpu_info_fw->gc_num_rb_per_se); | |
e2a75f88 | 1331 | adev->gfx.config.max_texture_channel_caches = |
b5ab16bf AD |
1332 | le32_to_cpu(gpu_info_fw->gc_num_tccs); |
1333 | adev->gfx.config.max_gprs = le32_to_cpu(gpu_info_fw->gc_num_gprs); | |
1334 | adev->gfx.config.max_gs_threads = le32_to_cpu(gpu_info_fw->gc_num_max_gs_thds); | |
1335 | adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gpu_info_fw->gc_gs_table_depth); | |
1336 | adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gpu_info_fw->gc_gsprim_buff_depth); | |
e2a75f88 | 1337 | adev->gfx.config.double_offchip_lds_buf = |
b5ab16bf AD |
1338 | le32_to_cpu(gpu_info_fw->gc_double_offchip_lds_buffer); |
1339 | adev->gfx.cu_info.wave_front_size = le32_to_cpu(gpu_info_fw->gc_wave_size); | |
51fd0370 HZ |
1340 | adev->gfx.cu_info.max_waves_per_simd = |
1341 | le32_to_cpu(gpu_info_fw->gc_max_waves_per_simd); | |
1342 | adev->gfx.cu_info.max_scratch_slots_per_cu = | |
1343 | le32_to_cpu(gpu_info_fw->gc_max_scratch_slots_per_cu); | |
1344 | adev->gfx.cu_info.lds_size = le32_to_cpu(gpu_info_fw->gc_lds_size); | |
e2a75f88 AD |
1345 | break; |
1346 | } | |
1347 | default: | |
1348 | dev_err(adev->dev, | |
1349 | "Unsupported gpu_info table %d\n", hdr->header.ucode_version); | |
1350 | err = -EINVAL; | |
1351 | goto out; | |
1352 | } | |
1353 | out: | |
e2a75f88 AD |
1354 | return err; |
1355 | } | |
1356 | ||
06ec9070 | 1357 | static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) |
d38ceaf9 | 1358 | { |
aaa36a97 | 1359 | int i, r; |
d38ceaf9 | 1360 | |
483ef985 | 1361 | amdgpu_device_enable_virtual_display(adev); |
a6be7570 | 1362 | |
d38ceaf9 | 1363 | switch (adev->asic_type) { |
aaa36a97 AD |
1364 | case CHIP_TOPAZ: |
1365 | case CHIP_TONGA: | |
48299f95 | 1366 | case CHIP_FIJI: |
2cc0c0b5 FC |
1367 | case CHIP_POLARIS11: |
1368 | case CHIP_POLARIS10: | |
c4642a47 | 1369 | case CHIP_POLARIS12: |
aaa36a97 | 1370 | case CHIP_CARRIZO: |
39bb0c92 SL |
1371 | case CHIP_STONEY: |
1372 | if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) | |
aaa36a97 AD |
1373 | adev->family = AMDGPU_FAMILY_CZ; |
1374 | else | |
1375 | adev->family = AMDGPU_FAMILY_VI; | |
1376 | ||
1377 | r = vi_set_ip_blocks(adev); | |
1378 | if (r) | |
1379 | return r; | |
1380 | break; | |
33f34802 KW |
1381 | #ifdef CONFIG_DRM_AMDGPU_SI |
1382 | case CHIP_VERDE: | |
1383 | case CHIP_TAHITI: | |
1384 | case CHIP_PITCAIRN: | |
1385 | case CHIP_OLAND: | |
1386 | case CHIP_HAINAN: | |
295d0daf | 1387 | adev->family = AMDGPU_FAMILY_SI; |
33f34802 KW |
1388 | r = si_set_ip_blocks(adev); |
1389 | if (r) | |
1390 | return r; | |
1391 | break; | |
1392 | #endif | |
a2e73f56 AD |
1393 | #ifdef CONFIG_DRM_AMDGPU_CIK |
1394 | case CHIP_BONAIRE: | |
1395 | case CHIP_HAWAII: | |
1396 | case CHIP_KAVERI: | |
1397 | case CHIP_KABINI: | |
1398 | case CHIP_MULLINS: | |
1399 | if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII)) | |
1400 | adev->family = AMDGPU_FAMILY_CI; | |
1401 | else | |
1402 | adev->family = AMDGPU_FAMILY_KV; | |
1403 | ||
1404 | r = cik_set_ip_blocks(adev); | |
1405 | if (r) | |
1406 | return r; | |
1407 | break; | |
1408 | #endif | |
2ca8a5d2 CZ |
1409 | case CHIP_VEGA10: |
1410 | case CHIP_RAVEN: | |
1411 | if (adev->asic_type == CHIP_RAVEN) | |
1412 | adev->family = AMDGPU_FAMILY_RV; | |
1413 | else | |
1414 | adev->family = AMDGPU_FAMILY_AI; | |
460826e6 KW |
1415 | |
1416 | r = soc15_set_ip_blocks(adev); | |
1417 | if (r) | |
1418 | return r; | |
1419 | break; | |
d38ceaf9 AD |
1420 | default: |
1421 | /* FIXME: not supported yet */ | |
1422 | return -EINVAL; | |
1423 | } | |
1424 | ||
e2a75f88 AD |
1425 | r = amdgpu_device_parse_gpu_info_fw(adev); |
1426 | if (r) | |
1427 | return r; | |
1428 | ||
1884734a | 1429 | amdgpu_amdkfd_device_probe(adev); |
1430 | ||
3149d9da XY |
1431 | if (amdgpu_sriov_vf(adev)) { |
1432 | r = amdgpu_virt_request_full_gpu(adev, true); | |
1433 | if (r) | |
5ffa61c1 | 1434 | return -EAGAIN; |
3149d9da XY |
1435 | } |
1436 | ||
d38ceaf9 AD |
1437 | for (i = 0; i < adev->num_ip_blocks; i++) { |
1438 | if ((amdgpu_ip_block_mask & (1 << i)) == 0) { | |
ed8cf00c HR |
1439 | DRM_ERROR("disabled ip block: %d <%s>\n", |
1440 | i, adev->ip_blocks[i].version->funcs->name); | |
a1255107 | 1441 | adev->ip_blocks[i].status.valid = false; |
d38ceaf9 | 1442 | } else { |
a1255107 AD |
1443 | if (adev->ip_blocks[i].version->funcs->early_init) { |
1444 | r = adev->ip_blocks[i].version->funcs->early_init((void *)adev); | |
2c1a2784 | 1445 | if (r == -ENOENT) { |
a1255107 | 1446 | adev->ip_blocks[i].status.valid = false; |
2c1a2784 | 1447 | } else if (r) { |
a1255107 AD |
1448 | DRM_ERROR("early_init of IP block <%s> failed %d\n", |
1449 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 1450 | return r; |
2c1a2784 | 1451 | } else { |
a1255107 | 1452 | adev->ip_blocks[i].status.valid = true; |
2c1a2784 | 1453 | } |
974e6b64 | 1454 | } else { |
a1255107 | 1455 | adev->ip_blocks[i].status.valid = true; |
d38ceaf9 | 1456 | } |
d38ceaf9 AD |
1457 | } |
1458 | } | |
1459 | ||
395d1fb9 NH |
1460 | adev->cg_flags &= amdgpu_cg_mask; |
1461 | adev->pg_flags &= amdgpu_pg_mask; | |
1462 | ||
d38ceaf9 AD |
1463 | return 0; |
1464 | } | |
1465 | ||
06ec9070 | 1466 | static int amdgpu_device_ip_init(struct amdgpu_device *adev) |
d38ceaf9 AD |
1467 | { |
1468 | int i, r; | |
1469 | ||
1470 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1471 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 1472 | continue; |
a1255107 | 1473 | r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev); |
2c1a2784 | 1474 | if (r) { |
a1255107 AD |
1475 | DRM_ERROR("sw_init of IP block <%s> failed %d\n", |
1476 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 1477 | return r; |
2c1a2784 | 1478 | } |
a1255107 | 1479 | adev->ip_blocks[i].status.sw = true; |
d38ceaf9 | 1480 | /* need to do gmc hw init early so we can allocate gpu mem */ |
a1255107 | 1481 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { |
06ec9070 | 1482 | r = amdgpu_device_vram_scratch_init(adev); |
2c1a2784 AD |
1483 | if (r) { |
1484 | DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r); | |
d38ceaf9 | 1485 | return r; |
2c1a2784 | 1486 | } |
a1255107 | 1487 | r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); |
2c1a2784 AD |
1488 | if (r) { |
1489 | DRM_ERROR("hw_init %d failed %d\n", i, r); | |
d38ceaf9 | 1490 | return r; |
2c1a2784 | 1491 | } |
06ec9070 | 1492 | r = amdgpu_device_wb_init(adev); |
2c1a2784 | 1493 | if (r) { |
06ec9070 | 1494 | DRM_ERROR("amdgpu_device_wb_init failed %d\n", r); |
d38ceaf9 | 1495 | return r; |
2c1a2784 | 1496 | } |
a1255107 | 1497 | adev->ip_blocks[i].status.hw = true; |
2493664f ML |
1498 | |
1499 | /* right after GMC hw init, we create CSA */ | |
1500 | if (amdgpu_sriov_vf(adev)) { | |
1501 | r = amdgpu_allocate_static_csa(adev); | |
1502 | if (r) { | |
1503 | DRM_ERROR("allocate CSA failed %d\n", r); | |
1504 | return r; | |
1505 | } | |
1506 | } | |
d38ceaf9 AD |
1507 | } |
1508 | } | |
1509 | ||
1510 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1511 | if (!adev->ip_blocks[i].status.sw) |
d38ceaf9 AD |
1512 | continue; |
1513 | /* gmc hw init is done early */ | |
a1255107 | 1514 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) |
d38ceaf9 | 1515 | continue; |
a1255107 | 1516 | r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev); |
2c1a2784 | 1517 | if (r) { |
a1255107 AD |
1518 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", |
1519 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 1520 | return r; |
2c1a2784 | 1521 | } |
a1255107 | 1522 | adev->ip_blocks[i].status.hw = true; |
d38ceaf9 AD |
1523 | } |
1524 | ||
1884734a | 1525 | amdgpu_amdkfd_device_init(adev); |
c6332b97 | 1526 | |
1527 | if (amdgpu_sriov_vf(adev)) | |
1528 | amdgpu_virt_release_full_gpu(adev, true); | |
1529 | ||
d38ceaf9 AD |
1530 | return 0; |
1531 | } | |
1532 | ||
06ec9070 | 1533 | static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev) |
0c49e0b8 CZ |
1534 | { |
1535 | memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM); | |
1536 | } | |
1537 | ||
06ec9070 | 1538 | static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev) |
0c49e0b8 CZ |
1539 | { |
1540 | return !!memcmp(adev->gart.ptr, adev->reset_magic, | |
1541 | AMDGPU_RESET_MAGIC_NUM); | |
1542 | } | |
1543 | ||
