]>
Commit | Line | Data |
---|---|---|
d38ceaf9 AD |
1 | /* |
2 | * Copyright 2008 Advanced Micro Devices, Inc. | |
3 | * Copyright 2008 Red Hat Inc. | |
4 | * Copyright 2009 Jerome Glisse. | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a | |
7 | * copy of this software and associated documentation files (the "Software"), | |
8 | * to deal in the Software without restriction, including without limitation | |
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
10 | * and/or sell copies of the Software, and to permit persons to whom the | |
11 | * Software is furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
22 | * OTHER DEALINGS IN THE SOFTWARE. | |
23 | * | |
24 | * Authors: Dave Airlie | |
25 | * Alex Deucher | |
26 | * Jerome Glisse | |
27 | */ | |
0875dc9e | 28 | #include <linux/kthread.h> |
d38ceaf9 AD |
29 | #include <linux/console.h> |
30 | #include <linux/slab.h> | |
31 | #include <linux/debugfs.h> | |
32 | #include <drm/drmP.h> | |
33 | #include <drm/drm_crtc_helper.h> | |
34 | #include <drm/amdgpu_drm.h> | |
35 | #include <linux/vgaarb.h> | |
36 | #include <linux/vga_switcheroo.h> | |
37 | #include <linux/efi.h> | |
38 | #include "amdgpu.h" | |
f4b373f4 | 39 | #include "amdgpu_trace.h" |
d38ceaf9 AD |
40 | #include "amdgpu_i2c.h" |
41 | #include "atom.h" | |
42 | #include "amdgpu_atombios.h" | |
d0dd7f0c | 43 | #include "amd_pcie.h" |
33f34802 KW |
44 | #ifdef CONFIG_DRM_AMDGPU_SI |
45 | #include "si.h" | |
46 | #endif | |
a2e73f56 AD |
47 | #ifdef CONFIG_DRM_AMDGPU_CIK |
48 | #include "cik.h" | |
49 | #endif | |
aaa36a97 | 50 | #include "vi.h" |
d38ceaf9 | 51 | #include "bif/bif_4_1_d.h" |
9accf2fd | 52 | #include <linux/pci.h> |
d38ceaf9 AD |
53 | |
54 | static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev); | |
55 | static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev); | |
56 | ||
57 | static const char *amdgpu_asic_name[] = { | |
da69c161 KW |
58 | "TAHITI", |
59 | "PITCAIRN", | |
60 | "VERDE", | |
61 | "OLAND", | |
62 | "HAINAN", | |
d38ceaf9 AD |
63 | "BONAIRE", |
64 | "KAVERI", | |
65 | "KABINI", | |
66 | "HAWAII", | |
67 | "MULLINS", | |
68 | "TOPAZ", | |
69 | "TONGA", | |
48299f95 | 70 | "FIJI", |
d38ceaf9 | 71 | "CARRIZO", |
139f4917 | 72 | "STONEY", |
2cc0c0b5 FC |
73 | "POLARIS10", |
74 | "POLARIS11", | |
d38ceaf9 AD |
75 | "LAST", |
76 | }; | |
77 | ||
78 | bool amdgpu_device_is_px(struct drm_device *dev) | |
79 | { | |
80 | struct amdgpu_device *adev = dev->dev_private; | |
81 | ||
2f7d10b3 | 82 | if (adev->flags & AMD_IS_PX) |
d38ceaf9 AD |
83 | return true; |
84 | return false; | |
85 | } | |
86 | ||
87 | /* | |
88 | * MMIO register access helper functions. | |
89 | */ | |
90 | uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg, | |
91 | bool always_indirect) | |
92 | { | |
f4b373f4 TSD |
93 | uint32_t ret; |
94 | ||
d38ceaf9 | 95 | if ((reg * 4) < adev->rmmio_size && !always_indirect) |
f4b373f4 | 96 | ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); |
d38ceaf9 AD |
97 | else { |
98 | unsigned long flags; | |
d38ceaf9 AD |
99 | |
100 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
101 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
102 | ret = readl(((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
103 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
d38ceaf9 | 104 | } |
f4b373f4 TSD |
105 | trace_amdgpu_mm_rreg(adev->pdev->device, reg, ret); |
106 | return ret; | |
d38ceaf9 AD |
107 | } |
108 | ||
109 | void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, | |
110 | bool always_indirect) | |
111 | { | |
f4b373f4 TSD |
112 | trace_amdgpu_mm_wreg(adev->pdev->device, reg, v); |
113 | ||
d38ceaf9 AD |
114 | if ((reg * 4) < adev->rmmio_size && !always_indirect) |
115 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); | |
116 | else { | |
117 | unsigned long flags; | |
118 | ||
119 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
120 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
121 | writel(v, ((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
122 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
123 | } | |
124 | } | |
125 | ||
126 | u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg) | |
127 | { | |
128 | if ((reg * 4) < adev->rio_mem_size) | |
129 | return ioread32(adev->rio_mem + (reg * 4)); | |
130 | else { | |
131 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
132 | return ioread32(adev->rio_mem + (mmMM_DATA * 4)); | |
133 | } | |
134 | } | |
135 | ||
136 | void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v) | |
137 | { | |
138 | ||
139 | if ((reg * 4) < adev->rio_mem_size) | |
140 | iowrite32(v, adev->rio_mem + (reg * 4)); | |
141 | else { | |
142 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
143 | iowrite32(v, adev->rio_mem + (mmMM_DATA * 4)); | |
144 | } | |
145 | } | |
146 | ||
147 | /** | |
148 | * amdgpu_mm_rdoorbell - read a doorbell dword | |
149 | * | |
150 | * @adev: amdgpu_device pointer | |
151 | * @index: doorbell index | |
152 | * | |
153 | * Returns the value in the doorbell aperture at the | |
154 | * requested doorbell index (CIK). | |
155 | */ | |
156 | u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index) | |
157 | { | |
158 | if (index < adev->doorbell.num_doorbells) { | |
159 | return readl(adev->doorbell.ptr + index); | |
160 | } else { | |
161 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
162 | return 0; | |
163 | } | |
164 | } | |
165 | ||
166 | /** | |
167 | * amdgpu_mm_wdoorbell - write a doorbell dword | |
168 | * | |
169 | * @adev: amdgpu_device pointer | |
170 | * @index: doorbell index | |
171 | * @v: value to write | |
172 | * | |
173 | * Writes @v to the doorbell aperture at the | |
174 | * requested doorbell index (CIK). | |
175 | */ | |
176 | void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v) | |
177 | { | |
178 | if (index < adev->doorbell.num_doorbells) { | |
179 | writel(v, adev->doorbell.ptr + index); | |
180 | } else { | |
181 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
182 | } | |
183 | } | |
184 | ||
185 | /** | |
186 | * amdgpu_invalid_rreg - dummy reg read function | |
187 | * | |
188 | * @adev: amdgpu device pointer | |
189 | * @reg: offset of register | |
190 | * | |
191 | * Dummy register read function. Used for register blocks | |
192 | * that certain asics don't have (all asics). | |
193 | * Returns the value in the register. | |
194 | */ | |
195 | static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg) | |
196 | { | |
197 | DRM_ERROR("Invalid callback to read register 0x%04X\n", reg); | |
198 | BUG(); | |
199 | return 0; | |
200 | } | |
201 | ||
202 | /** | |
203 | * amdgpu_invalid_wreg - dummy reg write function | |
204 | * | |
205 | * @adev: amdgpu device pointer | |
206 | * @reg: offset of register | |
207 | * @v: value to write to the register | |
208 | * | |
209 | * Dummy register read function. Used for register blocks | |
210 | * that certain asics don't have (all asics). | |
211 | */ | |
212 | static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) | |
213 | { | |
214 | DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n", | |
215 | reg, v); | |
216 | BUG(); | |
217 | } | |
218 | ||
219 | /** | |
220 | * amdgpu_block_invalid_rreg - dummy reg read function | |
221 | * | |
222 | * @adev: amdgpu device pointer | |
223 | * @block: offset of instance | |
224 | * @reg: offset of register | |
225 | * | |
226 | * Dummy register read function. Used for register blocks | |
227 | * that certain asics don't have (all asics). | |
228 | * Returns the value in the register. | |
229 | */ | |
230 | static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev, | |
231 | uint32_t block, uint32_t reg) | |
232 | { | |
233 | DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n", | |
234 | reg, block); | |
235 | BUG(); | |
236 | return 0; | |
237 | } | |
238 | ||
239 | /** | |
240 | * amdgpu_block_invalid_wreg - dummy reg write function | |
241 | * | |
242 | * @adev: amdgpu device pointer | |
243 | * @block: offset of instance | |
244 | * @reg: offset of register | |
245 | * @v: value to write to the register | |
246 | * | |
247 | * Dummy register read function. Used for register blocks | |
248 | * that certain asics don't have (all asics). | |
249 | */ | |
250 | static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev, | |
251 | uint32_t block, | |
252 | uint32_t reg, uint32_t v) | |
253 | { | |
254 | DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n", | |
255 | reg, block, v); | |
256 | BUG(); | |
257 | } | |
258 | ||
259 | static int amdgpu_vram_scratch_init(struct amdgpu_device *adev) | |
260 | { | |
261 | int r; | |
262 | ||
263 | if (adev->vram_scratch.robj == NULL) { | |
264 | r = amdgpu_bo_create(adev, AMDGPU_GPU_PAGE_SIZE, | |
857d913d AD |
265 | PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, |
266 | AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED, | |
72d7668b | 267 | NULL, NULL, &adev->vram_scratch.robj); |
d38ceaf9 AD |
268 | if (r) { |
269 | return r; | |
270 | } | |
271 | } | |
272 | ||
273 | r = amdgpu_bo_reserve(adev->vram_scratch.robj, false); | |
274 | if (unlikely(r != 0)) | |
275 | return r; | |
276 | r = amdgpu_bo_pin(adev->vram_scratch.robj, | |
277 | AMDGPU_GEM_DOMAIN_VRAM, &adev->vram_scratch.gpu_addr); | |
278 | if (r) { | |
279 | amdgpu_bo_unreserve(adev->vram_scratch.robj); | |
280 | return r; | |
281 | } | |
282 | r = amdgpu_bo_kmap(adev->vram_scratch.robj, | |
283 | (void **)&adev->vram_scratch.ptr); | |
284 | if (r) | |
285 | amdgpu_bo_unpin(adev->vram_scratch.robj); | |
286 | amdgpu_bo_unreserve(adev->vram_scratch.robj); | |
287 | ||
288 | return r; | |
289 | } | |
290 | ||
291 | static void amdgpu_vram_scratch_fini(struct amdgpu_device *adev) | |
292 | { | |
293 | int r; | |
294 | ||
295 | if (adev->vram_scratch.robj == NULL) { | |
296 | return; | |
297 | } | |
298 | r = amdgpu_bo_reserve(adev->vram_scratch.robj, false); | |
299 | if (likely(r == 0)) { | |
300 | amdgpu_bo_kunmap(adev->vram_scratch.robj); | |
301 | amdgpu_bo_unpin(adev->vram_scratch.robj); | |
302 | amdgpu_bo_unreserve(adev->vram_scratch.robj); | |
303 | } | |
304 | amdgpu_bo_unref(&adev->vram_scratch.robj); | |
305 | } | |
306 | ||
307 | /** | |
308 | * amdgpu_program_register_sequence - program an array of registers. | |
309 | * | |
310 | * @adev: amdgpu_device pointer | |
311 | * @registers: pointer to the register array | |
312 | * @array_size: size of the register array | |
313 | * | |
314 | * Programs an array or registers with and and or masks. | |
315 | * This is a helper for setting golden registers. | |
316 | */ | |
317 | void amdgpu_program_register_sequence(struct amdgpu_device *adev, | |
318 | const u32 *registers, | |
319 | const u32 array_size) | |
320 | { | |
321 | u32 tmp, reg, and_mask, or_mask; | |
322 | int i; | |
323 | ||
324 | if (array_size % 3) | |
325 | return; | |
326 | ||
327 | for (i = 0; i < array_size; i +=3) { | |
328 | reg = registers[i + 0]; | |
329 | and_mask = registers[i + 1]; | |
330 | or_mask = registers[i + 2]; | |
331 | ||
332 | if (and_mask == 0xffffffff) { | |
333 | tmp = or_mask; | |
334 | } else { | |
335 | tmp = RREG32(reg); | |
336 | tmp &= ~and_mask; | |
337 | tmp |= or_mask; | |
338 | } | |
339 | WREG32(reg, tmp); | |
340 | } | |
341 | } | |
342 | ||
343 | void amdgpu_pci_config_reset(struct amdgpu_device *adev) | |
344 | { | |
345 | pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA); | |
346 | } | |
347 | ||
348 | /* | |
349 | * GPU doorbell aperture helpers function. | |
350 | */ | |
351 | /** | |
352 | * amdgpu_doorbell_init - Init doorbell driver information. | |
353 | * | |
354 | * @adev: amdgpu_device pointer | |
355 | * | |
356 | * Init doorbell driver information (CIK) | |
357 | * Returns 0 on success, error on failure. | |
358 | */ | |
359 | static int amdgpu_doorbell_init(struct amdgpu_device *adev) | |
360 | { | |
361 | /* doorbell bar mapping */ | |
362 | adev->doorbell.base = pci_resource_start(adev->pdev, 2); | |
363 | adev->doorbell.size = pci_resource_len(adev->pdev, 2); | |
364 | ||
edf600da | 365 | adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32), |
d38ceaf9 AD |
366 | AMDGPU_DOORBELL_MAX_ASSIGNMENT+1); |
367 | if (adev->doorbell.num_doorbells == 0) | |
368 | return -EINVAL; | |
369 | ||
370 | adev->doorbell.ptr = ioremap(adev->doorbell.base, adev->doorbell.num_doorbells * sizeof(u32)); | |
371 | if (adev->doorbell.ptr == NULL) { | |
372 | return -ENOMEM; | |
373 | } | |
374 | DRM_INFO("doorbell mmio base: 0x%08X\n", (uint32_t)adev->doorbell.base); | |
375 | DRM_INFO("doorbell mmio size: %u\n", (unsigned)adev->doorbell.size); | |
376 | ||
377 | return 0; | |
378 | } | |
379 | ||
380 | /** | |
381 | * amdgpu_doorbell_fini - Tear down doorbell driver information. | |
