]> git.ipfire.org Git - thirdparty/kernel/stable.git/blame - drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
drm/amdgpu: amdgpu_kiq_reg_write_reg_wait() can be static
[thirdparty/kernel/stable.git] / drivers / gpu / drm / amd / amdgpu / amdgpu_device.c
CommitLineData
d38ceaf9
AD
1/*
2 * Copyright 2008 Advanced Micro Devices, Inc.
3 * Copyright 2008 Red Hat Inc.
4 * Copyright 2009 Jerome Glisse.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 *
24 * Authors: Dave Airlie
25 * Alex Deucher
26 * Jerome Glisse
27 */
b1ddf548 28#include <linux/power_supply.h>
0875dc9e 29#include <linux/kthread.h>
d38ceaf9
AD
30#include <linux/console.h>
31#include <linux/slab.h>
d38ceaf9
AD
32#include <drm/drmP.h>
33#include <drm/drm_crtc_helper.h>
4562236b 34#include <drm/drm_atomic_helper.h>
d38ceaf9
AD
35#include <drm/amdgpu_drm.h>
36#include <linux/vgaarb.h>
37#include <linux/vga_switcheroo.h>
38#include <linux/efi.h>
39#include "amdgpu.h"
f4b373f4 40#include "amdgpu_trace.h"
d38ceaf9
AD
41#include "amdgpu_i2c.h"
42#include "atom.h"
43#include "amdgpu_atombios.h"
a5bde2f9 44#include "amdgpu_atomfirmware.h"
d0dd7f0c 45#include "amd_pcie.h"
33f34802
KW
46#ifdef CONFIG_DRM_AMDGPU_SI
47#include "si.h"
48#endif
a2e73f56
AD
49#ifdef CONFIG_DRM_AMDGPU_CIK
50#include "cik.h"
51#endif
aaa36a97 52#include "vi.h"
460826e6 53#include "soc15.h"
d38ceaf9 54#include "bif/bif_4_1_d.h"
9accf2fd 55#include <linux/pci.h>
bec86378 56#include <linux/firmware.h>
89041940 57#include "amdgpu_vf_error.h"
d38ceaf9 58
ba997709 59#include "amdgpu_amdkfd.h"
d2f52ac8 60#include "amdgpu_pm.h"
d38ceaf9 61
e2a75f88 62MODULE_FIRMWARE("amdgpu/vega10_gpu_info.bin");
3f76dced 63MODULE_FIRMWARE("amdgpu/vega12_gpu_info.bin");
2d2e5e7e 64MODULE_FIRMWARE("amdgpu/raven_gpu_info.bin");
e2a75f88 65
2dc80b00
S
66#define AMDGPU_RESUME_MS 2000
67
d38ceaf9 68static const char *amdgpu_asic_name[] = {
da69c161
KW
69 "TAHITI",
70 "PITCAIRN",
71 "VERDE",
72 "OLAND",
73 "HAINAN",
d38ceaf9
AD
74 "BONAIRE",
75 "KAVERI",
76 "KABINI",
77 "HAWAII",
78 "MULLINS",
79 "TOPAZ",
80 "TONGA",
48299f95 81 "FIJI",
d38ceaf9 82 "CARRIZO",
139f4917 83 "STONEY",
2cc0c0b5
FC
84 "POLARIS10",
85 "POLARIS11",
c4642a47 86 "POLARIS12",
48ff108d 87 "VEGAM",
d4196f01 88 "VEGA10",
8fab806a 89 "VEGA12",
956fcddc 90 "VEGA20",
2ca8a5d2 91 "RAVEN",
d38ceaf9
AD
92 "LAST",
93};
94
5494d864
AD
95static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev);
96
e3ecdffa
AD
97/**
98 * amdgpu_device_is_px - Is the device is a dGPU with HG/PX power control
99 *
100 * @dev: drm_device pointer
101 *
102 * Returns true if the device is a dGPU with HG/PX power control,
103 * otherwise return false.
104 */
d38ceaf9
AD
105bool amdgpu_device_is_px(struct drm_device *dev)
106{
107 struct amdgpu_device *adev = dev->dev_private;
108
2f7d10b3 109 if (adev->flags & AMD_IS_PX)
d38ceaf9
AD
110 return true;
111 return false;
112}
113
114/*
115 * MMIO register access helper functions.
116 */
e3ecdffa
AD
117/**
118 * amdgpu_mm_rreg - read a memory mapped IO register
119 *
120 * @adev: amdgpu_device pointer
121 * @reg: dword aligned register offset
122 * @acc_flags: access flags which require special behavior
123 *
124 * Returns the 32 bit value from the offset specified.
125 */
d38ceaf9 126uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg,
15d72fd7 127 uint32_t acc_flags)
d38ceaf9 128{
f4b373f4
TSD
129 uint32_t ret;
130
43ca8efa 131 if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev))
bc992ba5 132 return amdgpu_virt_kiq_rreg(adev, reg);
bc992ba5 133
15d72fd7 134 if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
f4b373f4 135 ret = readl(((void __iomem *)adev->rmmio) + (reg * 4));
d38ceaf9
AD
136 else {
137 unsigned long flags;
d38ceaf9
AD
138
139 spin_lock_irqsave(&adev->mmio_idx_lock, flags);
140 writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4));
141 ret = readl(((void __iomem *)adev->rmmio) + (mmMM_DATA * 4));
142 spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
d38ceaf9 143 }
f4b373f4
TSD
144 trace_amdgpu_mm_rreg(adev->pdev->device, reg, ret);
145 return ret;
d38ceaf9
AD
146}
147
421a2a30
ML
148/*
149 * MMIO register read with bytes helper functions
150 * @offset:bytes offset from MMIO start
151 *
152*/
153
e3ecdffa
AD
154/**
155 * amdgpu_mm_rreg8 - read a memory mapped IO register
156 *
157 * @adev: amdgpu_device pointer
158 * @offset: byte aligned register offset
159 *
160 * Returns the 8 bit value from the offset specified.
161 */
421a2a30
ML
162uint8_t amdgpu_mm_rreg8(struct amdgpu_device *adev, uint32_t offset) {
163 if (offset < adev->rmmio_size)
164 return (readb(adev->rmmio + offset));
165 BUG();
166}
167
168/*
169 * MMIO register write with bytes helper functions
170 * @offset:bytes offset from MMIO start
171 * @value: the value want to be written to the register
172 *
173*/
e3ecdffa
AD
174/**
175 * amdgpu_mm_wreg8 - read a memory mapped IO register
176 *
177 * @adev: amdgpu_device pointer
178 * @offset: byte aligned register offset
179 * @value: 8 bit value to write
180 *
181 * Writes the value specified to the offset specified.
182 */
421a2a30
ML
183void amdgpu_mm_wreg8(struct amdgpu_device *adev, uint32_t offset, uint8_t value) {
184 if (offset < adev->rmmio_size)
185 writeb(value, adev->rmmio + offset);
186 else
187 BUG();
188}
189
e3ecdffa
AD
190/**
191 * amdgpu_mm_wreg - write to a memory mapped IO register
192 *
193 * @adev: amdgpu_device pointer
194 * @reg: dword aligned register offset
195 * @v: 32 bit value to write to the register
196 * @acc_flags: access flags which require special behavior
197 *
198 * Writes the value specified to the offset specified.
199 */
d38ceaf9 200void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v,
15d72fd7 201 uint32_t acc_flags)
d38ceaf9 202{
f4b373f4 203 trace_amdgpu_mm_wreg(adev->pdev->device, reg, v);
4e99a44e 204
47ed4e1c
KW
205 if (adev->asic_type >= CHIP_VEGA10 && reg == 0) {
206 adev->last_mm_index = v;
207 }
208
43ca8efa 209 if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev))
bc992ba5 210 return amdgpu_virt_kiq_wreg(adev, reg, v);
bc992ba5 211
15d72fd7 212 if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX))
d38ceaf9
AD
213 writel(v, ((void __iomem *)adev->rmmio) + (reg * 4));
214 else {
215 unsigned long flags;
216
217 spin_lock_irqsave(&adev->mmio_idx_lock, flags);
218 writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4));
219 writel(v, ((void __iomem *)adev->rmmio) + (mmMM_DATA * 4));
220 spin_unlock_irqrestore(&adev->mmio_idx_lock, flags);
221 }
47ed4e1c
KW
222
223 if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) {
224 udelay(500);
225 }
d38ceaf9
AD
226}
227
e3ecdffa
AD
228/**
229 * amdgpu_io_rreg - read an IO register
230 *
231 * @adev: amdgpu_device pointer
232 * @reg: dword aligned register offset
233 *
234 * Returns the 32 bit value from the offset specified.
235 */
d38ceaf9
AD
236u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg)
237{
238 if ((reg * 4) < adev->rio_mem_size)
239 return ioread32(adev->rio_mem + (reg * 4));
240 else {
241 iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
242 return ioread32(adev->rio_mem + (mmMM_DATA * 4));
243 }
244}
245
e3ecdffa
AD
246/**
247 * amdgpu_io_wreg - write to an IO register
248 *
249 * @adev: amdgpu_device pointer
250 * @reg: dword aligned register offset
251 * @v: 32 bit value to write to the register
252 *
253 * Writes the value specified to the offset specified.
254 */
d38ceaf9
AD
255void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
256{
47ed4e1c
KW
257 if (adev->asic_type >= CHIP_VEGA10 && reg == 0) {
258 adev->last_mm_index = v;
259 }
d38ceaf9
AD
260
261 if ((reg * 4) < adev->rio_mem_size)
262 iowrite32(v, adev->rio_mem + (reg * 4));
263 else {
264 iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4));
265 iowrite32(v, adev->rio_mem + (mmMM_DATA * 4));
266 }
47ed4e1c
KW
267
268 if (adev->asic_type >= CHIP_VEGA10 && reg == 1 && adev->last_mm_index == 0x5702C) {
269 udelay(500);
270 }
d38ceaf9
AD
271}
272
273/**
274 * amdgpu_mm_rdoorbell - read a doorbell dword
275 *
276 * @adev: amdgpu_device pointer
277 * @index: doorbell index
278 *
279 * Returns the value in the doorbell aperture at the
280 * requested doorbell index (CIK).
281 */
282u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index)
283{
284 if (index < adev->doorbell.num_doorbells) {
285 return readl(adev->doorbell.ptr + index);
286 } else {
287 DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
288 return 0;
289 }
290}
291
292/**
293 * amdgpu_mm_wdoorbell - write a doorbell dword
294 *
295 * @adev: amdgpu_device pointer
296 * @index: doorbell index
297 * @v: value to write
298 *
299 * Writes @v to the doorbell aperture at the
300 * requested doorbell index (CIK).
301 */
302void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v)
303{
304 if (index < adev->doorbell.num_doorbells) {
305 writel(v, adev->doorbell.ptr + index);
306 } else {
307 DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
308 }
309}
310
832be404
KW
311/**
312 * amdgpu_mm_rdoorbell64 - read a doorbell Qword
313 *
314 * @adev: amdgpu_device pointer
315 * @index: doorbell index
316 *
317 * Returns the value in the doorbell aperture at the
318 * requested doorbell index (VEGA10+).
319 */
320u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index)
321{
322 if (index < adev->doorbell.num_doorbells) {
323 return atomic64_read((atomic64_t *)(adev->doorbell.ptr + index));
324 } else {
325 DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
326 return 0;
327 }
328}
329
330/**
331 * amdgpu_mm_wdoorbell64 - write a doorbell Qword
332 *
333 * @adev: amdgpu_device pointer
334 * @index: doorbell index
335 * @v: value to write
336 *
337 * Writes @v to the doorbell aperture at the
338 * requested doorbell index (VEGA10+).
339 */
340void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v)
341{
342 if (index < adev->doorbell.num_doorbells) {
343 atomic64_set((atomic64_t *)(adev->doorbell.ptr + index), v);
344 } else {
345 DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
346 }
347}
348
d38ceaf9
AD
349/**
350 * amdgpu_invalid_rreg - dummy reg read function
351 *
352 * @adev: amdgpu device pointer
353 * @reg: offset of register
354 *
355 * Dummy register read function. Used for register blocks
356 * that certain asics don't have (all asics).
357 * Returns the value in the register.
358 */
359static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg)
360{
361 DRM_ERROR("Invalid callback to read register 0x%04X\n", reg);
362 BUG();
363 return 0;
364}
365
366/**
367 * amdgpu_invalid_wreg - dummy reg write function
368 *
369 * @adev: amdgpu device pointer
370 * @reg: offset of register
371 * @v: value to write to the register
372 *
373 * Dummy register read function. Used for register blocks
374 * that certain asics don't have (all asics).
375 */
376static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v)
377{
378 DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n",
379 reg, v);
380 BUG();
381}
382
383/**
384 * amdgpu_block_invalid_rreg - dummy reg read function
385 *
386 * @adev: amdgpu device pointer
387 * @block: offset of instance
388 * @reg: offset of register
389 *
390 * Dummy register read function. Used for register blocks
391 * that certain asics don't have (all asics).
392 * Returns the value in the register.
393 */
394static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev,
395 uint32_t block, uint32_t reg)
396{
397 DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n",
398 reg, block);
399 BUG();
400 return 0;
401}
402
403/**
404 * amdgpu_block_invalid_wreg - dummy reg write function
405 *
406 * @adev: amdgpu device pointer
407 * @block: offset of instance
408 * @reg: offset of register
409 * @v: value to write to the register
410 *
411 * Dummy register read function. Used for register blocks
412 * that certain asics don't have (all asics).
413 */
414static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev,
415 uint32_t block,
416 uint32_t reg, uint32_t v)
417{
418 DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n",
419 reg, block, v);
420 BUG();
421}
422
e3ecdffa
AD
423/**
424 * amdgpu_device_vram_scratch_init - allocate the VRAM scratch page
425 *
426 * @adev: amdgpu device pointer
427 *
428 * Allocates a scratch page of VRAM for use by various things in the
429 * driver.
430 */
06ec9070 431static int amdgpu_device_vram_scratch_init(struct amdgpu_device *adev)
d38ceaf9 432{
a4a02777
CK
433 return amdgpu_bo_create_kernel(adev, AMDGPU_GPU_PAGE_SIZE,
434 PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM,
435 &adev->vram_scratch.robj,
436 &adev->vram_scratch.gpu_addr,
437 (void **)&adev->vram_scratch.ptr);
d38ceaf9
AD
438}
439
e3ecdffa
AD
440/**
441 * amdgpu_device_vram_scratch_fini - Free the VRAM scratch page
442 *
443 * @adev: amdgpu device pointer
444 *
445 * Frees the VRAM scratch page.
446 */
06ec9070 447static void amdgpu_device_vram_scratch_fini(struct amdgpu_device *adev)
d38ceaf9 448{
078af1a3 449 amdgpu_bo_free_kernel(&adev->vram_scratch.robj, NULL, NULL);
d38ceaf9
AD
450}
451
452/**
9c3f2b54 453 * amdgpu_device_program_register_sequence - program an array of registers.
d38ceaf9
AD
454 *
455 * @adev: amdgpu_device pointer
456 * @registers: pointer to the register array
457 * @array_size: size of the register array
458 *
459 * Programs an array or registers with and and or masks.
460 * This is a helper for setting golden registers.
461 */
9c3f2b54
AD
462void amdgpu_device_program_register_sequence(struct amdgpu_device *adev,
463 const u32 *registers,
464 const u32 array_size)
d38ceaf9
AD
465{
466 u32 tmp, reg, and_mask, or_mask;
467 int i;
468
469 if (array_size % 3)
470 return;
471
472 for (i = 0; i < array_size; i +=3) {
473 reg = registers[i + 0];
474 and_mask = registers[i + 1];
475 or_mask = registers[i + 2];
476
477 if (and_mask == 0xffffffff) {
478 tmp = or_mask;
479 } else {
480 tmp = RREG32(reg);
481 tmp &= ~and_mask;
482 tmp |= or_mask;
483 }
484 WREG32(reg, tmp);
485 }
486}
487
e3ecdffa
AD
488/**
489 * amdgpu_device_pci_config_reset - reset the GPU
490 *
491 * @adev: amdgpu_device pointer
492 *
493 * Resets the GPU using the pci config reset sequence.
494 * Only applicable to asics prior to vega10.
495 */
8111c387 496void amdgpu_device_pci_config_reset(struct amdgpu_device *adev)
d38ceaf9
AD
497{
498 pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA);
499}
500
501/*
502 * GPU doorbell aperture helpers function.
503 */
504/**
06ec9070 505 * amdgpu_device_doorbell_init - Init doorbell driver information.
d38ceaf9
AD
506 *
507 * @adev: amdgpu_device pointer
508 *
509 * Init doorbell driver information (CIK)
510 * Returns 0 on success, error on failure.
511 */
06ec9070 512static int amdgpu_device_doorbell_init(struct amdgpu_device *adev)
d38ceaf9 513{
705e519e
CK
514 /* No doorbell on SI hardware generation */
515 if (adev->asic_type < CHIP_BONAIRE) {
516 adev->doorbell.base = 0;
517 adev->doorbell.size = 0;
518 adev->doorbell.num_doorbells = 0;
519 adev->doorbell.ptr = NULL;
520 return 0;
521 }
522
d6895ad3
CK
523 if (pci_resource_flags(adev->pdev, 2) & IORESOURCE_UNSET)
524 return -EINVAL;
525
d38ceaf9
AD
526 /* doorbell bar mapping */
527 adev->doorbell.base = pci_resource_start(adev->pdev, 2);
528 adev->doorbell.size = pci_resource_len(adev->pdev, 2);
529
edf600da 530 adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32),
d38ceaf9
AD
531 AMDGPU_DOORBELL_MAX_ASSIGNMENT+1);
532 if (adev->doorbell.num_doorbells == 0)
533 return -EINVAL;
534
8972e5d2
CK
535 adev->doorbell.ptr = ioremap(adev->doorbell.base,
536 adev->doorbell.num_doorbells *
537 sizeof(u32));
538 if (adev->doorbell.ptr == NULL)
d38ceaf9 539 return -ENOMEM;
d38ceaf9
AD
540
541 return 0;
542}
543
544/**
06ec9070 545 * amdgpu_device_doorbell_fini - Tear down doorbell driver information.
d38ceaf9
AD
546 *
547 * @adev: amdgpu_device pointer
548 *
549 * Tear down doorbell driver information (CIK)
550 */
06ec9070 551static void amdgpu_device_doorbell_fini(struct amdgpu_device *adev)
d38ceaf9
AD
552{
553 iounmap(adev->doorbell.ptr);
554 adev->doorbell.ptr = NULL;
555}
556
22cb0164 557
d38ceaf9
AD
558
559/*
06ec9070 560 * amdgpu_device_wb_*()
455a7bc2 561 * Writeback is the method by which the GPU updates special pages in memory
ea81a173 562 * with the status of certain GPU events (fences, ring pointers,etc.).
d38ceaf9
AD
563 */
564
565/**
06ec9070 566 * amdgpu_device_wb_fini - Disable Writeback and free memory
d38ceaf9
AD
567 *
568 * @adev: amdgpu_device pointer
569 *
570 * Disables Writeback and frees the Writeback memory (all asics).
