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 | */ | |
28 | #include <linux/console.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/debugfs.h> | |
31 | #include <drm/drmP.h> | |
32 | #include <drm/drm_crtc_helper.h> | |
33 | #include <drm/amdgpu_drm.h> | |
34 | #include <linux/vgaarb.h> | |
35 | #include <linux/vga_switcheroo.h> | |
36 | #include <linux/efi.h> | |
37 | #include "amdgpu.h" | |
38 | #include "amdgpu_i2c.h" | |
39 | #include "atom.h" | |
40 | #include "amdgpu_atombios.h" | |
a2e73f56 AD |
41 | #ifdef CONFIG_DRM_AMDGPU_CIK |
42 | #include "cik.h" | |
43 | #endif | |
aaa36a97 | 44 | #include "vi.h" |
d38ceaf9 AD |
45 | #include "bif/bif_4_1_d.h" |
46 | ||
47 | static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev); | |
48 | static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev); | |
49 | ||
50 | static const char *amdgpu_asic_name[] = { | |
51 | "BONAIRE", | |
52 | "KAVERI", | |
53 | "KABINI", | |
54 | "HAWAII", | |
55 | "MULLINS", | |
56 | "TOPAZ", | |
57 | "TONGA", | |
48299f95 | 58 | "FIJI", |
d38ceaf9 | 59 | "CARRIZO", |
139f4917 | 60 | "STONEY", |
d38ceaf9 AD |
61 | "LAST", |
62 | }; | |
63 | ||
64 | bool amdgpu_device_is_px(struct drm_device *dev) | |
65 | { | |
66 | struct amdgpu_device *adev = dev->dev_private; | |
67 | ||
2f7d10b3 | 68 | if (adev->flags & AMD_IS_PX) |
d38ceaf9 AD |
69 | return true; |
70 | return false; | |
71 | } | |
72 | ||
73 | /* | |
74 | * MMIO register access helper functions. | |
75 | */ | |
76 | uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg, | |
77 | bool always_indirect) | |
78 | { | |
79 | if ((reg * 4) < adev->rmmio_size && !always_indirect) | |
80 | return readl(((void __iomem *)adev->rmmio) + (reg * 4)); | |
81 | else { | |
82 | unsigned long flags; | |
83 | uint32_t ret; | |
84 | ||
85 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
86 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
87 | ret = readl(((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
88 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
89 | ||
90 | return ret; | |
91 | } | |
92 | } | |
93 | ||
94 | void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, | |
95 | bool always_indirect) | |
96 | { | |
97 | if ((reg * 4) < adev->rmmio_size && !always_indirect) | |
98 | writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); | |
99 | else { | |
100 | unsigned long flags; | |
101 | ||
102 | spin_lock_irqsave(&adev->mmio_idx_lock, flags); | |
103 | writel((reg * 4), ((void __iomem *)adev->rmmio) + (mmMM_INDEX * 4)); | |
104 | writel(v, ((void __iomem *)adev->rmmio) + (mmMM_DATA * 4)); | |
105 | spin_unlock_irqrestore(&adev->mmio_idx_lock, flags); | |
106 | } | |
107 | } | |
108 | ||
109 | u32 amdgpu_io_rreg(struct amdgpu_device *adev, u32 reg) | |
110 | { | |
111 | if ((reg * 4) < adev->rio_mem_size) | |
112 | return ioread32(adev->rio_mem + (reg * 4)); | |
113 | else { | |
114 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
115 | return ioread32(adev->rio_mem + (mmMM_DATA * 4)); | |
116 | } | |
117 | } | |
118 | ||
119 | void amdgpu_io_wreg(struct amdgpu_device *adev, u32 reg, u32 v) | |
120 | { | |
121 | ||
122 | if ((reg * 4) < adev->rio_mem_size) | |
123 | iowrite32(v, adev->rio_mem + (reg * 4)); | |
124 | else { | |
125 | iowrite32((reg * 4), adev->rio_mem + (mmMM_INDEX * 4)); | |
126 | iowrite32(v, adev->rio_mem + (mmMM_DATA * 4)); | |
127 | } | |
128 | } | |
129 | ||
130 | /** | |
131 | * amdgpu_mm_rdoorbell - read a doorbell dword | |
132 | * | |
133 | * @adev: amdgpu_device pointer | |
134 | * @index: doorbell index | |