382 | * | |
383 | * @adev: amdgpu_device pointer | |
384 | * | |
385 | * Tear down doorbell driver information (CIK) | |
386 | */ | |
387 | static void amdgpu_doorbell_fini(struct amdgpu_device *adev) | |
388 | { | |
389 | iounmap(adev->doorbell.ptr); | |
390 | adev->doorbell.ptr = NULL; | |
391 | } | |
392 | ||
393 | /** | |
394 | * amdgpu_doorbell_get_kfd_info - Report doorbell configuration required to | |
395 | * setup amdkfd | |
396 | * | |
397 | * @adev: amdgpu_device pointer | |
398 | * @aperture_base: output returning doorbell aperture base physical address | |
399 | * @aperture_size: output returning doorbell aperture size in bytes | |
400 | * @start_offset: output returning # of doorbell bytes reserved for amdgpu. | |
401 | * | |
402 | * amdgpu and amdkfd share the doorbell aperture. amdgpu sets it up, | |
403 | * takes doorbells required for its own rings and reports the setup to amdkfd. | |
404 | * amdgpu reserved doorbells are at the start of the doorbell aperture. | |
405 | */ | |
406 | void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev, | |
407 | phys_addr_t *aperture_base, | |
408 | size_t *aperture_size, | |
409 | size_t *start_offset) | |
410 | { | |
411 | /* | |
412 | * The first num_doorbells are used by amdgpu. | |
413 | * amdkfd takes whatever's left in the aperture. | |
414 | */ | |
415 | if (adev->doorbell.size > adev->doorbell.num_doorbells * sizeof(u32)) { | |
416 | *aperture_base = adev->doorbell.base; | |
417 | *aperture_size = adev->doorbell.size; | |
418 | *start_offset = adev->doorbell.num_doorbells * sizeof(u32); | |
419 | } else { | |
420 | *aperture_base = 0; | |
421 | *aperture_size = 0; | |
422 | *start_offset = 0; | |
423 | } | |
424 | } | |
425 | ||
426 | /* | |
427 | * amdgpu_wb_*() | |
428 | * Writeback is the the method by which the the GPU updates special pages | |
429 | * in memory with the status of certain GPU events (fences, ring pointers, | |
430 | * etc.). | |
431 | */ | |
432 | ||
433 | /** | |
434 | * amdgpu_wb_fini - Disable Writeback and free memory | |
435 | * | |
436 | * @adev: amdgpu_device pointer | |
437 | * | |
438 | * Disables Writeback and frees the Writeback memory (all asics). | |
439 | * Used at driver shutdown. | |
440 | */ | |
441 | static void amdgpu_wb_fini(struct amdgpu_device *adev) | |
442 | { | |
443 | if (adev->wb.wb_obj) { | |
444 | if (!amdgpu_bo_reserve(adev->wb.wb_obj, false)) { | |
445 | amdgpu_bo_kunmap(adev->wb.wb_obj); | |
446 | amdgpu_bo_unpin(adev->wb.wb_obj); | |
447 | amdgpu_bo_unreserve(adev->wb.wb_obj); | |
448 | } | |
449 | amdgpu_bo_unref(&adev->wb.wb_obj); | |
450 | adev->wb.wb = NULL; | |
451 | adev->wb.wb_obj = NULL; | |
452 | } | |
453 | } | |
454 | ||
455 | /** | |
456 | * amdgpu_wb_init- Init Writeback driver info and allocate memory | |
457 | * | |
458 | * @adev: amdgpu_device pointer | |
459 | * | |
460 | * Disables Writeback and frees the Writeback memory (all asics). | |
461 | * Used at driver startup. | |
462 | * Returns 0 on success or an -error on failure. | |
463 | */ | |
464 | static int amdgpu_wb_init(struct amdgpu_device *adev) | |
465 | { | |
466 | int r; | |
467 | ||
468 | if (adev->wb.wb_obj == NULL) { | |
469 | r = amdgpu_bo_create(adev, AMDGPU_MAX_WB * 4, PAGE_SIZE, true, | |
72d7668b CK |
470 | AMDGPU_GEM_DOMAIN_GTT, 0, NULL, NULL, |
471 | &adev->wb.wb_obj); | |
d38ceaf9 AD |
472 | if (r) { |
473 | dev_warn(adev->dev, "(%d) create WB bo failed\n", r); | |
474 | return r; | |
475 | } | |
476 | r = amdgpu_bo_reserve(adev->wb.wb_obj, false); | |
477 | if (unlikely(r != 0)) { | |
478 | amdgpu_wb_fini(adev); | |
479 | return r; | |
480 | } | |
481 | r = amdgpu_bo_pin(adev->wb.wb_obj, AMDGPU_GEM_DOMAIN_GTT, | |
482 | &adev->wb.gpu_addr); | |
483 | if (r) { | |
484 | amdgpu_bo_unreserve(adev->wb.wb_obj); | |
485 | dev_warn(adev->dev, "(%d) pin WB bo failed\n", r); | |
486 | amdgpu_wb_fini(adev); | |
487 | return r; | |
488 | } | |
489 | r = amdgpu_bo_kmap(adev->wb.wb_obj, (void **)&adev->wb.wb); | |
490 | amdgpu_bo_unreserve(adev->wb.wb_obj); | |
491 | if (r) { | |
492 | dev_warn(adev->dev, "(%d) map WB bo failed\n", r); | |
493 | amdgpu_wb_fini(adev); | |
494 | return r; | |
495 | } | |
496 | ||
497 | adev->wb.num_wb = AMDGPU_MAX_WB; | |
498 | memset(&adev->wb.used, 0, sizeof(adev->wb.used)); | |
499 | ||
500 | /* clear wb memory */ | |
501 | memset((char *)adev->wb.wb, 0, AMDGPU_GPU_PAGE_SIZE); | |
502 | } | |
503 | ||
504 | return 0; | |
505 | } | |
506 | ||
507 | /** | |
508 | * amdgpu_wb_get - Allocate a wb entry | |
509 | * | |
510 | * @adev: amdgpu_device pointer | |
511 | * @wb: wb index | |
512 | * | |
513 | * Allocate a wb slot for use by the driver (all asics). | |
514 | * Returns 0 on success or -EINVAL on failure. | |
515 | */ | |
516 | int amdgpu_wb_get(struct amdgpu_device *adev, u32 *wb) | |
517 | { | |
518 | unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb); | |
519 | if (offset < adev->wb.num_wb) { | |
520 | __set_bit(offset, adev->wb.used); | |
521 | *wb = offset; | |
522 | return 0; | |
523 | } else { | |
524 | return -EINVAL; | |
525 | } | |
526 | } | |
527 | ||
528 | /** | |
529 | * amdgpu_wb_free - Free a wb entry | |
530 | * | |
531 | * @adev: amdgpu_device pointer | |
532 | * @wb: wb index | |
533 | * | |
534 | * Free a wb slot allocated for use by the driver (all asics) | |
535 | */ | |
536 | void amdgpu_wb_free(struct amdgpu_device *adev, u32 wb) | |
537 | { | |
538 | if (wb < adev->wb.num_wb) | |
539 | __clear_bit(wb, adev->wb.used); | |
540 | } | |
541 | ||
542 | /** | |
543 | * amdgpu_vram_location - try to find VRAM location | |
544 | * @adev: amdgpu device structure holding all necessary informations | |
545 | * @mc: memory controller structure holding memory informations | |
546 | * @base: base address at which to put VRAM | |
547 | * | |
548 | * Function will place try to place VRAM at base address provided | |
549 | * as parameter (which is so far either PCI aperture address or | |
550 | * for IGP TOM base address). | |
551 | * | |
552 | * If there is not enough space to fit the unvisible VRAM in the 32bits | |
553 | * address space then we limit the VRAM size to the aperture. | |
554 | * | |
555 | * Note: We don't explicitly enforce VRAM start to be aligned on VRAM size, | |
556 | * this shouldn't be a problem as we are using the PCI aperture as a reference. | |
557 | * Otherwise this would be needed for rv280, all r3xx, and all r4xx, but | |
558 | * not IGP. | |
559 | * | |
560 | * Note: we use mc_vram_size as on some board we need to program the mc to | |
561 | * cover the whole aperture even if VRAM size is inferior to aperture size | |
562 | * Novell bug 204882 + along with lots of ubuntu ones | |
563 | * | |
564 | * Note: when limiting vram it's safe to overwritte real_vram_size because | |
565 | * we are not in case where real_vram_size is inferior to mc_vram_size (ie | |
566 | * note afected by bogus hw of Novell bug 204882 + along with lots of ubuntu | |
567 | * ones) | |
568 | * | |
569 | * Note: IGP TOM addr should be the same as the aperture addr, we don't | |
570 | * explicitly check for that thought. | |
571 | * | |
572 | * FIXME: when reducing VRAM size align new size on power of 2. | |
573 | */ | |
574 | void amdgpu_vram_location(struct amdgpu_device *adev, struct amdgpu_mc *mc, u64 base) | |
575 | { | |
576 | uint64_t limit = (uint64_t)amdgpu_vram_limit << 20; | |
577 | ||
578 | mc->vram_start = base; | |
579 | if (mc->mc_vram_size > (adev->mc.mc_mask - base + 1)) { | |
580 | dev_warn(adev->dev, "limiting VRAM to PCI aperture size\n"); | |
581 | mc->real_vram_size = mc->aper_size; | |
582 | mc->mc_vram_size = mc->aper_size; | |
583 | } | |
584 | mc->vram_end = mc->vram_start + mc->mc_vram_size - 1; | |
585 | if (limit && limit < mc->real_vram_size) | |
586 | mc->real_vram_size = limit; | |
587 | dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n", | |
588 | mc->mc_vram_size >> 20, mc->vram_start, | |
589 | mc->vram_end, mc->real_vram_size >> 20); | |
590 | } | |
591 | ||
592 | /** | |
593 | * amdgpu_gtt_location - try to find GTT location | |
594 | * @adev: amdgpu device structure holding all necessary informations | |
595 | * @mc: memory controller structure holding memory informations | |
596 | * | |
597 | * Function will place try to place GTT before or after VRAM. | |
598 | * | |
599 | * If GTT size is bigger than space left then we ajust GTT size. | |
600 | * Thus function will never fails. | |
601 | * | |
602 | * FIXME: when reducing GTT size align new size on power of 2. | |
603 | */ | |
604 | void amdgpu_gtt_location(struct amdgpu_device *adev, struct amdgpu_mc *mc) | |
605 | { | |
606 | u64 size_af, size_bf; | |
607 | ||
608 | size_af = ((adev->mc.mc_mask - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align; | |
609 | size_bf = mc->vram_start & ~mc->gtt_base_align; | |
610 | if (size_bf > size_af) { | |
611 | if (mc->gtt_size > size_bf) { | |
612 | dev_warn(adev->dev, "limiting GTT\n"); | |
613 | mc->gtt_size = size_bf; | |
614 | } | |
615 | mc->gtt_start = (mc->vram_start & ~mc->gtt_base_align) - mc->gtt_size; | |
616 | } else { | |
617 | if (mc->gtt_size > size_af) { | |
618 | dev_warn(adev->dev, "limiting GTT\n"); | |
619 | mc->gtt_size = size_af; | |
620 | } | |
621 | mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align; | |
622 | } | |
623 | mc->gtt_end = mc->gtt_start + mc->gtt_size - 1; | |
624 | dev_info(adev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n", | |
625 | mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end); | |
626 | } | |
627 | ||
628 | /* | |
629 | * GPU helpers function. | |
630 | */ | |
631 | /** | |
632 | * amdgpu_card_posted - check if the hw has already been initialized | |
633 | * | |
634 | * @adev: amdgpu_device pointer | |
635 | * | |
636 | * Check if the asic has been initialized (all asics). | |
637 | * Used at driver startup. | |
638 | * Returns true if initialized or false if not. | |
639 | */ | |
640 | bool amdgpu_card_posted(struct amdgpu_device *adev) | |
641 | { | |
642 | uint32_t reg; | |
643 | ||
644 | /* then check MEM_SIZE, in case the crtcs are off */ | |
645 | reg = RREG32(mmCONFIG_MEMSIZE); | |
646 | ||
647 | if (reg) | |
648 | return true; | |
649 | ||
650 | return false; | |
651 | ||
652 | } | |
653 | ||
d38ceaf9 AD |
654 | /** |
655 | * amdgpu_dummy_page_init - init dummy page used by the driver | |
656 | * | |
657 | * @adev: amdgpu_device pointer | |
658 | * | |
659 | * Allocate the dummy page used by the driver (all asics). | |
660 | * This dummy page is used by the driver as a filler for gart entries | |
661 | * when pages are taken out of the GART | |
662 | * Returns 0 on sucess, -ENOMEM on failure. | |
663 | */ | |
664 | int amdgpu_dummy_page_init(struct amdgpu_device *adev) | |
665 | { | |
666 | if (adev->dummy_page.page) | |
667 | return 0; | |
668 | adev->dummy_page.page = alloc_page(GFP_DMA32 | GFP_KERNEL | __GFP_ZERO); | |
669 | if (adev->dummy_page.page == NULL) | |
670 | return -ENOMEM; | |
671 | adev->dummy_page.addr = pci_map_page(adev->pdev, adev->dummy_page.page, | |
672 | 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
673 | if (pci_dma_mapping_error(adev->pdev, adev->dummy_page.addr)) { | |
674 | dev_err(&adev->pdev->dev, "Failed to DMA MAP the dummy page\n"); | |
675 | __free_page(adev->dummy_page.page); | |
676 | adev->dummy_page.page = NULL; | |
677 | return -ENOMEM; | |
678 | } | |
679 | return 0; | |
680 | } | |
681 | ||
682 | /** | |
683 | * amdgpu_dummy_page_fini - free dummy page used by the driver | |
684 | * | |
685 | * @adev: amdgpu_device pointer | |
686 | * | |
687 | * Frees the dummy page used by the driver (all asics). | |
688 | */ | |
689 | void amdgpu_dummy_page_fini(struct amdgpu_device *adev) | |
690 | { | |
691 | if (adev->dummy_page.page == NULL) | |
692 | return; | |
693 | pci_unmap_page(adev->pdev, adev->dummy_page.addr, | |
694 | PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
695 | __free_page(adev->dummy_page.page); | |
696 | adev->dummy_page.page = NULL; | |
697 | } | |
698 | ||
699 | ||
700 | /* ATOM accessor methods */ | |
701 | /* | |
702 | * ATOM is an interpreted byte code stored in tables in the vbios. The | |
703 | * driver registers callbacks to access registers and the interpreter | |
704 | * in the driver parses the tables and executes then to program specific | |
705 | * actions (set display modes, asic init, etc.). See amdgpu_atombios.c, | |
706 | * atombios.h, and atom.c | |
707 | */ | |
708 | ||
709 | /** | |
710 | * cail_pll_read - read PLL register | |
711 | * | |
712 | * @info: atom card_info pointer | |
713 | * @reg: PLL register offset | |
714 | * | |
715 | * Provides a PLL register accessor for the atom interpreter (r4xx+). | |