06ec9070 | 1544 | static int amdgpu_device_ip_late_set_cg_state(struct amdgpu_device *adev) |
d38ceaf9 AD |
1545 | { |
1546 | int i = 0, r; | |
1547 | ||
1548 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1549 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 1550 | continue; |
4a446d55 | 1551 | /* skip CG for VCE/UVD, it's handled specially */ |
a1255107 AD |
1552 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && |
1553 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE) { | |
4a446d55 | 1554 | /* enable clockgating to save power */ |
a1255107 AD |
1555 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev, |
1556 | AMD_CG_STATE_GATE); | |
4a446d55 AD |
1557 | if (r) { |
1558 | DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n", | |
a1255107 | 1559 | adev->ip_blocks[i].version->funcs->name, r); |
4a446d55 AD |
1560 | return r; |
1561 | } | |
b0b00ff1 | 1562 | } |
d38ceaf9 | 1563 | } |
2dc80b00 S |
1564 | return 0; |
1565 | } | |
1566 | ||
06ec9070 | 1567 | static int amdgpu_device_ip_late_init(struct amdgpu_device *adev) |
2dc80b00 S |
1568 | { |
1569 | int i = 0, r; | |
1570 | ||
1571 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1572 | if (!adev->ip_blocks[i].status.valid) | |
1573 | continue; | |
1574 | if (adev->ip_blocks[i].version->funcs->late_init) { | |
1575 | r = adev->ip_blocks[i].version->funcs->late_init((void *)adev); | |
1576 | if (r) { | |
1577 | DRM_ERROR("late_init of IP block <%s> failed %d\n", | |
1578 | adev->ip_blocks[i].version->funcs->name, r); | |
1579 | return r; | |
1580 | } | |
1581 | adev->ip_blocks[i].status.late_initialized = true; | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | mod_delayed_work(system_wq, &adev->late_init_work, | |
1586 | msecs_to_jiffies(AMDGPU_RESUME_MS)); | |
d38ceaf9 | 1587 | |
06ec9070 | 1588 | amdgpu_device_fill_reset_magic(adev); |
d38ceaf9 AD |
1589 | |
1590 | return 0; | |
1591 | } | |
1592 | ||
06ec9070 | 1593 | static int amdgpu_device_ip_fini(struct amdgpu_device *adev) |
d38ceaf9 AD |
1594 | { |
1595 | int i, r; | |
1596 | ||
1884734a | 1597 | amdgpu_amdkfd_device_fini(adev); |
3e96dbfd AD |
1598 | /* need to disable SMC first */ |
1599 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1600 | if (!adev->ip_blocks[i].status.hw) |
3e96dbfd | 1601 | continue; |
a1255107 | 1602 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) { |
3e96dbfd | 1603 | /* ungate blocks before hw fini so that we can shutdown the blocks safely */ |
a1255107 AD |
1604 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev, |
1605 | AMD_CG_STATE_UNGATE); | |
3e96dbfd AD |
1606 | if (r) { |
1607 | DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n", | |
a1255107 | 1608 | adev->ip_blocks[i].version->funcs->name, r); |
3e96dbfd AD |
1609 | return r; |
1610 | } | |
a1255107 | 1611 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); |
3e96dbfd AD |
1612 | /* XXX handle errors */ |
1613 | if (r) { | |
1614 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", | |
a1255107 | 1615 | adev->ip_blocks[i].version->funcs->name, r); |
3e96dbfd | 1616 | } |
a1255107 | 1617 | adev->ip_blocks[i].status.hw = false; |
3e96dbfd AD |
1618 | break; |
1619 | } | |
1620 | } | |
1621 | ||
d38ceaf9 | 1622 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 1623 | if (!adev->ip_blocks[i].status.hw) |
d38ceaf9 | 1624 | continue; |
a1255107 | 1625 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) { |
84e5b516 | 1626 | amdgpu_free_static_csa(adev); |
06ec9070 AD |
1627 | amdgpu_device_wb_fini(adev); |
1628 | amdgpu_device_vram_scratch_fini(adev); | |
d38ceaf9 | 1629 | } |
8201a67a RZ |
1630 | |
1631 | if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD && | |
1632 | adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE) { | |
1633 | /* ungate blocks before hw fini so that we can shutdown the blocks safely */ | |
1634 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev, | |
1635 | AMD_CG_STATE_UNGATE); | |
1636 | if (r) { | |
1637 | DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n", | |
1638 | adev->ip_blocks[i].version->funcs->name, r); | |
1639 | return r; | |
1640 | } | |
2c1a2784 | 1641 | } |
8201a67a | 1642 | |
a1255107 | 1643 | r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev); |
d38ceaf9 | 1644 | /* XXX handle errors */ |
2c1a2784 | 1645 | if (r) { |
a1255107 AD |
1646 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", |
1647 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 1648 | } |
8201a67a | 1649 | |
a1255107 | 1650 | adev->ip_blocks[i].status.hw = false; |
d38ceaf9 AD |
1651 | } |
1652 | ||
1653 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
a1255107 | 1654 | if (!adev->ip_blocks[i].status.sw) |
d38ceaf9 | 1655 | continue; |
a1255107 | 1656 | r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev); |
d38ceaf9 | 1657 | /* XXX handle errors */ |
2c1a2784 | 1658 | if (r) { |
a1255107 AD |
1659 | DRM_DEBUG("sw_fini of IP block <%s> failed %d\n", |
1660 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 1661 | } |
a1255107 AD |
1662 | adev->ip_blocks[i].status.sw = false; |
1663 | adev->ip_blocks[i].status.valid = false; | |
d38ceaf9 AD |
1664 | } |
1665 | ||
a6dcfd9c | 1666 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 1667 | if (!adev->ip_blocks[i].status.late_initialized) |
8a2eef1d | 1668 | continue; |
a1255107 AD |
1669 | if (adev->ip_blocks[i].version->funcs->late_fini) |
1670 | adev->ip_blocks[i].version->funcs->late_fini((void *)adev); | |
1671 | adev->ip_blocks[i].status.late_initialized = false; | |
a6dcfd9c ML |
1672 | } |
1673 | ||
030308fc | 1674 | if (amdgpu_sriov_vf(adev)) |
24136135 ML |
1675 | if (amdgpu_virt_release_full_gpu(adev, false)) |
1676 | DRM_ERROR("failed to release exclusive mode on fini\n"); | |
2493664f | 1677 | |
d38ceaf9 AD |
1678 | return 0; |
1679 | } | |
1680 | ||
06ec9070 | 1681 | static void amdgpu_device_ip_late_init_func_handler(struct work_struct *work) |
2dc80b00 S |
1682 | { |
1683 | struct amdgpu_device *adev = | |
1684 | container_of(work, struct amdgpu_device, late_init_work.work); | |
06ec9070 | 1685 | amdgpu_device_ip_late_set_cg_state(adev); |
2dc80b00 S |
1686 | } |
1687 | ||
cdd61df6 | 1688 | int amdgpu_device_ip_suspend(struct amdgpu_device *adev) |
d38ceaf9 AD |
1689 | { |
1690 | int i, r; | |
1691 | ||
e941ea99 XY |
1692 | if (amdgpu_sriov_vf(adev)) |
1693 | amdgpu_virt_request_full_gpu(adev, false); | |
1694 | ||
c5a93a28 FC |
1695 | /* ungate SMC block first */ |
1696 | r = amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_SMC, | |
1697 | AMD_CG_STATE_UNGATE); | |
1698 | if (r) { | |
1699 | DRM_ERROR("set_clockgating_state(ungate) SMC failed %d\n",r); | |
1700 | } | |
1701 | ||
d38ceaf9 | 1702 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
a1255107 | 1703 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 AD |
1704 | continue; |
1705 | /* ungate blocks so that suspend can properly shut them down */ | |
c5a93a28 | 1706 | if (i != AMD_IP_BLOCK_TYPE_SMC) { |
a1255107 AD |
1707 | r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev, |
1708 | AMD_CG_STATE_UNGATE); | |
c5a93a28 | 1709 | if (r) { |
a1255107 AD |
1710 | DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n", |
1711 | adev->ip_blocks[i].version->funcs->name, r); | |
c5a93a28 | 1712 | } |
2c1a2784 | 1713 | } |
d38ceaf9 | 1714 | /* XXX handle errors */ |
a1255107 | 1715 | r = adev->ip_blocks[i].version->funcs->suspend(adev); |
d38ceaf9 | 1716 | /* XXX handle errors */ |
2c1a2784 | 1717 | if (r) { |
a1255107 AD |
1718 | DRM_ERROR("suspend of IP block <%s> failed %d\n", |
1719 | adev->ip_blocks[i].version->funcs->name, r); | |
2c1a2784 | 1720 | } |
d38ceaf9 AD |
1721 | } |
1722 | ||
e941ea99 XY |
1723 | if (amdgpu_sriov_vf(adev)) |
1724 | amdgpu_virt_release_full_gpu(adev, false); | |
1725 | ||
d38ceaf9 AD |
1726 | return 0; |
1727 | } | |
1728 | ||
06ec9070 | 1729 | static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
1730 | { |
1731 | int i, r; | |
1732 | ||
2cb681b6 ML |
1733 | static enum amd_ip_block_type ip_order[] = { |
1734 | AMD_IP_BLOCK_TYPE_GMC, | |
1735 | AMD_IP_BLOCK_TYPE_COMMON, | |
2cb681b6 ML |
1736 | AMD_IP_BLOCK_TYPE_IH, |
1737 | }; | |
a90ad3c2 | 1738 | |
2cb681b6 ML |
1739 | for (i = 0; i < ARRAY_SIZE(ip_order); i++) { |
1740 | int j; | |
1741 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 1742 | |