571 * Used at driver shutdown.
572 */
06ec9070 573static void amdgpu_device_wb_fini(struct amdgpu_device *adev)
d38ceaf9
AD
574{
575 if (adev->wb.wb_obj) {
a76ed485
AD
576 amdgpu_bo_free_kernel(&adev->wb.wb_obj,
577 &adev->wb.gpu_addr,
578 (void **)&adev->wb.wb);
d38ceaf9
AD
579 adev->wb.wb_obj = NULL;
580 }
581}
582
583/**
06ec9070 584 * amdgpu_device_wb_init- Init Writeback driver info and allocate memory
d38ceaf9
AD
585 *
586 * @adev: amdgpu_device pointer
587 *
455a7bc2 588 * Initializes writeback and allocates writeback memory (all asics).
d38ceaf9
AD
589 * Used at driver startup.
590 * Returns 0 on success or an -error on failure.
591 */
06ec9070 592static int amdgpu_device_wb_init(struct amdgpu_device *adev)
d38ceaf9
AD
593{
594 int r;
595
596 if (adev->wb.wb_obj == NULL) {
97407b63
AD
597 /* AMDGPU_MAX_WB * sizeof(uint32_t) * 8 = AMDGPU_MAX_WB 256bit slots */
598 r = amdgpu_bo_create_kernel(adev, AMDGPU_MAX_WB * sizeof(uint32_t) * 8,
a76ed485
AD
599 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
600 &adev->wb.wb_obj, &adev->wb.gpu_addr,
601 (void **)&adev->wb.wb);
d38ceaf9
AD
602 if (r) {
603 dev_warn(adev->dev, "(%d) create WB bo failed\n", r);
604 return r;
605 }
d38ceaf9
AD
606
607 adev->wb.num_wb = AMDGPU_MAX_WB;
608 memset(&adev->wb.used, 0, sizeof(adev->wb.used));
609
610 /* clear wb memory */
73469585 611 memset((char *)adev->wb.wb, 0, AMDGPU_MAX_WB * sizeof(uint32_t) * 8);
d38ceaf9
AD
612 }
613
614 return 0;
615}
616
617/**
131b4b36 618 * amdgpu_device_wb_get - Allocate a wb entry
d38ceaf9
AD
619 *
620 * @adev: amdgpu_device pointer
621 * @wb: wb index
622 *
623 * Allocate a wb slot for use by the driver (all asics).
624 * Returns 0 on success or -EINVAL on failure.
625 */
131b4b36 626int amdgpu_device_wb_get(struct amdgpu_device *adev, u32 *wb)
d38ceaf9
AD
627{
628 unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb);
d38ceaf9 629
97407b63 630 if (offset < adev->wb.num_wb) {
7014285a 631 __set_bit(offset, adev->wb.used);
63ae07ca 632 *wb = offset << 3; /* convert to dw offset */
0915fdbc
ML
633 return 0;
634 } else {
635 return -EINVAL;
636 }
637}
638
d38ceaf9 639/**
131b4b36 640 * amdgpu_device_wb_free - Free a wb entry
d38ceaf9
AD
641 *
642 * @adev: amdgpu_device pointer
643 * @wb: wb index
644 *
645 * Free a wb slot allocated for use by the driver (all asics)
646 */
131b4b36 647void amdgpu_device_wb_free(struct amdgpu_device *adev, u32 wb)
d38ceaf9 648{
73469585 649 wb >>= 3;
d38ceaf9 650 if (wb < adev->wb.num_wb)
73469585 651 __clear_bit(wb, adev->wb.used);
d38ceaf9
AD
652}
653
654/**
2543e28a 655 * amdgpu_device_vram_location - try to find VRAM location
e3ecdffa 656 *
d38ceaf9
AD
657 * @adev: amdgpu device structure holding all necessary informations
658 * @mc: memory controller structure holding memory informations
659 * @base: base address at which to put VRAM
660 *
455a7bc2 661 * Function will try to place VRAM at base address provided
3d647c8f 662 * as parameter.
d38ceaf9 663 */
2543e28a 664void amdgpu_device_vram_location(struct amdgpu_device *adev,
770d13b1 665 struct amdgpu_gmc *mc, u64 base)
d38ceaf9
AD
666{
667 uint64_t limit = (uint64_t)amdgpu_vram_limit << 20;
668
669 mc->vram_start = base;
d38ceaf9
AD
670 mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
671 if (limit && limit < mc->real_vram_size)
672 mc->real_vram_size = limit;
673 dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
674 mc->mc_vram_size >> 20, mc->vram_start,
675 mc->vram_end, mc->real_vram_size >> 20);
676}
677
678/**
761f58e0 679 * amdgpu_device_gart_location - try to find GART location
e3ecdffa 680 *
d38ceaf9
AD
681 * @adev: amdgpu device structure holding all necessary informations
682 * @mc: memory controller structure holding memory informations
683 *
761f58e0 684 * Function will place try to place GART before or after VRAM.
d38ceaf9 685 *
761f58e0 686 * If GART size is bigger than space left then we ajust GART size.
d38ceaf9 687 * Thus function will never fails.
d38ceaf9 688 */
2543e28a 689void amdgpu_device_gart_location(struct amdgpu_device *adev,
770d13b1 690 struct amdgpu_gmc *mc)
d38ceaf9
AD
691{
692 u64 size_af, size_bf;
693
7951e376
RZ
694 mc->gart_size += adev->pm.smu_prv_buffer_size;
695
770d13b1 696 size_af = adev->gmc.mc_mask - mc->vram_end;
ed21c047 697 size_bf = mc->vram_start;
d38ceaf9 698 if (size_bf > size_af) {
6f02a696 699 if (mc->gart_size > size_bf) {
761f58e0 700 dev_warn(adev->dev, "limiting GART\n");
6f02a696 701 mc->gart_size = size_bf;
d38ceaf9 702 }
6f02a696 703 mc->gart_start = 0;
d38ceaf9 704 } else {
6f02a696 705 if (mc->gart_size > size_af) {
761f58e0 706 dev_warn(adev->dev, "limiting GART\n");
6f02a696 707 mc->gart_size = size_af;
d38ceaf9 708 }
b98f1b9e
CK
709 /* VCE doesn't like it when BOs cross a 4GB segment, so align
710 * the GART base on a 4GB boundary as well.
711 */
712 mc->gart_start = ALIGN(mc->vram_end + 1, 0x100000000ULL);
d38ceaf9 713 }
6f02a696 714 mc->gart_end = mc->gart_start + mc->gart_size - 1;
761f58e0 715 dev_info(adev->dev, "GART: %lluM 0x%016llX - 0x%016llX\n",
6f02a696 716 mc->gart_size >> 20, mc->gart_start, mc->gart_end);
d38ceaf9
AD
717}
718
d6895ad3
CK
719/**
720 * amdgpu_device_resize_fb_bar - try to resize FB BAR
721 *
722 * @adev: amdgpu_device pointer
723 *
724 * Try to resize FB BAR to make all VRAM CPU accessible. We try very hard not
725 * to fail, but if any of the BARs is not accessible after the size we abort
726 * driver loading by returning -ENODEV.
727 */
728int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev)
729{
770d13b1 730 u64 space_needed = roundup_pow_of_two(adev->gmc.real_vram_size);
d6895ad3 731 u32 rbar_size = order_base_2(((space_needed >> 20) | 1)) - 1;
31b8adab
CK
732 struct pci_bus *root;
733 struct resource *res;
734 unsigned i;
d6895ad3
CK
735 u16 cmd;
736 int r;
737
0c03b912 738 /* Bypass for VF */
739 if (amdgpu_sriov_vf(adev))
740 return 0;
741
31b8adab
CK
742 /* Check if the root BUS has 64bit memory resources */
743 root = adev->pdev->bus;
744 while (root->parent)
745 root = root->parent;
746
747 pci_bus_for_each_resource(root, res, i) {
0ebb7c54 748 if (res && res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) &&
31b8adab
CK
749 res->start > 0x100000000ull)
750 break;
751 }
752
753 /* Trying to resize is pointless without a root hub window above 4GB */
754 if (!res)
755 return 0;
756
d6895ad3
CK
757 /* Disable memory decoding while we change the BAR addresses and size */
758 pci_read_config_word(adev->pdev, PCI_COMMAND, &cmd);
759 pci_write_config_word(adev->pdev, PCI_COMMAND,
760 cmd & ~PCI_COMMAND_MEMORY);
761
762 /* Free the VRAM and doorbell BAR, we most likely need to move both. */
06ec9070 763 amdgpu_device_doorbell_fini(adev);
d6895ad3
CK
764 if (adev->asic_type >= CHIP_BONAIRE)
765 pci_release_resource(adev->pdev, 2);
766
767 pci_release_resource(adev->pdev, 0);
768
769 r = pci_resize_resource(adev->pdev, 0, rbar_size);
770 if (r == -ENOSPC)
771 DRM_INFO("Not enough PCI address space for a large BAR.");
772 else if (r && r != -ENOTSUPP)
773 DRM_ERROR("Problem resizing BAR0 (%d).", r);
774
775 pci_assign_unassigned_bus_resources(adev->pdev->bus);
776
777 /* When the doorbell or fb BAR isn't available we have no chance of
778 * using the device.
779 */
06ec9070 780 r = amdgpu_device_doorbell_init(adev);
d6895ad3
CK
781 if (r || (pci_resource_flags(adev->pdev, 0) & IORESOURCE_UNSET))
782 return -ENODEV;
783
784 pci_write_config_word(adev->pdev, PCI_COMMAND, cmd);
785
786 return 0;
787}
a05502e5 788
d38ceaf9
AD
789/*
790 * GPU helpers function.
791 */
792/**
39c640c0 793 * amdgpu_device_need_post - check if the hw need post or not
d38ceaf9
AD
794 *
795 * @adev: amdgpu_device pointer
796 *
c836fec5
JQ
797 * Check if the asic has been initialized (all asics) at driver startup
798 * or post is needed if hw reset is performed.
799 * Returns true if need or false if not.
d38ceaf9 800 */
39c640c0 801bool amdgpu_device_need_post(struct amdgpu_device *adev)
d38ceaf9
AD
802{
803 uint32_t reg;
804
bec86378
ML
805 if (amdgpu_sriov_vf(adev))
806 return false;
807
808 if (amdgpu_passthrough(adev)) {
1da2c326
ML
809 /* for FIJI: In whole GPU pass-through virtualization case, after VM reboot
810 * some old smc fw still need driver do vPost otherwise gpu hang, while
811 * those smc fw version above 22.15 doesn't have this flaw, so we force
812 * vpost executed for smc version below 22.15
bec86378
ML
813 */
814 if (adev->asic_type == CHIP_FIJI) {
815 int err;
816 uint32_t fw_ver;
817 err = request_firmware(&adev->pm.fw, "amdgpu/fiji_smc.bin", adev->dev);
818 /* force vPost if error occured */
819 if (err)
820 return true;
821
822 fw_ver = *((uint32_t *)adev->pm.fw->data + 69);
1da2c326
ML
823 if (fw_ver < 0x00160e00)
824 return true;
bec86378 825 }
bec86378 826 }
91fe77eb 827
828 if (adev->has_hw_reset) {
829 adev->has_hw_reset = false;
830 return true;
831 }
832
833 /* bios scratch used on CIK+ */
834 if (adev->asic_type >= CHIP_BONAIRE)
835 return amdgpu_atombios_scratch_need_asic_init(adev);
836
837 /* check MEM_SIZE for older asics */
838 reg = amdgpu_asic_get_config_memsize(adev);
839
840 if ((reg != 0) && (reg != 0xffffffff))
841 return false;
842
843 return true;
bec86378
ML
844}
845
d38ceaf9
AD
846/* if we get transitioned to only one device, take VGA back */
847/**
06ec9070 848 * amdgpu_device_vga_set_decode - enable/disable vga decode
d38ceaf9
AD
849 *
850 * @cookie: amdgpu_device pointer
851 * @state: enable/disable vga decode
852 *
853 * Enable/disable vga decode (all asics).
854 * Returns VGA resource flags.
855 */
06ec9070 856static unsigned int amdgpu_device_vga_set_decode(void *cookie, bool state)
d38ceaf9
AD
857{
858 struct amdgpu_device *adev = cookie;
859 amdgpu_asic_set_vga_state(adev, state);
860 if (state)
861 return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM |
862 VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
863 else
864 return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
865}
866
e3ecdffa
AD
867/**
868 * amdgpu_device_check_block_size - validate the vm block size
869 *
870 * @adev: amdgpu_device pointer
871 *
872 * Validates the vm block size specified via module parameter.
873 * The vm block size defines number of bits in page table versus page directory,
874 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
875 * page table and the remaining bits are in the page directory.
876 */
06ec9070 877static void amdgpu_device_check_block_size(struct amdgpu_device *adev)
a1adf8be
CZ
878{
879 /* defines number of bits in page table versus page directory,
880 * a page is 4KB so we have 12 bits offset, minimum 9 bits in the
881 * page table and the remaining bits are in the page directory */
bab4fee7
JZ
882 if (amdgpu_vm_block_size == -1)
883 return;
a1adf8be 884
bab4fee7 885 if (amdgpu_vm_block_size < 9) {
a1adf8be
CZ
886 dev_warn(adev->dev, "VM page table size (%d) too small\n",
887 amdgpu_vm_block_size);
97489129 888 amdgpu_vm_block_size = -1;
a1adf8be 889 }
a1adf8be
CZ
890}
891
e3ecdffa
AD
892/**
893 * amdgpu_device_check_vm_size - validate the vm size
894 *
895 * @adev: amdgpu_device pointer
896 *
897 * Validates the vm size in GB specified via module parameter.
898 * The VM size is the size of the GPU virtual memory space in GB.
899 */
06ec9070 900static void amdgpu_device_check_vm_size(struct amdgpu_device *adev)
83ca145d 901{
64dab074
AD
902 /* no need to check the default value */
903 if (amdgpu_vm_size == -1)
904 return;
905
83ca145d
ZJ
906 if (amdgpu_vm_size < 1) {
907 dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n",
908 amdgpu_vm_size);
f3368128 909 amdgpu_vm_size = -1;
83ca145d 910 }
83ca145d
ZJ
911}
912
7951e376
RZ
913static void amdgpu_device_check_smu_prv_buffer_size(struct amdgpu_device *adev)
914{
915 struct sysinfo si;
916 bool is_os_64 = (sizeof(void *) == 8) ? true : false;
917 uint64_t total_memory;
918 uint64_t dram_size_seven_GB = 0x1B8000000;
919 uint64_t dram_size_three_GB = 0xB8000000;
920
921 if (amdgpu_smu_memory_pool_size == 0)
922 return;
923
924 if (!is_os_64) {
925 DRM_WARN("Not 64-bit OS, feature not supported\n");
926 goto def_value;
927 }
928 si_meminfo(&si);
929 total_memory = (uint64_t)si.totalram * si.mem_unit;
930
931 if ((amdgpu_smu_memory_pool_size == 1) ||
932 (amdgpu_smu_memory_pool_size == 2)) {
933 if (total_memory < dram_size_three_GB)
934 goto def_value1;
935 } else if ((amdgpu_smu_memory_pool_size == 4) ||
936 (amdgpu_smu_memory_pool_size == 8)) {
937 if (total_memory < dram_size_seven_GB)
938 goto def_value1;
939 } else {
940 DRM_WARN("Smu memory pool size not supported\n");
941 goto def_value;
942 }
943 adev->pm.smu_prv_buffer_size = amdgpu_smu_memory_pool_size << 28;
944
945 return;
946
947def_value1:
948 DRM_WARN("No enough system memory\n");
949def_value:
950 adev->pm.smu_prv_buffer_size = 0;
951}
952
d38ceaf9 953/**
06ec9070 954 * amdgpu_device_check_arguments - validate module params
d38ceaf9
AD
955 *
956 * @adev: amdgpu_device pointer
957 *
958 * Validates certain module parameters and updates
959 * the associated values used by the driver (all asics).