716 | * Returns the value of the PLL register. | |
717 | */ | |
718 | static uint32_t cail_pll_read(struct card_info *info, uint32_t reg) | |
719 | { | |
720 | return 0; | |
721 | } | |
722 | ||
723 | /** | |
724 | * cail_pll_write - write PLL register | |
725 | * | |
726 | * @info: atom card_info pointer | |
727 | * @reg: PLL register offset | |
728 | * @val: value to write to the pll register | |
729 | * | |
730 | * Provides a PLL register accessor for the atom interpreter (r4xx+). | |
731 | */ | |
732 | static void cail_pll_write(struct card_info *info, uint32_t reg, uint32_t val) | |
733 | { | |
734 | ||
735 | } | |
736 | ||
737 | /** | |
738 | * cail_mc_read - read MC (Memory Controller) register | |
739 | * | |
740 | * @info: atom card_info pointer | |
741 | * @reg: MC register offset | |
742 | * | |
743 | * Provides an MC register accessor for the atom interpreter (r4xx+). | |
744 | * Returns the value of the MC register. | |
745 | */ | |
746 | static uint32_t cail_mc_read(struct card_info *info, uint32_t reg) | |
747 | { | |
748 | return 0; | |
749 | } | |
750 | ||
751 | /** | |
752 | * cail_mc_write - write MC (Memory Controller) register | |
753 | * | |
754 | * @info: atom card_info pointer | |
755 | * @reg: MC register offset | |
756 | * @val: value to write to the pll register | |
757 | * | |
758 | * Provides a MC register accessor for the atom interpreter (r4xx+). | |
759 | */ | |
760 | static void cail_mc_write(struct card_info *info, uint32_t reg, uint32_t val) | |
761 | { | |
762 | ||
763 | } | |
764 | ||
765 | /** | |
766 | * cail_reg_write - write MMIO register | |
767 | * | |
768 | * @info: atom card_info pointer | |
769 | * @reg: MMIO register offset | |
770 | * @val: value to write to the pll register | |
771 | * | |
772 | * Provides a MMIO register accessor for the atom interpreter (r4xx+). | |
773 | */ | |
774 | static void cail_reg_write(struct card_info *info, uint32_t reg, uint32_t val) | |
775 | { | |
776 | struct amdgpu_device *adev = info->dev->dev_private; | |
777 | ||
778 | WREG32(reg, val); | |
779 | } | |
780 | ||
781 | /** | |
782 | * cail_reg_read - read MMIO register | |
783 | * | |
784 | * @info: atom card_info pointer | |
785 | * @reg: MMIO register offset | |
786 | * | |
787 | * Provides an MMIO register accessor for the atom interpreter (r4xx+). | |
788 | * Returns the value of the MMIO register. | |
789 | */ | |
790 | static uint32_t cail_reg_read(struct card_info *info, uint32_t reg) | |
791 | { | |
792 | struct amdgpu_device *adev = info->dev->dev_private; | |
793 | uint32_t r; | |
794 | ||
795 | r = RREG32(reg); | |
796 | return r; | |
797 | } | |
798 | ||
799 | /** | |
800 | * cail_ioreg_write - write IO register | |
801 | * | |
802 | * @info: atom card_info pointer | |
803 | * @reg: IO register offset | |
804 | * @val: value to write to the pll register | |
805 | * | |
806 | * Provides a IO register accessor for the atom interpreter (r4xx+). | |
807 | */ | |
808 | static void cail_ioreg_write(struct card_info *info, uint32_t reg, uint32_t val) | |
809 | { | |
810 | struct amdgpu_device *adev = info->dev->dev_private; | |
811 | ||
812 | WREG32_IO(reg, val); | |
813 | } | |
814 | ||
815 | /** | |
816 | * cail_ioreg_read - read IO register | |
817 | * | |
818 | * @info: atom card_info pointer | |
819 | * @reg: IO register offset | |
820 | * | |
821 | * Provides an IO register accessor for the atom interpreter (r4xx+). | |
822 | * Returns the value of the IO register. | |
823 | */ | |
824 | static uint32_t cail_ioreg_read(struct card_info *info, uint32_t reg) | |
825 | { | |
826 | struct amdgpu_device *adev = info->dev->dev_private; | |
827 | uint32_t r; | |
828 | ||
829 | r = RREG32_IO(reg); | |
830 | return r; | |
831 | } | |
832 | ||
833 | /** | |
834 | * amdgpu_atombios_fini - free the driver info and callbacks for atombios | |
835 | * | |
836 | * @adev: amdgpu_device pointer | |
837 | * | |
838 | * Frees the driver info and register access callbacks for the ATOM | |
839 | * interpreter (r4xx+). | |
840 | * Called at driver shutdown. | |
841 | */ | |
842 | static void amdgpu_atombios_fini(struct amdgpu_device *adev) | |
843 | { | |
89e0ec9f | 844 | if (adev->mode_info.atom_context) { |
d38ceaf9 | 845 | kfree(adev->mode_info.atom_context->scratch); |
89e0ec9f ML |
846 | kfree(adev->mode_info.atom_context->iio); |
847 | } | |
d38ceaf9 AD |
848 | kfree(adev->mode_info.atom_context); |
849 | adev->mode_info.atom_context = NULL; | |
850 | kfree(adev->mode_info.atom_card_info); | |
851 | adev->mode_info.atom_card_info = NULL; | |
852 | } | |
853 | ||
854 | /** | |
855 | * amdgpu_atombios_init - init the driver info and callbacks for atombios | |
856 | * | |
857 | * @adev: amdgpu_device pointer | |
858 | * | |
859 | * Initializes the driver info and register access callbacks for the | |
860 | * ATOM interpreter (r4xx+). | |
861 | * Returns 0 on sucess, -ENOMEM on failure. | |
862 | * Called at driver startup. | |
863 | */ | |
864 | static int amdgpu_atombios_init(struct amdgpu_device *adev) | |
865 | { | |
866 | struct card_info *atom_card_info = | |
867 | kzalloc(sizeof(struct card_info), GFP_KERNEL); | |
868 | ||
869 | if (!atom_card_info) | |
870 | return -ENOMEM; | |
871 | ||
872 | adev->mode_info.atom_card_info = atom_card_info; | |
873 | atom_card_info->dev = adev->ddev; | |
874 | atom_card_info->reg_read = cail_reg_read; | |
875 | atom_card_info->reg_write = cail_reg_write; | |
876 | /* needed for iio ops */ | |
877 | if (adev->rio_mem) { | |
878 | atom_card_info->ioreg_read = cail_ioreg_read; | |
879 | atom_card_info->ioreg_write = cail_ioreg_write; | |
880 | } else { | |
881 | DRM_ERROR("Unable to find PCI I/O BAR; using MMIO for ATOM IIO\n"); | |
882 | atom_card_info->ioreg_read = cail_reg_read; | |
883 | atom_card_info->ioreg_write = cail_reg_write; | |
884 | } | |
885 | atom_card_info->mc_read = cail_mc_read; | |
886 | atom_card_info->mc_write = cail_mc_write; | |
887 | atom_card_info->pll_read = cail_pll_read; | |
888 | atom_card_info->pll_write = cail_pll_write; | |
889 | ||
890 | adev->mode_info.atom_context = amdgpu_atom_parse(atom_card_info, adev->bios); | |
891 | if (!adev->mode_info.atom_context) { | |
892 | amdgpu_atombios_fini(adev); | |
893 | return -ENOMEM; | |
894 | } | |
895 | ||
896 | mutex_init(&adev->mode_info.atom_context->mutex); | |
897 | amdgpu_atombios_scratch_regs_init(adev); | |
898 | amdgpu_atom_allocate_fb_scratch(adev->mode_info.atom_context); | |
899 | return 0; | |
900 | } | |
901 | ||
902 | /* if we get transitioned to only one device, take VGA back */ | |
903 | /** | |
904 | * amdgpu_vga_set_decode - enable/disable vga decode | |
905 | * | |
906 | * @cookie: amdgpu_device pointer | |
907 | * @state: enable/disable vga decode | |
908 | * | |
909 | * Enable/disable vga decode (all asics). | |
910 | * Returns VGA resource flags. | |
911 | */ | |
912 | static unsigned int amdgpu_vga_set_decode(void *cookie, bool state) | |
913 | { | |
914 | struct amdgpu_device *adev = cookie; | |
915 | amdgpu_asic_set_vga_state(adev, state); | |
916 | if (state) | |
917 | return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM | | |
918 | VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
919 | else | |
920 | return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
921 | } | |
922 | ||
923 | /** | |
924 | * amdgpu_check_pot_argument - check that argument is a power of two | |
925 | * | |
926 | * @arg: value to check | |
927 | * | |
928 | * Validates that a certain argument is a power of two (all asics). | |
929 | * Returns true if argument is valid. | |
930 | */ | |
931 | static bool amdgpu_check_pot_argument(int arg) | |
932 | { | |
933 | return (arg & (arg - 1)) == 0; | |
934 | } | |
935 | ||
936 | /** | |
937 | * amdgpu_check_arguments - validate module params | |
938 | * | |
939 | * @adev: amdgpu_device pointer | |
940 | * | |
941 | * Validates certain module parameters and updates | |
942 | * the associated values used by the driver (all asics). | |
943 | */ | |
944 | static void amdgpu_check_arguments(struct amdgpu_device *adev) | |
945 | { | |
5b011235 CZ |
946 | if (amdgpu_sched_jobs < 4) { |
947 | dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n", | |
948 | amdgpu_sched_jobs); | |
949 | amdgpu_sched_jobs = 4; | |
950 | } else if (!amdgpu_check_pot_argument(amdgpu_sched_jobs)){ | |
951 | dev_warn(adev->dev, "sched jobs (%d) must be a power of 2\n", | |
952 | amdgpu_sched_jobs); | |
953 | amdgpu_sched_jobs = roundup_pow_of_two(amdgpu_sched_jobs); | |
954 | } | |
d38ceaf9 AD |
955 | |
956 | if (amdgpu_gart_size != -1) { | |
c4e1a13a | 957 | /* gtt size must be greater or equal to 32M */ |
d38ceaf9 AD |
958 | if (amdgpu_gart_size < 32) { |
959 | dev_warn(adev->dev, "gart size (%d) too small\n", | |
960 | amdgpu_gart_size); | |
961 | amdgpu_gart_size = -1; | |
d38ceaf9 AD |
962 | } |
963 | } | |
964 | ||
965 | if (!amdgpu_check_pot_argument(amdgpu_vm_size)) { | |
966 | dev_warn(adev->dev, "VM size (%d) must be a power of 2\n", | |
967 | amdgpu_vm_size); | |
8dacc127 | 968 | amdgpu_vm_size = 8; |
d38ceaf9 AD |
969 | } |
970 | ||
971 | if (amdgpu_vm_size < 1) { | |
972 | dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n", | |
973 | amdgpu_vm_size); | |
8dacc127 | 974 | amdgpu_vm_size = 8; |
d38ceaf9 AD |
975 | } |
976 | ||
977 | /* | |
978 | * Max GPUVM size for Cayman, SI and CI are 40 bits. | |
979 | */ | |
980 | if (amdgpu_vm_size > 1024) { | |
981 | dev_warn(adev->dev, "VM size (%d) too large, max is 1TB\n", | |
982 | amdgpu_vm_size); | |
8dacc127 | 983 | amdgpu_vm_size = 8; |
d38ceaf9 AD |
984 | } |
985 | ||
986 | /* defines number of bits in page table versus page directory, | |
987 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
988 | * page table and the remaining bits are in the page directory */ | |
989 | if (amdgpu_vm_block_size == -1) { | |
990 | ||
991 | /* Total bits covered by PD + PTs */ | |
992 | unsigned bits = ilog2(amdgpu_vm_size) + 18; | |
993 | ||
994 | /* Make sure the PD is 4K in size up to 8GB address space. | |
995 | Above that split equal between PD and PTs */ | |
996 | if (amdgpu_vm_size <= 8) | |
997 | amdgpu_vm_block_size = bits - 9; | |
998 | else | |
999 | amdgpu_vm_block_size = (bits + 3) / 2; | |
1000 | ||
1001 | } else if (amdgpu_vm_block_size < 9) { | |
1002 | dev_warn(adev->dev, "VM page table size (%d) too small\n", | |
1003 | amdgpu_vm_block_size); | |
1004 | amdgpu_vm_block_size = 9; | |
1005 | } | |
1006 | ||
1007 | if (amdgpu_vm_block_size > 24 || | |
1008 | (amdgpu_vm_size * 1024) < (1ull << amdgpu_vm_block_size)) { | |
1009 | dev_warn(adev->dev, "VM page table size (%d) too large\n", | |
1010 | amdgpu_vm_block_size); | |
1011 | amdgpu_vm_block_size = 9; | |
1012 | } | |
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 | */ | |
1024 | static 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) { | |
1032 | unsigned d3_delay = dev->pdev->d3_delay; | |
1033 | ||
1034 | printk(KERN_INFO "amdgpu: switched on\n"); | |
1035 | /* don't suspend or resume card normally */ | |
1036 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
1037 | ||
810ddc3a | 1038 | amdgpu_device_resume(dev, true, true); |
d38ceaf9 AD |
1039 | |
1040 | dev->pdev->d3_delay = d3_delay; | |
1041 | ||
1042 | dev->switch_power_state = DRM_SWITCH_POWER_ON; | |
1043 | drm_kms_helper_poll_enable(dev); | |
1044 | } else { | |
1045 | printk(KERN_INFO "amdgpu: switched off\n"); | |
1046 | drm_kms_helper_poll_disable(dev); | |
1047 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
810ddc3a | 1048 | amdgpu_device_suspend(dev, true, true); |
d38ceaf9 AD |
1049 | dev->switch_power_state = DRM_SWITCH_POWER_OFF; |
1050 | } | |
1051 | } | |
1052 | ||
1053 | /** | |
1054 | * amdgpu_switcheroo_can_switch - see if switcheroo state can change | |
1055 | * | |
1056 | * @pdev: pci dev pointer | |
1057 | * | |
1058 | * Callback for the switcheroo driver. Check of the switcheroo | |
1059 | * state can be changed. | |
1060 | * Returns true if the state can be changed, false if not. | |
1061 | */ | |
1062 | static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev) | |
1063 | { | |
1064 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1065 | ||
1066 | /* | |
1067 | * FIXME: open_count is protected by drm_global_mutex but that would lead to | |
1068 | * locking inversion with the driver load path. And the access here is | |
1069 | * completely racy anyway. So don't bother with locking for now. | |
1070 | */ | |
1071 | return dev->open_count == 0; | |
1072 | } | |
1073 | ||
1074 | static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = { | |
1075 | .set_gpu_state = amdgpu_switcheroo_set_state, | |
1076 | .reprobe = NULL, | |
1077 | .can_switch = amdgpu_switcheroo_can_switch, | |
1078 | }; | |
1079 | ||