2cb681b6 ML |
1743 | for (j = 0; j < adev->num_ip_blocks; j++) { |
1744 | block = &adev->ip_blocks[j]; | |
1745 | ||
1746 | if (block->version->type != ip_order[i] || | |
1747 | !block->status.valid) | |
1748 | continue; | |
1749 | ||
1750 | r = block->version->funcs->hw_init(adev); | |
1751 | DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"successed"); | |
a90ad3c2 ML |
1752 | } |
1753 | } | |
1754 | ||
1755 | return 0; | |
1756 | } | |
1757 | ||
06ec9070 | 1758 | static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev) |
a90ad3c2 ML |
1759 | { |
1760 | int i, r; | |
1761 | ||
2cb681b6 ML |
1762 | static enum amd_ip_block_type ip_order[] = { |
1763 | AMD_IP_BLOCK_TYPE_SMC, | |
ef4c166d | 1764 | AMD_IP_BLOCK_TYPE_PSP, |
2cb681b6 ML |
1765 | AMD_IP_BLOCK_TYPE_DCE, |
1766 | AMD_IP_BLOCK_TYPE_GFX, | |
1767 | AMD_IP_BLOCK_TYPE_SDMA, | |
257deb8c FM |
1768 | AMD_IP_BLOCK_TYPE_UVD, |
1769 | AMD_IP_BLOCK_TYPE_VCE | |
2cb681b6 | 1770 | }; |
a90ad3c2 | 1771 | |
2cb681b6 ML |
1772 | for (i = 0; i < ARRAY_SIZE(ip_order); i++) { |
1773 | int j; | |
1774 | struct amdgpu_ip_block *block; | |
a90ad3c2 | 1775 | |
2cb681b6 ML |
1776 | for (j = 0; j < adev->num_ip_blocks; j++) { |
1777 | block = &adev->ip_blocks[j]; | |
1778 | ||
1779 | if (block->version->type != ip_order[i] || | |
1780 | !block->status.valid) | |
1781 | continue; | |
1782 | ||
1783 | r = block->version->funcs->hw_init(adev); | |
1784 | DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"successed"); | |
a90ad3c2 ML |
1785 | } |
1786 | } | |
1787 | ||
1788 | return 0; | |
1789 | } | |
1790 | ||
06ec9070 | 1791 | static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev) |
d38ceaf9 AD |
1792 | { |
1793 | int i, r; | |
1794 | ||
a90ad3c2 ML |
1795 | for (i = 0; i < adev->num_ip_blocks; i++) { |
1796 | if (!adev->ip_blocks[i].status.valid) | |
1797 | continue; | |
a90ad3c2 ML |
1798 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
1799 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || | |
fcf0649f CZ |
1800 | adev->ip_blocks[i].version->type == |
1801 | AMD_IP_BLOCK_TYPE_IH) { | |
1802 | r = adev->ip_blocks[i].version->funcs->resume(adev); | |
1803 | if (r) { | |
1804 | DRM_ERROR("resume of IP block <%s> failed %d\n", | |
1805 | adev->ip_blocks[i].version->funcs->name, r); | |
1806 | return r; | |
1807 | } | |
a90ad3c2 ML |
1808 | } |
1809 | } | |
1810 | ||
1811 | return 0; | |
1812 | } | |
1813 | ||
06ec9070 | 1814 | static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev) |
d38ceaf9 AD |
1815 | { |
1816 | int i, r; | |
1817 | ||
1818 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 1819 | if (!adev->ip_blocks[i].status.valid) |
d38ceaf9 | 1820 | continue; |
fcf0649f CZ |
1821 | if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON || |
1822 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC || | |
1823 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH ) | |
1824 | continue; | |
a1255107 | 1825 | r = adev->ip_blocks[i].version->funcs->resume(adev); |
2c1a2784 | 1826 | if (r) { |
a1255107 AD |
1827 | DRM_ERROR("resume of IP block <%s> failed %d\n", |
1828 | adev->ip_blocks[i].version->funcs->name, r); | |
d38ceaf9 | 1829 | return r; |
2c1a2784 | 1830 | } |
d38ceaf9 AD |
1831 | } |
1832 | ||
1833 | return 0; | |
1834 | } | |
1835 | ||
06ec9070 | 1836 | static int amdgpu_device_ip_resume(struct amdgpu_device *adev) |
fcf0649f CZ |
1837 | { |
1838 | int r; | |
1839 | ||
06ec9070 | 1840 | r = amdgpu_device_ip_resume_phase1(adev); |
fcf0649f CZ |
1841 | if (r) |
1842 | return r; | |
06ec9070 | 1843 | r = amdgpu_device_ip_resume_phase2(adev); |
fcf0649f CZ |
1844 | |
1845 | return r; | |
1846 | } | |
1847 | ||
4e99a44e | 1848 | static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev) |
048765ad | 1849 | { |
6867e1b5 ML |
1850 | if (amdgpu_sriov_vf(adev)) { |
1851 | if (adev->is_atom_fw) { | |
1852 | if (amdgpu_atomfirmware_gpu_supports_virtualization(adev)) | |
1853 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; | |
1854 | } else { | |
1855 | if (amdgpu_atombios_has_gpu_virtualization_table(adev)) | |
1856 | adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS; | |
1857 | } | |
1858 | ||
1859 | if (!(adev->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS)) | |
1860 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0); | |
a5bde2f9 | 1861 | } |
048765ad AR |
1862 | } |
1863 | ||
4562236b HW |
1864 | bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type) |
1865 | { | |
1866 | switch (asic_type) { | |
1867 | #if defined(CONFIG_DRM_AMD_DC) | |
1868 | case CHIP_BONAIRE: | |
1869 | case CHIP_HAWAII: | |
0d6fbccb | 1870 | case CHIP_KAVERI: |
4562236b HW |
1871 | case CHIP_CARRIZO: |
1872 | case CHIP_STONEY: | |
1873 | case CHIP_POLARIS11: | |
1874 | case CHIP_POLARIS10: | |
2c8ad2d5 | 1875 | case CHIP_POLARIS12: |
4562236b HW |
1876 | case CHIP_TONGA: |
1877 | case CHIP_FIJI: | |
1878 | #if defined(CONFIG_DRM_AMD_DC_PRE_VEGA) | |
1879 | return amdgpu_dc != 0; | |
4562236b | 1880 | #endif |
17b7cf8c AD |
1881 | case CHIP_KABINI: |
1882 | case CHIP_MULLINS: | |
1883 | return amdgpu_dc > 0; | |
42f8ffa1 HW |
1884 | case CHIP_VEGA10: |
1885 | #if defined(CONFIG_DRM_AMD_DC_DCN1_0) | |
fd187853 | 1886 | case CHIP_RAVEN: |
42f8ffa1 | 1887 | #endif |
fd187853 | 1888 | return amdgpu_dc != 0; |
4562236b HW |
1889 | #endif |
1890 | default: | |
1891 | return false; | |
1892 | } | |
1893 | } | |
1894 | ||
1895 | /** | |
1896 | * amdgpu_device_has_dc_support - check if dc is supported | |
1897 | * | |
1898 | * @adev: amdgpu_device_pointer | |
1899 | * | |
1900 | * Returns true for supported, false for not supported | |
1901 | */ | |
1902 | bool amdgpu_device_has_dc_support(struct amdgpu_device *adev) | |
1903 | { | |
2555039d XY |
1904 | if (amdgpu_sriov_vf(adev)) |
1905 | return false; | |
1906 | ||
4562236b HW |
1907 | return amdgpu_device_asic_has_dc_support(adev->asic_type); |
1908 | } | |
1909 | ||
d38ceaf9 AD |
1910 | /** |
1911 | * amdgpu_device_init - initialize the driver | |
1912 | * | |
1913 | * @adev: amdgpu_device pointer | |
1914 | * @pdev: drm dev pointer | |
1915 | * @pdev: pci dev pointer | |
1916 | * @flags: driver flags | |
1917 | * | |
1918 | * Initializes the driver info and hw (all asics). | |
1919 | * Returns 0 for success or an error on failure. | |
1920 | * Called at driver startup. | |
1921 | */ | |
1922 | int amdgpu_device_init(struct amdgpu_device *adev, | |
1923 | struct drm_device *ddev, | |
1924 | struct pci_dev *pdev, | |
1925 | uint32_t flags) | |
1926 | { | |
1927 | int r, i; | |
1928 | bool runtime = false; | |
95844d20 | 1929 | u32 max_MBps; |
d38ceaf9 AD |
1930 | |
1931 | adev->shutdown = false; | |
1932 | adev->dev = &pdev->dev; | |
1933 | adev->ddev = ddev; | |
1934 | adev->pdev = pdev; | |
1935 | adev->flags = flags; | |
2f7d10b3 | 1936 | adev->asic_type = flags & AMD_ASIC_MASK; |
d38ceaf9 | 1937 | adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT; |
6f02a696 | 1938 | adev->mc.gart_size = 512 * 1024 * 1024; |
d38ceaf9 AD |
1939 | adev->accel_working = false; |
1940 | adev->num_rings = 0; | |
1941 | adev->mman.buffer_funcs = NULL; | |
1942 | adev->mman.buffer_funcs_ring = NULL; | |
1943 | adev->vm_manager.vm_pte_funcs = NULL; | |
2d55e45a | 1944 | adev->vm_manager.vm_pte_num_rings = 0; |
d38ceaf9 | 1945 | adev->gart.gart_funcs = NULL; |
f54d1867 | 1946 | adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS); |
b8866c26 | 1947 | bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); |
d38ceaf9 AD |
1948 | |
1949 | adev->smc_rreg = &amdgpu_invalid_rreg; | |
1950 | adev->smc_wreg = &amdgpu_invalid_wreg; | |
1951 | adev->pcie_rreg = &amdgpu_invalid_rreg; | |
1952 | adev->pcie_wreg = &amdgpu_invalid_wreg; | |
36b9a952 HR |
1953 | adev->pciep_rreg = &amdgpu_invalid_rreg; |
1954 | adev->pciep_wreg = &amdgpu_invalid_wreg; | |
d38ceaf9 AD |
1955 | adev->uvd_ctx_rreg = &amdgpu_invalid_rreg; |
1956 | adev->uvd_ctx_wreg = &amdgpu_invalid_wreg; | |
1957 | adev->didt_rreg = &amdgpu_invalid_rreg; | |
1958 | adev->didt_wreg = &amdgpu_invalid_wreg; | |
ccdbb20a RZ |
1959 | adev->gc_cac_rreg = &amdgpu_invalid_rreg; |
1960 | adev->gc_cac_wreg = &amdgpu_invalid_wreg; | |
d38ceaf9 AD |
1961 | adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg; |
1962 | adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg; | |
1963 | ||
3e39ab90 AD |