135 | * | |
136 | * Returns the value in the doorbell aperture at the | |
137 | * requested doorbell index (CIK). | |
138 | */ | |
139 | u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index) | |
140 | { | |
141 | if (index < adev->doorbell.num_doorbells) { | |
142 | return readl(adev->doorbell.ptr + index); | |
143 | } else { | |
144 | DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index); | |
145 | return 0; | |
146 | } | |
147 | } | |
148 | ||
149 | /** | |
150 | * amdgpu_mm_wdoorbell - write a doorbell dword | |
151 | * | |
152 | * @adev: amdgpu_device pointer | |
153 | * @index: doorbell index | |
154 | * @v: value to write | |
155 | * | |
156 | * Writes @v to the doorbell aperture at the | |
157 | * requested doorbell index (CIK). | |
158 | */ | |
159 | void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v) | |
160 | { | |
161 | if (index < adev->doorbell.num_doorbells) { | |
162 | writel(v, adev->doorbell.ptr + index); | |
163 | } else { | |
164 | DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index); | |
165 | } | |
166 | } | |
167 | ||
168 | /** | |
169 | * amdgpu_invalid_rreg - dummy reg read function | |
170 | * | |
171 | * @adev: amdgpu device pointer | |
172 | * @reg: offset of register | |
173 | * | |
174 | * Dummy register read function. Used for register blocks | |
175 | * that certain asics don't have (all asics). | |
176 | * Returns the value in the register. | |
177 | */ | |
178 | static uint32_t amdgpu_invalid_rreg(struct amdgpu_device *adev, uint32_t reg) | |
179 | { | |
180 | DRM_ERROR("Invalid callback to read register 0x%04X\n", reg); | |
181 | BUG(); | |
182 | return 0; | |
183 | } | |
184 | ||
185 | /** | |
186 | * amdgpu_invalid_wreg - dummy reg write function | |
187 | * | |
188 | * @adev: amdgpu device pointer | |
189 | * @reg: offset of register | |
190 | * @v: value to write to the register | |
191 | * | |
192 | * Dummy register read function. Used for register blocks | |
193 | * that certain asics don't have (all asics). | |
194 | */ | |
195 | static void amdgpu_invalid_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) | |
196 | { | |
197 | DRM_ERROR("Invalid callback to write register 0x%04X with 0x%08X\n", | |
198 | reg, v); | |
199 | BUG(); | |
200 | } | |
201 | ||
202 | /** | |
203 | * amdgpu_block_invalid_rreg - dummy reg read function | |
204 | * | |
205 | * @adev: amdgpu device pointer | |
206 | * @block: offset of instance | |
207 | * @reg: offset of register | |
208 | * | |
209 | * Dummy register read function. Used for register blocks | |
210 | * that certain asics don't have (all asics). | |
211 | * Returns the value in the register. | |
212 | */ | |
213 | static uint32_t amdgpu_block_invalid_rreg(struct amdgpu_device *adev, | |
214 | uint32_t block, uint32_t reg) | |
215 | { | |
216 | DRM_ERROR("Invalid callback to read register 0x%04X in block 0x%04X\n", | |
217 | reg, block); | |
218 | BUG(); | |
219 | return 0; | |
220 | } | |
221 | ||
222 | /** | |
223 | * amdgpu_block_invalid_wreg - dummy reg write function | |
224 | * | |
225 | * @adev: amdgpu device pointer | |
226 | * @block: offset of instance | |
227 | * @reg: offset of register | |
228 | * @v: value to write to the register | |
229 | * | |
230 | * Dummy register read function. Used for register blocks | |
231 | * that certain asics don't have (all asics). | |
232 | */ | |
233 | static void amdgpu_block_invalid_wreg(struct amdgpu_device *adev, | |
234 | uint32_t block, | |
235 | uint32_t reg, uint32_t v) | |
236 | { | |
237 | DRM_ERROR("Invalid block callback to write register 0x%04X in block 0x%04X with 0x%08X\n", | |