1080 | int amdgpu_set_clockgating_state(struct amdgpu_device *adev, | |
5fc3aeeb | 1081 | enum amd_ip_block_type block_type, |
1082 | enum amd_clockgating_state state) | |
d38ceaf9 AD |
1083 | { |
1084 | int i, r = 0; | |
1085 | ||
1086 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
9ecbe7f5 AD |
1087 | if (!adev->ip_block_status[i].valid) |
1088 | continue; | |
d38ceaf9 | 1089 | if (adev->ip_blocks[i].type == block_type) { |
5fc3aeeb | 1090 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, |
d38ceaf9 AD |
1091 | state); |
1092 | if (r) | |
1093 | return r; | |
a225bf1c | 1094 | break; |
d38ceaf9 AD |
1095 | } |
1096 | } | |
1097 | return r; | |
1098 | } | |
1099 | ||
1100 | int amdgpu_set_powergating_state(struct amdgpu_device *adev, | |
5fc3aeeb | 1101 | enum amd_ip_block_type block_type, |
1102 | enum amd_powergating_state state) | |
d38ceaf9 AD |
1103 | { |
1104 | int i, r = 0; | |
1105 | ||
1106 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
9ecbe7f5 AD |
1107 | if (!adev->ip_block_status[i].valid) |
1108 | continue; | |
d38ceaf9 | 1109 | if (adev->ip_blocks[i].type == block_type) { |
5fc3aeeb | 1110 | r = adev->ip_blocks[i].funcs->set_powergating_state((void *)adev, |
d38ceaf9 AD |
1111 | state); |
1112 | if (r) | |
1113 | return r; | |
a225bf1c | 1114 | break; |
d38ceaf9 AD |
1115 | } |
1116 | } | |
1117 | return r; | |
1118 | } | |
1119 | ||
5dbbb60b AD |
1120 | int amdgpu_wait_for_idle(struct amdgpu_device *adev, |
1121 | enum amd_ip_block_type block_type) | |
1122 | { | |
1123 | int i, r; | |
1124 | ||
1125 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
9ecbe7f5 AD |
1126 | if (!adev->ip_block_status[i].valid) |
1127 | continue; | |
5dbbb60b AD |
1128 | if (adev->ip_blocks[i].type == block_type) { |
1129 | r = adev->ip_blocks[i].funcs->wait_for_idle((void *)adev); | |
1130 | if (r) | |
1131 | return r; | |
1132 | break; | |
1133 | } | |
1134 | } | |
1135 | return 0; | |
1136 | ||
1137 | } | |
1138 | ||
1139 | bool amdgpu_is_idle(struct amdgpu_device *adev, | |
1140 | enum amd_ip_block_type block_type) | |
1141 | { | |
1142 | int i; | |
1143 | ||
1144 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
9ecbe7f5 AD |
1145 | if (!adev->ip_block_status[i].valid) |
1146 | continue; | |
5dbbb60b AD |
1147 | if (adev->ip_blocks[i].type == block_type) |
1148 | return adev->ip_blocks[i].funcs->is_idle((void *)adev); | |
1149 | } | |
1150 | return true; | |
1151 | ||
1152 | } | |
1153 | ||
d38ceaf9 AD |
1154 | const struct amdgpu_ip_block_version * amdgpu_get_ip_block( |
1155 | struct amdgpu_device *adev, | |
5fc3aeeb | 1156 | enum amd_ip_block_type type) |
d38ceaf9 AD |
1157 | { |
1158 | int i; | |
1159 | ||
1160 | for (i = 0; i < adev->num_ip_blocks; i++) | |
1161 | if (adev->ip_blocks[i].type == type) | |
1162 | return &adev->ip_blocks[i]; | |
1163 | ||
1164 | return NULL; | |
1165 | } | |
1166 | ||
1167 | /** | |
1168 | * amdgpu_ip_block_version_cmp | |
1169 | * | |
1170 | * @adev: amdgpu_device pointer | |
5fc3aeeb | 1171 | * @type: enum amd_ip_block_type |
d38ceaf9 AD |
1172 | * @major: major version |
1173 | * @minor: minor version | |
1174 | * | |
1175 | * return 0 if equal or greater | |
1176 | * return 1 if smaller or the ip_block doesn't exist | |
1177 | */ | |
1178 | int amdgpu_ip_block_version_cmp(struct amdgpu_device *adev, | |
5fc3aeeb | 1179 | enum amd_ip_block_type type, |
d38ceaf9 AD |
1180 | u32 major, u32 minor) |
1181 | { | |
1182 | const struct amdgpu_ip_block_version *ip_block; | |
1183 | ip_block = amdgpu_get_ip_block(adev, type); | |
1184 | ||
1185 | if (ip_block && ((ip_block->major > major) || | |
1186 | ((ip_block->major == major) && | |
1187 | (ip_block->minor >= minor)))) | |
1188 | return 0; | |
1189 | ||
1190 | return 1; | |
1191 | } | |
1192 | ||
9accf2fd ED |
1193 | static void amdgpu_whether_enable_virtual_display(struct amdgpu_device *adev) |
1194 | { | |
1195 | adev->enable_virtual_display = false; | |
1196 | ||
1197 | if (amdgpu_virtual_display) { | |
1198 | struct drm_device *ddev = adev->ddev; | |
1199 | const char *pci_address_name = pci_name(ddev->pdev); | |
1200 | char *pciaddstr, *pciaddstr_tmp, *pciaddname; | |
1201 | ||
1202 | pciaddstr = kstrdup(amdgpu_virtual_display, GFP_KERNEL); | |
1203 | pciaddstr_tmp = pciaddstr; | |
1204 | while ((pciaddname = strsep(&pciaddstr_tmp, ";"))) { | |
1205 | if (!strcmp(pci_address_name, pciaddname)) { | |
1206 | adev->enable_virtual_display = true; | |
1207 | break; | |
1208 | } | |
1209 | } | |
1210 | ||
1211 | DRM_INFO("virtual display string:%s, %s:virtual_display:%d\n", | |
1212 | amdgpu_virtual_display, pci_address_name, | |
1213 | adev->enable_virtual_display); | |
1214 | ||
1215 | kfree(pciaddstr); | |
1216 | } | |
1217 | } | |
1218 | ||
d38ceaf9 AD |
1219 | static int amdgpu_early_init(struct amdgpu_device *adev) |
1220 | { | |
aaa36a97 | 1221 | int i, r; |
d38ceaf9 | 1222 | |
9accf2fd | 1223 | amdgpu_whether_enable_virtual_display(adev); |
a6be7570 | 1224 | |
d38ceaf9 | 1225 | switch (adev->asic_type) { |
aaa36a97 AD |
1226 | case CHIP_TOPAZ: |
1227 | case CHIP_TONGA: | |
48299f95 | 1228 | case CHIP_FIJI: |
2cc0c0b5 FC |
1229 | case CHIP_POLARIS11: |
1230 | case CHIP_POLARIS10: | |
aaa36a97 | 1231 | case CHIP_CARRIZO: |
39bb0c92 SL |
1232 | case CHIP_STONEY: |
1233 | if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) | |
aaa36a97 AD |
1234 | adev->family = AMDGPU_FAMILY_CZ; |
1235 | else | |
1236 | adev->family = AMDGPU_FAMILY_VI; | |
1237 | ||
1238 | r = vi_set_ip_blocks(adev); | |
1239 | if (r) | |
1240 | return r; | |
1241 | break; | |
33f34802 KW |
1242 | #ifdef CONFIG_DRM_AMDGPU_SI |
1243 | case CHIP_VERDE: | |
1244 | case CHIP_TAHITI: | |
1245 | case CHIP_PITCAIRN: | |
1246 | case CHIP_OLAND: | |
1247 | case CHIP_HAINAN: | |
1248 | r = si_set_ip_blocks(adev); | |
1249 | if (r) | |
1250 | return r; | |
1251 | break; | |
1252 | #endif | |
a2e73f56 AD |
1253 | #ifdef CONFIG_DRM_AMDGPU_CIK |
1254 | case CHIP_BONAIRE: | |
1255 | case CHIP_HAWAII: | |
1256 | case CHIP_KAVERI: | |
1257 | case CHIP_KABINI: | |
1258 | case CHIP_MULLINS: | |
1259 | if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII)) | |
1260 | adev->family = AMDGPU_FAMILY_CI; | |
1261 | else | |
1262 | adev->family = AMDGPU_FAMILY_KV; | |
1263 | ||
1264 | r = cik_set_ip_blocks(adev); | |
1265 | if (r) | |
1266 | return r; | |
1267 | break; | |
1268 | #endif | |
d38ceaf9 AD |
1269 | default: |
1270 | /* FIXME: not supported yet */ | |
1271 | return -EINVAL; | |
1272 | } | |
1273 | ||
8faf0e08 AD |
1274 | adev->ip_block_status = kcalloc(adev->num_ip_blocks, |
1275 | sizeof(struct amdgpu_ip_block_status), GFP_KERNEL); | |
1276 | if (adev->ip_block_status == NULL) | |
d8d090b7 | 1277 | return -ENOMEM; |
d38ceaf9 AD |
1278 | |
1279 | if (adev->ip_blocks == NULL) { | |
1280 | DRM_ERROR("No IP blocks found!\n"); | |
1281 | return r; | |
1282 | } | |
1283 | ||
1284 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1285 | if ((amdgpu_ip_block_mask & (1 << i)) == 0) { | |
1286 | DRM_ERROR("disabled ip block: %d\n", i); | |
8faf0e08 | 1287 | adev->ip_block_status[i].valid = false; |
d38ceaf9 AD |
1288 | } else { |
1289 | if (adev->ip_blocks[i].funcs->early_init) { | |
5fc3aeeb | 1290 | r = adev->ip_blocks[i].funcs->early_init((void *)adev); |
2c1a2784 | 1291 | if (r == -ENOENT) { |
8faf0e08 | 1292 | adev->ip_block_status[i].valid = false; |
2c1a2784 | 1293 | } else if (r) { |
88a907d6 | 1294 | DRM_ERROR("early_init of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
d38ceaf9 | 1295 | return r; |
2c1a2784 | 1296 | } else { |
8faf0e08 | 1297 | adev->ip_block_status[i].valid = true; |
2c1a2784 | 1298 | } |
974e6b64 | 1299 | } else { |
8faf0e08 | 1300 | adev->ip_block_status[i].valid = true; |
d38ceaf9 | 1301 | } |
d38ceaf9 AD |
1302 | } |
1303 | } | |
1304 | ||
395d1fb9 NH |
1305 | adev->cg_flags &= amdgpu_cg_mask; |
1306 | adev->pg_flags &= amdgpu_pg_mask; | |
1307 | ||
d38ceaf9 AD |
1308 | return 0; |
1309 | } | |
1310 | ||
1311 | static int amdgpu_init(struct amdgpu_device *adev) | |
1312 | { | |
1313 | int i, r; | |
1314 | ||
1315 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1316 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 | 1317 | continue; |
5fc3aeeb | 1318 | r = adev->ip_blocks[i].funcs->sw_init((void *)adev); |
2c1a2784 | 1319 | if (r) { |
822b2cef | 1320 | DRM_ERROR("sw_init of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
d38ceaf9 | 1321 | return r; |
2c1a2784 | 1322 | } |
8faf0e08 | 1323 | adev->ip_block_status[i].sw = true; |
d38ceaf9 | 1324 | /* need to do gmc hw init early so we can allocate gpu mem */ |
5fc3aeeb | 1325 | if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) { |
d38ceaf9 | 1326 | r = amdgpu_vram_scratch_init(adev); |
2c1a2784 AD |
1327 | if (r) { |
1328 | DRM_ERROR("amdgpu_vram_scratch_init failed %d\n", r); | |
d38ceaf9 | 1329 | return r; |
2c1a2784 | 1330 | } |
5fc3aeeb | 1331 | r = adev->ip_blocks[i].funcs->hw_init((void *)adev); |
2c1a2784 AD |
1332 | if (r) { |
1333 | DRM_ERROR("hw_init %d failed %d\n", i, r); | |
d38ceaf9 | 1334 | return r; |
2c1a2784 | 1335 | } |
d38ceaf9 | 1336 | r = amdgpu_wb_init(adev); |
2c1a2784 AD |
1337 | if (r) { |
1338 | DRM_ERROR("amdgpu_wb_init failed %d\n", r); | |
d38ceaf9 | 1339 | return r; |
2c1a2784 | 1340 | } |
8faf0e08 | 1341 | adev->ip_block_status[i].hw = true; |
d38ceaf9 AD |
1342 | } |
1343 | } | |
1344 | ||
1345 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1346 | if (!adev->ip_block_status[i].sw) |
d38ceaf9 AD |
1347 | continue; |
1348 | /* gmc hw init is done early */ | |
5fc3aeeb | 1349 | if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) |
d38ceaf9 | 1350 | continue; |
5fc3aeeb | 1351 | r = adev->ip_blocks[i].funcs->hw_init((void *)adev); |
2c1a2784 | 1352 | if (r) { |
822b2cef | 1353 | DRM_ERROR("hw_init of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
d38ceaf9 | 1354 | return r; |
2c1a2784 | 1355 | } |
8faf0e08 | 1356 | adev->ip_block_status[i].hw = true; |
d38ceaf9 AD |
1357 | } |
1358 | ||
1359 | return 0; | |
1360 | } | |
1361 | ||
1362 | static int amdgpu_late_init(struct amdgpu_device *adev) | |
1363 | { | |
1364 | int i = 0, r; | |
1365 | ||
1366 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1367 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 AD |
1368 | continue; |
1369 | /* enable clockgating to save power */ | |
5fc3aeeb | 1370 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, |
1371 | AMD_CG_STATE_GATE); | |
2c1a2784 | 1372 | if (r) { |
822b2cef | 1373 | DRM_ERROR("set_clockgating_state(gate) of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
d38ceaf9 | 1374 | return r; |
2c1a2784 | 1375 | } |
d38ceaf9 | 1376 | if (adev->ip_blocks[i].funcs->late_init) { |
5fc3aeeb | 1377 | r = adev->ip_blocks[i].funcs->late_init((void *)adev); |
2c1a2784 | 1378 | if (r) { |
822b2cef | 1379 | DRM_ERROR("late_init of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
d38ceaf9 | 1380 | return r; |
2c1a2784 | 1381 | } |
d38ceaf9 AD |
1382 | } |
1383 | } | |
1384 | ||
1385 | return 0; | |
1386 | } | |
1387 | ||
1388 | static int amdgpu_fini(struct amdgpu_device *adev) | |
1389 | { | |
1390 | int i, r; | |
1391 | ||
1392 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
8faf0e08 | 1393 | if (!adev->ip_block_status[i].hw) |
d38ceaf9 | 1394 | continue; |
5fc3aeeb | 1395 | if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) { |
d38ceaf9 AD |
1396 | amdgpu_wb_fini(adev); |
1397 | amdgpu_vram_scratch_fini(adev); | |
1398 | } | |
1399 | /* ungate blocks before hw fini so that we can shutdown the blocks safely */ | |
5fc3aeeb | 1400 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, |
1401 | AMD_CG_STATE_UNGATE); | |
2c1a2784 | 1402 | if (r) { |
822b2cef | 1403 | DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
d38ceaf9 | 1404 | return r; |
2c1a2784 | 1405 | } |
5fc3aeeb | 1406 | r = adev->ip_blocks[i].funcs->hw_fini((void *)adev); |
d38ceaf9 | 1407 | /* XXX handle errors */ |
2c1a2784 | 1408 | if (r) { |
822b2cef | 1409 | DRM_DEBUG("hw_fini of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
2c1a2784 | 1410 | } |
8faf0e08 | 1411 | adev->ip_block_status[i].hw = false; |
d38ceaf9 AD |
1412 | } |
1413 | ||
1414 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
8faf0e08 | 1415 | if (!adev->ip_block_status[i].sw) |
d38ceaf9 | 1416 | continue; |
5fc3aeeb | 1417 | r = adev->ip_blocks[i].funcs->sw_fini((void *)adev); |
d38ceaf9 | 1418 | /* XXX handle errors */ |
2c1a2784 | 1419 | if (r) { |
822b2cef | 1420 | DRM_DEBUG("sw_fini of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
2c1a2784 | 1421 | } |
8faf0e08 AD |
1422 | adev->ip_block_status[i].sw = false; |
1423 | adev->ip_block_status[i].valid = false; | |
d38ceaf9 AD |
1424 | } |
1425 | ||