1964 | DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n", |
1965 | amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device, | |
1966 | pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision); | |
d38ceaf9 AD |
1967 | |
1968 | /* mutex initialization are all done here so we | |
1969 | * can recall function without having locking issues */ | |
d38ceaf9 | 1970 | atomic_set(&adev->irq.ih.lock, 0); |
0e5ca0d1 | 1971 | mutex_init(&adev->firmware.mutex); |
d38ceaf9 AD |
1972 | mutex_init(&adev->pm.mutex); |
1973 | mutex_init(&adev->gfx.gpu_clock_mutex); | |
1974 | mutex_init(&adev->srbm_mutex); | |
b8866c26 | 1975 | mutex_init(&adev->gfx.pipe_reserve_mutex); |
d38ceaf9 | 1976 | mutex_init(&adev->grbm_idx_mutex); |
d38ceaf9 | 1977 | mutex_init(&adev->mn_lock); |
e23b74aa | 1978 | mutex_init(&adev->virt.vf_errors.lock); |
d38ceaf9 | 1979 | hash_init(adev->mn_hash); |
13a752e3 | 1980 | mutex_init(&adev->lock_reset); |
d38ceaf9 | 1981 | |
06ec9070 | 1982 | amdgpu_device_check_arguments(adev); |
d38ceaf9 | 1983 | |
d38ceaf9 AD |
1984 | spin_lock_init(&adev->mmio_idx_lock); |
1985 | spin_lock_init(&adev->smc_idx_lock); | |
1986 | spin_lock_init(&adev->pcie_idx_lock); | |
1987 | spin_lock_init(&adev->uvd_ctx_idx_lock); | |
1988 | spin_lock_init(&adev->didt_idx_lock); | |
ccdbb20a | 1989 | spin_lock_init(&adev->gc_cac_idx_lock); |
16abb5d2 | 1990 | spin_lock_init(&adev->se_cac_idx_lock); |
d38ceaf9 | 1991 | spin_lock_init(&adev->audio_endpt_idx_lock); |
95844d20 | 1992 | spin_lock_init(&adev->mm_stats.lock); |
d38ceaf9 | 1993 | |
0c4e7fa5 CZ |
1994 | INIT_LIST_HEAD(&adev->shadow_list); |
1995 | mutex_init(&adev->shadow_list_lock); | |
1996 | ||
795f2813 AR |
1997 | INIT_LIST_HEAD(&adev->ring_lru_list); |
1998 | spin_lock_init(&adev->ring_lru_list_lock); | |
1999 | ||
06ec9070 AD |
2000 | INIT_DELAYED_WORK(&adev->late_init_work, |
2001 | amdgpu_device_ip_late_init_func_handler); | |
2dc80b00 | 2002 | |
0fa49558 AX |
2003 | /* Registers mapping */ |
2004 | /* TODO: block userspace mapping of io register */ | |
da69c161 KW |
2005 | if (adev->asic_type >= CHIP_BONAIRE) { |
2006 | adev->rmmio_base = pci_resource_start(adev->pdev, 5); | |
2007 | adev->rmmio_size = pci_resource_len(adev->pdev, 5); | |
2008 | } else { | |
2009 | adev->rmmio_base = pci_resource_start(adev->pdev, 2); | |
2010 | adev->rmmio_size = pci_resource_len(adev->pdev, 2); | |
2011 | } | |
d38ceaf9 | 2012 | |
d38ceaf9 AD |
2013 | adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size); |
2014 | if (adev->rmmio == NULL) { | |
2015 | return -ENOMEM; | |
2016 | } | |
2017 | DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base); | |
2018 | DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size); | |
2019 | ||
705e519e | 2020 | /* doorbell bar mapping */ |
06ec9070 | 2021 | amdgpu_device_doorbell_init(adev); |
d38ceaf9 AD |
2022 | |
2023 | /* io port mapping */ | |
2024 | for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { | |
2025 | if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) { | |
2026 | adev->rio_mem_size = pci_resource_len(adev->pdev, i); | |
2027 | adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size); | |
2028 | break; | |
2029 | } | |
2030 | } | |
2031 | if (adev->rio_mem == NULL) | |
b64a18c5 | 2032 | DRM_INFO("PCI I/O BAR is not found.\n"); |
d38ceaf9 AD |
2033 | |
2034 | /* early init functions */ | |
06ec9070 | 2035 | r = amdgpu_device_ip_early_init(adev); |
d38ceaf9 AD |
2036 | if (r) |
2037 | return r; | |
2038 | ||
2039 | /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */ | |
2040 | /* this will fail for cards that aren't VGA class devices, just | |
2041 | * ignore it */ | |
06ec9070 | 2042 | vga_client_register(adev->pdev, adev, NULL, amdgpu_device_vga_set_decode); |
d38ceaf9 AD |
2043 | |
2044 | if (amdgpu_runtime_pm == 1) | |
2045 | runtime = true; | |
e9bef455 | 2046 | if (amdgpu_device_is_px(ddev)) |
d38ceaf9 | 2047 | runtime = true; |
84c8b22e LW |
2048 | if (!pci_is_thunderbolt_attached(adev->pdev)) |
2049 | vga_switcheroo_register_client(adev->pdev, | |
2050 | &amdgpu_switcheroo_ops, runtime); | |
d38ceaf9 AD |
2051 | if (runtime) |
2052 | vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain); | |
2053 | ||
2054 | /* Read BIOS */ | |
83ba126a AD |
2055 | if (!amdgpu_get_bios(adev)) { |
2056 | r = -EINVAL; | |
2057 | goto failed; | |
2058 | } | |
f7e9e9fe | 2059 | |
d38ceaf9 | 2060 | r = amdgpu_atombios_init(adev); |
2c1a2784 AD |
2061 | if (r) { |
2062 | dev_err(adev->dev, "amdgpu_atombios_init failed\n"); | |
e23b74aa | 2063 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0); |
83ba126a | 2064 | goto failed; |
2c1a2784 | 2065 | } |
d38ceaf9 | 2066 | |
4e99a44e ML |
2067 | /* detect if we are with an SRIOV vbios */ |
2068 | amdgpu_device_detect_sriov_bios(adev); | |
048765ad | 2069 | |
d38ceaf9 | 2070 | /* Post card if necessary */ |
91fe77eb | 2071 | if (amdgpu_need_post(adev)) { |
d38ceaf9 | 2072 | if (!adev->bios) { |
bec86378 | 2073 | dev_err(adev->dev, "no vBIOS found\n"); |
83ba126a AD |
2074 | r = -EINVAL; |
2075 | goto failed; | |
d38ceaf9 | 2076 | } |
bec86378 | 2077 | DRM_INFO("GPU posting now...\n"); |
4e99a44e ML |
2078 | r = amdgpu_atom_asic_init(adev->mode_info.atom_context); |
2079 | if (r) { | |
2080 | dev_err(adev->dev, "gpu post error!\n"); | |
2081 | goto failed; | |
2082 | } | |
d38ceaf9 AD |
2083 | } |
2084 | ||
88b64e95 AD |
2085 | if (adev->is_atom_fw) { |
2086 | /* Initialize clocks */ | |
2087 | r = amdgpu_atomfirmware_get_clock_info(adev); | |
2088 | if (r) { | |
2089 | dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n"); | |
e23b74aa | 2090 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); |
88b64e95 AD |
2091 | goto failed; |
2092 | } | |
2093 | } else { | |
a5bde2f9 AD |
2094 | /* Initialize clocks */ |
2095 | r = amdgpu_atombios_get_clock_info(adev); | |
2096 | if (r) { | |
2097 | dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n"); | |
e23b74aa | 2098 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0); |
89041940 | 2099 | goto failed; |
a5bde2f9 AD |
2100 | } |
2101 | /* init i2c buses */ | |
4562236b HW |
2102 | if (!amdgpu_device_has_dc_support(adev)) |
2103 | amdgpu_atombios_i2c_init(adev); | |
2c1a2784 | 2104 | } |
d38ceaf9 AD |
2105 | |
2106 | /* Fence driver */ | |
2107 | r = amdgpu_fence_driver_init(adev); | |
2c1a2784 AD |
2108 | if (r) { |
2109 | dev_err(adev->dev, "amdgpu_fence_driver_init failed\n"); | |
e23b74aa | 2110 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0); |
83ba126a | 2111 | goto failed; |
2c1a2784 | 2112 | } |
d38ceaf9 AD |
2113 | |
2114 | /* init the mode config */ | |
2115 | drm_mode_config_init(adev->ddev); | |
2116 | ||
06ec9070 | 2117 | r = amdgpu_device_ip_init(adev); |
d38ceaf9 | 2118 | if (r) { |
8840a387 | 2119 | /* failed in exclusive mode due to timeout */ |
2120 | if (amdgpu_sriov_vf(adev) && | |
2121 | !amdgpu_sriov_runtime(adev) && | |
2122 | amdgpu_virt_mmio_blocked(adev) && | |
2123 | !amdgpu_virt_wait_reset(adev)) { | |
2124 | dev_err(adev->dev, "VF exclusive mode timeout\n"); | |
1daee8b4 PD |
2125 | /* Don't send request since VF is inactive. */ |
2126 | adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME; | |
2127 | adev->virt.ops = NULL; | |
8840a387 | 2128 | r = -EAGAIN; |
2129 | goto failed; | |
2130 | } | |
06ec9070 | 2131 | dev_err(adev->dev, "amdgpu_device_ip_init failed\n"); |
e23b74aa | 2132 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0); |
06ec9070 | 2133 | amdgpu_device_ip_fini(adev); |
83ba126a | 2134 | goto failed; |
d38ceaf9 AD |
2135 | } |
2136 | ||
2137 | adev->accel_working = true; | |
2138 | ||
e59c0205 AX |
2139 | amdgpu_vm_check_compute_bug(adev); |
2140 | ||
95844d20 MO |
2141 | /* Initialize the buffer migration limit. */ |
2142 | if (amdgpu_moverate >= 0) | |
2143 | max_MBps = amdgpu_moverate; | |
2144 | else | |
2145 | max_MBps = 8; /* Allow 8 MB/s. */ | |
2146 | /* Get a log2 for easy divisions. */ | |
2147 | adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps)); | |
2148 | ||
d38ceaf9 AD |
2149 | r = amdgpu_ib_pool_init(adev); |
2150 | if (r) { | |
2151 | dev_err(adev->dev, "IB initialization failed (%d).\n", r); | |
e23b74aa | 2152 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r); |
83ba126a | 2153 | goto failed; |
d38ceaf9 AD |
2154 | } |
2155 | ||
2156 | r = amdgpu_ib_ring_tests(adev); | |