960 */
06ec9070 961static void amdgpu_device_check_arguments(struct amdgpu_device *adev)
d38ceaf9 962{
5b011235
CZ
963 if (amdgpu_sched_jobs < 4) {
964 dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n",
965 amdgpu_sched_jobs);
966 amdgpu_sched_jobs = 4;
76117507 967 } else if (!is_power_of_2(amdgpu_sched_jobs)){
5b011235
CZ
968 dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n",
969 amdgpu_sched_jobs);
970 amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs);
971 }
d38ceaf9 972
83e74db6 973 if (amdgpu_gart_size != -1 && amdgpu_gart_size < 32) {
f9321cc4
CK
974 /* gart size must be greater or equal to 32M */
975 dev_warn(adev->dev, "gart size (%d) too small\n",
976 amdgpu_gart_size);
83e74db6 977 amdgpu_gart_size = -1;
d38ceaf9
AD
978 }
979
36d38372 980 if (amdgpu_gtt_size != -1 && amdgpu_gtt_size < 32) {
c4e1a13a 981 /* gtt size must be greater or equal to 32M */
36d38372
CK
982 dev_warn(adev->dev, "gtt size (%d) too small\n",
983 amdgpu_gtt_size);
984 amdgpu_gtt_size = -1;
d38ceaf9
AD
985 }
986
d07f14be
RH
987 /* valid range is between 4 and 9 inclusive */
988 if (amdgpu_vm_fragment_size != -1 &&
989 (amdgpu_vm_fragment_size > 9 || amdgpu_vm_fragment_size < 4)) {
990 dev_warn(adev->dev, "valid range is between 4 and 9\n");
991 amdgpu_vm_fragment_size = -1;
992 }
993
7951e376
RZ
994 amdgpu_device_check_smu_prv_buffer_size(adev);
995
06ec9070 996 amdgpu_device_check_vm_size(adev);
d38ceaf9 997
06ec9070 998 amdgpu_device_check_block_size(adev);
6a7f76e7 999
526bae37 1000 if (amdgpu_vram_page_split != -1 && (amdgpu_vram_page_split < 16 ||
76117507 1001 !is_power_of_2(amdgpu_vram_page_split))) {
6a7f76e7
CK
1002 dev_warn(adev->dev, "invalid VRAM page split (%d)\n",
1003 amdgpu_vram_page_split);
1004 amdgpu_vram_page_split = 1024;
1005 }
8854695a
AG
1006
1007 if (amdgpu_lockup_timeout == 0) {
1008 dev_warn(adev->dev, "lockup_timeout msut be > 0, adjusting to 10000\n");
1009 amdgpu_lockup_timeout = 10000;
1010 }
19aede77
AD
1011
1012 adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type);
d38ceaf9
AD
1013}
1014
1015/**
1016 * amdgpu_switcheroo_set_state - set switcheroo state
1017 *
1018 * @pdev: pci dev pointer
1694467b 1019 * @state: vga_switcheroo state
d38ceaf9
AD
1020 *
1021 * Callback for the switcheroo driver. Suspends or resumes the
1022 * the asics before or after it is powered up using ACPI methods.
1023 */
1024static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_state state)
1025{
1026 struct drm_device *dev = pci_get_drvdata(pdev);
1027
1028 if (amdgpu_device_is_px(dev) && state == VGA_SWITCHEROO_OFF)
1029 return;
1030
1031 if (state == VGA_SWITCHEROO_ON) {
7ca85295 1032 pr_info("amdgpu: switched on\n");
d38ceaf9
AD
1033 /* don't suspend or resume card normally */
1034 dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
1035
810ddc3a 1036 amdgpu_device_resume(dev, true, true);
d38ceaf9 1037
d38ceaf9
AD
1038 dev->switch_power_state = DRM_SWITCH_POWER_ON;
1039 drm_kms_helper_poll_enable(dev);
1040 } else {
7ca85295 1041 pr_info("amdgpu: switched off\n");
d38ceaf9
AD
1042 drm_kms_helper_poll_disable(dev);
1043 dev->switch_power_state = DRM_SWITCH_POWER_CHANGING;
810ddc3a 1044 amdgpu_device_suspend(dev, true, true);
d38ceaf9
AD
1045 dev->switch_power_state = DRM_SWITCH_POWER_OFF;
1046 }
1047}
1048
1049/**
1050 * amdgpu_switcheroo_can_switch - see if switcheroo state can change
1051 *
1052 * @pdev: pci dev pointer
1053 *
1054 * Callback for the switcheroo driver. Check of the switcheroo
1055 * state can be changed.
1056 * Returns true if the state can be changed, false if not.
1057 */
1058static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev)
1059{
1060 struct drm_device *dev = pci_get_drvdata(pdev);
1061
1062 /*
1063 * FIXME: open_count is protected by drm_global_mutex but that would lead to
1064 * locking inversion with the driver load path. And the access here is
1065 * completely racy anyway. So don't bother with locking for now.
1066 */
1067 return dev->open_count == 0;
1068}
1069
1070static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = {
1071 .set_gpu_state = amdgpu_switcheroo_set_state,
1072 .reprobe = NULL,
1073 .can_switch = amdgpu_switcheroo_can_switch,
1074};
1075
e3ecdffa
AD
1076/**
1077 * amdgpu_device_ip_set_clockgating_state - set the CG state
1078 *
87e3f136 1079 * @dev: amdgpu_device pointer
e3ecdffa
AD
1080 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
1081 * @state: clockgating state (gate or ungate)
1082 *
1083 * Sets the requested clockgating state for all instances of
1084 * the hardware IP specified.
1085 * Returns the error code from the last instance.
1086 */
43fa561f 1087int amdgpu_device_ip_set_clockgating_state(void *dev,
2990a1fc
AD
1088 enum amd_ip_block_type block_type,
1089 enum amd_clockgating_state state)
d38ceaf9 1090{
43fa561f 1091 struct amdgpu_device *adev = dev;
d38ceaf9
AD
1092 int i, r = 0;
1093
1094 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 1095 if (!adev->ip_blocks[i].status.valid)
9ecbe7f5 1096 continue;
c722865a
RZ
1097 if (adev->ip_blocks[i].version->type != block_type)
1098 continue;
1099 if (!adev->ip_blocks[i].version->funcs->set_clockgating_state)
1100 continue;
1101 r = adev->ip_blocks[i].version->funcs->set_clockgating_state(
1102 (void *)adev, state);
1103 if (r)
1104 DRM_ERROR("set_clockgating_state of IP block <%s> failed %d\n",
1105 adev->ip_blocks[i].version->funcs->name, r);
d38ceaf9
AD
1106 }
1107 return r;
1108}
1109
e3ecdffa
AD
1110/**
1111 * amdgpu_device_ip_set_powergating_state - set the PG state
1112 *
87e3f136 1113 * @dev: amdgpu_device pointer
e3ecdffa
AD
1114 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
1115 * @state: powergating state (gate or ungate)
1116 *
1117 * Sets the requested powergating state for all instances of
1118 * the hardware IP specified.
1119 * Returns the error code from the last instance.
1120 */
43fa561f 1121int amdgpu_device_ip_set_powergating_state(void *dev,
2990a1fc
AD
1122 enum amd_ip_block_type block_type,
1123 enum amd_powergating_state state)
d38ceaf9 1124{
43fa561f 1125 struct amdgpu_device *adev = dev;
d38ceaf9
AD
1126 int i, r = 0;
1127
1128 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 1129 if (!adev->ip_blocks[i].status.valid)
9ecbe7f5 1130 continue;
c722865a
RZ
1131 if (adev->ip_blocks[i].version->type != block_type)
1132 continue;
1133 if (!adev->ip_blocks[i].version->funcs->set_powergating_state)
1134 continue;
1135 r = adev->ip_blocks[i].version->funcs->set_powergating_state(
1136 (void *)adev, state);
1137 if (r)
1138 DRM_ERROR("set_powergating_state of IP block <%s> failed %d\n",
1139 adev->ip_blocks[i].version->funcs->name, r);
d38ceaf9
AD
1140 }
1141 return r;
1142}
1143
e3ecdffa
AD
1144/**
1145 * amdgpu_device_ip_get_clockgating_state - get the CG state
1146 *
1147 * @adev: amdgpu_device pointer
1148 * @flags: clockgating feature flags
1149 *
1150 * Walks the list of IPs on the device and updates the clockgating
1151 * flags for each IP.
1152 * Updates @flags with the feature flags for each hardware IP where
1153 * clockgating is enabled.
1154 */
2990a1fc
AD
1155void amdgpu_device_ip_get_clockgating_state(struct amdgpu_device *adev,
1156 u32 *flags)
6cb2d4e4
HR
1157{
1158 int i;
1159
1160 for (i = 0; i < adev->num_ip_blocks; i++) {
1161 if (!adev->ip_blocks[i].status.valid)
1162 continue;
1163 if (adev->ip_blocks[i].version->funcs->get_clockgating_state)
1164 adev->ip_blocks[i].version->funcs->get_clockgating_state((void *)adev, flags);
1165 }
1166}
1167
e3ecdffa
AD
1168/**
1169 * amdgpu_device_ip_wait_for_idle - wait for idle
1170 *
1171 * @adev: amdgpu_device pointer
1172 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
1173 *
1174 * Waits for the request hardware IP to be idle.
1175 * Returns 0 for success or a negative error code on failure.
1176 */
2990a1fc
AD
1177int amdgpu_device_ip_wait_for_idle(struct amdgpu_device *adev,
1178 enum amd_ip_block_type block_type)
5dbbb60b
AD
1179{
1180 int i, r;
1181
1182 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 1183 if (!adev->ip_blocks[i].status.valid)
9ecbe7f5 1184 continue;
a1255107
AD
1185 if (adev->ip_blocks[i].version->type == block_type) {
1186 r = adev->ip_blocks[i].version->funcs->wait_for_idle((void *)adev);
5dbbb60b
AD
1187 if (r)
1188 return r;
1189 break;
1190 }
1191 }
1192 return 0;
1193
1194}
1195
e3ecdffa
AD
1196/**
1197 * amdgpu_device_ip_is_idle - is the hardware IP idle
1198 *
1199 * @adev: amdgpu_device pointer
1200 * @block_type: Type of hardware IP (SMU, GFX, UVD, etc.)
1201 *
1202 * Check if the hardware IP is idle or not.
1203 * Returns true if it the IP is idle, false if not.
1204 */
2990a1fc
AD
1205bool amdgpu_device_ip_is_idle(struct amdgpu_device *adev,
1206 enum amd_ip_block_type block_type)
5dbbb60b
AD
1207{
1208 int i;
1209
1210 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 1211 if (!adev->ip_blocks[i].status.valid)
9ecbe7f5 1212 continue;
a1255107
AD
1213 if (adev->ip_blocks[i].version->type == block_type)
1214 return adev->ip_blocks[i].version->funcs->is_idle((void *)adev);
5dbbb60b
AD
1215 }
1216 return true;
1217
1218}
1219
e3ecdffa
AD
1220/**
1221 * amdgpu_device_ip_get_ip_block - get a hw IP pointer
1222 *
1223 * @adev: amdgpu_device pointer
87e3f136 1224 * @type: Type of hardware IP (SMU, GFX, UVD, etc.)
e3ecdffa
AD
1225 *
1226 * Returns a pointer to the hardware IP block structure
1227 * if it exists for the asic, otherwise NULL.
1228 */
2990a1fc
AD
1229struct amdgpu_ip_block *
1230amdgpu_device_ip_get_ip_block(struct amdgpu_device *adev,
1231 enum amd_ip_block_type type)
d38ceaf9
AD
1232{
1233 int i;
1234
1235 for (i = 0; i < adev->num_ip_blocks; i++)
a1255107 1236 if (adev->ip_blocks[i].version->type == type)
d38ceaf9
AD
1237 return &adev->ip_blocks[i];
1238
1239 return NULL;
1240}
1241
1242/**
2990a1fc 1243 * amdgpu_device_ip_block_version_cmp
d38ceaf9
AD
1244 *
1245 * @adev: amdgpu_device pointer
5fc3aeeb 1246 * @type: enum amd_ip_block_type
d38ceaf9
AD
1247 * @major: major version
1248 * @minor: minor version
1249 *
1250 * return 0 if equal or greater
1251 * return 1 if smaller or the ip_block doesn't exist
1252 */
2990a1fc
AD
1253int amdgpu_device_ip_block_version_cmp(struct amdgpu_device *adev,
1254 enum amd_ip_block_type type,
1255 u32 major, u32 minor)
d38ceaf9 1256{
2990a1fc 1257 struct amdgpu_ip_block *ip_block = amdgpu_device_ip_get_ip_block(adev, type);
d38ceaf9 1258
a1255107
AD
1259 if (ip_block && ((ip_block->version->major > major) ||
1260 ((ip_block->version->major == major) &&
1261 (ip_block->version->minor >= minor))))
d38ceaf9
AD
1262 return 0;
1263
1264 return 1;
1265}
1266
a1255107 1267/**
2990a1fc 1268 * amdgpu_device_ip_block_add
a1255107
AD
1269 *
1270 * @adev: amdgpu_device pointer
1271 * @ip_block_version: pointer to the IP to add
1272 *
1273 * Adds the IP block driver information to the collection of IPs
1274 * on the asic.
1275 */
2990a1fc
AD
1276int amdgpu_device_ip_block_add(struct amdgpu_device *adev,
1277 const struct amdgpu_ip_block_version *ip_block_version)
a1255107
AD
1278{
1279 if (!ip_block_version)
1280 return -EINVAL;
1281
e966a725 1282 DRM_INFO("add ip block number %d <%s>\n", adev->num_ip_blocks,
a0bae357
HR
1283 ip_block_version->funcs->name);
1284
a1255107
AD
1285 adev->ip_blocks[adev->num_ip_blocks++].version = ip_block_version;
1286
1287 return 0;
1288}
1289
e3ecdffa
AD
1290/**
1291 * amdgpu_device_enable_virtual_display - enable virtual display feature
1292 *
1293 * @adev: amdgpu_device pointer
1294 *
1295 * Enabled the virtual display feature if the user has enabled it via
1296 * the module parameter virtual_display. This feature provides a virtual
1297 * display hardware on headless boards or in virtualized environments.
1298 * This function parses and validates the configuration string specified by
1299 * the user and configues the virtual display configuration (number of
1300 * virtual connectors, crtcs, etc.) specified.
1301 */
483ef985 1302static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev)
9accf2fd
ED
1303{
1304 adev->enable_virtual_display = false;
1305
1306 if (amdgpu_virtual_display) {
1307 struct drm_device *ddev = adev->ddev;
1308 const char *pci_address_name = pci_name(ddev->pdev);
0f66356d 1309 char *pciaddstr, *pciaddstr_tmp, *pciaddname_tmp, *pciaddname;
9accf2fd
ED
1310
1311 pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL);
1312 pciaddstr_tmp = pciaddstr;
0f66356d
ED
1313 while ((pciaddname_tmp = strsep(&pciaddstr_tmp, ";"))) {
1314 pciaddname = strsep(&pciaddname_tmp, ",");
967de2a9
YT
1315 if (!strcmp("all", pciaddname)
1316 || !strcmp(pci_address_name, pciaddname)) {
0f66356d
ED
1317 long num_crtc;
1318 int res = -1;
1319
9accf2fd 1320 adev->enable_virtual_display = true;
0f66356d
ED
1321
1322 if (pciaddname_tmp)
1323 res = kstrtol(pciaddname_tmp, 10,
1324 &num_crtc);
1325
1326 if (!res) {
1327 if (num_crtc < 1)
1328 num_crtc = 1;
1329 if (num_crtc > 6)
1330 num_crtc = 6;
1331 adev->mode_info.num_crtc = num_crtc;
1332 } else {
1333 adev->mode_info.num_crtc = 1;
1334 }
9accf2fd
ED
1335 break;
1336 }
1337 }
1338
0f66356d
ED
1339 DRM_INFO("virtual display string:%s, %s:virtual_display:%d, num_crtc:%d\n",
1340 amdgpu_virtual_display, pci_address_name,
1341 adev->enable_virtual_display, adev->mode_info.num_crtc);
9accf2fd
ED
1342
1343 kfree(pciaddstr);
1344 }
1345}
1346
e3ecdffa
AD
1347/**
1348 * amdgpu_device_parse_gpu_info_fw - parse gpu info firmware
1349 *
1350 * @adev: amdgpu_device pointer
1351 *
1352 * Parses the asic configuration parameters specified in the gpu info
1353 * firmware and makes them availale to the driver for use in configuring
1354 * the asic.
1355 * Returns 0 on success, -EINVAL on failure.