a6dcfd9c ML |
1426 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
1427 | if (adev->ip_blocks[i].funcs->late_fini) | |
1428 | adev->ip_blocks[i].funcs->late_fini((void *)adev); | |
1429 | } | |
1430 | ||
d38ceaf9 AD |
1431 | return 0; |
1432 | } | |
1433 | ||
1434 | static int amdgpu_suspend(struct amdgpu_device *adev) | |
1435 | { | |
1436 | int i, r; | |
1437 | ||
c5a93a28 FC |
1438 | /* ungate SMC block first */ |
1439 | r = amdgpu_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_SMC, | |
1440 | AMD_CG_STATE_UNGATE); | |
1441 | if (r) { | |
1442 | DRM_ERROR("set_clockgating_state(ungate) SMC failed %d\n",r); | |
1443 | } | |
1444 | ||
d38ceaf9 | 1445 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { |
8faf0e08 | 1446 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 AD |
1447 | continue; |
1448 | /* ungate blocks so that suspend can properly shut them down */ | |
c5a93a28 FC |
1449 | if (i != AMD_IP_BLOCK_TYPE_SMC) { |
1450 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, | |
1451 | AMD_CG_STATE_UNGATE); | |
1452 | if (r) { | |
822b2cef | 1453 | DRM_ERROR("set_clockgating_state(ungate) of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
c5a93a28 | 1454 | } |
2c1a2784 | 1455 | } |
d38ceaf9 AD |
1456 | /* XXX handle errors */ |
1457 | r = adev->ip_blocks[i].funcs->suspend(adev); | |
1458 | /* XXX handle errors */ | |
2c1a2784 | 1459 | if (r) { |
822b2cef | 1460 | DRM_ERROR("suspend of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
2c1a2784 | 1461 | } |
d38ceaf9 AD |
1462 | } |
1463 | ||
1464 | return 0; | |
1465 | } | |
1466 | ||
1467 | static int amdgpu_resume(struct amdgpu_device *adev) | |
1468 | { | |
1469 | int i, r; | |
1470 | ||
1471 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1472 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 AD |
1473 | continue; |
1474 | r = adev->ip_blocks[i].funcs->resume(adev); | |
2c1a2784 | 1475 | if (r) { |
822b2cef | 1476 | DRM_ERROR("resume of IP block <%s> failed %d\n", adev->ip_blocks[i].funcs->name, r); |
d38ceaf9 | 1477 | return r; |
2c1a2784 | 1478 | } |
d38ceaf9 AD |
1479 | } |
1480 | ||
1481 | return 0; | |
1482 | } | |
1483 | ||
048765ad AR |
1484 | static bool amdgpu_device_is_virtual(void) |
1485 | { | |
1486 | #ifdef CONFIG_X86 | |
1487 | return boot_cpu_has(X86_FEATURE_HYPERVISOR); | |
1488 | #else | |
1489 | return false; | |
1490 | #endif | |
1491 | } | |
1492 | ||
d38ceaf9 AD |
1493 | /** |
1494 | * amdgpu_device_init - initialize the driver | |
1495 | * | |
1496 | * @adev: amdgpu_device pointer | |
1497 | * @pdev: drm dev pointer | |
1498 | * @pdev: pci dev pointer | |
1499 | * @flags: driver flags | |
1500 | * | |
1501 | * Initializes the driver info and hw (all asics). | |
1502 | * Returns 0 for success or an error on failure. | |
1503 | * Called at driver startup. | |
1504 | */ | |
1505 | int amdgpu_device_init(struct amdgpu_device *adev, | |
1506 | struct drm_device *ddev, | |
1507 | struct pci_dev *pdev, | |
1508 | uint32_t flags) | |
1509 | { | |
1510 | int r, i; | |
1511 | bool runtime = false; | |
95844d20 | 1512 | u32 max_MBps; |
d38ceaf9 AD |
1513 | |
1514 | adev->shutdown = false; | |
1515 | adev->dev = &pdev->dev; | |
1516 | adev->ddev = ddev; | |
1517 | adev->pdev = pdev; | |
1518 | adev->flags = flags; | |
2f7d10b3 | 1519 | adev->asic_type = flags & AMD_ASIC_MASK; |
d38ceaf9 AD |
1520 | adev->is_atom_bios = false; |
1521 | adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT; | |
1522 | adev->mc.gtt_size = 512 * 1024 * 1024; | |
1523 | adev->accel_working = false; | |
1524 | adev->num_rings = 0; | |
1525 | adev->mman.buffer_funcs = NULL; | |
1526 | adev->mman.buffer_funcs_ring = NULL; | |
1527 | adev->vm_manager.vm_pte_funcs = NULL; | |
2d55e45a | 1528 | adev->vm_manager.vm_pte_num_rings = 0; |
d38ceaf9 AD |
1529 | adev->gart.gart_funcs = NULL; |
1530 | adev->fence_context = fence_context_alloc(AMDGPU_MAX_RINGS); | |
1531 | ||
1532 | adev->smc_rreg = &amdgpu_invalid_rreg; | |
1533 | adev->smc_wreg = &amdgpu_invalid_wreg; | |
1534 | adev->pcie_rreg = &amdgpu_invalid_rreg; | |
1535 | adev->pcie_wreg = &amdgpu_invalid_wreg; | |
36b9a952 HR |
1536 | adev->pciep_rreg = &amdgpu_invalid_rreg; |
1537 | adev->pciep_wreg = &amdgpu_invalid_wreg; | |
d38ceaf9 AD |
1538 | adev->uvd_ctx_rreg = &amdgpu_invalid_rreg; |
1539 | adev->uvd_ctx_wreg = &amdgpu_invalid_wreg; | |
1540 | adev->didt_rreg = &amdgpu_invalid_rreg; | |
1541 | adev->didt_wreg = &amdgpu_invalid_wreg; | |
ccdbb20a RZ |
1542 | adev->gc_cac_rreg = &amdgpu_invalid_rreg; |
1543 | adev->gc_cac_wreg = &amdgpu_invalid_wreg; | |
d38ceaf9 AD |
1544 | adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg; |
1545 | adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg; | |
1546 | ||
ccdbb20a | 1547 | |
3e39ab90 AD |
1548 | DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n", |
1549 | amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device, | |
1550 | pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision); | |
d38ceaf9 AD |
1551 | |
1552 | /* mutex initialization are all done here so we | |
1553 | * can recall function without having locking issues */ | |
8d0a7cea | 1554 | mutex_init(&adev->vm_manager.lock); |
d38ceaf9 | 1555 | atomic_set(&adev->irq.ih.lock, 0); |
d38ceaf9 AD |
1556 | mutex_init(&adev->pm.mutex); |
1557 | mutex_init(&adev->gfx.gpu_clock_mutex); | |
1558 | mutex_init(&adev->srbm_mutex); | |
1559 | mutex_init(&adev->grbm_idx_mutex); | |
d38ceaf9 AD |
1560 | mutex_init(&adev->mn_lock); |
1561 | hash_init(adev->mn_hash); | |
1562 | ||
1563 | amdgpu_check_arguments(adev); | |
1564 | ||
1565 | /* Registers mapping */ | |
1566 | /* TODO: block userspace mapping of io register */ | |
1567 | spin_lock_init(&adev->mmio_idx_lock); | |
1568 | spin_lock_init(&adev->smc_idx_lock); | |
1569 | spin_lock_init(&adev->pcie_idx_lock); | |
1570 | spin_lock_init(&adev->uvd_ctx_idx_lock); | |
1571 | spin_lock_init(&adev->didt_idx_lock); | |
ccdbb20a | 1572 | spin_lock_init(&adev->gc_cac_idx_lock); |
d38ceaf9 | 1573 | spin_lock_init(&adev->audio_endpt_idx_lock); |
95844d20 | 1574 | spin_lock_init(&adev->mm_stats.lock); |
d38ceaf9 | 1575 | |
0c4e7fa5 CZ |
1576 | INIT_LIST_HEAD(&adev->shadow_list); |
1577 | mutex_init(&adev->shadow_list_lock); | |
1578 | ||
da69c161 KW |
1579 | if (adev->asic_type >= CHIP_BONAIRE) { |
1580 | adev->rmmio_base = pci_resource_start(adev->pdev, 5); | |
1581 | adev->rmmio_size = pci_resource_len(adev->pdev, 5); | |
1582 | } else { | |
1583 | adev->rmmio_base = pci_resource_start(adev->pdev, 2); | |
1584 | adev->rmmio_size = pci_resource_len(adev->pdev, 2); | |
1585 | } | |
d38ceaf9 AD |
1586 | adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size); |
1587 | if (adev->rmmio == NULL) { | |
1588 | return -ENOMEM; | |
1589 | } | |
1590 | DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base); | |
1591 | DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size); | |
1592 | ||
da69c161 KW |
1593 | if (adev->asic_type >= CHIP_BONAIRE) |
1594 | /* doorbell bar mapping */ | |
1595 | amdgpu_doorbell_init(adev); | |
d38ceaf9 AD |
1596 | |
1597 | /* io port mapping */ | |
1598 | for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { | |
1599 | if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) { | |
1600 | adev->rio_mem_size = pci_resource_len(adev->pdev, i); | |
1601 | adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size); | |
1602 | break; | |
1603 | } | |
1604 | } | |
1605 | if (adev->rio_mem == NULL) | |
1606 | DRM_ERROR("Unable to find PCI I/O BAR\n"); | |
1607 | ||
1608 | /* early init functions */ | |
1609 | r = amdgpu_early_init(adev); | |
1610 | if (r) | |
1611 | return r; | |
1612 | ||
1613 | /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */ | |
1614 | /* this will fail for cards that aren't VGA class devices, just | |
1615 | * ignore it */ | |
1616 | vga_client_register(adev->pdev, adev, NULL, amdgpu_vga_set_decode); | |
1617 | ||
1618 | if (amdgpu_runtime_pm == 1) | |
1619 | runtime = true; | |
e9bef455 | 1620 | if (amdgpu_device_is_px(ddev)) |
d38ceaf9 AD |
1621 | runtime = true; |
1622 | vga_switcheroo_register_client(adev->pdev, &amdgpu_switcheroo_ops, runtime); | |
1623 | if (runtime) | |
1624 | vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain); | |
1625 | ||
1626 | /* Read BIOS */ | |
83ba126a AD |
1627 | if (!amdgpu_get_bios(adev)) { |
1628 | r = -EINVAL; | |
1629 | goto failed; | |
1630 | } | |
d38ceaf9 AD |
1631 | /* Must be an ATOMBIOS */ |
1632 | if (!adev->is_atom_bios) { | |
1633 | dev_err(adev->dev, "Expecting atombios for GPU\n"); | |
83ba126a AD |
1634 | r = -EINVAL; |
1635 | goto failed; | |
d38ceaf9 AD |
1636 | } |
1637 | r = amdgpu_atombios_init(adev); | |
2c1a2784 AD |
1638 | if (r) { |
1639 | dev_err(adev->dev, "amdgpu_atombios_init failed\n"); | |
83ba126a | 1640 | goto failed; |
2c1a2784 | 1641 | } |
d38ceaf9 | 1642 | |
7e471e6f AD |
1643 | /* See if the asic supports SR-IOV */ |
1644 | adev->virtualization.supports_sr_iov = | |
1645 | amdgpu_atombios_has_gpu_virtualization_table(adev); | |
1646 | ||
048765ad AR |
1647 | /* Check if we are executing in a virtualized environment */ |
1648 | adev->virtualization.is_virtual = amdgpu_device_is_virtual(); | |
1649 | adev->virtualization.caps = amdgpu_asic_get_virtual_caps(adev); | |
1650 | ||
d38ceaf9 | 1651 | /* Post card if necessary */ |
048765ad AR |
1652 | if (!amdgpu_card_posted(adev) || |
1653 | (adev->virtualization.is_virtual && | |
48a70e1c | 1654 | !(adev->virtualization.caps & AMDGPU_VIRT_CAPS_SRIOV_EN))) { |
d38ceaf9 AD |
1655 | if (!adev->bios) { |
1656 | dev_err(adev->dev, "Card not posted and no BIOS - ignoring\n"); | |
83ba126a AD |
1657 | r = -EINVAL; |
1658 | goto failed; | |
d38ceaf9 AD |
1659 | } |
1660 | DRM_INFO("GPU not posted. posting now...\n"); | |
1661 | amdgpu_atom_asic_init(adev->mode_info.atom_context); | |
1662 | } | |
1663 | ||
1664 | /* Initialize clocks */ | |
1665 | r = amdgpu_atombios_get_clock_info(adev); | |
2c1a2784 AD |
1666 | if (r) { |
1667 | dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n"); | |
83ba126a | 1668 | goto failed; |
2c1a2784 | 1669 | } |
d38ceaf9 AD |
1670 | /* init i2c buses */ |
1671 | amdgpu_atombios_i2c_init(adev); | |
1672 | ||
1673 | /* Fence driver */ | |
1674 | r = amdgpu_fence_driver_init(adev); | |
2c1a2784 AD |
1675 | if (r) { |
1676 | dev_err(adev->dev, "amdgpu_fence_driver_init failed\n"); | |
83ba126a | 1677 | goto failed; |
2c1a2784 | 1678 | } |
d38ceaf9 AD |
1679 | |
1680 | /* init the mode config */ | |
1681 | drm_mode_config_init(adev->ddev); | |
1682 | ||
1683 | r = amdgpu_init(adev); | |
1684 | if (r) { | |
2c1a2784 | 1685 | dev_err(adev->dev, "amdgpu_init failed\n"); |
d38ceaf9 | 1686 | amdgpu_fini(adev); |
83ba126a | 1687 | goto failed; |
d38ceaf9 AD |
1688 | } |
1689 | ||
1690 | adev->accel_working = true; | |
1691 | ||
95844d20 MO |
1692 | /* Initialize the buffer migration limit. */ |
1693 | if (amdgpu_moverate >= 0) | |
1694 | max_MBps = amdgpu_moverate; | |
1695 | else | |
1696 | max_MBps = 8; /* Allow 8 MB/s. */ | |
1697 | /* Get a log2 for easy divisions. */ | |
1698 | adev->mm_stats.log2_max_MBps = ilog2(max(1u, max_MBps)); | |
1699 | ||
d38ceaf9 AD |
1700 | amdgpu_fbdev_init(adev); |
1701 | ||
1702 | r = amdgpu_ib_pool_init(adev); | |
1703 | if (r) { | |
1704 | dev_err(adev->dev, "IB initialization failed (%d).\n", r); | |
83ba126a | 1705 | goto failed; |
d38ceaf9 AD |
1706 | } |
1707 | ||
1708 | r = amdgpu_ib_ring_tests(adev); | |
1709 | if (r) | |
1710 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
1711 | ||
1712 | r = amdgpu_gem_debugfs_init(adev); | |
1713 | if (r) { | |
1714 | DRM_ERROR("registering gem debugfs failed (%d).\n", r); | |
1715 | } | |
1716 | ||
1717 | r = amdgpu_debugfs_regs_init(adev); | |
1718 | if (r) { | |
1719 | DRM_ERROR("registering register debugfs failed (%d).\n", r); | |
1720 | } | |
1721 | ||
50ab2533 HR |
1722 | r = amdgpu_debugfs_firmware_init(adev); |
1723 | if (r) { | |
1724 | DRM_ERROR("registering firmware debugfs failed (%d).\n", r); | |
1725 | return r; | |
1726 | } | |
1727 | ||
d38ceaf9 AD |
1728 | if ((amdgpu_testing & 1)) { |
1729 | if (adev->accel_working) | |
1730 | amdgpu_test_moves(adev); | |
1731 | else | |
1732 | DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n"); | |
1733 | } | |
1734 | if ((amdgpu_testing & 2)) { | |
1735 | if (adev->accel_working) | |
1736 | amdgpu_test_syncing(adev); | |
1737 | else | |
1738 | DRM_INFO("amdgpu: acceleration disabled, skipping sync tests\n"); | |
1739 | } | |
1740 | if (amdgpu_benchmarking) { | |
1741 | if (adev->accel_working) | |
1742 | amdgpu_benchmark(adev, amdgpu_benchmarking); | |
1743 | else | |
1744 | DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n"); | |
1745 | } | |