2157 | if (r) | |
2158 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
2159 | ||
2dc8f81e HC |
2160 | if (amdgpu_sriov_vf(adev)) |
2161 | amdgpu_virt_init_data_exchange(adev); | |
2162 | ||
9bc92b9c ML |
2163 | amdgpu_fbdev_init(adev); |
2164 | ||
d2f52ac8 RZ |
2165 | r = amdgpu_pm_sysfs_init(adev); |
2166 | if (r) | |
2167 | DRM_ERROR("registering pm debugfs failed (%d).\n", r); | |
2168 | ||
75758255 | 2169 | r = amdgpu_debugfs_gem_init(adev); |
3f14e623 | 2170 | if (r) |
d38ceaf9 | 2171 | DRM_ERROR("registering gem debugfs failed (%d).\n", r); |
d38ceaf9 AD |
2172 | |
2173 | r = amdgpu_debugfs_regs_init(adev); | |
3f14e623 | 2174 | if (r) |
d38ceaf9 | 2175 | DRM_ERROR("registering register debugfs failed (%d).\n", r); |
d38ceaf9 | 2176 | |
50ab2533 | 2177 | r = amdgpu_debugfs_firmware_init(adev); |
3f14e623 | 2178 | if (r) |
50ab2533 | 2179 | DRM_ERROR("registering firmware debugfs failed (%d).\n", r); |
50ab2533 | 2180 | |
763efb6c | 2181 | r = amdgpu_debugfs_init(adev); |
db95e218 | 2182 | if (r) |
763efb6c | 2183 | DRM_ERROR("Creating debugfs files failed (%d).\n", r); |
db95e218 | 2184 | |
d38ceaf9 AD |
2185 | if ((amdgpu_testing & 1)) { |
2186 | if (adev->accel_working) | |
2187 | amdgpu_test_moves(adev); | |
2188 | else | |
2189 | DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n"); | |
2190 | } | |
d38ceaf9 AD |
2191 | if (amdgpu_benchmarking) { |
2192 | if (adev->accel_working) | |
2193 | amdgpu_benchmark(adev, amdgpu_benchmarking); | |
2194 | else | |
2195 | DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n"); | |
2196 | } | |
2197 | ||
2198 | /* enable clockgating, etc. after ib tests, etc. since some blocks require | |
2199 | * explicit gating rather than handling it automatically. | |
2200 | */ | |
06ec9070 | 2201 | r = amdgpu_device_ip_late_init(adev); |
2c1a2784 | 2202 | if (r) { |
06ec9070 | 2203 | dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n"); |
e23b74aa | 2204 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r); |
83ba126a | 2205 | goto failed; |
2c1a2784 | 2206 | } |
d38ceaf9 AD |
2207 | |
2208 | return 0; | |
83ba126a AD |
2209 | |
2210 | failed: | |
89041940 | 2211 | amdgpu_vf_error_trans_all(adev); |
83ba126a AD |
2212 | if (runtime) |
2213 | vga_switcheroo_fini_domain_pm_ops(adev->dev); | |
8840a387 | 2214 | |
83ba126a | 2215 | return r; |
d38ceaf9 AD |
2216 | } |
2217 | ||
d38ceaf9 AD |
2218 | /** |
2219 | * amdgpu_device_fini - tear down the driver | |
2220 | * | |
2221 | * @adev: amdgpu_device pointer | |
2222 | * | |
2223 | * Tear down the driver info (all asics). | |
2224 | * Called at driver shutdown. | |
2225 | */ | |
2226 | void amdgpu_device_fini(struct amdgpu_device *adev) | |
2227 | { | |
2228 | int r; | |
2229 | ||
2230 | DRM_INFO("amdgpu: finishing device.\n"); | |
2231 | adev->shutdown = true; | |
db2c2a97 PD |
2232 | if (adev->mode_info.mode_config_initialized) |
2233 | drm_crtc_force_disable_all(adev->ddev); | |
b9141cd3 | 2234 | |
d38ceaf9 AD |
2235 | amdgpu_ib_pool_fini(adev); |
2236 | amdgpu_fence_driver_fini(adev); | |
2237 | amdgpu_fbdev_fini(adev); | |
06ec9070 | 2238 | r = amdgpu_device_ip_fini(adev); |
ab4fe3e1 HR |
2239 | if (adev->firmware.gpu_info_fw) { |
2240 | release_firmware(adev->firmware.gpu_info_fw); | |
2241 | adev->firmware.gpu_info_fw = NULL; | |
2242 | } | |
d38ceaf9 | 2243 | adev->accel_working = false; |
2dc80b00 | 2244 | cancel_delayed_work_sync(&adev->late_init_work); |
d38ceaf9 | 2245 | /* free i2c buses */ |
4562236b HW |
2246 | if (!amdgpu_device_has_dc_support(adev)) |
2247 | amdgpu_i2c_fini(adev); | |
d38ceaf9 AD |
2248 | amdgpu_atombios_fini(adev); |
2249 | kfree(adev->bios); | |
2250 | adev->bios = NULL; | |
84c8b22e LW |
2251 | if (!pci_is_thunderbolt_attached(adev->pdev)) |
2252 | vga_switcheroo_unregister_client(adev->pdev); | |
83ba126a AD |
2253 | if (adev->flags & AMD_IS_PX) |
2254 | vga_switcheroo_fini_domain_pm_ops(adev->dev); | |
d38ceaf9 AD |
2255 | vga_client_register(adev->pdev, NULL, NULL, NULL); |
2256 | if (adev->rio_mem) | |
2257 | pci_iounmap(adev->pdev, adev->rio_mem); | |
2258 | adev->rio_mem = NULL; | |
2259 | iounmap(adev->rmmio); | |
2260 | adev->rmmio = NULL; | |
06ec9070 | 2261 | amdgpu_device_doorbell_fini(adev); |
d2f52ac8 | 2262 | amdgpu_pm_sysfs_fini(adev); |
d38ceaf9 | 2263 | amdgpu_debugfs_regs_cleanup(adev); |
d38ceaf9 AD |
2264 | } |
2265 | ||
2266 | ||
2267 | /* | |
2268 | * Suspend & resume. | |
2269 | */ | |
2270 | /** | |
810ddc3a | 2271 | * amdgpu_device_suspend - initiate device suspend |
d38ceaf9 AD |
2272 | * |
2273 | * @pdev: drm dev pointer | |
2274 | * @state: suspend state | |
2275 | * | |
2276 | * Puts the hw in the suspend state (all asics). | |
2277 | * Returns 0 for success or an error on failure. | |
2278 | * Called at driver suspend. | |
2279 | */ | |
810ddc3a | 2280 | int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon) |
d38ceaf9 AD |
2281 | { |
2282 | struct amdgpu_device *adev; | |
2283 | struct drm_crtc *crtc; | |
2284 | struct drm_connector *connector; | |
5ceb54c6 | 2285 | int r; |
d38ceaf9 AD |
2286 | |
2287 | if (dev == NULL || dev->dev_private == NULL) { | |
2288 | return -ENODEV; | |
2289 | } | |
2290 | ||
2291 | adev = dev->dev_private; | |
2292 | ||
2293 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
2294 | return 0; | |
2295 | ||
2296 | drm_kms_helper_poll_disable(dev); | |
2297 | ||
4562236b HW |
2298 | if (!amdgpu_device_has_dc_support(adev)) { |
2299 | /* turn off display hw */ | |
2300 | drm_modeset_lock_all(dev); | |
2301 | list_for_each_entry(connector, &dev->mode_config.connector_list, head) { | |
2302 | drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF); | |
2303 | } | |
2304 | drm_modeset_unlock_all(dev); | |
d38ceaf9 AD |
2305 | } |
2306 | ||
ba997709 YZ |
2307 | amdgpu_amdkfd_suspend(adev); |
2308 | ||
756e6880 | 2309 | /* unpin the front buffers and cursors */ |
d38ceaf9 | 2310 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { |
756e6880 | 2311 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); |
d38ceaf9 AD |
2312 | struct amdgpu_framebuffer *rfb = to_amdgpu_framebuffer(crtc->primary->fb); |
2313 | struct amdgpu_bo *robj; | |
2314 | ||
756e6880 AD |
2315 | if (amdgpu_crtc->cursor_bo) { |
2316 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); | |
7a6901d7 | 2317 | r = amdgpu_bo_reserve(aobj, true); |
756e6880 AD |
2318 | if (r == 0) { |
2319 | amdgpu_bo_unpin(aobj); | |
2320 | amdgpu_bo_unreserve(aobj); | |
2321 | } | |
2322 | } | |
2323 | ||
d38ceaf9 AD |
2324 | if (rfb == NULL || rfb->obj == NULL) { |
2325 | continue; | |
2326 | } | |
2327 | robj = gem_to_amdgpu_bo(rfb->obj); | |
2328 | /* don't unpin kernel fb objects */ | |
2329 | if (!amdgpu_fbdev_robj_is_fb(adev, robj)) { | |
7a6901d7 | 2330 | r = amdgpu_bo_reserve(robj, true); |
d38ceaf9 AD |
2331 | if (r == 0) { |
2332 | amdgpu_bo_unpin(robj); | |
2333 | amdgpu_bo_unreserve(robj); | |
2334 | } | |
2335 | } | |
2336 | } | |
2337 | /* evict vram memory */ | |
2338 | amdgpu_bo_evict_vram(adev); | |
2339 | ||
5ceb54c6 | 2340 | amdgpu_fence_driver_suspend(adev); |
d38ceaf9 | 2341 | |
cdd61df6 | 2342 | r = amdgpu_device_ip_suspend(adev); |
d38ceaf9 | 2343 | |
a0a71e49 AD |
2344 | /* evict remaining vram memory |
2345 | * This second call to evict vram is to evict the gart page table | |
2346 | * using the CPU. | |
2347 | */ | |
d38ceaf9 AD |
2348 | amdgpu_bo_evict_vram(adev); |
2349 | ||
2350 | pci_save_state(dev->pdev); | |
2351 | if (suspend) { | |
2352 | /* Shut down the device */ | |
2353 | pci_disable_device(dev->pdev); | |
2354 | pci_set_power_state(dev->pdev, PCI_D3hot); | |
74b0b157 | 2355 | } else { |
2356 | r = amdgpu_asic_reset(adev); | |
2357 | if (r) | |
2358 | DRM_ERROR("amdgpu asic reset failed\n"); | |
d38ceaf9 AD |
2359 | } |
2360 | ||
2361 | if (fbcon) { | |
2362 | console_lock(); | |
2363 | amdgpu_fbdev_set_suspend(adev, 1); | |
2364 | console_unlock(); | |
2365 | } | |
2366 | return 0; | |
2367 | } | |
2368 | ||
2369 | /** | |
810ddc3a | 2370 | * amdgpu_device_resume - initiate device resume |
d38ceaf9 AD |
2371 | * |
2372 | * @pdev: drm dev pointer | |
2373 | * | |
2374 | * Bring the hw back to operating state (all asics). | |
2375 | * Returns 0 for success or an error on failure. | |
2376 | * Called at driver resume. | |
2377 | */ | |