238 | reg, block, v); | |
239 | BUG(); | |
240 | } | |
241 | ||
242 | static int amdgpu_vram_scratch_init(struct amdgpu_device *adev) | |
243 | { | |
244 | int r; | |
245 | ||
246 | if (adev->vram_scratch.robj == NULL) { | |
247 | r = amdgpu_bo_create(adev, AMDGPU_GPU_PAGE_SIZE, | |
857d913d AD |
248 | PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, |
249 | AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED, | |
72d7668b | 250 | NULL, NULL, &adev->vram_scratch.robj); |
d38ceaf9 AD |
251 | if (r) { |
252 | return r; | |
253 | } | |
254 | } | |
255 | ||
256 | r = amdgpu_bo_reserve(adev->vram_scratch.robj, false); | |
257 | if (unlikely(r != 0)) | |
258 | return r; | |
259 | r = amdgpu_bo_pin(adev->vram_scratch.robj, | |
260 | AMDGPU_GEM_DOMAIN_VRAM, &adev->vram_scratch.gpu_addr); | |
261 | if (r) { | |
262 | amdgpu_bo_unreserve(adev->vram_scratch.robj); | |
263 | return r; | |
264 | } | |
265 | r = amdgpu_bo_kmap(adev->vram_scratch.robj, | |
266 | (void **)&adev->vram_scratch.ptr); | |
267 | if (r) | |
268 | amdgpu_bo_unpin(adev->vram_scratch.robj); | |
269 | amdgpu_bo_unreserve(adev->vram_scratch.robj); | |
270 | ||
271 | return r; | |
272 | } | |
273 | ||
274 | static void amdgpu_vram_scratch_fini(struct amdgpu_device *adev) | |
275 | { | |
276 | int r; | |
277 | ||
278 | if (adev->vram_scratch.robj == NULL) { | |
279 | return; | |
280 | } | |
281 | r = amdgpu_bo_reserve(adev->vram_scratch.robj, false); | |
282 | if (likely(r == 0)) { | |
283 | amdgpu_bo_kunmap(adev->vram_scratch.robj); | |
284 | amdgpu_bo_unpin(adev->vram_scratch.robj); | |
285 | amdgpu_bo_unreserve(adev->vram_scratch.robj); | |
286 | } | |
287 | amdgpu_bo_unref(&adev->vram_scratch.robj); | |
288 | } | |
289 | ||
290 | /** | |
291 | * amdgpu_program_register_sequence - program an array of registers. | |
292 | * | |
293 | * @adev: amdgpu_device pointer | |
294 | * @registers: pointer to the register array | |
295 | * @array_size: size of the register array | |
296 | * | |
297 | * Programs an array or registers with and and or masks. | |
298 | * This is a helper for setting golden registers. | |
299 | */ | |
300 | void amdgpu_program_register_sequence(struct amdgpu_device *adev, | |
301 | const u32 *registers, | |
302 | const u32 array_size) | |
303 | { | |
304 | u32 tmp, reg, and_mask, or_mask; | |
305 | int i; | |
306 | ||
307 | if (array_size % 3) | |
308 | return; | |
309 | ||
310 | for (i = 0; i < array_size; i +=3) { | |
311 | reg = registers[i + 0]; | |
312 | and_mask = registers[i + 1]; | |
313 | or_mask = registers[i + 2]; | |
314 | ||
315 | if (and_mask == 0xffffffff) { | |
316 | tmp = or_mask; | |
317 | } else { | |
318 | tmp = RREG32(reg); | |
319 | tmp &= ~and_mask; | |
320 | tmp |= or_mask; | |
321 | } | |
322 | WREG32(reg, tmp); | |
323 | } | |
324 | } | |
325 | ||
326 | void amdgpu_pci_config_reset(struct amdgpu_device *adev) | |
327 | { | |
328 | pci_write_config_dword(adev->pdev, 0x7c, AMDGPU_ASIC_RESET_DATA); | |
329 | } | |
330 | ||
331 | /* | |
332 | * GPU doorbell aperture helpers function. | |
333 | */ | |
334 | /** | |
335 | * amdgpu_doorbell_init - Init doorbell driver information. | |
336 | * | |
337 | * @adev: amdgpu_device pointer | |
338 | * | |
339 | * Init doorbell driver information (CIK) | |
340 | * Returns 0 on success, error on failure. | |
341 | */ | |
342 | static int amdgpu_doorbell_init(struct amdgpu_device *adev) | |
343 | { | |
344 | /* doorbell bar mapping */ | |
345 | adev->doorbell.base = pci_resource_start(adev->pdev, 2); | |