1746 | ||
1747 | /* enable clockgating, etc. after ib tests, etc. since some blocks require | |
1748 | * explicit gating rather than handling it automatically. | |
1749 | */ | |
1750 | r = amdgpu_late_init(adev); | |
2c1a2784 AD |
1751 | if (r) { |
1752 | dev_err(adev->dev, "amdgpu_late_init failed\n"); | |
83ba126a | 1753 | goto failed; |
2c1a2784 | 1754 | } |
d38ceaf9 AD |
1755 | |
1756 | return 0; | |
83ba126a AD |
1757 | |
1758 | failed: | |
1759 | if (runtime) | |
1760 | vga_switcheroo_fini_domain_pm_ops(adev->dev); | |
1761 | return r; | |
d38ceaf9 AD |
1762 | } |
1763 | ||
1764 | static void amdgpu_debugfs_remove_files(struct amdgpu_device *adev); | |
1765 | ||
1766 | /** | |
1767 | * amdgpu_device_fini - tear down the driver | |
1768 | * | |
1769 | * @adev: amdgpu_device pointer | |
1770 | * | |
1771 | * Tear down the driver info (all asics). | |
1772 | * Called at driver shutdown. | |
1773 | */ | |
1774 | void amdgpu_device_fini(struct amdgpu_device *adev) | |
1775 | { | |
1776 | int r; | |
1777 | ||
1778 | DRM_INFO("amdgpu: finishing device.\n"); | |
1779 | adev->shutdown = true; | |
1780 | /* evict vram memory */ | |
1781 | amdgpu_bo_evict_vram(adev); | |
1782 | amdgpu_ib_pool_fini(adev); | |
1783 | amdgpu_fence_driver_fini(adev); | |
84b89bdc | 1784 | drm_crtc_force_disable_all(adev->ddev); |
d38ceaf9 AD |
1785 | amdgpu_fbdev_fini(adev); |
1786 | r = amdgpu_fini(adev); | |
8faf0e08 AD |
1787 | kfree(adev->ip_block_status); |
1788 | adev->ip_block_status = NULL; | |
d38ceaf9 AD |
1789 | adev->accel_working = false; |
1790 | /* free i2c buses */ | |
1791 | amdgpu_i2c_fini(adev); | |
1792 | amdgpu_atombios_fini(adev); | |
1793 | kfree(adev->bios); | |
1794 | adev->bios = NULL; | |
1795 | vga_switcheroo_unregister_client(adev->pdev); | |
83ba126a AD |
1796 | if (adev->flags & AMD_IS_PX) |
1797 | vga_switcheroo_fini_domain_pm_ops(adev->dev); | |
d38ceaf9 AD |
1798 | vga_client_register(adev->pdev, NULL, NULL, NULL); |
1799 | if (adev->rio_mem) | |
1800 | pci_iounmap(adev->pdev, adev->rio_mem); | |
1801 | adev->rio_mem = NULL; | |
1802 | iounmap(adev->rmmio); | |
1803 | adev->rmmio = NULL; | |
da69c161 KW |
1804 | if (adev->asic_type >= CHIP_BONAIRE) |
1805 | amdgpu_doorbell_fini(adev); | |
d38ceaf9 AD |
1806 | amdgpu_debugfs_regs_cleanup(adev); |
1807 | amdgpu_debugfs_remove_files(adev); | |
1808 | } | |
1809 | ||
1810 | ||
1811 | /* | |
1812 | * Suspend & resume. | |
1813 | */ | |
1814 | /** | |
810ddc3a | 1815 | * amdgpu_device_suspend - initiate device suspend |
d38ceaf9 AD |
1816 | * |
1817 | * @pdev: drm dev pointer | |
1818 | * @state: suspend state | |
1819 | * | |
1820 | * Puts the hw in the suspend state (all asics). | |
1821 | * Returns 0 for success or an error on failure. | |
1822 | * Called at driver suspend. | |
1823 | */ | |
810ddc3a | 1824 | int amdgpu_device_suspend(struct drm_device *dev, bool suspend, bool fbcon) |
d38ceaf9 AD |
1825 | { |
1826 | struct amdgpu_device *adev; | |
1827 | struct drm_crtc *crtc; | |
1828 | struct drm_connector *connector; | |
5ceb54c6 | 1829 | int r; |
d38ceaf9 AD |
1830 | |
1831 | if (dev == NULL || dev->dev_private == NULL) { | |
1832 | return -ENODEV; | |
1833 | } | |
1834 | ||
1835 | adev = dev->dev_private; | |
1836 | ||
1837 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
1838 | return 0; | |
1839 | ||
1840 | drm_kms_helper_poll_disable(dev); | |
1841 | ||
1842 | /* turn off display hw */ | |
4c7fbc39 | 1843 | drm_modeset_lock_all(dev); |
d38ceaf9 AD |
1844 | list_for_each_entry(connector, &dev->mode_config.connector_list, head) { |
1845 | drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF); | |
1846 | } | |
4c7fbc39 | 1847 | drm_modeset_unlock_all(dev); |
d38ceaf9 | 1848 | |
756e6880 | 1849 | /* unpin the front buffers and cursors */ |
d38ceaf9 | 1850 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { |
756e6880 | 1851 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); |
d38ceaf9 AD |
1852 | struct amdgpu_framebuffer *rfb = to_amdgpu_framebuffer(crtc->primary->fb); |
1853 | struct amdgpu_bo *robj; | |
1854 | ||
756e6880 AD |
1855 | if (amdgpu_crtc->cursor_bo) { |
1856 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); | |
1857 | r = amdgpu_bo_reserve(aobj, false); | |
1858 | if (r == 0) { | |
1859 | amdgpu_bo_unpin(aobj); | |
1860 | amdgpu_bo_unreserve(aobj); | |
1861 | } | |
1862 | } | |
1863 | ||
d38ceaf9 AD |
1864 | if (rfb == NULL || rfb->obj == NULL) { |
1865 | continue; | |
1866 | } | |
1867 | robj = gem_to_amdgpu_bo(rfb->obj); | |
1868 | /* don't unpin kernel fb objects */ | |
1869 | if (!amdgpu_fbdev_robj_is_fb(adev, robj)) { | |
1870 | r = amdgpu_bo_reserve(robj, false); | |
1871 | if (r == 0) { | |
1872 | amdgpu_bo_unpin(robj); | |
1873 | amdgpu_bo_unreserve(robj); | |
1874 | } | |
1875 | } | |
1876 | } | |
1877 | /* evict vram memory */ | |
1878 | amdgpu_bo_evict_vram(adev); | |
1879 | ||
5ceb54c6 | 1880 | amdgpu_fence_driver_suspend(adev); |
d38ceaf9 AD |
1881 | |
1882 | r = amdgpu_suspend(adev); | |
1883 | ||
1884 | /* evict remaining vram memory */ | |
1885 | amdgpu_bo_evict_vram(adev); | |
1886 | ||
1887 | pci_save_state(dev->pdev); | |
1888 | if (suspend) { | |
1889 | /* Shut down the device */ | |
1890 | pci_disable_device(dev->pdev); | |
1891 | pci_set_power_state(dev->pdev, PCI_D3hot); | |
1892 | } | |
1893 | ||
1894 | if (fbcon) { | |
1895 | console_lock(); | |
1896 | amdgpu_fbdev_set_suspend(adev, 1); | |
1897 | console_unlock(); | |
1898 | } | |
1899 | return 0; | |
1900 | } | |
1901 | ||
1902 | /** | |
810ddc3a | 1903 | * amdgpu_device_resume - initiate device resume |
d38ceaf9 AD |
1904 | * |
1905 | * @pdev: drm dev pointer | |
1906 | * | |
1907 | * Bring the hw back to operating state (all asics). | |
1908 | * Returns 0 for success or an error on failure. | |
1909 | * Called at driver resume. | |
1910 | */ | |
810ddc3a | 1911 | int amdgpu_device_resume(struct drm_device *dev, bool resume, bool fbcon) |
d38ceaf9 AD |
1912 | { |
1913 | struct drm_connector *connector; | |
1914 | struct amdgpu_device *adev = dev->dev_private; | |
756e6880 | 1915 | struct drm_crtc *crtc; |
d38ceaf9 AD |
1916 | int r; |
1917 | ||
1918 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
1919 | return 0; | |
1920 | ||
1921 | if (fbcon) { | |
1922 | console_lock(); | |
1923 | } | |
1924 | if (resume) { | |
1925 | pci_set_power_state(dev->pdev, PCI_D0); | |
1926 | pci_restore_state(dev->pdev); | |
1927 | if (pci_enable_device(dev->pdev)) { | |
1928 | if (fbcon) | |
1929 | console_unlock(); | |
1930 | return -1; | |
1931 | } | |
1932 | } | |
1933 | ||
1934 | /* post card */ | |
ca198528 FC |
1935 | if (!amdgpu_card_posted(adev)) |
1936 | amdgpu_atom_asic_init(adev->mode_info.atom_context); | |
d38ceaf9 AD |
1937 | |
1938 | r = amdgpu_resume(adev); | |
ca198528 FC |
1939 | if (r) |
1940 | DRM_ERROR("amdgpu_resume failed (%d).\n", r); | |
d38ceaf9 | 1941 | |
5ceb54c6 AD |
1942 | amdgpu_fence_driver_resume(adev); |
1943 | ||
ca198528 FC |
1944 | if (resume) { |
1945 | r = amdgpu_ib_ring_tests(adev); | |
1946 | if (r) | |
1947 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
1948 | } | |
d38ceaf9 AD |
1949 | |
1950 | r = amdgpu_late_init(adev); | |
1951 | if (r) | |
1952 | return r; | |
1953 | ||
756e6880 AD |
1954 | /* pin cursors */ |
1955 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
1956 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); | |
1957 | ||
1958 | if (amdgpu_crtc->cursor_bo) { | |
1959 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); | |
1960 | r = amdgpu_bo_reserve(aobj, false); | |
1961 | if (r == 0) { | |
1962 | r = amdgpu_bo_pin(aobj, | |
1963 | AMDGPU_GEM_DOMAIN_VRAM, | |
1964 | &amdgpu_crtc->cursor_addr); | |
1965 | if (r != 0) | |
1966 | DRM_ERROR("Failed to pin cursor BO (%d)\n", r); | |
1967 | amdgpu_bo_unreserve(aobj); | |
1968 | } | |
1969 | } | |
1970 | } | |
1971 | ||
d38ceaf9 AD |
1972 | /* blat the mode back in */ |
1973 | if (fbcon) { | |
1974 | drm_helper_resume_force_mode(dev); | |
1975 | /* turn on display hw */ | |
4c7fbc39 | 1976 | drm_modeset_lock_all(dev); |
d38ceaf9 AD |
1977 | list_for_each_entry(connector, &dev->mode_config.connector_list, head) { |
1978 | drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON); | |
1979 | } | |
4c7fbc39 | 1980 | drm_modeset_unlock_all(dev); |
d38ceaf9 AD |
1981 | } |
1982 | ||
1983 | drm_kms_helper_poll_enable(dev); | |
23a1a9e5 L |
1984 | |
1985 | /* | |
1986 | * Most of the connector probing functions try to acquire runtime pm | |
1987 | * refs to ensure that the GPU is powered on when connector polling is | |
1988 | * performed. Since we're calling this from a runtime PM callback, | |
1989 | * trying to acquire rpm refs will cause us to deadlock. | |
1990 | * | |
1991 | * Since we're guaranteed to be holding the rpm lock, it's safe to | |
1992 | * temporarily disable the rpm helpers so this doesn't deadlock us. | |
1993 | */ | |
1994 | #ifdef CONFIG_PM | |
1995 | dev->dev->power.disable_depth++; | |
1996 | #endif | |
54fb2a5c | 1997 | drm_helper_hpd_irq_event(dev); |
23a1a9e5 L |
1998 | #ifdef CONFIG_PM |
1999 | dev->dev->power.disable_depth--; | |
2000 | #endif | |
d38ceaf9 AD |
2001 | |
2002 | if (fbcon) { | |
2003 | amdgpu_fbdev_set_suspend(adev, 0); | |
2004 | console_unlock(); | |
2005 | } | |
2006 | ||
2007 | return 0; | |
2008 | } | |
2009 | ||
63fbf42f CZ |
2010 | static bool amdgpu_check_soft_reset(struct amdgpu_device *adev) |
2011 | { | |
2012 | int i; | |
2013 | bool asic_hang = false; | |
2014 | ||
2015 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
2016 | if (!adev->ip_block_status[i].valid) | |
2017 | continue; | |
2018 | if (adev->ip_blocks[i].funcs->check_soft_reset) | |
2019 | adev->ip_blocks[i].funcs->check_soft_reset(adev); | |
2020 | if (adev->ip_block_status[i].hang) { | |
2021 | DRM_INFO("IP block:%d is hang!\n", i); | |
2022 | asic_hang = true; | |
2023 | } | |
2024 | } | |
2025 | return asic_hang; | |
2026 | } | |
2027 | ||
d31a501e CZ |
2028 | int amdgpu_pre_soft_reset(struct amdgpu_device *adev) |
2029 | { | |
2030 | int i, r = 0; | |
2031 | ||
2032 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
2033 | if (!adev->ip_block_status[i].valid) | |
2034 | continue; | |
35d782fe CZ |
2035 | if (adev->ip_block_status[i].hang && |
2036 | adev->ip_blocks[i].funcs->pre_soft_reset) { | |
d31a501e CZ |
2037 | r = adev->ip_blocks[i].funcs->pre_soft_reset(adev); |
2038 | if (r) | |
2039 | return r; | |
2040 | } | |
2041 | } | |
2042 | ||
2043 | return 0; | |
2044 | } | |
2045 | ||
35d782fe CZ |
2046 | static bool amdgpu_need_full_reset(struct amdgpu_device *adev) |
2047 | { | |
2048 | if (adev->ip_block_status[AMD_IP_BLOCK_TYPE_GMC].hang || | |
35d782fe | 2049 | adev->ip_block_status[AMD_IP_BLOCK_TYPE_SMC].hang || |
35d782fe CZ |
2050 | adev->ip_block_status[AMD_IP_BLOCK_TYPE_ACP].hang || |
2051 | adev->ip_block_status[AMD_IP_BLOCK_TYPE_DCE].hang) { | |
2052 | DRM_INFO("Some block need full reset!\n"); | |
2053 | return true; | |
2054 | } | |
2055 | return false; | |
2056 | } | |
2057 | ||
2058 | static int amdgpu_soft_reset(struct amdgpu_device *adev) | |
2059 | { | |
2060 | int i, r = 0; | |
2061 | ||
2062 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
2063 | if (!adev->ip_block_status[i].valid) | |
2064 | continue; | |
2065 | if (adev->ip_block_status[i].hang && | |
2066 | adev->ip_blocks[i].funcs->soft_reset) { | |
2067 | r = adev->ip_blocks[i].funcs->soft_reset(adev); | |
2068 | if (r) | |
2069 | return r; | |
2070 | } | |
2071 | } | |
2072 | ||
2073 | return 0; | |
2074 | } | |
2075 | ||
2076 | static int amdgpu_post_soft_reset(struct amdgpu_device *adev) | |
2077 | { | |
2078 | int i, r = 0; | |
2079 | ||
2080 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
2081 | if (!adev->ip_block_status[i].valid) | |
2082 | continue; | |
2083 | if (adev->ip_block_status[i].hang && | |
2084 | adev->ip_blocks[i].funcs->post_soft_reset) | |
2085 | r = adev->ip_blocks[i].funcs->post_soft_reset(adev); | |
2086 | if (r) | |
2087 | return r; | |
2088 | } | |
2089 | ||
2090 | return 0; | |
2091 | } | |
2092 | ||
3ad81f16 CZ |
2093 | bool amdgpu_need_backup(struct amdgpu_device *adev) |
2094 | { | |
2095 | if (adev->flags & AMD_IS_APU) | |
2096 | return false; | |
2097 | ||
2098 | return amdgpu_lockup_timeout > 0 ? true : false; | |
2099 | } | |
2100 | ||
53cdccd5 CZ |
2101 | static int amdgpu_recover_vram_from_shadow(struct amdgpu_device *adev, |
2102 | struct amdgpu_ring *ring, | |
2103 | struct amdgpu_bo *bo, | |
2104 | struct fence **fence) | |
2105 | { | |
2106 | uint32_t domain; | |
2107 | int r; | |
2108 | ||
2109 | if (!bo->shadow) | |
2110 | return 0; | |
2111 | ||
2112 | r = amdgpu_bo_reserve(bo, false); | |
2113 | if (r) | |
2114 | return r; | |