810ddc3a | 2378 | int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon) |
d38ceaf9 AD |
2379 | { |
2380 | struct drm_connector *connector; | |
2381 | struct amdgpu_device *adev = dev->dev_private; | |
756e6880 | 2382 | struct drm_crtc *crtc; |
03161a6e | 2383 | int r = 0; |
d38ceaf9 AD |
2384 | |
2385 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
2386 | return 0; | |
2387 | ||
74b0b157 | 2388 | if (fbcon) |
d38ceaf9 | 2389 | console_lock(); |
74b0b157 | 2390 | |
d38ceaf9 AD |
2391 | if (resume) { |
2392 | pci_set_power_state(dev->pdev, PCI_D0); | |
2393 | pci_restore_state(dev->pdev); | |
74b0b157 | 2394 | r = pci_enable_device(dev->pdev); |
03161a6e HR |
2395 | if (r) |
2396 | goto unlock; | |
d38ceaf9 AD |
2397 | } |
2398 | ||
2399 | /* post card */ | |
c836fec5 | 2400 | if (amdgpu_need_post(adev)) { |
74b0b157 | 2401 | r = amdgpu_atom_asic_init(adev->mode_info.atom_context); |
2402 | if (r) | |
2403 | DRM_ERROR("amdgpu asic init failed\n"); | |
2404 | } | |
d38ceaf9 | 2405 | |
06ec9070 | 2406 | r = amdgpu_device_ip_resume(adev); |
e6707218 | 2407 | if (r) { |
06ec9070 | 2408 | DRM_ERROR("amdgpu_device_ip_resume failed (%d).\n", r); |
03161a6e | 2409 | goto unlock; |
e6707218 | 2410 | } |
5ceb54c6 AD |
2411 | amdgpu_fence_driver_resume(adev); |
2412 | ||
ca198528 FC |
2413 | if (resume) { |
2414 | r = amdgpu_ib_ring_tests(adev); | |
2415 | if (r) | |
2416 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
2417 | } | |
d38ceaf9 | 2418 | |
06ec9070 | 2419 | r = amdgpu_device_ip_late_init(adev); |
03161a6e HR |
2420 | if (r) |
2421 | goto unlock; | |
d38ceaf9 | 2422 | |
756e6880 AD |
2423 | /* pin cursors */ |
2424 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
2425 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); | |
2426 | ||
2427 | if (amdgpu_crtc->cursor_bo) { | |
2428 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); | |
7a6901d7 | 2429 | r = amdgpu_bo_reserve(aobj, true); |
756e6880 AD |
2430 | if (r == 0) { |
2431 | r = amdgpu_bo_pin(aobj, | |
2432 | AMDGPU_GEM_DOMAIN_VRAM, | |
2433 | &amdgpu_crtc->cursor_addr); | |
2434 | if (r != 0) | |
2435 | DRM_ERROR("Failed to pin cursor BO (%d)\n", r); | |
2436 | amdgpu_bo_unreserve(aobj); | |
2437 | } | |
2438 | } | |
2439 | } | |
ba997709 YZ |
2440 | r = amdgpu_amdkfd_resume(adev); |
2441 | if (r) | |
2442 | return r; | |
756e6880 | 2443 | |
d38ceaf9 AD |
2444 | /* blat the mode back in */ |
2445 | if (fbcon) { | |
4562236b HW |
2446 | if (!amdgpu_device_has_dc_support(adev)) { |
2447 | /* pre DCE11 */ | |
2448 | drm_helper_resume_force_mode(dev); | |
2449 | ||
2450 | /* turn on display hw */ | |
2451 | drm_modeset_lock_all(dev); | |
2452 | list_for_each_entry(connector, &dev->mode_config.connector_list, head) { | |
2453 | drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON); | |
2454 | } | |
2455 | drm_modeset_unlock_all(dev); | |
2456 | } else { | |
2457 | /* | |
2458 | * There is no equivalent atomic helper to turn on | |
2459 | * display, so we defined our own function for this, | |
2460 | * once suspend resume is supported by the atomic | |
2461 | * framework this will be reworked | |
2462 | */ | |
2463 | amdgpu_dm_display_resume(adev); | |
d38ceaf9 AD |
2464 | } |
2465 | } | |
2466 | ||
2467 | drm_kms_helper_poll_enable(dev); | |
23a1a9e5 L |
2468 | |
2469 | /* | |
2470 | * Most of the connector probing functions try to acquire runtime pm | |
2471 | * refs to ensure that the GPU is powered on when connector polling is | |
2472 | * performed. Since we're calling this from a runtime PM callback, | |
2473 | * trying to acquire rpm refs will cause us to deadlock. | |
2474 | * | |
2475 | * Since we're guaranteed to be holding the rpm lock, it's safe to | |
2476 | * temporarily disable the rpm helpers so this doesn't deadlock us. | |
2477 | */ | |
2478 | #ifdef CONFIG_PM | |
2479 | dev->dev->power.disable_depth++; | |
2480 | #endif | |
4562236b HW |
2481 | if (!amdgpu_device_has_dc_support(adev)) |
2482 | drm_helper_hpd_irq_event(dev); | |
2483 | else | |
2484 | drm_kms_helper_hotplug_event(dev); | |
23a1a9e5 L |
2485 | #ifdef CONFIG_PM |
2486 | dev->dev->power.disable_depth--; | |
2487 | #endif | |
d38ceaf9 | 2488 | |
03161a6e | 2489 | if (fbcon) |
d38ceaf9 | 2490 | amdgpu_fbdev_set_suspend(adev, 0); |
03161a6e HR |
2491 | |
2492 | unlock: | |
2493 | if (fbcon) | |
d38ceaf9 | 2494 | console_unlock(); |
d38ceaf9 | 2495 | |
03161a6e | 2496 | return r; |
d38ceaf9 AD |
2497 | } |
2498 | ||
06ec9070 | 2499 | static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev) |
63fbf42f CZ |
2500 | { |
2501 | int i; | |
2502 | bool asic_hang = false; | |
2503 | ||
f993d628 ML |
2504 | if (amdgpu_sriov_vf(adev)) |
2505 | return true; | |
2506 | ||
63fbf42f | 2507 | for (i = 0; i < adev->num_ip_blocks; i++) { |
a1255107 | 2508 | if (!adev->ip_blocks[i].status.valid) |
63fbf42f | 2509 | continue; |
a1255107 AD |
2510 | if (adev->ip_blocks[i].version->funcs->check_soft_reset) |
2511 | adev->ip_blocks[i].status.hang = | |
2512 | adev->ip_blocks[i].version->funcs->check_soft_reset(adev); | |
2513 | if (adev->ip_blocks[i].status.hang) { | |
2514 | DRM_INFO("IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name); | |
63fbf42f CZ |
2515 | asic_hang = true; |
2516 | } | |
2517 | } | |
2518 | return asic_hang; | |
2519 | } | |
2520 | ||
06ec9070 | 2521 | static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev) |
d31a501e CZ |
2522 | { |
2523 | int i, r = 0; | |
2524 | ||
2525 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 2526 | if (!adev->ip_blocks[i].status.valid) |
d31a501e | 2527 | continue; |
a1255107 AD |
2528 | if (adev->ip_blocks[i].status.hang && |
2529 | adev->ip_blocks[i].version->funcs->pre_soft_reset) { | |
2530 | r = adev->ip_blocks[i].version->funcs->pre_soft_reset(adev); | |
d31a501e CZ |
2531 | if (r) |
2532 | return r; | |
2533 | } | |
2534 | } | |
2535 | ||
2536 | return 0; | |
2537 | } | |
2538 | ||
06ec9070 | 2539 | static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev) |
35d782fe | 2540 | { |
da146d3b AD |
2541 | int i; |
2542 | ||
2543 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 2544 | if (!adev->ip_blocks[i].status.valid) |
da146d3b | 2545 | continue; |
a1255107 AD |
2546 | if ((adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) || |
2547 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) || | |
2548 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_ACP) || | |
98512bb8 KW |
2549 | (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) || |
2550 | adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) { | |
a1255107 | 2551 | if (adev->ip_blocks[i].status.hang) { |
da146d3b AD |
2552 | DRM_INFO("Some block need full reset!\n"); |
2553 | return true; | |
2554 | } | |
2555 | } | |
35d782fe CZ |
2556 | } |
2557 | return false; | |
2558 | } | |
2559 | ||
06ec9070 | 2560 | static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
2561 | { |
2562 | int i, r = 0; | |
2563 | ||
2564 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 2565 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 2566 | continue; |
a1255107 AD |
2567 | if (adev->ip_blocks[i].status.hang && |
2568 | adev->ip_blocks[i].version->funcs->soft_reset) { | |
2569 | r = adev->ip_blocks[i].version->funcs->soft_reset(adev); | |
35d782fe CZ |
2570 | if (r) |
2571 | return r; | |
2572 | } | |
2573 | } | |
2574 | ||
2575 | return 0; | |
2576 | } | |
2577 | ||
06ec9070 | 2578 | static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev) |
35d782fe CZ |
2579 | { |
2580 | int i, r = 0; | |
2581 | ||
2582 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
a1255107 | 2583 | if (!adev->ip_blocks[i].status.valid) |
35d782fe | 2584 | continue; |
a1255107 AD |
2585 | if (adev->ip_blocks[i].status.hang && |
2586 | adev->ip_blocks[i].version->funcs->post_soft_reset) | |
2587 | r = adev->ip_blocks[i].version->funcs->post_soft_reset(adev); | |
35d782fe CZ |
2588 | if (r) |
2589 | return r; | |
2590 | } | |
2591 | ||
2592 | return 0; | |
2593 | } | |
2594 | ||
3ad81f16 CZ |
2595 | bool amdgpu_need_backup(struct amdgpu_device *adev) |
2596 | { | |
2597 | if (adev->flags & AMD_IS_APU) | |
2598 | return false; | |
2599 | ||
8854695a | 2600 | return amdgpu_gpu_recovery; |
3ad81f16 CZ |
2601 | } |
2602 | ||
06ec9070 AD |
2603 | static int amdgpu_device_recover_vram_from_shadow(struct amdgpu_device *adev, |