1356 */
e2a75f88
AD
1357static int amdgpu_device_parse_gpu_info_fw(struct amdgpu_device *adev)
1358{
e2a75f88
AD
1359 const char *chip_name;
1360 char fw_name[30];
1361 int err;
1362 const struct gpu_info_firmware_header_v1_0 *hdr;
1363
ab4fe3e1
HR
1364 adev->firmware.gpu_info_fw = NULL;
1365
e2a75f88
AD
1366 switch (adev->asic_type) {
1367 case CHIP_TOPAZ:
1368 case CHIP_TONGA:
1369 case CHIP_FIJI:
e2a75f88 1370 case CHIP_POLARIS10:
cc07f18d 1371 case CHIP_POLARIS11:
e2a75f88 1372 case CHIP_POLARIS12:
cc07f18d 1373 case CHIP_VEGAM:
e2a75f88
AD
1374 case CHIP_CARRIZO:
1375 case CHIP_STONEY:
1376#ifdef CONFIG_DRM_AMDGPU_SI
1377 case CHIP_VERDE:
1378 case CHIP_TAHITI:
1379 case CHIP_PITCAIRN:
1380 case CHIP_OLAND:
1381 case CHIP_HAINAN:
1382#endif
1383#ifdef CONFIG_DRM_AMDGPU_CIK
1384 case CHIP_BONAIRE:
1385 case CHIP_HAWAII:
1386 case CHIP_KAVERI:
1387 case CHIP_KABINI:
1388 case CHIP_MULLINS:
1389#endif
27c0bc71 1390 case CHIP_VEGA20:
e2a75f88
AD
1391 default:
1392 return 0;
1393 case CHIP_VEGA10:
1394 chip_name = "vega10";
1395 break;
3f76dced
AD
1396 case CHIP_VEGA12:
1397 chip_name = "vega12";
1398 break;
2d2e5e7e
AD
1399 case CHIP_RAVEN:
1400 chip_name = "raven";
1401 break;
e2a75f88
AD
1402 }
1403
1404 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_gpu_info.bin", chip_name);
ab4fe3e1 1405 err = request_firmware(&adev->firmware.gpu_info_fw, fw_name, adev->dev);
e2a75f88
AD
1406 if (err) {
1407 dev_err(adev->dev,
1408 "Failed to load gpu_info firmware \"%s\"\n",
1409 fw_name);
1410 goto out;
1411 }
ab4fe3e1 1412 err = amdgpu_ucode_validate(adev->firmware.gpu_info_fw);
e2a75f88
AD
1413 if (err) {
1414 dev_err(adev->dev,
1415 "Failed to validate gpu_info firmware \"%s\"\n",
1416 fw_name);
1417 goto out;
1418 }
1419
ab4fe3e1 1420 hdr = (const struct gpu_info_firmware_header_v1_0 *)adev->firmware.gpu_info_fw->data;
e2a75f88
AD
1421 amdgpu_ucode_print_gpu_info_hdr(&hdr->header);
1422
1423 switch (hdr->version_major) {
1424 case 1:
1425 {
1426 const struct gpu_info_firmware_v1_0 *gpu_info_fw =
ab4fe3e1 1427 (const struct gpu_info_firmware_v1_0 *)(adev->firmware.gpu_info_fw->data +
e2a75f88
AD
1428 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1429
b5ab16bf
AD
1430 adev->gfx.config.max_shader_engines = le32_to_cpu(gpu_info_fw->gc_num_se);
1431 adev->gfx.config.max_cu_per_sh = le32_to_cpu(gpu_info_fw->gc_num_cu_per_sh);
1432 adev->gfx.config.max_sh_per_se = le32_to_cpu(gpu_info_fw->gc_num_sh_per_se);
1433 adev->gfx.config.max_backends_per_se = le32_to_cpu(gpu_info_fw->gc_num_rb_per_se);
e2a75f88 1434 adev->gfx.config.max_texture_channel_caches =
b5ab16bf
AD
1435 le32_to_cpu(gpu_info_fw->gc_num_tccs);
1436 adev->gfx.config.max_gprs = le32_to_cpu(gpu_info_fw->gc_num_gprs);
1437 adev->gfx.config.max_gs_threads = le32_to_cpu(gpu_info_fw->gc_num_max_gs_thds);
1438 adev->gfx.config.gs_vgt_table_depth = le32_to_cpu(gpu_info_fw->gc_gs_table_depth);
1439 adev->gfx.config.gs_prim_buffer_depth = le32_to_cpu(gpu_info_fw->gc_gsprim_buff_depth);
e2a75f88 1440 adev->gfx.config.double_offchip_lds_buf =
b5ab16bf
AD
1441 le32_to_cpu(gpu_info_fw->gc_double_offchip_lds_buffer);
1442 adev->gfx.cu_info.wave_front_size = le32_to_cpu(gpu_info_fw->gc_wave_size);
51fd0370
HZ
1443 adev->gfx.cu_info.max_waves_per_simd =
1444 le32_to_cpu(gpu_info_fw->gc_max_waves_per_simd);
1445 adev->gfx.cu_info.max_scratch_slots_per_cu =
1446 le32_to_cpu(gpu_info_fw->gc_max_scratch_slots_per_cu);
1447 adev->gfx.cu_info.lds_size = le32_to_cpu(gpu_info_fw->gc_lds_size);
e2a75f88
AD
1448 break;
1449 }
1450 default:
1451 dev_err(adev->dev,
1452 "Unsupported gpu_info table %d\n", hdr->header.ucode_version);
1453 err = -EINVAL;
1454 goto out;
1455 }
1456out:
e2a75f88
AD
1457 return err;
1458}
1459
e3ecdffa
AD
1460/**
1461 * amdgpu_device_ip_early_init - run early init for hardware IPs
1462 *
1463 * @adev: amdgpu_device pointer
1464 *
1465 * Early initialization pass for hardware IPs. The hardware IPs that make
1466 * up each asic are discovered each IP's early_init callback is run. This
1467 * is the first stage in initializing the asic.
1468 * Returns 0 on success, negative error code on failure.
1469 */
06ec9070 1470static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
d38ceaf9 1471{
aaa36a97 1472 int i, r;
d38ceaf9 1473
483ef985 1474 amdgpu_device_enable_virtual_display(adev);
a6be7570 1475
d38ceaf9 1476 switch (adev->asic_type) {
aaa36a97
AD
1477 case CHIP_TOPAZ:
1478 case CHIP_TONGA:
48299f95 1479 case CHIP_FIJI:
2cc0c0b5 1480 case CHIP_POLARIS10:
32cc7e53 1481 case CHIP_POLARIS11:
c4642a47 1482 case CHIP_POLARIS12:
32cc7e53 1483 case CHIP_VEGAM:
aaa36a97 1484 case CHIP_CARRIZO:
39bb0c92
SL
1485 case CHIP_STONEY:
1486 if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY)
aaa36a97
AD
1487 adev->family = AMDGPU_FAMILY_CZ;
1488 else
1489 adev->family = AMDGPU_FAMILY_VI;
1490
1491 r = vi_set_ip_blocks(adev);
1492 if (r)
1493 return r;
1494 break;
33f34802
KW
1495#ifdef CONFIG_DRM_AMDGPU_SI
1496 case CHIP_VERDE:
1497 case CHIP_TAHITI:
1498 case CHIP_PITCAIRN:
1499 case CHIP_OLAND:
1500 case CHIP_HAINAN:
295d0daf 1501 adev->family = AMDGPU_FAMILY_SI;
33f34802
KW
1502 r = si_set_ip_blocks(adev);
1503 if (r)
1504 return r;
1505 break;
1506#endif
a2e73f56
AD
1507#ifdef CONFIG_DRM_AMDGPU_CIK
1508 case CHIP_BONAIRE:
1509 case CHIP_HAWAII:
1510 case CHIP_KAVERI:
1511 case CHIP_KABINI:
1512 case CHIP_MULLINS:
1513 if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII))
1514 adev->family = AMDGPU_FAMILY_CI;
1515 else
1516 adev->family = AMDGPU_FAMILY_KV;
1517
1518 r = cik_set_ip_blocks(adev);
1519 if (r)
1520 return r;
1521 break;
1522#endif
e48a3cd9
AD
1523 case CHIP_VEGA10:
1524 case CHIP_VEGA12:
e4bd8170 1525 case CHIP_VEGA20:
e48a3cd9 1526 case CHIP_RAVEN:
2ca8a5d2
CZ
1527 if (adev->asic_type == CHIP_RAVEN)
1528 adev->family = AMDGPU_FAMILY_RV;
1529 else
1530 adev->family = AMDGPU_FAMILY_AI;
460826e6
KW
1531
1532 r = soc15_set_ip_blocks(adev);
1533 if (r)
1534 return r;
1535 break;
d38ceaf9
AD
1536 default:
1537 /* FIXME: not supported yet */
1538 return -EINVAL;
1539 }
1540
e2a75f88
AD
1541 r = amdgpu_device_parse_gpu_info_fw(adev);
1542 if (r)
1543 return r;
1544
1884734a 1545 amdgpu_amdkfd_device_probe(adev);
1546
3149d9da
XY
1547 if (amdgpu_sriov_vf(adev)) {
1548 r = amdgpu_virt_request_full_gpu(adev, true);
1549 if (r)
5ffa61c1 1550 return -EAGAIN;
3149d9da
XY
1551 }
1552
00f54b97
HR
1553 adev->powerplay.pp_feature = amdgpu_pp_feature_mask;
1554
d38ceaf9
AD
1555 for (i = 0; i < adev->num_ip_blocks; i++) {
1556 if ((amdgpu_ip_block_mask & (1 << i)) == 0) {
ed8cf00c
HR
1557 DRM_ERROR("disabled ip block: %d <%s>\n",
1558 i, adev->ip_blocks[i].version->funcs->name);
a1255107 1559 adev->ip_blocks[i].status.valid = false;
d38ceaf9 1560 } else {
a1255107
AD
1561 if (adev->ip_blocks[i].version->funcs->early_init) {
1562 r = adev->ip_blocks[i].version->funcs->early_init((void *)adev);
2c1a2784 1563 if (r == -ENOENT) {
a1255107 1564 adev->ip_blocks[i].status.valid = false;
2c1a2784 1565 } else if (r) {
a1255107
AD
1566 DRM_ERROR("early_init of IP block <%s> failed %d\n",
1567 adev->ip_blocks[i].version->funcs->name, r);
d38ceaf9 1568 return r;
2c1a2784 1569 } else {
a1255107 1570 adev->ip_blocks[i].status.valid = true;
2c1a2784 1571 }
974e6b64 1572 } else {
a1255107 1573 adev->ip_blocks[i].status.valid = true;
d38ceaf9 1574 }
d38ceaf9
AD
1575 }
1576 }
1577
395d1fb9
NH
1578 adev->cg_flags &= amdgpu_cg_mask;
1579 adev->pg_flags &= amdgpu_pg_mask;
1580
d38ceaf9
AD
1581 return 0;
1582}
1583
e3ecdffa
AD
1584/**
1585 * amdgpu_device_ip_init - run init for hardware IPs
1586 *
1587 * @adev: amdgpu_device pointer
1588 *
1589 * Main initialization pass for hardware IPs. The list of all the hardware
1590 * IPs that make up the asic is walked and the sw_init and hw_init callbacks
1591 * are run. sw_init initializes the software state associated with each IP
1592 * and hw_init initializes the hardware associated with each IP.
1593 * Returns 0 on success, negative error code on failure.
1594 */
06ec9070 1595static int amdgpu_device_ip_init(struct amdgpu_device *adev)
d38ceaf9
AD
1596{
1597 int i, r;
1598
1599 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 1600 if (!adev->ip_blocks[i].status.valid)
d38ceaf9 1601 continue;
a1255107 1602 r = adev->ip_blocks[i].version->funcs->sw_init((void *)adev);
2c1a2784 1603 if (r) {
a1255107
AD
1604 DRM_ERROR("sw_init of IP block <%s> failed %d\n",
1605 adev->ip_blocks[i].version->funcs->name, r);
d38ceaf9 1606 return r;
2c1a2784 1607 }
a1255107 1608 adev->ip_blocks[i].status.sw = true;
bfca0289 1609
d38ceaf9 1610 /* need to do gmc hw init early so we can allocate gpu mem */
a1255107 1611 if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
06ec9070 1612 r = amdgpu_device_vram_scratch_init(adev);
2c1a2784
AD
1613 if (r) {
1614 DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r);
d38ceaf9 1615 return r;
2c1a2784 1616 }
a1255107 1617 r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev);
2c1a2784
AD
1618 if (r) {
1619 DRM_ERROR("hw_init %d failed %d\n", i, r);
d38ceaf9 1620 return r;
2c1a2784 1621 }
06ec9070 1622 r = amdgpu_device_wb_init(adev);
2c1a2784 1623 if (r) {
06ec9070 1624 DRM_ERROR("amdgpu_device_wb_init failed %d\n", r);
d38ceaf9 1625 return r;
2c1a2784 1626 }
a1255107 1627 adev->ip_blocks[i].status.hw = true;
2493664f
ML
1628
1629 /* right after GMC hw init, we create CSA */
1630 if (amdgpu_sriov_vf(adev)) {
1631 r = amdgpu_allocate_static_csa(adev);
1632 if (r) {
1633 DRM_ERROR("allocate CSA failed %d\n", r);
1634 return r;
1635 }
1636 }
d38ceaf9
AD
1637 }
1638 }
1639
1640 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 1641 if (!adev->ip_blocks[i].status.sw)
d38ceaf9 1642 continue;
bfca0289 1643 if (adev->ip_blocks[i].status.hw)
d38ceaf9 1644 continue;
a1255107 1645 r = adev->ip_blocks[i].version->funcs->hw_init((void *)adev);
2c1a2784 1646 if (r) {
a1255107
AD
1647 DRM_ERROR("hw_init of IP block <%s> failed %d\n",
1648 adev->ip_blocks[i].version->funcs->name, r);
d38ceaf9 1649 return r;
2c1a2784 1650 }
a1255107 1651 adev->ip_blocks[i].status.hw = true;
d38ceaf9
AD
1652 }
1653
1884734a 1654 amdgpu_amdkfd_device_init(adev);
c6332b97 1655
1656 if (amdgpu_sriov_vf(adev))
1657 amdgpu_virt_release_full_gpu(adev, true);
1658
d38ceaf9
AD
1659 return 0;
1660}
1661
e3ecdffa
AD
1662/**
1663 * amdgpu_device_fill_reset_magic - writes reset magic to gart pointer
1664 *
1665 * @adev: amdgpu_device pointer
1666 *
1667 * Writes a reset magic value to the gart pointer in VRAM. The driver calls
1668 * this function before a GPU reset. If the value is retained after a
1669 * GPU reset, VRAM has not been lost. Some GPU resets may destry VRAM contents.
1670 */
06ec9070 1671static void amdgpu_device_fill_reset_magic(struct amdgpu_device *adev)
0c49e0b8
CZ
1672{
1673 memcpy(adev->reset_magic, adev->gart.ptr, AMDGPU_RESET_MAGIC_NUM);
1674}
1675
e3ecdffa
AD
1676/**
1677 * amdgpu_device_check_vram_lost - check if vram is valid
1678 *
1679 * @adev: amdgpu_device pointer
1680 *
1681 * Checks the reset magic value written to the gart pointer in VRAM.
1682 * The driver calls this after a GPU reset to see if the contents of
1683 * VRAM is lost or now.
1684 * returns true if vram is lost, false if not.
1685 */
06ec9070 1686static bool amdgpu_device_check_vram_lost(struct amdgpu_device *adev)
0c49e0b8
CZ
1687{
1688 return !!memcmp(adev->gart.ptr, adev->reset_magic,
1689 AMDGPU_RESET_MAGIC_NUM);
1690}
1691
e3ecdffa 1692/**
1112a46b 1693 * amdgpu_device_set_cg_state - set clockgating for amdgpu device
e3ecdffa
AD
1694 *
1695 * @adev: amdgpu_device pointer
1696 *
e3ecdffa 1697 * The list of all the hardware IPs that make up the asic is walked and the
1112a46b
RZ
1698 * set_clockgating_state callbacks are run.
1699 * Late initialization pass enabling clockgating for hardware IPs.
1700 * Fini or suspend, pass disabling clockgating for hardware IPs.
e3ecdffa
AD
1701 * Returns 0 on success, negative error code on failure.
1702 */
fdd34271 1703
1112a46b
RZ
1704static int amdgpu_device_set_cg_state(struct amdgpu_device *adev,
1705 enum amd_clockgating_state state)
d38ceaf9 1706{
1112a46b 1707 int i, j, r;
d38ceaf9 1708
4a2ba394
SL
1709 if (amdgpu_emu_mode == 1)
1710 return 0;
1711
1112a46b
RZ
1712 for (j = 0; j < adev->num_ip_blocks; j++) {
1713 i = state == AMD_CG_STATE_GATE ? j : adev->num_ip_blocks - j - 1;
a1255107 1714 if (!adev->ip_blocks[i].status.valid)
d38ceaf9 1715 continue;
4a446d55 1716 /* skip CG for VCE/UVD, it's handled specially */
a1255107 1717 if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
57716327 1718 adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
34319b32 1719 adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
57716327 1720 adev->ip_blocks[i].version->funcs->set_clockgating_state) {
4a446d55 1721 /* enable clockgating to save power */
a1255107 1722 r = adev->ip_blocks[i].version->funcs->set_clockgating_state((void *)adev,
1112a46b 1723 state);
4a446d55
AD
1724 if (r) {
1725 DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n",
a1255107 1726 adev->ip_blocks[i].version->funcs->name, r);
4a446d55
AD
1727 return r;
1728 }
b0b00ff1 1729 }
d38ceaf9 1730 }
06b18f61 1731
c9f96fd5
RZ
1732 return 0;
1733}
1734
1112a46b 1735static int amdgpu_device_set_pg_state(struct amdgpu_device *adev, enum amd_powergating_state state)
c9f96fd5 1736{
1112a46b 1737 int i, j, r;
06b18f61 1738
c9f96fd5
RZ
1739 if (amdgpu_emu_mode == 1)
1740 return 0;
1741
1112a46b
RZ
1742 for (j = 0; j < adev->num_ip_blocks; j++) {
1743 i = state == AMD_PG_STATE_GATE ? j : adev->num_ip_blocks - j - 1;
c9f96fd5
RZ
1744 if (!adev->ip_blocks[i].status.valid)
1745 continue;
1746 /* skip CG for VCE/UVD, it's handled specially */
1747 if (adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_UVD &&
1748 adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCE &&
1749 adev->ip_blocks[i].version->type != AMD_IP_BLOCK_TYPE_VCN &&
1750 adev->ip_blocks[i].version->funcs->set_powergating_state) {
1751 /* enable powergating to save power */
1752 r = adev->ip_blocks[i].version->funcs->set_powergating_state((void *)adev,
1112a46b 1753 state);
c9f96fd5
RZ
1754 if (r) {
1755 DRM_ERROR("set_powergating_state(gate) of IP block <%s> failed %d\n",
1756 adev->ip_blocks[i].version->funcs->name, r);
1757 return r;
1758 }
1759 }
1760 }
2dc80b00
S
1761 return 0;
1762}
1763
e3ecdffa
AD
1764/**
1765 * amdgpu_device_ip_late_init - run late init for hardware IPs
1766 *
1767 * @adev: amdgpu_device pointer
1768 *
1769 * Late initialization pass for hardware IPs. The list of all the hardware
1770 * IPs that make up the asic is walked and the late_init callbacks are run.
1771 * late_init covers any special initialization that an IP requires
1772 * after all of the have been initialized or something that needs to happen
1773 * late in the init process.
1774 * Returns 0 on success, negative error code on failure.