2115 | domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type); | |
2116 | /* if bo has been evicted, then no need to recover */ | |
2117 | if (domain == AMDGPU_GEM_DOMAIN_VRAM) { | |
2118 | r = amdgpu_bo_restore_from_shadow(adev, ring, bo, | |
2119 | NULL, fence, true); | |
2120 | if (r) { | |
2121 | DRM_ERROR("recover page table failed!\n"); | |
2122 | goto err; | |
2123 | } | |
2124 | } | |
2125 | err: | |
2126 | amdgpu_bo_unreserve(bo); | |
2127 | return r; | |
2128 | } | |
2129 | ||
d38ceaf9 AD |
2130 | /** |
2131 | * amdgpu_gpu_reset - reset the asic | |
2132 | * | |
2133 | * @adev: amdgpu device pointer | |
2134 | * | |
2135 | * Attempt the reset the GPU if it has hung (all asics). | |
2136 | * Returns 0 for success or an error on failure. | |
2137 | */ | |
2138 | int amdgpu_gpu_reset(struct amdgpu_device *adev) | |
2139 | { | |
d38ceaf9 AD |
2140 | int i, r; |
2141 | int resched; | |
35d782fe | 2142 | bool need_full_reset; |
d38ceaf9 | 2143 | |
63fbf42f CZ |
2144 | if (!amdgpu_check_soft_reset(adev)) { |
2145 | DRM_INFO("No hardware hang detected. Did some blocks stall?\n"); | |
2146 | return 0; | |
2147 | } | |
2148 | ||
d94aed5a | 2149 | atomic_inc(&adev->gpu_reset_counter); |
d38ceaf9 | 2150 | |
a3c47d6b CZ |
2151 | /* block TTM */ |
2152 | resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev); | |
2153 | ||
0875dc9e CZ |
2154 | /* block scheduler */ |
2155 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
2156 | struct amdgpu_ring *ring = adev->rings[i]; | |
2157 | ||
2158 | if (!ring) | |
2159 | continue; | |
2160 | kthread_park(ring->sched.thread); | |
aa1c8900 | 2161 | amd_sched_hw_job_reset(&ring->sched); |
0875dc9e | 2162 | } |
2200edac CZ |
2163 | /* after all hw jobs are reset, hw fence is meaningless, so force_completion */ |
2164 | amdgpu_fence_driver_force_completion(adev); | |
d38ceaf9 | 2165 | |
35d782fe | 2166 | need_full_reset = amdgpu_need_full_reset(adev); |
d38ceaf9 | 2167 | |
35d782fe CZ |
2168 | if (!need_full_reset) { |
2169 | amdgpu_pre_soft_reset(adev); | |
2170 | r = amdgpu_soft_reset(adev); | |
2171 | amdgpu_post_soft_reset(adev); | |
2172 | if (r || amdgpu_check_soft_reset(adev)) { | |
2173 | DRM_INFO("soft reset failed, will fallback to full reset!\n"); | |
2174 | need_full_reset = true; | |
2175 | } | |
f1aa7e08 CZ |
2176 | } |
2177 | ||
35d782fe CZ |
2178 | if (need_full_reset) { |
2179 | /* save scratch */ | |
2180 | amdgpu_atombios_scratch_regs_save(adev); | |
2181 | r = amdgpu_suspend(adev); | |
bfa99269 | 2182 | |
35d782fe CZ |
2183 | retry: |
2184 | /* Disable fb access */ | |
2185 | if (adev->mode_info.num_crtc) { | |
2186 | struct amdgpu_mode_mc_save save; | |
2187 | amdgpu_display_stop_mc_access(adev, &save); | |
2188 | amdgpu_wait_for_idle(adev, AMD_IP_BLOCK_TYPE_GMC); | |
2189 | } | |
2190 | ||
2191 | r = amdgpu_asic_reset(adev); | |
2192 | /* post card */ | |
2193 | amdgpu_atom_asic_init(adev->mode_info.atom_context); | |
2194 | ||
2195 | if (!r) { | |
2196 | dev_info(adev->dev, "GPU reset succeeded, trying to resume\n"); | |
2197 | r = amdgpu_resume(adev); | |
2198 | } | |
2199 | /* restore scratch */ | |
2200 | amdgpu_atombios_scratch_regs_restore(adev); | |
d38ceaf9 | 2201 | } |
d38ceaf9 | 2202 | if (!r) { |
e72cfd58 | 2203 | amdgpu_irq_gpu_reset_resume_helper(adev); |
1f465087 CZ |
2204 | r = amdgpu_ib_ring_tests(adev); |
2205 | if (r) { | |
2206 | dev_err(adev->dev, "ib ring test failed (%d).\n", r); | |
40019dc4 | 2207 | r = amdgpu_suspend(adev); |
53cdccd5 | 2208 | need_full_reset = true; |
40019dc4 | 2209 | goto retry; |
1f465087 | 2210 | } |
53cdccd5 CZ |
2211 | /** |
2212 | * recovery vm page tables, since we cannot depend on VRAM is | |
2213 | * consistent after gpu full reset. | |
2214 | */ | |
2215 | if (need_full_reset && amdgpu_need_backup(adev)) { | |
2216 | struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; | |
2217 | struct amdgpu_bo *bo, *tmp; | |
2218 | struct fence *fence = NULL, *next = NULL; | |
2219 | ||
2220 | DRM_INFO("recover vram bo from shadow\n"); | |
2221 | mutex_lock(&adev->shadow_list_lock); | |
2222 | list_for_each_entry_safe(bo, tmp, &adev->shadow_list, shadow_list) { | |
2223 | amdgpu_recover_vram_from_shadow(adev, ring, bo, &next); | |
2224 | if (fence) { | |
2225 | r = fence_wait(fence, false); | |
2226 | if (r) { | |
2227 | WARN(r, "recovery from shadow isn't comleted\n"); | |
2228 | break; | |
2229 | } | |
2230 | } | |
1f465087 | 2231 | |
53cdccd5 CZ |
2232 | fence_put(fence); |
2233 | fence = next; | |
2234 | } | |
2235 | mutex_unlock(&adev->shadow_list_lock); | |
2236 | if (fence) { | |
2237 | r = fence_wait(fence, false); | |
2238 | if (r) | |
2239 | WARN(r, "recovery from shadow isn't comleted\n"); | |
2240 | } | |
2241 | fence_put(fence); | |
2242 | } | |
d38ceaf9 AD |
2243 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
2244 | struct amdgpu_ring *ring = adev->rings[i]; | |
2245 | if (!ring) | |
2246 | continue; | |
53cdccd5 | 2247 | |
aa1c8900 | 2248 | amd_sched_job_recovery(&ring->sched); |
0875dc9e | 2249 | kthread_unpark(ring->sched.thread); |
d38ceaf9 | 2250 | } |
d38ceaf9 | 2251 | } else { |
2200edac | 2252 | dev_err(adev->dev, "asic resume failed (%d).\n", r); |
d38ceaf9 | 2253 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { |
0875dc9e CZ |
2254 | if (adev->rings[i]) { |
2255 | kthread_unpark(adev->rings[i]->sched.thread); | |
0875dc9e | 2256 | } |
d38ceaf9 AD |
2257 | } |
2258 | } | |
2259 | ||
2260 | drm_helper_resume_force_mode(adev->ddev); | |
2261 | ||
2262 | ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched); | |
2263 | if (r) { | |
2264 | /* bad news, how to tell it to userspace ? */ | |
2265 | dev_info(adev->dev, "GPU reset failed\n"); | |
2266 | } | |
2267 | ||
d38ceaf9 AD |
2268 | return r; |
2269 | } | |
2270 | ||
d0dd7f0c AD |
2271 | void amdgpu_get_pcie_info(struct amdgpu_device *adev) |
2272 | { | |
2273 | u32 mask; | |
2274 | int ret; | |
2275 | ||
cd474ba0 AD |
2276 | if (amdgpu_pcie_gen_cap) |
2277 | adev->pm.pcie_gen_mask = amdgpu_pcie_gen_cap; | |
d0dd7f0c | 2278 | |
cd474ba0 AD |
2279 | if (amdgpu_pcie_lane_cap) |
2280 | adev->pm.pcie_mlw_mask = amdgpu_pcie_lane_cap; | |
d0dd7f0c | 2281 | |
cd474ba0 AD |
2282 | /* covers APUs as well */ |
2283 | if (pci_is_root_bus(adev->pdev->bus)) { | |
2284 | if (adev->pm.pcie_gen_mask == 0) | |
2285 | adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; | |
2286 | if (adev->pm.pcie_mlw_mask == 0) | |
2287 | adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; | |
d0dd7f0c | 2288 | return; |
cd474ba0 | 2289 | } |
d0dd7f0c | 2290 | |
cd474ba0 AD |
2291 | if (adev->pm.pcie_gen_mask == 0) { |
2292 | ret = drm_pcie_get_speed_cap_mask(adev->ddev, &mask); | |
2293 | if (!ret) { | |
2294 | adev->pm.pcie_gen_mask = (CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN1 | | |
2295 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN2 | | |
2296 | CAIL_ASIC_PCIE_LINK_SPEED_SUPPORT_GEN3); | |
2297 | ||
2298 | if (mask & DRM_PCIE_SPEED_25) | |
2299 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN1; | |
2300 | if (mask & DRM_PCIE_SPEED_50) | |
2301 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN2; | |
2302 | if (mask & DRM_PCIE_SPEED_80) | |
2303 | adev->pm.pcie_gen_mask |= CAIL_PCIE_LINK_SPEED_SUPPORT_GEN3; | |
2304 | } else { | |
2305 | adev->pm.pcie_gen_mask = AMDGPU_DEFAULT_PCIE_GEN_MASK; | |
2306 | } | |
2307 | } | |
2308 | if (adev->pm.pcie_mlw_mask == 0) { | |
2309 | ret = drm_pcie_get_max_link_width(adev->ddev, &mask); | |
2310 | if (!ret) { | |
2311 | switch (mask) { | |
2312 | case 32: | |
2313 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X32 | | |
2314 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | | |
2315 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
2316 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2317 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2318 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2319 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2320 | break; | |
2321 | case 16: | |
2322 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X16 | | |
2323 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
2324 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2325 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2326 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2327 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2328 | break; | |
2329 | case 12: | |
2330 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X12 | | |
2331 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2332 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2333 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2334 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2335 | break; | |
2336 | case 8: | |
2337 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X8 | | |
2338 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2339 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2340 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2341 | break; | |
2342 | case 4: | |
2343 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X4 | | |
2344 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2345 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2346 | break; | |
2347 | case 2: | |
2348 | adev->pm.pcie_mlw_mask = (CAIL_PCIE_LINK_WIDTH_SUPPORT_X2 | | |
2349 | CAIL_PCIE_LINK_WIDTH_SUPPORT_X1); | |
2350 | break; | |
2351 | case 1: | |
2352 | adev->pm.pcie_mlw_mask = CAIL_PCIE_LINK_WIDTH_SUPPORT_X1; | |
2353 | break; | |
2354 | default: | |
2355 | break; | |
2356 | } | |
2357 | } else { | |
2358 | adev->pm.pcie_mlw_mask = AMDGPU_DEFAULT_PCIE_MLW_MASK; | |
d0dd7f0c AD |
2359 | } |
2360 | } | |
2361 | } | |
d38ceaf9 AD |
2362 | |
2363 | /* | |
2364 | * Debugfs | |
2365 | */ | |
2366 | int amdgpu_debugfs_add_files(struct amdgpu_device *adev, | |
06ab6832 | 2367 | const struct drm_info_list *files, |
d38ceaf9 AD |
2368 | unsigned nfiles) |
2369 | { | |
2370 | unsigned i; | |
2371 | ||
2372 | for (i = 0; i < adev->debugfs_count; i++) { | |
2373 | if (adev->debugfs[i].files == files) { | |
2374 | /* Already registered */ | |
2375 | return 0; | |
2376 | } | |
2377 | } | |
2378 | ||
2379 | i = adev->debugfs_count + 1; | |
2380 | if (i > AMDGPU_DEBUGFS_MAX_COMPONENTS) { | |
2381 | DRM_ERROR("Reached maximum number of debugfs components.\n"); | |
2382 | DRM_ERROR("Report so we increase " | |
2383 | "AMDGPU_DEBUGFS_MAX_COMPONENTS.\n"); | |
2384 | return -EINVAL; | |
2385 | } | |
2386 | adev->debugfs[adev->debugfs_count].files = files; | |
2387 | adev->debugfs[adev->debugfs_count].num_files = nfiles; | |
2388 | adev->debugfs_count = i; | |
2389 | #if defined(CONFIG_DEBUG_FS) | |
2390 | drm_debugfs_create_files(files, nfiles, | |
2391 | adev->ddev->control->debugfs_root, | |
2392 | adev->ddev->control); | |
2393 | drm_debugfs_create_files(files, nfiles, | |
2394 | adev->ddev->primary->debugfs_root, | |
2395 | adev->ddev->primary); | |
2396 | #endif | |
2397 | return 0; | |
2398 | } | |
2399 | ||
2400 | static void amdgpu_debugfs_remove_files(struct amdgpu_device *adev) | |
2401 | { | |
2402 | #if defined(CONFIG_DEBUG_FS) | |
2403 | unsigned i; | |
2404 | ||
2405 | for (i = 0; i < adev->debugfs_count; i++) { | |
2406 | drm_debugfs_remove_files(adev->debugfs[i].files, | |
2407 | adev->debugfs[i].num_files, | |
2408 | adev->ddev->control); | |
2409 | drm_debugfs_remove_files(adev->debugfs[i].files, | |
2410 | adev->debugfs[i].num_files, | |
2411 | adev->ddev->primary); | |
2412 | } | |
2413 | #endif | |
2414 | } | |
2415 | ||
2416 | #if defined(CONFIG_DEBUG_FS) | |
2417 | ||
2418 | static ssize_t amdgpu_debugfs_regs_read(struct file *f, char __user *buf, | |
2419 | size_t size, loff_t *pos) | |
2420 | { | |
2421 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2422 | ssize_t result = 0; | |
2423 | int r; | |
bd12267d | 2424 | bool pm_pg_lock, use_bank; |
56628159 | 2425 | unsigned instance_bank, sh_bank, se_bank; |
d38ceaf9 AD |
2426 | |
2427 | if (size & 0x3 || *pos & 0x3) | |
2428 | return -EINVAL; | |
2429 | ||
bd12267d TSD |
2430 | /* are we reading registers for which a PG lock is necessary? */ |
2431 | pm_pg_lock = (*pos >> 23) & 1; | |
2432 | ||
56628159 TSD |
2433 | if (*pos & (1ULL << 62)) { |
2434 | se_bank = (*pos >> 24) & 0x3FF; | |
2435 | sh_bank = (*pos >> 34) & 0x3FF; | |
2436 | instance_bank = (*pos >> 44) & 0x3FF; | |
2437 | use_bank = 1; | |
56628159 TSD |
2438 | } else { |
2439 | use_bank = 0; | |
2440 | } | |
2441 | ||
bd12267d TSD |
2442 | *pos &= 0x3FFFF; |
2443 | ||
56628159 TSD |
2444 | if (use_bank) { |
2445 | if (sh_bank >= adev->gfx.config.max_sh_per_se || | |
2446 | se_bank >= adev->gfx.config.max_shader_engines) | |
2447 | return -EINVAL; | |
2448 | mutex_lock(&adev->grbm_idx_mutex); | |
2449 | amdgpu_gfx_select_se_sh(adev, se_bank, | |