2604 | struct amdgpu_ring *ring, | |
2605 | struct amdgpu_bo *bo, | |
2606 | struct dma_fence **fence) | |
53cdccd5 CZ |
2607 | { |
2608 | uint32_t domain; | |
2609 | int r; | |
2610 | ||
23d2e504 RH |
2611 | if (!bo->shadow) |
2612 | return 0; | |
2613 | ||
1d284797 | 2614 | r = amdgpu_bo_reserve(bo, true); |
23d2e504 RH |
2615 | if (r) |
2616 | return r; | |
2617 | domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type); | |
2618 | /* if bo has been evicted, then no need to recover */ | |
2619 | if (domain == AMDGPU_GEM_DOMAIN_VRAM) { | |
82521316 RH |
2620 | r = amdgpu_bo_validate(bo->shadow); |
2621 | if (r) { | |
2622 | DRM_ERROR("bo validate failed!\n"); | |
2623 | goto err; | |
2624 | } | |
2625 | ||
23d2e504 | 2626 | r = amdgpu_bo_restore_from_shadow(adev, ring, bo, |
53cdccd5 | 2627 | NULL, fence, true); |
23d2e504 RH |
2628 | if (r) { |
2629 | DRM_ERROR("recover page table failed!\n"); | |
2630 | goto err; | |
2631 | } | |
2632 | } | |
53cdccd5 | 2633 | err: |
23d2e504 RH |
2634 | amdgpu_bo_unreserve(bo); |
2635 | return r; | |
53cdccd5 CZ |
2636 | } |
2637 | ||
5740682e | 2638 | /* |
06ec9070 | 2639 | * amdgpu_device_reset - reset ASIC/GPU for bare-metal or passthrough |
a90ad3c2 ML |
2640 | * |
2641 | * @adev: amdgpu device pointer | |
5740682e | 2642 | * @reset_flags: output param tells caller the reset result |
a90ad3c2 | 2643 | * |
5740682e ML |
2644 | * attempt to do soft-reset or full-reset and reinitialize Asic |
2645 | * return 0 means successed otherwise failed | |
2646 | */ | |
06ec9070 AD |
2647 | static int amdgpu_device_reset(struct amdgpu_device *adev, |
2648 | uint64_t* reset_flags) | |
a90ad3c2 | 2649 | { |
5740682e ML |
2650 | bool need_full_reset, vram_lost = 0; |
2651 | int r; | |
a90ad3c2 | 2652 | |
06ec9070 | 2653 | need_full_reset = amdgpu_device_ip_need_full_reset(adev); |
a90ad3c2 | 2654 | |
5740682e | 2655 | if (!need_full_reset) { |
06ec9070 AD |
2656 | amdgpu_device_ip_pre_soft_reset(adev); |
2657 | r = amdgpu_device_ip_soft_reset(adev); | |
2658 | amdgpu_device_ip_post_soft_reset(adev); | |
2659 | if (r || amdgpu_device_ip_check_soft_reset(adev)) { | |
5740682e ML |
2660 | DRM_INFO("soft reset failed, will fallback to full reset!\n"); |
2661 | need_full_reset = true; | |
2662 | } | |
a90ad3c2 | 2663 | |
5740682e | 2664 | } |
a90ad3c2 | 2665 | |
5740682e | 2666 | if (need_full_reset) { |
cdd61df6 | 2667 | r = amdgpu_device_ip_suspend(adev); |
a90ad3c2 | 2668 | |
5740682e | 2669 | retry: |
5740682e | 2670 | r = amdgpu_asic_reset(adev); |
5740682e ML |
2671 | /* post card */ |
2672 | amdgpu_atom_asic_init(adev->mode_info.atom_context); | |
65781c78 | 2673 | |
5740682e ML |
2674 | if (!r) { |
2675 | dev_info(adev->dev, "GPU reset succeeded, trying to resume\n"); | |
06ec9070 | 2676 | r = amdgpu_device_ip_resume_phase1(adev); |
5740682e ML |
2677 | if (r) |
2678 | goto out; | |
65781c78 | 2679 | |
06ec9070 | 2680 | vram_lost = amdgpu_device_check_vram_lost(adev); |
5740682e ML |
2681 | if (vram_lost) { |
2682 | DRM_ERROR("VRAM is lost!\n"); | |
2683 | atomic_inc(&adev->vram_lost_counter); | |
2684 | } | |
2685 | ||
c1c7ce8f CK |
2686 | r = amdgpu_gtt_mgr_recover( |
2687 | &adev->mman.bdev.man[TTM_PL_TT]); | |
5740682e ML |
2688 | if (r) |
2689 | goto out; | |
2690 | ||
06ec9070 | 2691 | r = amdgpu_device_ip_resume_phase2(adev); |
5740682e ML |
2692 | if (r) |
2693 | goto out; | |
2694 | ||
2695 | if (vram_lost) | |
06ec9070 | 2696 | amdgpu_device_fill_reset_magic(adev); |
65781c78 | 2697 | } |
5740682e | 2698 | } |
65781c78 | 2699 | |
5740682e ML |
2700 | out: |
2701 | if (!r) { | |
2702 | amdgpu_irq_gpu_reset_resume_helper(adev); | |
2703 | r = amdgpu_ib_ring_tests(adev); | |
2704 | if (r) { | |
2705 | dev_err(adev->dev, "ib ring test failed (%d).\n", r); | |
cdd61df6 | 2706 | r = amdgpu_device_ip_suspend(adev); |
5740682e ML |
2707 | need_full_reset = true; |
2708 | goto retry; | |
2709 | } | |
2710 | } | |
65781c78 | 2711 | |
5740682e ML |
2712 | if (reset_flags) { |
2713 | if (vram_lost) | |
2714 | (*reset_flags) |= AMDGPU_RESET_INFO_VRAM_LOST; | |
a90ad3c2 | 2715 | |
5740682e ML |
2716 | if (need_full_reset) |
2717 | (*reset_flags) |= AMDGPU_RESET_INFO_FULLRESET; | |
65781c78 | 2718 | } |
a90ad3c2 | 2719 | |
5740682e ML |
2720 | return r; |
2721 | } | |
a90ad3c2 | 2722 | |
5740682e | 2723 | /* |
06ec9070 | 2724 | * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf |
5740682e ML |
2725 | * |
2726 | * @adev: amdgpu device pointer | |
2727 | * @reset_flags: output param tells caller the reset result | |
2728 | * | |
2729 | * do VF FLR and reinitialize Asic | |
2730 | * return 0 means successed otherwise failed | |
2731 | */ | |
06ec9070 AD |
2732 | static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, |
2733 | uint64_t *reset_flags, | |
2734 | bool from_hypervisor) | |
5740682e ML |
2735 | { |
2736 | int r; | |
2737 | ||
2738 | if (from_hypervisor) | |
2739 | r = amdgpu_virt_request_full_gpu(adev, true); | |
2740 | else | |
2741 | r = amdgpu_virt_reset_gpu(adev); | |
2742 | if (r) | |
2743 | return r; | |
a90ad3c2 ML |
2744 | |
2745 | /* Resume IP prior to SMC */ | |
06ec9070 | 2746 | r = amdgpu_device_ip_reinit_early_sriov(adev); |
5740682e ML |
2747 | if (r) |
2748 | goto error; | |
a90ad3c2 ML |
2749 | |
2750 | /* we need recover gart prior to run SMC/CP/SDMA resume */ | |
c1c7ce8f | 2751 | amdgpu_gtt_mgr_recover(&adev->mman.bdev.man[TTM_PL_TT]); |
a90ad3c2 ML |
2752 | |
2753 | /* now we are okay to resume SMC/CP/SDMA */ | |
06ec9070 | 2754 | r = amdgpu_device_ip_reinit_late_sriov(adev); |
5740682e ML |
2755 | if (r) |
2756 | goto error; | |
a90ad3c2 ML |
2757 | |
2758 | amdgpu_irq_gpu_reset_resume_helper(adev); | |
5740682e ML |
2759 | r = amdgpu_ib_ring_tests(adev); |
2760 | if (r) | |
a90ad3c2 ML |
2761 | dev_err(adev->dev, "[GPU_RESET] ib ring test failed (%d).\n", r); |
2762 | ||
5740682e | 2763 | error: |
a90ad3c2 ML |
2764 | /* release full control of GPU after ib test */ |
2765 | amdgpu_virt_release_full_gpu(adev, true); | |
2766 | ||
5740682e | 2767 | if (reset_flags) { |
75bc6099 ML |
2768 | if (adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) { |
2769 | (*reset_flags) |= AMDGPU_RESET_INFO_VRAM_LOST; | |
2770 | atomic_inc(&adev->vram_lost_counter); | |
2771 | } | |
a90ad3c2 | 2772 | |
5740682e ML |
2773 | /* VF FLR or hotlink reset is always full-reset */ |
2774 | (*reset_flags) |= AMDGPU_RESET_INFO_FULLRESET; | |
a90ad3c2 ML |
2775 | } |
2776 | ||
2777 | return r; | |
2778 | } | |
2779 | ||
d38ceaf9 | 2780 | /** |
5740682e | 2781 | * amdgpu_gpu_recover - reset the asic and recover scheduler |
d38ceaf9 AD |
2782 | * |
2783 | * @adev: amdgpu device pointer | |
5740682e | 2784 | * @job: which job trigger hang |
dcebf026 | 2785 | * @force forces reset regardless of amdgpu_gpu_recovery |
d38ceaf9 | 2786 | * |
5740682e | 2787 | * Attempt to reset the GPU if it has hung (all asics). |
d38ceaf9 AD |
2788 | * Returns 0 for success or an error on failure. |
2789 | */ | |
dcebf026 | 2790 | int amdgpu_gpu_recover(struct amdgpu_device *adev, struct amdgpu_job *job, bool force) |
d38ceaf9 | 2791 | { |
4562236b | 2792 | struct drm_atomic_state *state = NULL; |
5740682e ML |
2793 | uint64_t reset_flags = 0; |
2794 | int i, r, resched; | |
fb140b29 | 2795 | |
06ec9070 | 2796 | if (!amdgpu_device_ip_check_soft_reset(adev)) { |
63fbf42f CZ |
2797 | DRM_INFO("No hardware hang detected. Did some blocks stall?\n"); |
2798 | return 0; | |
2799 | } | |
d38ceaf9 | 2800 | |
dcebf026 AG |
2801 | if (!force && (amdgpu_gpu_recovery == 0 || |
2802 | (amdgpu_gpu_recovery == -1 && !amdgpu_sriov_vf(adev)))) { | |
2803 | DRM_INFO("GPU recovery disabled.\n"); | |
2804 | return 0; | |
2805 | } | |
2806 | ||
5740682e ML |
2807 | dev_info(adev->dev, "GPU reset begin!\n"); |
2808 | ||
13a752e3 | 2809 | mutex_lock(&adev->lock_reset); |
d94aed5a | 2810 | atomic_inc(&adev->gpu_reset_counter); |
13a752e3 | 2811 | adev->in_gpu_reset = 1; |
d38ceaf9 | 2812 | |
a3c47d6b CZ |
2813 | /* block TTM */ |
2814 | resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev); | |
4562236b HW |
2815 | /* store modesetting */ |
2816 | if (amdgpu_device_has_dc_support(adev)) | |
2817 | state = drm_atomic_helper_suspend(adev->ddev); | |
a3c47d6b | 2818 | |
0875dc9e CZ |
2819 | /* block scheduler */ |
2820 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