1775 */
06ec9070 1776static int amdgpu_device_ip_late_init(struct amdgpu_device *adev)
2dc80b00
S
1777{
1778 int i = 0, r;
1779
1780 for (i = 0; i < adev->num_ip_blocks; i++) {
1781 if (!adev->ip_blocks[i].status.valid)
1782 continue;
1783 if (adev->ip_blocks[i].version->funcs->late_init) {
1784 r = adev->ip_blocks[i].version->funcs->late_init((void *)adev);
1785 if (r) {
1786 DRM_ERROR("late_init of IP block <%s> failed %d\n",
1787 adev->ip_blocks[i].version->funcs->name, r);
1788 return r;
1789 }
1790 adev->ip_blocks[i].status.late_initialized = true;
1791 }
1792 }
1793
1112a46b
RZ
1794 amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE);
1795 amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE);
916ac57f 1796
2c773de2
S
1797 queue_delayed_work(system_wq, &adev->late_init_work,
1798 msecs_to_jiffies(AMDGPU_RESUME_MS));
d38ceaf9 1799
06ec9070 1800 amdgpu_device_fill_reset_magic(adev);
d38ceaf9
AD
1801
1802 return 0;
1803}
1804
e3ecdffa
AD
1805/**
1806 * amdgpu_device_ip_fini - run fini for hardware IPs
1807 *
1808 * @adev: amdgpu_device pointer
1809 *
1810 * Main teardown pass for hardware IPs. The list of all the hardware
1811 * IPs that make up the asic is walked and the hw_fini and sw_fini callbacks
1812 * are run. hw_fini tears down the hardware associated with each IP
1813 * and sw_fini tears down any software state associated with each IP.
1814 * Returns 0 on success, negative error code on failure.
1815 */
06ec9070 1816static int amdgpu_device_ip_fini(struct amdgpu_device *adev)
d38ceaf9
AD
1817{
1818 int i, r;
1819
1884734a 1820 amdgpu_amdkfd_device_fini(adev);
05df1f01
RZ
1821
1822 amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
fdd34271
RZ
1823 amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
1824
3e96dbfd
AD
1825 /* need to disable SMC first */
1826 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 1827 if (!adev->ip_blocks[i].status.hw)
3e96dbfd 1828 continue;
fdd34271 1829 if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) {
a1255107 1830 r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
3e96dbfd
AD
1831 /* XXX handle errors */
1832 if (r) {
1833 DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
a1255107 1834 adev->ip_blocks[i].version->funcs->name, r);
3e96dbfd 1835 }
a1255107 1836 adev->ip_blocks[i].status.hw = false;
3e96dbfd
AD
1837 break;
1838 }
1839 }
1840
d38ceaf9 1841 for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
a1255107 1842 if (!adev->ip_blocks[i].status.hw)
d38ceaf9 1843 continue;
8201a67a 1844
a1255107 1845 r = adev->ip_blocks[i].version->funcs->hw_fini((void *)adev);
d38ceaf9 1846 /* XXX handle errors */
2c1a2784 1847 if (r) {
a1255107
AD
1848 DRM_DEBUG("hw_fini of IP block <%s> failed %d\n",
1849 adev->ip_blocks[i].version->funcs->name, r);
2c1a2784 1850 }
8201a67a 1851
a1255107 1852 adev->ip_blocks[i].status.hw = false;
d38ceaf9
AD
1853 }
1854
9950cda2 1855
d38ceaf9 1856 for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
a1255107 1857 if (!adev->ip_blocks[i].status.sw)
d38ceaf9 1858 continue;
c12aba3a
ML
1859
1860 if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) {
1861 amdgpu_free_static_csa(adev);
1862 amdgpu_device_wb_fini(adev);
1863 amdgpu_device_vram_scratch_fini(adev);
1864 }
1865
a1255107 1866 r = adev->ip_blocks[i].version->funcs->sw_fini((void *)adev);
d38ceaf9 1867 /* XXX handle errors */
2c1a2784 1868 if (r) {
a1255107
AD
1869 DRM_DEBUG("sw_fini of IP block <%s> failed %d\n",
1870 adev->ip_blocks[i].version->funcs->name, r);
2c1a2784 1871 }
a1255107
AD
1872 adev->ip_blocks[i].status.sw = false;
1873 adev->ip_blocks[i].status.valid = false;
d38ceaf9
AD
1874 }
1875
a6dcfd9c 1876 for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
a1255107 1877 if (!adev->ip_blocks[i].status.late_initialized)
8a2eef1d 1878 continue;
a1255107
AD
1879 if (adev->ip_blocks[i].version->funcs->late_fini)
1880 adev->ip_blocks[i].version->funcs->late_fini((void *)adev);
1881 adev->ip_blocks[i].status.late_initialized = false;
a6dcfd9c
ML
1882 }
1883
030308fc 1884 if (amdgpu_sriov_vf(adev))
24136135
ML
1885 if (amdgpu_virt_release_full_gpu(adev, false))
1886 DRM_ERROR("failed to release exclusive mode on fini\n");
2493664f 1887
d38ceaf9
AD
1888 return 0;
1889}
1890
e3ecdffa 1891/**
1112a46b 1892 * amdgpu_device_ip_late_init_func_handler - work handler for ib test
e3ecdffa 1893 *
1112a46b 1894 * @work: work_struct.
e3ecdffa 1895 */
06ec9070 1896static void amdgpu_device_ip_late_init_func_handler(struct work_struct *work)
2dc80b00
S
1897{
1898 struct amdgpu_device *adev =
1899 container_of(work, struct amdgpu_device, late_init_work.work);
916ac57f
RZ
1900 int r;
1901
1902 r = amdgpu_ib_ring_tests(adev);
1903 if (r)
1904 DRM_ERROR("ib ring test failed (%d).\n", r);
2dc80b00
S
1905}
1906
1e317b99
RZ
1907static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work)
1908{
1909 struct amdgpu_device *adev =
1910 container_of(work, struct amdgpu_device, gfx.gfx_off_delay_work.work);
1911
1912 mutex_lock(&adev->gfx.gfx_off_mutex);
1913 if (!adev->gfx.gfx_off_state && !adev->gfx.gfx_off_req_count) {
1914 if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true))
1915 adev->gfx.gfx_off_state = true;
1916 }
1917 mutex_unlock(&adev->gfx.gfx_off_mutex);
1918}
1919
e3ecdffa 1920/**
e7854a03 1921 * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1)
e3ecdffa
AD
1922 *
1923 * @adev: amdgpu_device pointer
1924 *
1925 * Main suspend function for hardware IPs. The list of all the hardware
1926 * IPs that make up the asic is walked, clockgating is disabled and the
1927 * suspend callbacks are run. suspend puts the hardware and software state
1928 * in each IP into a state suitable for suspend.
1929 * Returns 0 on success, negative error code on failure.
1930 */
e7854a03
AD
1931static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev)
1932{
1933 int i, r;
1934
1935 if (amdgpu_sriov_vf(adev))
1936 amdgpu_virt_request_full_gpu(adev, false);
1937
05df1f01 1938 amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE);
fdd34271 1939 amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE);
05df1f01 1940
e7854a03
AD
1941 for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
1942 if (!adev->ip_blocks[i].status.valid)
1943 continue;
1944 /* displays are handled separately */
1945 if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) {
e7854a03
AD
1946 /* XXX handle errors */
1947 r = adev->ip_blocks[i].version->funcs->suspend(adev);
1948 /* XXX handle errors */
1949 if (r) {
1950 DRM_ERROR("suspend of IP block <%s> failed %d\n",
1951 adev->ip_blocks[i].version->funcs->name, r);
1952 }
1953 }
1954 }
1955
1956 if (amdgpu_sriov_vf(adev))
1957 amdgpu_virt_release_full_gpu(adev, false);
1958
1959 return 0;
1960}
1961
1962/**
1963 * amdgpu_device_ip_suspend_phase2 - run suspend for hardware IPs (phase 2)
1964 *
1965 * @adev: amdgpu_device pointer
1966 *
1967 * Main suspend function for hardware IPs. The list of all the hardware
1968 * IPs that make up the asic is walked, clockgating is disabled and the
1969 * suspend callbacks are run. suspend puts the hardware and software state
1970 * in each IP into a state suitable for suspend.
1971 * Returns 0 on success, negative error code on failure.
1972 */
1973static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev)
d38ceaf9
AD
1974{
1975 int i, r;
1976
1977 for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
a1255107 1978 if (!adev->ip_blocks[i].status.valid)
d38ceaf9 1979 continue;
e7854a03
AD
1980 /* displays are handled in phase1 */
1981 if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE)
1982 continue;
d38ceaf9 1983 /* XXX handle errors */
a1255107 1984 r = adev->ip_blocks[i].version->funcs->suspend(adev);
d38ceaf9 1985 /* XXX handle errors */
2c1a2784 1986 if (r) {
a1255107
AD
1987 DRM_ERROR("suspend of IP block <%s> failed %d\n",
1988 adev->ip_blocks[i].version->funcs->name, r);
2c1a2784 1989 }
d38ceaf9
AD
1990 }
1991
1992 return 0;
1993}
1994
e7854a03
AD
1995/**
1996 * amdgpu_device_ip_suspend - run suspend for hardware IPs
1997 *
1998 * @adev: amdgpu_device pointer
1999 *
2000 * Main suspend function for hardware IPs. The list of all the hardware
2001 * IPs that make up the asic is walked, clockgating is disabled and the
2002 * suspend callbacks are run. suspend puts the hardware and software state
2003 * in each IP into a state suitable for suspend.
2004 * Returns 0 on success, negative error code on failure.
2005 */
2006int amdgpu_device_ip_suspend(struct amdgpu_device *adev)
2007{
2008 int r;
2009
2010 r = amdgpu_device_ip_suspend_phase1(adev);
2011 if (r)
2012 return r;
2013 r = amdgpu_device_ip_suspend_phase2(adev);
2014
2015 return r;
2016}
2017
06ec9070 2018static int amdgpu_device_ip_reinit_early_sriov(struct amdgpu_device *adev)
a90ad3c2
ML
2019{
2020 int i, r;
2021
2cb681b6
ML
2022 static enum amd_ip_block_type ip_order[] = {
2023 AMD_IP_BLOCK_TYPE_GMC,
2024 AMD_IP_BLOCK_TYPE_COMMON,
2cb681b6
ML
2025 AMD_IP_BLOCK_TYPE_IH,
2026 };
a90ad3c2 2027
2cb681b6
ML
2028 for (i = 0; i < ARRAY_SIZE(ip_order); i++) {
2029 int j;
2030 struct amdgpu_ip_block *block;
a90ad3c2 2031
2cb681b6
ML
2032 for (j = 0; j < adev->num_ip_blocks; j++) {
2033 block = &adev->ip_blocks[j];
2034
2035 if (block->version->type != ip_order[i] ||
2036 !block->status.valid)
2037 continue;
2038
2039 r = block->version->funcs->hw_init(adev);
3f48c681 2040 DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded");
c41d1cf6
ML
2041 if (r)
2042 return r;
a90ad3c2
ML
2043 }
2044 }
2045
2046 return 0;
2047}
2048
06ec9070 2049static int amdgpu_device_ip_reinit_late_sriov(struct amdgpu_device *adev)
a90ad3c2
ML
2050{
2051 int i, r;
2052
2cb681b6
ML
2053 static enum amd_ip_block_type ip_order[] = {
2054 AMD_IP_BLOCK_TYPE_SMC,
ef4c166d 2055 AMD_IP_BLOCK_TYPE_PSP,
2cb681b6
ML
2056 AMD_IP_BLOCK_TYPE_DCE,
2057 AMD_IP_BLOCK_TYPE_GFX,
2058 AMD_IP_BLOCK_TYPE_SDMA,
257deb8c
FM
2059 AMD_IP_BLOCK_TYPE_UVD,
2060 AMD_IP_BLOCK_TYPE_VCE
2cb681b6 2061 };
a90ad3c2 2062
2cb681b6
ML
2063 for (i = 0; i < ARRAY_SIZE(ip_order); i++) {
2064 int j;
2065 struct amdgpu_ip_block *block;
a90ad3c2 2066
2cb681b6
ML
2067 for (j = 0; j < adev->num_ip_blocks; j++) {
2068 block = &adev->ip_blocks[j];
2069
2070 if (block->version->type != ip_order[i] ||
2071 !block->status.valid)
2072 continue;
2073
2074 r = block->version->funcs->hw_init(adev);
3f48c681 2075 DRM_INFO("RE-INIT: %s %s\n", block->version->funcs->name, r?"failed":"succeeded");
c41d1cf6
ML
2076 if (r)
2077 return r;
a90ad3c2
ML
2078 }
2079 }
2080
2081 return 0;
2082}
2083
e3ecdffa
AD
2084/**
2085 * amdgpu_device_ip_resume_phase1 - run resume for hardware IPs
2086 *
2087 * @adev: amdgpu_device pointer
2088 *
2089 * First resume function for hardware IPs. The list of all the hardware
2090 * IPs that make up the asic is walked and the resume callbacks are run for
2091 * COMMON, GMC, and IH. resume puts the hardware into a functional state
2092 * after a suspend and updates the software state as necessary. This
2093 * function is also used for restoring the GPU after a GPU reset.
2094 * Returns 0 on success, negative error code on failure.
2095 */
06ec9070 2096static int amdgpu_device_ip_resume_phase1(struct amdgpu_device *adev)
d38ceaf9
AD
2097{
2098 int i, r;
2099
a90ad3c2
ML
2100 for (i = 0; i < adev->num_ip_blocks; i++) {
2101 if (!adev->ip_blocks[i].status.valid)
2102 continue;
a90ad3c2 2103 if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
e3ecdffa
AD
2104 adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
2105 adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH) {
fcf0649f
CZ
2106 r = adev->ip_blocks[i].version->funcs->resume(adev);
2107 if (r) {
2108 DRM_ERROR("resume of IP block <%s> failed %d\n",
2109 adev->ip_blocks[i].version->funcs->name, r);
2110 return r;
2111 }
a90ad3c2
ML
2112 }
2113 }
2114
2115 return 0;
2116}
2117
e3ecdffa
AD
2118/**
2119 * amdgpu_device_ip_resume_phase2 - run resume for hardware IPs
2120 *
2121 * @adev: amdgpu_device pointer
2122 *
2123 * First resume function for hardware IPs. The list of all the hardware
2124 * IPs that make up the asic is walked and the resume callbacks are run for
2125 * all blocks except COMMON, GMC, and IH. resume puts the hardware into a
2126 * functional state after a suspend and updates the software state as
2127 * necessary. This function is also used for restoring the GPU after a GPU
2128 * reset.
2129 * Returns 0 on success, negative error code on failure.
2130 */
06ec9070 2131static int amdgpu_device_ip_resume_phase2(struct amdgpu_device *adev)
d38ceaf9
AD
2132{
2133 int i, r;
2134
2135 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 2136 if (!adev->ip_blocks[i].status.valid)
d38ceaf9 2137 continue;
fcf0649f 2138 if (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_COMMON ||
e3ecdffa
AD
2139 adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC ||
2140 adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_IH)
fcf0649f 2141 continue;
a1255107 2142 r = adev->ip_blocks[i].version->funcs->resume(adev);
2c1a2784 2143 if (r) {
a1255107
AD
2144 DRM_ERROR("resume of IP block <%s> failed %d\n",
2145 adev->ip_blocks[i].version->funcs->name, r);
d38ceaf9 2146 return r;
2c1a2784 2147 }
d38ceaf9
AD
2148 }
2149
2150 return 0;
2151}
2152
e3ecdffa
AD
2153/**
2154 * amdgpu_device_ip_resume - run resume for hardware IPs
2155 *
2156 * @adev: amdgpu_device pointer
2157 *
2158 * Main resume function for hardware IPs. The hardware IPs
2159 * are split into two resume functions because they are
2160 * are also used in in recovering from a GPU reset and some additional
2161 * steps need to be take between them. In this case (S3/S4) they are
2162 * run sequentially.
2163 * Returns 0 on success, negative error code on failure.
2164 */
06ec9070 2165static int amdgpu_device_ip_resume(struct amdgpu_device *adev)
fcf0649f
CZ
2166{
2167 int r;
2168
06ec9070 2169 r = amdgpu_device_ip_resume_phase1(adev);
fcf0649f
CZ
2170 if (r)
2171 return r;
06ec9070 2172 r = amdgpu_device_ip_resume_phase2(adev);
fcf0649f
CZ
2173
2174 return r;
2175}
2176
e3ecdffa
AD
2177/**
2178 * amdgpu_device_detect_sriov_bios - determine if the board supports SR-IOV
2179 *
2180 * @adev: amdgpu_device pointer
2181 *
2182 * Query the VBIOS data tables to determine if the board supports SR-IOV.
2183 */
4e99a44e 2184static void amdgpu_device_detect_sriov_bios(struct amdgpu_device *adev)
048765ad 2185{
6867e1b5
ML
2186 if (amdgpu_sriov_vf(adev)) {
2187 if (adev->is_atom_fw) {
2188 if (amdgpu_atomfirmware_gpu_supports_virtualization(adev))
2189 adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
2190 } else {
2191 if (amdgpu_atombios_has_gpu_virtualization_table(adev))
2192 adev->virt.caps |= AMDGPU_SRIOV_CAPS_SRIOV_VBIOS;
2193 }
2194
2195 if (!(adev->virt.caps & AMDGPU_SRIOV_CAPS_SRIOV_VBIOS))
2196 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_NO_VBIOS, 0, 0);
a5bde2f9 2197 }
048765ad
AR
2198}
2199
e3ecdffa
AD
2200/**
2201 * amdgpu_device_asic_has_dc_support - determine if DC supports the asic
2202 *
2203 * @asic_type: AMD asic type
2204 *
2205 * Check if there is DC (new modesetting infrastructre) support for an asic.
2206 * returns true if DC has support, false if not.
2207 */
4562236b
HW
2208bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type)
2209{
2210 switch (asic_type) {
2211#if defined(CONFIG_DRM_AMD_DC)
2212 case CHIP_BONAIRE:
0d6fbccb 2213 case CHIP_KAVERI:
367e6687
AD
2214 case CHIP_KABINI:
2215 case CHIP_MULLINS:
d9fda248
HW
2216 /*
2217 * We have systems in the wild with these ASICs that require
2218 * LVDS and VGA support which is not supported with DC.
2219 *
2220 * Fallback to the non-DC driver here by default so as not to
2221 * cause regressions.