2450 | sh_bank, instance_bank); | |
2451 | } | |
2452 | ||
bd12267d TSD |
2453 | if (pm_pg_lock) |
2454 | mutex_lock(&adev->pm.mutex); | |
2455 | ||
d38ceaf9 AD |
2456 | while (size) { |
2457 | uint32_t value; | |
2458 | ||
2459 | if (*pos > adev->rmmio_size) | |
56628159 | 2460 | goto end; |
d38ceaf9 AD |
2461 | |
2462 | value = RREG32(*pos >> 2); | |
2463 | r = put_user(value, (uint32_t *)buf); | |
56628159 TSD |
2464 | if (r) { |
2465 | result = r; | |
2466 | goto end; | |
2467 | } | |
d38ceaf9 AD |
2468 | |
2469 | result += 4; | |
2470 | buf += 4; | |
2471 | *pos += 4; | |
2472 | size -= 4; | |
2473 | } | |
2474 | ||
56628159 TSD |
2475 | end: |
2476 | if (use_bank) { | |
2477 | amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); | |
2478 | mutex_unlock(&adev->grbm_idx_mutex); | |
2479 | } | |
2480 | ||
bd12267d TSD |
2481 | if (pm_pg_lock) |
2482 | mutex_unlock(&adev->pm.mutex); | |
2483 | ||
d38ceaf9 AD |
2484 | return result; |
2485 | } | |
2486 | ||
2487 | static ssize_t amdgpu_debugfs_regs_write(struct file *f, const char __user *buf, | |
2488 | size_t size, loff_t *pos) | |
2489 | { | |
2490 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2491 | ssize_t result = 0; | |
2492 | int r; | |
2493 | ||
2494 | if (size & 0x3 || *pos & 0x3) | |
2495 | return -EINVAL; | |
2496 | ||
2497 | while (size) { | |
2498 | uint32_t value; | |
2499 | ||
2500 | if (*pos > adev->rmmio_size) | |
2501 | return result; | |
2502 | ||
2503 | r = get_user(value, (uint32_t *)buf); | |
2504 | if (r) | |
2505 | return r; | |
2506 | ||
2507 | WREG32(*pos >> 2, value); | |
2508 | ||
2509 | result += 4; | |
2510 | buf += 4; | |
2511 | *pos += 4; | |
2512 | size -= 4; | |
2513 | } | |
2514 | ||
2515 | return result; | |
2516 | } | |
2517 | ||
adcec288 TSD |
2518 | static ssize_t amdgpu_debugfs_regs_pcie_read(struct file *f, char __user *buf, |
2519 | size_t size, loff_t *pos) | |
2520 | { | |
2521 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2522 | ssize_t result = 0; | |
2523 | int r; | |
2524 | ||
2525 | if (size & 0x3 || *pos & 0x3) | |
2526 | return -EINVAL; | |
2527 | ||
2528 | while (size) { | |
2529 | uint32_t value; | |
2530 | ||
2531 | value = RREG32_PCIE(*pos >> 2); | |
2532 | r = put_user(value, (uint32_t *)buf); | |
2533 | if (r) | |
2534 | return r; | |
2535 | ||
2536 | result += 4; | |
2537 | buf += 4; | |
2538 | *pos += 4; | |
2539 | size -= 4; | |
2540 | } | |
2541 | ||
2542 | return result; | |
2543 | } | |
2544 | ||
2545 | static ssize_t amdgpu_debugfs_regs_pcie_write(struct file *f, const char __user *buf, | |
2546 | size_t size, loff_t *pos) | |
2547 | { | |
2548 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2549 | ssize_t result = 0; | |
2550 | int r; | |
2551 | ||
2552 | if (size & 0x3 || *pos & 0x3) | |
2553 | return -EINVAL; | |
2554 | ||
2555 | while (size) { | |
2556 | uint32_t value; | |
2557 | ||
2558 | r = get_user(value, (uint32_t *)buf); | |
2559 | if (r) | |
2560 | return r; | |
2561 | ||
2562 | WREG32_PCIE(*pos >> 2, value); | |
2563 | ||
2564 | result += 4; | |
2565 | buf += 4; | |
2566 | *pos += 4; | |
2567 | size -= 4; | |
2568 | } | |
2569 | ||
2570 | return result; | |
2571 | } | |
2572 | ||
2573 | static ssize_t amdgpu_debugfs_regs_didt_read(struct file *f, char __user *buf, | |
2574 | size_t size, loff_t *pos) | |
2575 | { | |
2576 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2577 | ssize_t result = 0; | |
2578 | int r; | |
2579 | ||
2580 | if (size & 0x3 || *pos & 0x3) | |
2581 | return -EINVAL; | |
2582 | ||
2583 | while (size) { | |
2584 | uint32_t value; | |
2585 | ||
2586 | value = RREG32_DIDT(*pos >> 2); | |
2587 | r = put_user(value, (uint32_t *)buf); | |
2588 | if (r) | |
2589 | return r; | |
2590 | ||
2591 | result += 4; | |
2592 | buf += 4; | |
2593 | *pos += 4; | |
2594 | size -= 4; | |
2595 | } | |
2596 | ||
2597 | return result; | |
2598 | } | |
2599 | ||
2600 | static ssize_t amdgpu_debugfs_regs_didt_write(struct file *f, const char __user *buf, | |
2601 | size_t size, loff_t *pos) | |
2602 | { | |
2603 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2604 | ssize_t result = 0; | |
2605 | int r; | |
2606 | ||
2607 | if (size & 0x3 || *pos & 0x3) | |
2608 | return -EINVAL; | |
2609 | ||
2610 | while (size) { | |
2611 | uint32_t value; | |
2612 | ||
2613 | r = get_user(value, (uint32_t *)buf); | |
2614 | if (r) | |
2615 | return r; | |
2616 | ||
2617 | WREG32_DIDT(*pos >> 2, value); | |
2618 | ||
2619 | result += 4; | |
2620 | buf += 4; | |
2621 | *pos += 4; | |
2622 | size -= 4; | |
2623 | } | |
2624 | ||
2625 | return result; | |
2626 | } | |
2627 | ||
2628 | static ssize_t amdgpu_debugfs_regs_smc_read(struct file *f, char __user *buf, | |
2629 | size_t size, loff_t *pos) | |
2630 | { | |
2631 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2632 | ssize_t result = 0; | |
2633 | int r; | |
2634 | ||
2635 | if (size & 0x3 || *pos & 0x3) | |
2636 | return -EINVAL; | |
2637 | ||
2638 | while (size) { | |
2639 | uint32_t value; | |
2640 | ||
6fc0deaf | 2641 | value = RREG32_SMC(*pos); |
adcec288 TSD |
2642 | r = put_user(value, (uint32_t *)buf); |
2643 | if (r) | |
2644 | return r; | |
2645 | ||
2646 | result += 4; | |
2647 | buf += 4; | |
2648 | *pos += 4; | |
2649 | size -= 4; | |
2650 | } | |
2651 | ||
2652 | return result; | |
2653 | } | |
2654 | ||
2655 | static ssize_t amdgpu_debugfs_regs_smc_write(struct file *f, const char __user *buf, | |
2656 | size_t size, loff_t *pos) | |
2657 | { | |
2658 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2659 | ssize_t result = 0; | |
2660 | int r; | |
2661 | ||
2662 | if (size & 0x3 || *pos & 0x3) | |
2663 | return -EINVAL; | |
2664 | ||
2665 | while (size) { | |
2666 | uint32_t value; | |
2667 | ||
2668 | r = get_user(value, (uint32_t *)buf); | |
2669 | if (r) | |
2670 | return r; | |
2671 | ||
6fc0deaf | 2672 | WREG32_SMC(*pos, value); |
adcec288 TSD |
2673 | |
2674 | result += 4; | |
2675 | buf += 4; | |
2676 | *pos += 4; | |
2677 | size -= 4; | |
2678 | } | |
2679 | ||
2680 | return result; | |
2681 | } | |
2682 | ||
1e051413 TSD |
2683 | static ssize_t amdgpu_debugfs_gca_config_read(struct file *f, char __user *buf, |
2684 | size_t size, loff_t *pos) | |
2685 | { | |
2686 | struct amdgpu_device *adev = f->f_inode->i_private; | |
2687 | ssize_t result = 0; | |
2688 | int r; | |
2689 | uint32_t *config, no_regs = 0; | |
2690 | ||
2691 | if (size & 0x3 || *pos & 0x3) | |
2692 | return -EINVAL; | |
2693 | ||
2694 | config = kmalloc(256 * sizeof(*config), GFP_KERNEL); | |
2695 | if (!config) | |
2696 | return -ENOMEM; | |
2697 | ||
2698 | /* version, increment each time something is added */ | |
e9f11dc8 | 2699 | config[no_regs++] = 2; |
1e051413 TSD |
2700 | config[no_regs++] = adev->gfx.config.max_shader_engines; |
2701 | config[no_regs++] = adev->gfx.config.max_tile_pipes; | |
2702 | config[no_regs++] = adev->gfx.config.max_cu_per_sh; | |
2703 | config[no_regs++] = adev->gfx.config.max_sh_per_se; | |
2704 | config[no_regs++] = adev->gfx.config.max_backends_per_se; | |
2705 | config[no_regs++] = adev->gfx.config.max_texture_channel_caches; | |
2706 | config[no_regs++] = adev->gfx.config.max_gprs; | |
2707 | config[no_regs++] = adev->gfx.config.max_gs_threads; | |
2708 | config[no_regs++] = adev->gfx.config.max_hw_contexts; | |
2709 | config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_frontend; | |
2710 | config[no_regs++] = adev->gfx.config.sc_prim_fifo_size_backend; | |
2711 | config[no_regs++] = adev->gfx.config.sc_hiz_tile_fifo_size; | |
2712 | config[no_regs++] = adev->gfx.config.sc_earlyz_tile_fifo_size; | |
2713 | config[no_regs++] = adev->gfx.config.num_tile_pipes; | |
2714 | config[no_regs++] = adev->gfx.config.backend_enable_mask; | |
2715 | config[no_regs++] = adev->gfx.config.mem_max_burst_length_bytes; | |
2716 | config[no_regs++] = adev->gfx.config.mem_row_size_in_kb; | |
2717 | config[no_regs++] = adev->gfx.config.shader_engine_tile_size; | |
2718 | config[no_regs++] = adev->gfx.config.num_gpus; | |
2719 | config[no_regs++] = adev->gfx.config.multi_gpu_tile_size; | |
2720 | config[no_regs++] = adev->gfx.config.mc_arb_ramcfg; | |
2721 | config[no_regs++] = adev->gfx.config.gb_addr_config; | |
2722 | config[no_regs++] = adev->gfx.config.num_rbs; | |
2723 | ||
89a8f309 TSD |
2724 | /* rev==1 */ |
2725 | config[no_regs++] = adev->rev_id; | |
2726 | config[no_regs++] = adev->pg_flags; | |
2727 | config[no_regs++] = adev->cg_flags; | |
2728 | ||
e9f11dc8 TSD |
2729 | /* rev==2 */ |
2730 | config[no_regs++] = adev->family; | |
2731 | config[no_regs++] = adev->external_rev_id; | |
2732 | ||
1e051413 TSD |
2733 | while (size && (*pos < no_regs * 4)) { |
2734 | uint32_t value; | |
2735 | ||
2736 | value = config[*pos >> 2]; | |
2737 | r = put_user(value, (uint32_t *)buf); | |
2738 | if (r) { | |
2739 | kfree(config); | |
2740 | return r; | |
2741 | } | |
2742 | ||
2743 | result += 4; | |
2744 | buf += 4; | |
2745 | *pos += 4; | |
2746 | size -= 4; | |
2747 | } | |
2748 | ||
2749 | kfree(config); | |
2750 | return result; | |
2751 | } | |
2752 | ||
2753 | ||
d38ceaf9 AD |
2754 | static const struct file_operations amdgpu_debugfs_regs_fops = { |
2755 | .owner = THIS_MODULE, | |
2756 | .read = amdgpu_debugfs_regs_read, | |
2757 | .write = amdgpu_debugfs_regs_write, | |
2758 | .llseek = default_llseek | |
2759 | }; | |
adcec288 TSD |
2760 | static const struct file_operations amdgpu_debugfs_regs_didt_fops = { |
2761 | .owner = THIS_MODULE, | |
2762 | .read = amdgpu_debugfs_regs_didt_read, | |
2763 | .write = amdgpu_debugfs_regs_didt_write, | |
2764 | .llseek = default_llseek | |
2765 | }; | |
2766 | static const struct file_operations amdgpu_debugfs_regs_pcie_fops = { | |
2767 | .owner = THIS_MODULE, | |
2768 | .read = amdgpu_debugfs_regs_pcie_read, | |
2769 | .write = amdgpu_debugfs_regs_pcie_write, | |
2770 | .llseek = default_llseek | |
2771 | }; | |
2772 | static const struct file_operations amdgpu_debugfs_regs_smc_fops = { | |
2773 | .owner = THIS_MODULE, | |
2774 | .read = amdgpu_debugfs_regs_smc_read, | |
2775 | .write = amdgpu_debugfs_regs_smc_write, | |
2776 | .llseek = default_llseek | |
2777 | }; | |
2778 | ||
1e051413 TSD |
2779 | static const struct file_operations amdgpu_debugfs_gca_config_fops = { |
2780 | .owner = THIS_MODULE, | |
2781 | .read = amdgpu_debugfs_gca_config_read, | |
2782 | .llseek = default_llseek | |
2783 | }; | |
2784 | ||
adcec288 TSD |
2785 | static const struct file_operations *debugfs_regs[] = { |
2786 | &amdgpu_debugfs_regs_fops, | |
2787 | &amdgpu_debugfs_regs_didt_fops, | |
2788 | &amdgpu_debugfs_regs_pcie_fops, | |
2789 | &amdgpu_debugfs_regs_smc_fops, | |
1e051413 | 2790 | &amdgpu_debugfs_gca_config_fops, |
adcec288 TSD |
2791 | }; |
2792 | ||
2793 | static const char *debugfs_regs_names[] = { | |
2794 | "amdgpu_regs", | |
2795 | "amdgpu_regs_didt", | |
2796 | "amdgpu_regs_pcie", | |
2797 | "amdgpu_regs_smc", | |
1e051413 | 2798 | "amdgpu_gca_config", |
adcec288 | 2799 | }; |
d38ceaf9 AD |
2800 | |
2801 | static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev) | |
2802 | { | |
2803 | struct drm_minor *minor = adev->ddev->primary; | |
2804 | struct dentry *ent, *root = minor->debugfs_root; | |
adcec288 TSD |
2805 | unsigned i, j; |
2806 | ||
2807 | for (i = 0; i < ARRAY_SIZE(debugfs_regs); i++) { | |
2808 | ent = debugfs_create_file(debugfs_regs_names[i], | |
2809 | S_IFREG | S_IRUGO, root, | |
2810 | adev, debugfs_regs[i]); | |
2811 | if (IS_ERR(ent)) { | |
2812 | for (j = 0; j < i; j++) { | |
2813 | debugfs_remove(adev->debugfs_regs[i]); | |
2814 | adev->debugfs_regs[i] = NULL; | |
2815 | } | |
2816 | return PTR_ERR(ent); | |
2817 | } | |
d38ceaf9 | 2818 | |
adcec288 TSD |
2819 | if (!i) |
2820 | i_size_write(ent->d_inode, adev->rmmio_size); | |
2821 | adev->debugfs_regs[i] = ent; | |
2822 | } | |
d38ceaf9 AD |
2823 | |
2824 | return 0; | |
2825 | } | |
2826 | ||
2827 | static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev) | |
2828 | { | |
adcec288 TSD |
2829 | unsigned i; |
2830 | ||
2831 | for (i = 0; i < ARRAY_SIZE(debugfs_regs); i++) { | |
2832 | if (adev->debugfs_regs[i]) { | |
2833 | debugfs_remove(adev->debugfs_regs[i]); | |
2834 | adev->debugfs_regs[i] = NULL; | |
2835 | } | |
2836 | } | |
d38ceaf9 AD |
2837 | } |
2838 | ||
2839 | int amdgpu_debugfs_init(struct drm_minor *minor) | |
2840 | { | |
2841 | return 0; | |
2842 | } | |
2843 | ||
2844 | void amdgpu_debugfs_cleanup(struct drm_minor *minor) | |
2845 | { | |
2846 | } | |
7cebc728 AK |
2847 | #else |
2848 | static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev) | |
2849 | { | |
2850 | return 0; | |
2851 | } | |
2852 | static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev) { } | |
d38ceaf9 | 2853 | #endif |