2821 | struct amdgpu_ring *ring = adev->rings[i]; | |
2822 | ||
51687759 | 2823 | if (!ring || !ring->sched.thread) |
0875dc9e | 2824 | continue; |
5740682e ML |
2825 | |
2826 | /* only focus on the ring hit timeout if &job not NULL */ | |
2827 | if (job && job->ring->idx != i) | |
2828 | continue; | |
2829 | ||
0875dc9e | 2830 | kthread_park(ring->sched.thread); |
1b1f42d8 | 2831 | drm_sched_hw_job_reset(&ring->sched, &job->base); |
5740682e | 2832 | |
2f9d4084 ML |
2833 | /* after all hw jobs are reset, hw fence is meaningless, so force_completion */ |
2834 | amdgpu_fence_driver_force_completion(ring); | |
0875dc9e | 2835 | } |
d38ceaf9 | 2836 | |
5740682e | 2837 | if (amdgpu_sriov_vf(adev)) |
06ec9070 | 2838 | r = amdgpu_device_reset_sriov(adev, &reset_flags, job ? false : true); |
5740682e | 2839 | else |
06ec9070 | 2840 | r = amdgpu_device_reset(adev, &reset_flags); |
35d782fe | 2841 | |
d38ceaf9 | 2842 | if (!r) { |
5740682e ML |
2843 | if (((reset_flags & AMDGPU_RESET_INFO_FULLRESET) && !(adev->flags & AMD_IS_APU)) || |
2844 | (reset_flags & AMDGPU_RESET_INFO_VRAM_LOST)) { | |
53cdccd5 CZ |
2845 | struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; |
2846 | struct amdgpu_bo *bo, *tmp; | |
f54d1867 | 2847 | struct dma_fence *fence = NULL, *next = NULL; |
53cdccd5 CZ |
2848 | |
2849 | DRM_INFO("recover vram bo from shadow\n"); | |
2850 | mutex_lock(&adev->shadow_list_lock); | |
2851 | list_for_each_entry_safe(bo, tmp, &adev->shadow_list, shadow_list) { | |
236763d3 | 2852 | next = NULL; |
06ec9070 | 2853 | amdgpu_device_recover_vram_from_shadow(adev, ring, bo, &next); |
53cdccd5 | 2854 | if (fence) { |
f54d1867 | 2855 | r = dma_fence_wait(fence, false); |
53cdccd5 | 2856 | if (r) { |
1d7b17b0 | 2857 | WARN(r, "recovery from shadow isn't completed\n"); |
53cdccd5 CZ |
2858 | break; |
2859 | } | |
2860 | } | |
1f465087 | 2861 | |
f54d1867 | 2862 | dma_fence_put(fence); |
53cdccd5 CZ |
2863 | fence = next; |
2864 | } | |
2865 | mutex_unlock(&adev->shadow_list_lock); | |
2866 | if (fence) { | |
f54d1867 | 2867 | r = dma_fence_wait(fence, false); |
53cdccd5 | 2868 | if (r) |
1d7b17b0 | 2869 | WARN(r, "recovery from shadow isn't completed\n"); |
53cdccd5 | 2870 | } |
f54d1867 | 2871 | dma_fence_put(fence); |
53cdccd5 | 2872 | } |
5740682e | 2873 | |
d38ceaf9 AD |
2874 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
2875 | struct amdgpu_ring *ring = adev->rings[i]; | |
51687759 CZ |
2876 | |
2877 | if (!ring || !ring->sched.thread) | |
d38ceaf9 | 2878 | continue; |
53cdccd5 | 2879 | |
5740682e ML |
2880 | /* only focus on the ring hit timeout if &job not NULL */ |
2881 | if (job && job->ring->idx != i) | |
2882 | continue; | |
2883 | ||
1b1f42d8 | 2884 | drm_sched_job_recovery(&ring->sched); |
0875dc9e | 2885 | kthread_unpark(ring->sched.thread); |
d38ceaf9 | 2886 | } |
d38ceaf9 | 2887 | } else { |
d38ceaf9 | 2888 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
5740682e ML |
2889 | struct amdgpu_ring *ring = adev->rings[i]; |
2890 | ||
2891 | if (!ring || !ring->sched.thread) | |
2892 | continue; | |
2893 | ||
2894 | /* only focus on the ring hit timeout if &job not NULL */ | |
2895 | if (job && job->ring->idx != i) | |
2896 | continue; | |
2897 | ||
2898 | kthread_unpark(adev->rings[i]->sched.thread); | |
d38ceaf9 AD |
2899 | } |
2900 | } | |
2901 | ||
4562236b | 2902 | if (amdgpu_device_has_dc_support(adev)) { |
5740682e ML |
2903 | if (drm_atomic_helper_resume(adev->ddev, state)) |
2904 | dev_info(adev->dev, "drm resume failed:%d\n", r); | |
4562236b | 2905 | amdgpu_dm_display_resume(adev); |
5740682e | 2906 | } else { |
4562236b | 2907 | drm_helper_resume_force_mode(adev->ddev); |
5740682e | 2908 | } |
d38ceaf9 AD |
2909 | |
2910 | ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched); | |
5740682e | 2911 | |
89041940 | 2912 | if (r) { |
d38ceaf9 | 2913 | /* bad news, how to tell it to userspace ? */ |
5740682e ML |
2914 | dev_info(adev->dev, "GPU reset(%d) failed\n", atomic_read(&adev->gpu_reset_counter)); |
2915 | amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r); | |
2916 | } else { | |
2917 | dev_info(adev->dev, "GPU reset(%d) successed!\n",atomic_read(&adev->gpu_reset_counter)); | |
89041940 | 2918 | } |
d38ceaf9 | 2919 | |
89041940 | 2920 | amdgpu_vf_error_trans_all(adev); |
13a752e3 ML |
2921 | adev->in_gpu_reset = 0; |
2922 | mutex_unlock(&adev->lock_reset); | |
d38ceaf9 AD |
2923 | return r; |
2924 | } | |
2925 | ||
d0dd7f0c AD |
2926 | void amdgpu_get_pcie_info(struct amdgpu_device *adev) |
2927 | { | |
2928 | u32 mask; | |
2929 | int ret; | |
2930 | ||
cd474ba0 AD |
2931 | if (amdgpu_pcie_gen_cap) |
2932 | adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap; | |
d0dd7f0c | 2933 | |
cd474ba0 AD |
2934 | if (amdgpu_pcie_lane_cap) |
2935 | adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap; | |
d0dd7f0c | 2936 | |
cd474ba0 AD |
2937 | /* covers APUs as well */ |
2938 | if (pci_is_root_bus(adev->pdev->bus)) { | |
2939 | if (adev->pm.pcie_gen_mask == 0) | |
2940 | adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; | |
2941 | if (adev->pm.pcie_mlw_mask == 0) | |
2942 | adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; | |
d0dd7f0c | 2943 | return; |
cd474ba0 | 2944 | } |
d0dd7f0c | 2945 | |
cd474ba0 AD |
2946 | if (adev->pm.pcie_gen_mask == 0) { |
2947 | ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask); | |
2948 | if (!ret) { | |
2949 | adev->pm.pcie_gen_mask = (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
2950 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
2951 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
2952 | ||
2953 | if (mask & DRM_PCIE_SPEED_25) | |
2954 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
2955 | if (mask & DRM_PCIE_SPEED_50) | |
2956 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2; | |
2957 | if (mask & DRM_PCIE_SPEED_80) | |
2958 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3; | |
2959 | } else { | |
2960 | adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; | |
2961 | } | |
2962 | } | |
2963 | if (adev->pm.pcie_mlw_mask == 0) { | |
2964 | ret = drm_pcie_get_max_link_width(adev->ddev, &mask); | |
2965 | if (!ret) { | |
2966 | switch (mask) { | |
2967 | case 32: | |
2968 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 | | |
2969 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | | |
2970 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
2971 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2972 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2973 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2974 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2975 | break; | |
2976 | case 16: | |
2977 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | | |
2978 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
2979 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2980 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2981 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2982 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2983 | break; | |
2984 | case 12: | |
2985 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
2986 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2987 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2988 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2989 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2990 | break; | |
2991 | case 8: | |
2992 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2993 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2994 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2995 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2996 | break; | |
2997 | case 4: | |
2998 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2999 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
3000 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
3001 | break; | |
3002 | case 2: | |
3003 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
3004 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
3005 | break; | |
3006 | case 1: | |
3007 | adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1; | |
3008 | break; | |
3009 | default: | |
3010 | break; | |
3011 | } | |
3012 | } else { | |
3013 | adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; | |
d0dd7f0c AD |
3014 | } |
3015 | } | |
3016 | } | |
d38ceaf9 | 3017 |