2222 */
2223 return amdgpu_dc > 0;
2224 case CHIP_HAWAII:
4562236b
HW
2225 case CHIP_CARRIZO:
2226 case CHIP_STONEY:
4562236b 2227 case CHIP_POLARIS10:
675fd32b 2228 case CHIP_POLARIS11:
2c8ad2d5 2229 case CHIP_POLARIS12:
675fd32b 2230 case CHIP_VEGAM:
4562236b
HW
2231 case CHIP_TONGA:
2232 case CHIP_FIJI:
42f8ffa1 2233 case CHIP_VEGA10:
dca7b401 2234 case CHIP_VEGA12:
c6034aa2 2235 case CHIP_VEGA20:
dc37a9a0 2236#if defined(CONFIG_DRM_AMD_DC_DCN1_0)
fd187853 2237 case CHIP_RAVEN:
42f8ffa1 2238#endif
fd187853 2239 return amdgpu_dc != 0;
4562236b
HW
2240#endif
2241 default:
2242 return false;
2243 }
2244}
2245
2246/**
2247 * amdgpu_device_has_dc_support - check if dc is supported
2248 *
2249 * @adev: amdgpu_device_pointer
2250 *
2251 * Returns true for supported, false for not supported
2252 */
2253bool amdgpu_device_has_dc_support(struct amdgpu_device *adev)
2254{
2555039d
XY
2255 if (amdgpu_sriov_vf(adev))
2256 return false;
2257
4562236b
HW
2258 return amdgpu_device_asic_has_dc_support(adev->asic_type);
2259}
2260
d38ceaf9
AD
2261/**
2262 * amdgpu_device_init - initialize the driver
2263 *
2264 * @adev: amdgpu_device pointer
87e3f136 2265 * @ddev: drm dev pointer
d38ceaf9
AD
2266 * @pdev: pci dev pointer
2267 * @flags: driver flags
2268 *
2269 * Initializes the driver info and hw (all asics).
2270 * Returns 0 for success or an error on failure.
2271 * Called at driver startup.
2272 */
2273int amdgpu_device_init(struct amdgpu_device *adev,
2274 struct drm_device *ddev,
2275 struct pci_dev *pdev,
2276 uint32_t flags)
2277{
2278 int r, i;
2279 bool runtime = false;
95844d20 2280 u32 max_MBps;
d38ceaf9
AD
2281
2282 adev->shutdown = false;
2283 adev->dev = &pdev->dev;
2284 adev->ddev = ddev;
2285 adev->pdev = pdev;
2286 adev->flags = flags;
2f7d10b3 2287 adev->asic_type = flags & AMD_ASIC_MASK;
d38ceaf9 2288 adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT;
593aa2d2
SL
2289 if (amdgpu_emu_mode == 1)
2290 adev->usec_timeout *= 2;
770d13b1 2291 adev->gmc.gart_size = 512 * 1024 * 1024;
d38ceaf9
AD
2292 adev->accel_working = false;
2293 adev->num_rings = 0;
2294 adev->mman.buffer_funcs = NULL;
2295 adev->mman.buffer_funcs_ring = NULL;
2296 adev->vm_manager.vm_pte_funcs = NULL;
3798e9a6 2297 adev->vm_manager.vm_pte_num_rqs = 0;
132f34e4 2298 adev->gmc.gmc_funcs = NULL;
f54d1867 2299 adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS);
b8866c26 2300 bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
d38ceaf9
AD
2301
2302 adev->smc_rreg = &amdgpu_invalid_rreg;
2303 adev->smc_wreg = &amdgpu_invalid_wreg;
2304 adev->pcie_rreg = &amdgpu_invalid_rreg;
2305 adev->pcie_wreg = &amdgpu_invalid_wreg;
36b9a952
HR
2306 adev->pciep_rreg = &amdgpu_invalid_rreg;
2307 adev->pciep_wreg = &amdgpu_invalid_wreg;
d38ceaf9
AD
2308 adev->uvd_ctx_rreg = &amdgpu_invalid_rreg;
2309 adev->uvd_ctx_wreg = &amdgpu_invalid_wreg;
2310 adev->didt_rreg = &amdgpu_invalid_rreg;
2311 adev->didt_wreg = &amdgpu_invalid_wreg;
ccdbb20a
RZ
2312 adev->gc_cac_rreg = &amdgpu_invalid_rreg;
2313 adev->gc_cac_wreg = &amdgpu_invalid_wreg;
d38ceaf9
AD
2314 adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg;
2315 adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg;
2316
3e39ab90
AD
2317 DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n",
2318 amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device,
2319 pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision);
d38ceaf9
AD
2320
2321 /* mutex initialization are all done here so we
2322 * can recall function without having locking issues */
d38ceaf9 2323 atomic_set(&adev->irq.ih.lock, 0);
0e5ca0d1 2324 mutex_init(&adev->firmware.mutex);
d38ceaf9
AD
2325 mutex_init(&adev->pm.mutex);
2326 mutex_init(&adev->gfx.gpu_clock_mutex);
2327 mutex_init(&adev->srbm_mutex);
b8866c26 2328 mutex_init(&adev->gfx.pipe_reserve_mutex);
d23ee13f 2329 mutex_init(&adev->gfx.gfx_off_mutex);
d38ceaf9 2330 mutex_init(&adev->grbm_idx_mutex);
d38ceaf9 2331 mutex_init(&adev->mn_lock);
e23b74aa 2332 mutex_init(&adev->virt.vf_errors.lock);
d38ceaf9 2333 hash_init(adev->mn_hash);
13a752e3 2334 mutex_init(&adev->lock_reset);
d38ceaf9 2335
06ec9070 2336 amdgpu_device_check_arguments(adev);
d38ceaf9 2337
d38ceaf9
AD
2338 spin_lock_init(&adev->mmio_idx_lock);
2339 spin_lock_init(&adev->smc_idx_lock);
2340 spin_lock_init(&adev->pcie_idx_lock);
2341 spin_lock_init(&adev->uvd_ctx_idx_lock);
2342 spin_lock_init(&adev->didt_idx_lock);
ccdbb20a 2343 spin_lock_init(&adev->gc_cac_idx_lock);
16abb5d2 2344 spin_lock_init(&adev->se_cac_idx_lock);
d38ceaf9 2345 spin_lock_init(&adev->audio_endpt_idx_lock);
95844d20 2346 spin_lock_init(&adev->mm_stats.lock);
d38ceaf9 2347
0c4e7fa5
CZ
2348 INIT_LIST_HEAD(&adev->shadow_list);
2349 mutex_init(&adev->shadow_list_lock);
2350
795f2813
AR
2351 INIT_LIST_HEAD(&adev->ring_lru_list);
2352 spin_lock_init(&adev->ring_lru_list_lock);
2353
06ec9070
AD
2354 INIT_DELAYED_WORK(&adev->late_init_work,
2355 amdgpu_device_ip_late_init_func_handler);
1e317b99
RZ
2356 INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
2357 amdgpu_device_delay_enable_gfx_off);
2dc80b00 2358
d23ee13f 2359 adev->gfx.gfx_off_req_count = 1;
b1ddf548
RZ
2360 adev->pm.ac_power = power_supply_is_system_supplied() > 0 ? true : false;
2361
0fa49558
AX
2362 /* Registers mapping */
2363 /* TODO: block userspace mapping of io register */
da69c161
KW
2364 if (adev->asic_type >= CHIP_BONAIRE) {
2365 adev->rmmio_base = pci_resource_start(adev->pdev, 5);
2366 adev->rmmio_size = pci_resource_len(adev->pdev, 5);
2367 } else {
2368 adev->rmmio_base = pci_resource_start(adev->pdev, 2);
2369 adev->rmmio_size = pci_resource_len(adev->pdev, 2);
2370 }
d38ceaf9 2371
d38ceaf9
AD
2372 adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size);
2373 if (adev->rmmio == NULL) {
2374 return -ENOMEM;
2375 }
2376 DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base);
2377 DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size);
2378
705e519e 2379 /* doorbell bar mapping */
06ec9070 2380 amdgpu_device_doorbell_init(adev);
d38ceaf9
AD
2381
2382 /* io port mapping */
2383 for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
2384 if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) {
2385 adev->rio_mem_size = pci_resource_len(adev->pdev, i);
2386 adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size);
2387 break;
2388 }
2389 }
2390 if (adev->rio_mem == NULL)
b64a18c5 2391 DRM_INFO("PCI I/O BAR is not found.\n");
d38ceaf9 2392
5494d864
AD
2393 amdgpu_device_get_pcie_info(adev);
2394
d38ceaf9 2395 /* early init functions */
06ec9070 2396 r = amdgpu_device_ip_early_init(adev);
d38ceaf9
AD
2397 if (r)
2398 return r;
2399
2400 /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */
2401 /* this will fail for cards that aren't VGA class devices, just
2402 * ignore it */
06ec9070 2403 vga_client_register(adev->pdev, adev, NULL, amdgpu_device_vga_set_decode);
d38ceaf9 2404
e9bef455 2405 if (amdgpu_device_is_px(ddev))
d38ceaf9 2406 runtime = true;
84c8b22e
LW
2407 if (!pci_is_thunderbolt_attached(adev->pdev))
2408 vga_switcheroo_register_client(adev->pdev,
2409 &amdgpu_switcheroo_ops, runtime);
d38ceaf9
AD
2410 if (runtime)
2411 vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain);
2412
9475a943
SL
2413 if (amdgpu_emu_mode == 1) {
2414 /* post the asic on emulation mode */
2415 emu_soc_asic_init(adev);
bfca0289 2416 goto fence_driver_init;
9475a943 2417 }
bfca0289 2418
d38ceaf9 2419 /* Read BIOS */
83ba126a
AD
2420 if (!amdgpu_get_bios(adev)) {
2421 r = -EINVAL;
2422 goto failed;
2423 }
f7e9e9fe 2424
d38ceaf9 2425 r = amdgpu_atombios_init(adev);
2c1a2784
AD
2426 if (r) {
2427 dev_err(adev->dev, "amdgpu_atombios_init failed\n");
e23b74aa 2428 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0);
83ba126a 2429 goto failed;
2c1a2784 2430 }
d38ceaf9 2431
4e99a44e
ML
2432 /* detect if we are with an SRIOV vbios */
2433 amdgpu_device_detect_sriov_bios(adev);
048765ad 2434
d38ceaf9 2435 /* Post card if necessary */
39c640c0 2436 if (amdgpu_device_need_post(adev)) {
d38ceaf9 2437 if (!adev->bios) {
bec86378 2438 dev_err(adev->dev, "no vBIOS found\n");
83ba126a
AD
2439 r = -EINVAL;
2440 goto failed;
d38ceaf9 2441 }
bec86378 2442 DRM_INFO("GPU posting now...\n");
4e99a44e
ML
2443 r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
2444 if (r) {
2445 dev_err(adev->dev, "gpu post error!\n");
2446 goto failed;
2447 }
d38ceaf9
AD
2448 }
2449
88b64e95
AD
2450 if (adev->is_atom_fw) {
2451 /* Initialize clocks */
2452 r = amdgpu_atomfirmware_get_clock_info(adev);
2453 if (r) {
2454 dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n");
e23b74aa 2455 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
88b64e95
AD
2456 goto failed;
2457 }
2458 } else {
a5bde2f9
AD
2459 /* Initialize clocks */
2460 r = amdgpu_atombios_get_clock_info(adev);
2461 if (r) {
2462 dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n");
e23b74aa 2463 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
89041940 2464 goto failed;
a5bde2f9
AD
2465 }
2466 /* init i2c buses */
4562236b
HW
2467 if (!amdgpu_device_has_dc_support(adev))
2468 amdgpu_atombios_i2c_init(adev);
2c1a2784 2469 }
d38ceaf9 2470
bfca0289 2471fence_driver_init:
d38ceaf9
AD
2472 /* Fence driver */
2473 r = amdgpu_fence_driver_init(adev);
2c1a2784
AD
2474 if (r) {
2475 dev_err(adev->dev, "amdgpu_fence_driver_init failed\n");
e23b74aa 2476 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_FENCE_INIT_FAIL, 0, 0);
83ba126a 2477 goto failed;
2c1a2784 2478 }
d38ceaf9
AD
2479
2480 /* init the mode config */
2481 drm_mode_config_init(adev->ddev);
2482
06ec9070 2483 r = amdgpu_device_ip_init(adev);
d38ceaf9 2484 if (r) {
8840a387 2485 /* failed in exclusive mode due to timeout */
2486 if (amdgpu_sriov_vf(adev) &&
2487 !amdgpu_sriov_runtime(adev) &&
2488 amdgpu_virt_mmio_blocked(adev) &&
2489 !amdgpu_virt_wait_reset(adev)) {
2490 dev_err(adev->dev, "VF exclusive mode timeout\n");
1daee8b4
PD
2491 /* Don't send request since VF is inactive. */
2492 adev->virt.caps &= ~AMDGPU_SRIOV_CAPS_RUNTIME;
2493 adev->virt.ops = NULL;
8840a387 2494 r = -EAGAIN;
2495 goto failed;
2496 }
06ec9070 2497 dev_err(adev->dev, "amdgpu_device_ip_init failed\n");
e23b74aa 2498 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_INIT_FAIL, 0, 0);
83ba126a 2499 goto failed;
d38ceaf9
AD
2500 }
2501
2502 adev->accel_working = true;
2503
e59c0205
AX
2504 amdgpu_vm_check_compute_bug(adev);
2505
95844d20
MO
2506 /* Initialize the buffer migration limit. */
2507 if (amdgpu_moverate >= 0)
2508 max_MBps = amdgpu_moverate;
2509 else
2510 max_MBps = 8; /* Allow 8 MB/s. */
2511 /* Get a log2 for easy divisions. */
2512 adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps));
2513
d38ceaf9
AD
2514 r = amdgpu_ib_pool_init(adev);
2515 if (r) {
2516 dev_err(adev->dev, "IB initialization failed (%d).\n", r);
e23b74aa 2517 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_IB_INIT_FAIL, 0, r);
83ba126a 2518 goto failed;
d38ceaf9
AD
2519 }
2520
2dc8f81e
HC
2521 if (amdgpu_sriov_vf(adev))
2522 amdgpu_virt_init_data_exchange(adev);
2523
9bc92b9c
ML
2524 amdgpu_fbdev_init(adev);
2525
d2f52ac8
RZ
2526 r = amdgpu_pm_sysfs_init(adev);
2527 if (r)
2528 DRM_ERROR("registering pm debugfs failed (%d).\n", r);
2529
75758255 2530 r = amdgpu_debugfs_gem_init(adev);
3f14e623 2531 if (r)
d38ceaf9 2532 DRM_ERROR("registering gem debugfs failed (%d).\n", r);
d38ceaf9
AD
2533
2534 r = amdgpu_debugfs_regs_init(adev);
3f14e623 2535 if (r)
d38ceaf9 2536 DRM_ERROR("registering register debugfs failed (%d).\n", r);
d38ceaf9 2537
50ab2533 2538 r = amdgpu_debugfs_firmware_init(adev);
3f14e623 2539 if (r)
50ab2533 2540 DRM_ERROR("registering firmware debugfs failed (%d).\n", r);
50ab2533 2541
763efb6c 2542 r = amdgpu_debugfs_init(adev);
db95e218 2543 if (r)
763efb6c 2544 DRM_ERROR("Creating debugfs files failed (%d).\n", r);
db95e218 2545
d38ceaf9
AD
2546 if ((amdgpu_testing & 1)) {
2547 if (adev->accel_working)
2548 amdgpu_test_moves(adev);
2549 else
2550 DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n");
2551 }
d38ceaf9
AD
2552 if (amdgpu_benchmarking) {
2553 if (adev->accel_working)
2554 amdgpu_benchmark(adev, amdgpu_benchmarking);
2555 else
2556 DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n");
2557 }
2558
2559 /* enable clockgating, etc. after ib tests, etc. since some blocks require
2560 * explicit gating rather than handling it automatically.
2561 */
06ec9070 2562 r = amdgpu_device_ip_late_init(adev);
2c1a2784 2563 if (r) {
06ec9070 2564 dev_err(adev->dev, "amdgpu_device_ip_late_init failed\n");
e23b74aa 2565 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_AMDGPU_LATE_INIT_FAIL, 0, r);
83ba126a 2566 goto failed;
2c1a2784 2567 }
d38ceaf9
AD
2568
2569 return 0;
83ba126a
AD
2570
2571failed:
89041940 2572 amdgpu_vf_error_trans_all(adev);
83ba126a
AD
2573 if (runtime)
2574 vga_switcheroo_fini_domain_pm_ops(adev->dev);
8840a387 2575
83ba126a 2576 return r;
d38ceaf9
AD
2577}
2578
d38ceaf9
AD
2579/**
2580 * amdgpu_device_fini - tear down the driver
2581 *
2582 * @adev: amdgpu_device pointer
2583 *
2584 * Tear down the driver info (all asics).
2585 * Called at driver shutdown.
2586 */
2587void amdgpu_device_fini(struct amdgpu_device *adev)
2588{
2589 int r;
2590
2591 DRM_INFO("amdgpu: finishing device.\n");
2592 adev->shutdown = true;
e5b03032
ML
2593 /* disable all interrupts */
2594 amdgpu_irq_disable_all(adev);
ff97cba8
ML
2595 if (adev->mode_info.mode_config_initialized){
2596 if (!amdgpu_device_has_dc_support(adev))
2597 drm_crtc_force_disable_all(adev->ddev);
2598 else
2599 drm_atomic_helper_shutdown(adev->ddev);
2600 }
d38ceaf9
AD
2601 amdgpu_ib_pool_fini(adev);
2602 amdgpu_fence_driver_fini(adev);
58e955d9 2603 amdgpu_pm_sysfs_fini(adev);
d38ceaf9 2604 amdgpu_fbdev_fini(adev);
06ec9070 2605 r = amdgpu_device_ip_fini(adev);
ab4fe3e1
HR
2606 if (adev->firmware.gpu_info_fw) {
2607 release_firmware(adev->firmware.gpu_info_fw);
2608 adev->firmware.gpu_info_fw = NULL;
2609 }
d38ceaf9 2610 adev->accel_working = false;
2dc80b00 2611 cancel_delayed_work_sync(&adev->late_init_work);
d38ceaf9 2612 /* free i2c buses */
4562236b
HW
2613 if (!amdgpu_device_has_dc_support(adev))
2614 amdgpu_i2c_fini(adev);
bfca0289
SL
2615
2616 if (amdgpu_emu_mode != 1)
2617 amdgpu_atombios_fini(adev);
2618
d38ceaf9
AD
2619 kfree(adev->bios);
2620 adev->bios = NULL;
84c8b22e
LW
2621 if (!pci_is_thunderbolt_attached(adev->pdev))
2622 vga_switcheroo_unregister_client(adev->pdev);
83ba126a
AD
2623 if (adev->flags & AMD_IS_PX)
2624 vga_switcheroo_fini_domain_pm_ops(adev->dev);
d38ceaf9
AD
2625 vga_client_register(adev->pdev, NULL, NULL, NULL);
2626 if (adev->rio_mem)
2627 pci_iounmap(adev->pdev, adev->rio_mem);
2628 adev->rio_mem = NULL;
2629 iounmap(adev->rmmio);
2630 adev->rmmio = NULL;
06ec9070 2631 amdgpu_device_doorbell_fini(adev);
d38ceaf9 2632 amdgpu_debugfs_regs_cleanup(adev);
d38ceaf9
AD
2633}
2634
2635
2636/*
2637 * Suspend & resume.
2638 */
2639/**
810ddc3a 2640 * amdgpu_device_suspend - initiate device suspend
d38ceaf9 2641 *
87e3f136
DP
2642 * @dev: drm dev pointer
2643 * @suspend: suspend state
2644 * @fbcon : notify the fbdev of suspend
d38ceaf9
AD
2645 *
2646 * Puts the hw in the suspend state (all asics).
2647 * Returns 0 for success or an error on failure.
2648 * Called at driver suspend.
2649 */
810ddc3a 2650int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon)
d38ceaf9
AD
2651{
2652 struct amdgpu_device *adev;
2653 struct drm_crtc *crtc;
2654 struct drm_connector *connector;
5ceb54c6 2655 int r;
d38ceaf9
AD
2656
2657 if (dev == NULL || dev->dev_private == NULL) {
2658 return -ENODEV;
2659 }
2660
2661 adev = dev->dev_private;
2662
2663 if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
2664 return 0;
2665
2666 drm_kms_helper_poll_disable(dev);
2667
5f818173
S
2668 if (fbcon)
2669 amdgpu_fbdev_set_suspend(adev, 1);
2670
a5459475
RZ
2671 cancel_delayed_work_sync(&adev->late_init_work);
2672
4562236b
HW
2673 if (!amdgpu_device_has_dc_support(adev)) {
2674 /* turn off display hw */
2675 drm_modeset_lock_all(dev);
2676 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
2677 drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF);
2678 }
2679 drm_modeset_unlock_all(dev);
fe1053b7
AD
2680 /* unpin the front buffers and cursors */
2681 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
2682 struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
2683 struct drm_framebuffer *fb = crtc->primary->fb;
2684 struct amdgpu_bo *robj;
2685
2686 if (amdgpu_crtc->cursor_bo) {
2687 struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
2688 r = amdgpu_bo_reserve(aobj, true);
2689 if (r == 0) {
2690 amdgpu_bo_unpin(aobj);
2691 amdgpu_bo_unreserve(aobj);
2692 }
756e6880 2693 }
756e6880 2694
fe1053b7
AD
2695 if (fb == NULL || fb->obj[0] == NULL) {
2696 continue;
2697 }
2698 robj = gem_to_amdgpu_bo(fb->obj[0]);
2699 /* don't unpin kernel fb objects */
2700 if (!amdgpu_fbdev_robj_is_fb(adev, robj)) {
2701 r = amdgpu_bo_reserve(robj, true);
2702 if (r == 0) {
2703 amdgpu_bo_unpin(robj);
2704 amdgpu_bo_unreserve(robj);
2705 }
d38ceaf9
AD
2706 }
2707 }
2708 }
fe1053b7
AD
2709
2710 amdgpu_amdkfd_suspend(adev);
2711
2712 r = amdgpu_device_ip_suspend_phase1(adev);
2713
d38ceaf9
AD
2714 /* evict vram memory */
2715 amdgpu_bo_evict_vram(adev);
2716
5ceb54c6 2717 amdgpu_fence_driver_suspend(adev);
d38ceaf9 2718
fe1053b7 2719 r = amdgpu_device_ip_suspend_phase2(adev);
d38ceaf9 2720
a0a71e49
AD
2721 /* evict remaining vram memory
2722 * This second call to evict vram is to evict the gart page table
2723 * using the CPU.
2724 */
d38ceaf9
AD
2725 amdgpu_bo_evict_vram(adev);
2726
2727 pci_save_state(dev->pdev);
2728 if (suspend) {
2729 /* Shut down the device */
2730 pci_disable_device(dev->pdev);
2731 pci_set_power_state(dev->pdev, PCI_D3hot);
74b0b157 2732 } else {
2733 r = amdgpu_asic_reset(adev);
2734 if (r)
2735 DRM_ERROR("amdgpu asic reset failed\n");
d38ceaf9
AD
2736 }
2737
d38ceaf9
AD
2738 return 0;
2739}
2740
2741/**
810ddc3a 2742 * amdgpu_device_resume - initiate device resume
d38ceaf9 2743 *
87e3f136
DP
2744 * @dev: drm dev pointer
2745 * @resume: resume state
2746 * @fbcon : notify the fbdev of resume
d38ceaf9
AD
2747 *
2748 * Bring the hw back to operating state (all asics).
2749 * Returns 0 for success or an error on failure.
2750 * Called at driver resume.
2751 */
810ddc3a 2752int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon)
d38ceaf9
AD
2753{
2754 struct drm_connector *connector;
2755 struct amdgpu_device *adev = dev->dev_private;
756e6880 2756 struct drm_crtc *crtc;
03161a6e 2757 int r = 0;
d38ceaf9
AD
2758
2759 if (dev->switch_power_state == DRM_SWITCH_POWER_OFF)
2760 return 0;
2761
d38ceaf9
AD
2762 if (resume) {
2763 pci_set_power_state(dev->pdev, PCI_D0);
2764 pci_restore_state(dev->pdev);
74b0b157 2765 r = pci_enable_device(dev->pdev);
03161a6e 2766 if (r)
4d3b9ae5 2767 return r;
d38ceaf9
AD
2768 }
2769
2770 /* post card */
39c640c0 2771 if (amdgpu_device_need_post(adev)) {
74b0b157 2772 r = amdgpu_atom_asic_init(adev->mode_info.atom_context);
2773 if (r)
2774 DRM_ERROR("amdgpu asic init failed\n");
2775 }
d38ceaf9 2776
06ec9070 2777 r = amdgpu_device_ip_resume(adev);
e6707218 2778 if (r) {
06ec9070 2779 DRM_ERROR("amdgpu_device_ip_resume failed (%d).\n", r);
4d3b9ae5 2780 return r;
e6707218 2781 }
5ceb54c6
AD
2782 amdgpu_fence_driver_resume(adev);
2783
d38ceaf9 2784
06ec9070 2785 r = amdgpu_device_ip_late_init(adev);
03161a6e 2786 if (r)
4d3b9ae5 2787 return r;
d38ceaf9 2788
fe1053b7
AD
2789 if (!amdgpu_device_has_dc_support(adev)) {
2790 /* pin cursors */
2791 list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
2792 struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc);
2793
2794 if (amdgpu_crtc->cursor_bo) {
2795 struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo);
2796 r = amdgpu_bo_reserve(aobj, true);
2797 if (r == 0) {
2798 r = amdgpu_bo_pin(aobj, AMDGPU_GEM_DOMAIN_VRAM);
2799 if (r != 0)
2800 DRM_ERROR("Failed to pin cursor BO (%d)\n", r);
2801 amdgpu_crtc->cursor_addr = amdgpu_bo_gpu_offset(aobj);
2802 amdgpu_bo_unreserve(aobj);
2803 }
756e6880
AD
2804 }
2805 }
2806 }
ba997709
YZ
2807 r = amdgpu_amdkfd_resume(adev);
2808 if (r)
2809 return r;
756e6880 2810
96a5d8d4
LL
2811 /* Make sure IB tests flushed */
2812 flush_delayed_work(&adev->late_init_work);
2813
d38ceaf9
AD
2814 /* blat the mode back in */
2815 if (fbcon) {
4562236b
HW
2816 if (!amdgpu_device_has_dc_support(adev)) {
2817 /* pre DCE11 */
2818 drm_helper_resume_force_mode(dev);
2819
2820 /* turn on display hw */
2821 drm_modeset_lock_all(dev);
2822 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
2823 drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON);
2824 }
2825 drm_modeset_unlock_all(dev);
d38ceaf9 2826 }
4d3b9ae5 2827 amdgpu_fbdev_set_suspend(adev, 0);
d38ceaf9
AD
2828 }
2829
2830 drm_kms_helper_poll_enable(dev);
23a1a9e5
L
2831
2832 /*
2833 * Most of the connector probing functions try to acquire runtime pm
2834 * refs to ensure that the GPU is powered on when connector polling is
2835 * performed. Since we're calling this from a runtime PM callback,
2836 * trying to acquire rpm refs will cause us to deadlock.
2837 *
2838 * Since we're guaranteed to be holding the rpm lock, it's safe to
2839 * temporarily disable the rpm helpers so this doesn't deadlock us.
2840 */
2841#ifdef CONFIG_PM
2842 dev->dev->power.disable_depth++;
2843#endif
4562236b
HW
2844 if (!amdgpu_device_has_dc_support(adev))
2845 drm_helper_hpd_irq_event(dev);
2846 else
2847 drm_kms_helper_hotplug_event(dev);
23a1a9e5
L
2848#ifdef CONFIG_PM
2849 dev->dev->power.disable_depth--;
2850#endif
4d3b9ae5 2851 return 0;
d38ceaf9
AD
2852}
2853
e3ecdffa
AD
2854/**
2855 * amdgpu_device_ip_check_soft_reset - did soft reset succeed
2856 *
2857 * @adev: amdgpu_device pointer
2858 *
2859 * The list of all the hardware IPs that make up the asic is walked and
2860 * the check_soft_reset callbacks are run. check_soft_reset determines
2861 * if the asic is still hung or not.
2862 * Returns true if any of the IPs are still in a hung state, false if not.
2863 */
06ec9070 2864static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev)
63fbf42f
CZ
2865{
2866 int i;
2867 bool asic_hang = false;
2868
f993d628
ML
2869 if (amdgpu_sriov_vf(adev))
2870 return true;
2871
8bc04c29
AD
2872 if (amdgpu_asic_need_full_reset(adev))
2873 return true;
2874
63fbf42f 2875 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 2876 if (!adev->ip_blocks[i].status.valid)
63fbf42f 2877 continue;
a1255107
AD
2878 if (adev->ip_blocks[i].version->funcs->check_soft_reset)
2879 adev->ip_blocks[i].status.hang =
2880 adev->ip_blocks[i].version->funcs->check_soft_reset(adev);
2881 if (adev->ip_blocks[i].status.hang) {
2882 DRM_INFO("IP block:%s is hung!\n", adev->ip_blocks[i].version->funcs->name);
63fbf42f
CZ
2883 asic_hang = true;
2884 }
2885 }
2886 return asic_hang;
2887}
2888
e3ecdffa
AD
2889/**
2890 * amdgpu_device_ip_pre_soft_reset - prepare for soft reset
2891 *
2892 * @adev: amdgpu_device pointer
2893 *
2894 * The list of all the hardware IPs that make up the asic is walked and the
2895 * pre_soft_reset callbacks are run if the block is hung. pre_soft_reset
2896 * handles any IP specific hardware or software state changes that are
2897 * necessary for a soft reset to succeed.
2898 * Returns 0 on success, negative error code on failure.
2899 */
06ec9070 2900static int amdgpu_device_ip_pre_soft_reset(struct amdgpu_device *adev)
d31a501e
CZ
2901{
2902 int i, r = 0;
2903
2904 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 2905 if (!adev->ip_blocks[i].status.valid)
d31a501e 2906 continue;
a1255107
AD
2907 if (adev->ip_blocks[i].status.hang &&
2908 adev->ip_blocks[i].version->funcs->pre_soft_reset) {
2909 r = adev->ip_blocks[i].version->funcs->pre_soft_reset(adev);
d31a501e
CZ
2910 if (r)
2911 return r;
2912 }
2913 }
2914
2915 return 0;
2916}
2917
e3ecdffa
AD
2918/**
2919 * amdgpu_device_ip_need_full_reset - check if a full asic reset is needed
2920 *
2921 * @adev: amdgpu_device pointer
2922 *
2923 * Some hardware IPs cannot be soft reset. If they are hung, a full gpu
2924 * reset is necessary to recover.
2925 * Returns true if a full asic reset is required, false if not.
2926 */
06ec9070 2927static bool amdgpu_device_ip_need_full_reset(struct amdgpu_device *adev)
35d782fe 2928{
da146d3b
AD
2929 int i;
2930
8bc04c29
AD
2931 if (amdgpu_asic_need_full_reset(adev))
2932 return true;
2933
da146d3b 2934 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 2935 if (!adev->ip_blocks[i].status.valid)
da146d3b 2936 continue;
a1255107
AD
2937 if ((adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_GMC) ||
2938 (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_SMC) ||
2939 (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_ACP) ||
98512bb8
KW
2940 (adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_DCE) ||
2941 adev->ip_blocks[i].version->type == AMD_IP_BLOCK_TYPE_PSP) {
a1255107 2942 if (adev->ip_blocks[i].status.hang) {
da146d3b
AD
2943 DRM_INFO("Some block need full reset!\n");
2944 return true;
2945 }
2946 }
35d782fe
CZ
2947 }
2948 return false;
2949}
2950
e3ecdffa
AD
2951/**
2952 * amdgpu_device_ip_soft_reset - do a soft reset
2953 *
2954 * @adev: amdgpu_device pointer
2955 *
2956 * The list of all the hardware IPs that make up the asic is walked and the
2957 * soft_reset callbacks are run if the block is hung. soft_reset handles any
2958 * IP specific hardware or software state changes that are necessary to soft
2959 * reset the IP.
2960 * Returns 0 on success, negative error code on failure.
2961 */
06ec9070 2962static int amdgpu_device_ip_soft_reset(struct amdgpu_device *adev)
35d782fe
CZ
2963{
2964 int i, r = 0;
2965
2966 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 2967 if (!adev->ip_blocks[i].status.valid)
35d782fe 2968 continue;
a1255107
AD
2969 if (adev->ip_blocks[i].status.hang &&
2970 adev->ip_blocks[i].version->funcs->soft_reset) {
2971 r = adev->ip_blocks[i].version->funcs->soft_reset(adev);
35d782fe
CZ
2972 if (r)
2973 return r;
2974 }
2975 }
2976
2977 return 0;
2978}
2979
e3ecdffa
AD
2980/**
2981 * amdgpu_device_ip_post_soft_reset - clean up from soft reset
2982 *
2983 * @adev: amdgpu_device pointer
2984 *
2985 * The list of all the hardware IPs that make up the asic is walked and the
2986 * post_soft_reset callbacks are run if the asic was hung. post_soft_reset
2987 * handles any IP specific hardware or software state changes that are
2988 * necessary after the IP has been soft reset.
2989 * Returns 0 on success, negative error code on failure.
2990 */
06ec9070 2991static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev)
35d782fe
CZ
2992{
2993 int i, r = 0;
2994
2995 for (i = 0; i < adev->num_ip_blocks; i++) {
a1255107 2996 if (!adev->ip_blocks[i].status.valid)
35d782fe 2997 continue;
a1255107
AD
2998 if (adev->ip_blocks[i].status.hang &&
2999 adev->ip_blocks[i].version->funcs->post_soft_reset)
3000 r = adev->ip_blocks[i].version->funcs->post_soft_reset(adev);
35d782fe
CZ
3001 if (r)
3002 return r;
3003 }
3004
3005 return 0;
3006}
3007
e3ecdffa
AD
3008/**
3009 * amdgpu_device_recover_vram_from_shadow - restore shadowed VRAM buffers
3010 *
3011 * @adev: amdgpu_device pointer
3012 * @ring: amdgpu_ring for the engine handling the buffer operations
3013 * @bo: amdgpu_bo buffer whose shadow is being restored
3014 * @fence: dma_fence associated with the operation
3015 *
3016 * Restores the VRAM buffer contents from the shadow in GTT. Used to
3017 * restore things like GPUVM page tables after a GPU reset where
3018 * the contents of VRAM might be lost.
3019 * Returns 0 on success, negative error code on failure.
3020 */
06ec9070
AD
3021static int amdgpu_device_recover_vram_from_shadow(struct amdgpu_device *adev,
3022 struct amdgpu_ring *ring,
3023 struct amdgpu_bo *bo,
3024 struct dma_fence **fence)
53cdccd5
CZ
3025{
3026 uint32_t domain;
3027 int r;
3028
23d2e504
RH
3029 if (!bo->shadow)
3030 return 0;
3031
1d284797 3032 r = amdgpu_bo_reserve(bo, true);
23d2e504
RH
3033 if (r)
3034 return r;
3035 domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
3036 /* if bo has been evicted, then no need to recover */
3037 if (domain == AMDGPU_GEM_DOMAIN_VRAM) {
82521316
RH
3038 r = amdgpu_bo_validate(bo->shadow);
3039 if (r) {
3040 DRM_ERROR("bo validate failed!\n");
3041 goto err;
3042 }
3043
23d2e504 3044 r = amdgpu_bo_restore_from_shadow(adev, ring, bo,
53cdccd5 3045 NULL, fence, true);
23d2e504
RH
3046 if (r) {
3047 DRM_ERROR("recover page table failed!\n");
3048 goto err;
3049 }
3050 }
53cdccd5 3051err:
23d2e504
RH
3052 amdgpu_bo_unreserve(bo);
3053 return r;
53cdccd5
CZ
3054}
3055
e3ecdffa
AD
3056/**
3057 * amdgpu_device_handle_vram_lost - Handle the loss of VRAM contents
3058 *
3059 * @adev: amdgpu_device pointer
3060 *
3061 * Restores the contents of VRAM buffers from the shadows in GTT. Used to
3062 * restore things like GPUVM page tables after a GPU reset where
3063 * the contents of VRAM might be lost.
3064 * Returns 0 on success, 1 on failure.
3065 */
c41d1cf6
ML
3066static int amdgpu_device_handle_vram_lost(struct amdgpu_device *adev)
3067{
3068 struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
3069 struct amdgpu_bo *bo, *tmp;
3070 struct dma_fence *fence = NULL, *next = NULL;
3071 long r = 1;
3072 int i = 0;
3073 long tmo;
3074
3075 if (amdgpu_sriov_runtime(adev))
b045d3af 3076 tmo = msecs_to_jiffies(8000);
c41d1cf6
ML
3077 else
3078 tmo = msecs_to_jiffies(100);
3079
3080 DRM_INFO("recover vram bo from shadow start\n");
3081 mutex_lock(&adev->shadow_list_lock);
3082 list_for_each_entry_safe(bo, tmp, &adev->shadow_list, shadow_list) {
3083 next = NULL;
3084 amdgpu_device_recover_vram_from_shadow(adev, ring, bo, &next);
3085 if (fence) {
3086 r = dma_fence_wait_timeout(fence, false, tmo);
3087 if (r == 0)
3088 pr_err("wait fence %p[%d] timeout\n", fence, i);
3089 else if (r < 0)
3090 pr_err("wait fence %p[%d] interrupted\n", fence, i);
3091 if (r < 1) {
3092 dma_fence_put(fence);
3093 fence = next;
3094 break;
3095 }
3096 i++;
3097 }
3098
3099 dma_fence_put(fence);
3100 fence = next;
3101 }
3102 mutex_unlock(&adev->shadow_list_lock);
3103
3104 if (fence) {
3105 r = dma_fence_wait_timeout(fence, false, tmo);
3106 if (r == 0)
3107 pr_err("wait fence %p[%d] timeout\n", fence, i);
3108 else if (r < 0)
3109 pr_err("wait fence %p[%d] interrupted\n", fence, i);
3110
3111 }
3112 dma_fence_put(fence);
3113
3114 if (r > 0)
3115 DRM_INFO("recover vram bo from shadow done\n");
3116 else
3117 DRM_ERROR("recover vram bo from shadow failed\n");
3118
e3ecdffa 3119 return (r > 0) ? 0 : 1;
c41d1cf6
ML
3120}
3121
e3ecdffa 3122/**
06ec9070 3123 * amdgpu_device_reset - reset ASIC/GPU for bare-metal or passthrough
a90ad3c2
ML
3124 *
3125 * @adev: amdgpu device pointer
a90ad3c2 3126 *
5740682e 3127 * attempt to do soft-reset or full-reset and reinitialize Asic
3f48c681 3128 * return 0 means succeeded otherwise failed
e3ecdffa 3129 */
c41d1cf6 3130static int amdgpu_device_reset(struct amdgpu_device *adev)
a90ad3c2 3131{
5740682e
ML
3132 bool need_full_reset, vram_lost = 0;
3133 int r;
a90ad3c2 3134
06ec9070 3135 need_full_reset = amdgpu_device_ip_need_full_reset(adev);
a90ad3c2 3136
5740682e 3137 if (!need_full_reset) {
06ec9070
AD
3138 amdgpu_device_ip_pre_soft_reset(adev);
3139 r = amdgpu_device_ip_soft_reset(adev);
3140 amdgpu_device_ip_post_soft_reset(adev);
3141 if (r || amdgpu_device_ip_check_soft_reset(adev)) {
5740682e
ML
3142 DRM_INFO("soft reset failed, will fallback to full reset!\n");
3143 need_full_reset = true;
3144 }
5740682e 3145 }
a90ad3c2 3146
5740682e 3147 if (need_full_reset) {
cdd61df6 3148 r = amdgpu_device_ip_suspend(adev);
a90ad3c2 3149
5740682e 3150retry:
5740682e 3151 r = amdgpu_asic_reset(adev);
5740682e
ML
3152 /* post card */
3153 amdgpu_atom_asic_init(adev->mode_info.atom_context);
65781c78 3154
5740682e
ML
3155 if (!r) {
3156 dev_info(adev->dev, "GPU reset succeeded, trying to resume\n");
06ec9070 3157 r = amdgpu_device_ip_resume_phase1(adev);
5740682e
ML
3158 if (r)
3159 goto out;
65781c78 3160
06ec9070 3161 vram_lost = amdgpu_device_check_vram_lost(adev);
5740682e
ML
3162 if (vram_lost) {
3163 DRM_ERROR("VRAM is lost!\n");
3164 atomic_inc(&adev->vram_lost_counter);
3165 }
3166
c1c7ce8f
CK
3167 r = amdgpu_gtt_mgr_recover(
3168 &adev->mman.bdev.man[TTM_PL_TT]);
5740682e
ML
3169 if (r)
3170 goto out;
3171
06ec9070 3172 r = amdgpu_device_ip_resume_phase2(adev);
5740682e
ML
3173 if (r)
3174 goto out;
3175
3176 if (vram_lost)
06ec9070 3177 amdgpu_device_fill_reset_magic(adev);
65781c78 3178 }
5740682e 3179 }
65781c78 3180
5740682e
ML
3181out:
3182 if (!r) {
3183 amdgpu_irq_gpu_reset_resume_helper(adev);
3184 r = amdgpu_ib_ring_tests(adev);
3185 if (r) {
3186 dev_err(adev->dev, "ib ring test failed (%d).\n", r);
cdd61df6 3187 r = amdgpu_device_ip_suspend(adev);
5740682e
ML
3188 need_full_reset = true;
3189 goto retry;
3190 }
3191 }
65781c78 3192
c41d1cf6
ML
3193 if (!r && ((need_full_reset && !(adev->flags & AMD_IS_APU)) || vram_lost))
3194 r = amdgpu_device_handle_vram_lost(adev);
a90ad3c2 3195
5740682e
ML
3196 return r;
3197}
a90ad3c2 3198
e3ecdffa 3199/**
06ec9070 3200 * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf
5740682e
ML
3201 *
3202 * @adev: amdgpu device pointer
87e3f136 3203 * @from_hypervisor: request from hypervisor
5740682e
ML
3204 *
3205 * do VF FLR and reinitialize Asic
3f48c681 3206 * return 0 means succeeded otherwise failed
e3ecdffa
AD
3207 */
3208static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
3209 bool from_hypervisor)
5740682e
ML
3210{
3211 int r;
3212
3213 if (from_hypervisor)
3214 r = amdgpu_virt_request_full_gpu(adev, true);
3215 else
3216 r = amdgpu_virt_reset_gpu(adev);
3217 if (r)
3218 return r;
a90ad3c2
ML
3219
3220 /* Resume IP prior to SMC */
06ec9070 3221 r = amdgpu_device_ip_reinit_early_sriov(adev);
5740682e
ML
3222 if (r)
3223 goto error;
a90ad3c2
ML
3224
3225 /* we need recover gart prior to run SMC/CP/SDMA resume */
c1c7ce8f 3226 amdgpu_gtt_mgr_recover(&adev->mman.bdev.man[TTM_PL_TT]);
a90ad3c2
ML
3227
3228 /* now we are okay to resume SMC/CP/SDMA */
06ec9070 3229 r = amdgpu_device_ip_reinit_late_sriov(adev);
5740682e
ML
3230 if (r)
3231 goto error;
a90ad3c2
ML
3232
3233 amdgpu_irq_gpu_reset_resume_helper(adev);
5740682e 3234 r = amdgpu_ib_ring_tests(adev);
a90ad3c2 3235
abc34253
ED
3236error:
3237 amdgpu_virt_release_full_gpu(adev, true);
c41d1cf6
ML
3238 if (!r && adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) {
3239 atomic_inc(&adev->vram_lost_counter);
3240 r = amdgpu_device_handle_vram_lost(adev);
a90ad3c2
ML
3241 }
3242
3243 return r;
3244}
3245
d38ceaf9 3246/**
5f152b5e 3247 * amdgpu_device_gpu_recover - reset the asic and recover scheduler
d38ceaf9
AD
3248 *
3249 * @adev: amdgpu device pointer
5740682e 3250 * @job: which job trigger hang
87e3f136 3251 * @force: forces reset regardless of amdgpu_gpu_recovery
d38ceaf9 3252 *
5740682e 3253 * Attempt to reset the GPU if it has hung (all asics).
d38ceaf9
AD
3254 * Returns 0 for success or an error on failure.
3255 */
5f152b5e
AD
3256int amdgpu_device_gpu_recover(struct amdgpu_device *adev,
3257 struct amdgpu_job *job, bool force)
d38ceaf9 3258{
5740682e 3259 int i, r, resched;
fb140b29 3260
54bc1398 3261 if (!force && !amdgpu_device_ip_check_soft_reset(adev)) {
63fbf42f
CZ
3262 DRM_INFO("No hardware hang detected. Did some blocks stall?\n");
3263 return 0;
3264 }
d38ceaf9 3265
dcebf026
AG
3266 if (!force && (amdgpu_gpu_recovery == 0 ||
3267 (amdgpu_gpu_recovery == -1 && !amdgpu_sriov_vf(adev)))) {
3268 DRM_INFO("GPU recovery disabled.\n");
3269 return 0;
3270 }
3271
5740682e
ML
3272 dev_info(adev->dev, "GPU reset begin!\n");
3273
13a752e3 3274 mutex_lock(&adev->lock_reset);
d94aed5a 3275 atomic_inc(&adev->gpu_reset_counter);
13a752e3 3276 adev->in_gpu_reset = 1;
d38ceaf9 3277
5c6dd71e
SL
3278 /* Block kfd */
3279 amdgpu_amdkfd_pre_reset(adev);
3280
a3c47d6b
CZ
3281 /* block TTM */
3282 resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev);
71182665 3283
71182665 3284 /* block all schedulers and reset given job's ring */
0875dc9e
CZ
3285 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
3286 struct amdgpu_ring *ring = adev->rings[i];
3287
51687759 3288 if (!ring || !ring->sched.thread)
0875dc9e 3289 continue;
5740682e 3290
71182665
ML
3291 kthread_park(ring->sched.thread);
3292
3320b8d2 3293 if (job && job->base.sched == &ring->sched)
5740682e
ML
3294 continue;
3295
67ccea60 3296 drm_sched_hw_job_reset(&ring->sched, job ? &job->base : NULL);
5740682e 3297
2f9d4084
ML
3298 /* after all hw jobs are reset, hw fence is meaningless, so force_completion */
3299 amdgpu_fence_driver_force_completion(ring);
0875dc9e 3300 }
d38ceaf9 3301
5740682e 3302 if (amdgpu_sriov_vf(adev))
c41d1cf6 3303 r = amdgpu_device_reset_sriov(adev, job ? false : true);
5740682e 3304 else
c41d1cf6 3305 r = amdgpu_device_reset(adev);
5740682e 3306
71182665
ML
3307 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
3308 struct amdgpu_ring *ring = adev->rings[i];
53cdccd5 3309
71182665
ML
3310 if (!ring || !ring->sched.thread)
3311 continue;
5740682e 3312
71182665
ML
3313 /* only need recovery sched of the given job's ring
3314 * or all rings (in the case @job is NULL)
3315 * after above amdgpu_reset accomplished
3316 */
3320b8d2 3317 if ((!job || job->base.sched == &ring->sched) && !r)
1b1f42d8 3318 drm_sched_job_recovery(&ring->sched);
5740682e 3319
71182665 3320 kthread_unpark(ring->sched.thread);
d38ceaf9
AD
3321 }
3322
bf830604 3323 if (!amdgpu_device_has_dc_support(adev)) {
4562236b 3324 drm_helper_resume_force_mode(adev->ddev);
5740682e 3325 }
d38ceaf9
AD
3326
3327 ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched);
5740682e 3328
89041940 3329 if (r) {
d38ceaf9 3330 /* bad news, how to tell it to userspace ? */
5740682e
ML
3331 dev_info(adev->dev, "GPU reset(%d) failed\n", atomic_read(&adev->gpu_reset_counter));
3332 amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r);
3333 } else {
3f48c681 3334 dev_info(adev->dev, "GPU reset(%d) succeeded!\n",atomic_read(&adev->gpu_reset_counter));
89041940 3335 }
d38ceaf9 3336
5c6dd71e
SL
3337 /*unlock kfd */
3338 amdgpu_amdkfd_post_reset(adev);
89041940 3339 amdgpu_vf_error_trans_all(adev);
13a752e3
ML
3340 adev->in_gpu_reset = 0;
3341 mutex_unlock(&adev->lock_reset);
d38ceaf9
AD
3342 return r;
3343}
3344
e3ecdffa
AD
3345/**
3346 * amdgpu_device_get_pcie_info - fence pcie info about the PCIE slot
3347 *
3348 * @adev: amdgpu_device pointer
3349 *
3350 * Fetchs and stores in the driver the PCIE capabilities (gen speed
3351 * and lanes) of the slot the device is in. Handles APUs and
3352 * virtualized environments where PCIE config space may not be available.
3353 */
5494d864 3354static void amdgpu_device_get_pcie_info(struct amdgpu_device *adev)
d0dd7f0c 3355{
5d9a6330
AD
3356 struct pci_dev *pdev;
3357 enum pci_bus_speed speed_cap;
3358 enum pcie_link_width link_width;
d0dd7f0c 3359
cd474ba0
AD
3360 if (amdgpu_pcie_gen_cap)
3361 adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap;
d0dd7f0c 3362
cd474ba0
AD
3363 if (amdgpu_pcie_lane_cap)
3364 adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap;
d0dd7f0c 3365
cd474ba0
AD
3366 /* covers APUs as well */
3367 if (pci_is_root_bus(adev->pdev->bus)) {
3368 if (adev->pm.pcie_gen_mask == 0)
3369 adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK;
3370 if (adev->pm.pcie_mlw_mask == 0)
3371 adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK;
d0dd7f0c 3372 return;
cd474ba0 3373 }
d0dd7f0c 3374
cd474ba0 3375 if (adev->pm.pcie_gen_mask == 0) {
5d9a6330
AD
3376 /* asic caps */
3377 pdev = adev->pdev;
3378 speed_cap = pcie_get_speed_cap(pdev);
3379 if (speed_cap == PCI_SPEED_UNKNOWN) {
3380 adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
cd474ba0
AD
3381 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
3382 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3);
cd474ba0 3383 } else {
5d9a6330
AD
3384 if (speed_cap == PCIE_SPEED_16_0GT)
3385 adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3386 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
3387 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3 |
3388 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN4);
3389 else if (speed_cap == PCIE_SPEED_8_0GT)
3390 adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3391 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 |
3392 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3);
3393 else if (speed_cap == PCIE_SPEED_5_0GT)
3394 adev->pm.pcie_gen_mask |= (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3395 CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2);
3396 else
3397 adev->pm.pcie_gen_mask |= CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1;
3398 }
3399 /* platform caps */
3400 pdev = adev->ddev->pdev->bus->self;
3401 speed_cap = pcie_get_speed_cap(pdev);
3402 if (speed_cap == PCI_SPEED_UNKNOWN) {
3403 adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3404 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2);
3405 } else {
3406 if (speed_cap == PCIE_SPEED_16_0GT)
3407 adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3408 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 |
3409 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3 |
3410 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN4);
3411 else if (speed_cap == PCIE_SPEED_8_0GT)
3412 adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3413 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2 |
3414 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3);
3415 else if (speed_cap == PCIE_SPEED_5_0GT)
3416 adev->pm.pcie_gen_mask |= (CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1 |
3417 CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2);
3418 else
3419 adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1;
3420
cd474ba0
AD
3421 }
3422 }
3423 if (adev->pm.pcie_mlw_mask == 0) {
5d9a6330
AD
3424 pdev = adev->ddev->pdev->bus->self;
3425 link_width = pcie_get_width_cap(pdev);
3426 if (link_width == PCIE_LNK_WIDTH_UNKNOWN) {
3427 adev->pm.pcie_mlw_mask |= AMDGPU_DEFAULT_PCIE_MLW_MASK;
3428 } else {
3429 switch (link_width) {
3430 case PCIE_LNK_X32:
cd474ba0
AD
3431 adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 |
3432 CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
3433 CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
3434 CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
3435 CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
3436 CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
3437 CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
3438 break;
5d9a6330 3439 case PCIE_LNK_X16:
cd474ba0
AD
3440 adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 |
3441 CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
3442 CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
3443 CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
3444 CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
3445 CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
3446 break;
5d9a6330 3447 case PCIE_LNK_X12:
cd474ba0
AD
3448 adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 |
3449 CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
3450 CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
3451 CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
3452 CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
3453 break;
5d9a6330 3454 case PCIE_LNK_X8:
cd474ba0
AD
3455 adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 |
3456 CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
3457 CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
3458 CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
3459 break;
5d9a6330 3460 case PCIE_LNK_X4:
cd474ba0
AD
3461 adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 |
3462 CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
3463 CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
3464 break;
5d9a6330 3465 case PCIE_LNK_X2:
cd474ba0
AD
3466 adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 |
3467 CAIL_PCIE_LINK_WIDTH_SUPPORT_X1);
3468 break;
5d9a6330 3469 case PCIE_LNK_X1:
cd474ba0
AD
3470 adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1;
3471 break;
3472 default:
3473 break;
3474 }
d0dd7f0c
AD
3475 }
3476 }
3477}
d38ceaf9 3478