346 | adev->doorbell.size = pci_resource_len(adev->pdev, 2); | |
347 | ||
348 | adev->doorbell.num_doorbells = min_t(u32, adev->doorbell.size / sizeof(u32), | |
349 | AMDGPU_DOORBELL_MAX_ASSIGNMENT+1); | |
350 | if (adev->doorbell.num_doorbells == 0) | |
351 | return -EINVAL; | |
352 | ||
353 | adev->doorbell.ptr = ioremap(adev->doorbell.base, adev->doorbell.num_doorbells * sizeof(u32)); | |
354 | if (adev->doorbell.ptr == NULL) { | |
355 | return -ENOMEM; | |
356 | } | |
357 | DRM_INFO("doorbell mmio base: 0x%08X\n", (uint32_t)adev->doorbell.base); | |
358 | DRM_INFO("doorbell mmio size: %u\n", (unsigned)adev->doorbell.size); | |
359 | ||
360 | return 0; | |
361 | } | |
362 | ||
363 | /** | |
364 | * amdgpu_doorbell_fini - Tear down doorbell driver information. | |
365 | * | |
366 | * @adev: amdgpu_device pointer | |
367 | * | |
368 | * Tear down doorbell driver information (CIK) | |
369 | */ | |
370 | static void amdgpu_doorbell_fini(struct amdgpu_device *adev) | |
371 | { | |
372 | iounmap(adev->doorbell.ptr); | |
373 | adev->doorbell.ptr = NULL; | |
374 | } | |
375 | ||
376 | /** | |
377 | * amdgpu_doorbell_get_kfd_info - Report doorbell configuration required to | |
378 | * setup amdkfd | |
379 | * | |
380 | * @adev: amdgpu_device pointer | |
381 | * @aperture_base: output returning doorbell aperture base physical address | |
382 | * @aperture_size: output returning doorbell aperture size in bytes | |
383 | * @start_offset: output returning # of doorbell bytes reserved for amdgpu. | |
384 | * | |
385 | * amdgpu and amdkfd share the doorbell aperture. amdgpu sets it up, | |
386 | * takes doorbells required for its own rings and reports the setup to amdkfd. | |
387 | * amdgpu reserved doorbells are at the start of the doorbell aperture. | |
388 | */ | |
389 | void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev, | |
390 | phys_addr_t *aperture_base, | |
391 | size_t *aperture_size, | |
392 | size_t *start_offset) | |
393 | { | |
394 | /* | |
395 | * The first num_doorbells are used by amdgpu. | |
396 | * amdkfd takes whatever's left in the aperture. | |
397 | */ | |
398 | if (adev->doorbell.size > adev->doorbell.num_doorbells * sizeof(u32)) { | |
399 | *aperture_base = adev->doorbell.base; | |
400 | *aperture_size = adev->doorbell.size; | |
401 | *start_offset = adev->doorbell.num_doorbells * sizeof(u32); | |
402 | } else { | |
403 | *aperture_base = 0; | |
404 | *aperture_size = 0; | |
405 | *start_offset = 0; | |
406 | } | |
407 | } | |
408 | ||
409 | /* | |
410 | * amdgpu_wb_*() | |
411 | * Writeback is the the method by which the the GPU updates special pages | |
412 | * in memory with the status of certain GPU events (fences, ring pointers, | |
413 | * etc.). | |
414 | */ | |
415 | ||
416 | /** | |
417 | * amdgpu_wb_fini - Disable Writeback and free memory | |
418 | * | |
419 | * @adev: amdgpu_device pointer | |
420 | * | |
421 | * Disables Writeback and frees the Writeback memory (all asics). | |
422 | * Used at driver shutdown. | |
423 | */ | |
424 | static void amdgpu_wb_fini(struct amdgpu_device *adev) | |
425 | { | |
426 | if (adev->wb.wb_obj) { | |
427 | if (!amdgpu_bo_reserve(adev->wb.wb_obj, false)) { | |
428 | amdgpu_bo_kunmap(adev->wb.wb_obj); | |
429 | amdgpu_bo_unpin(adev->wb.wb_obj); | |
430 | amdgpu_bo_unreserve(adev->wb.wb_obj); | |
431 | } | |
432 | amdgpu_bo_unref(&adev->wb.wb_obj); | |
433 | adev->wb.wb = NULL; | |
434 | adev->wb.wb_obj = NULL; | |
435 | } | |
436 | } | |
437 | ||
438 | /** | |
439 | * amdgpu_wb_init- Init Writeback driver info and allocate memory | |
440 | * | |
441 | * @adev: amdgpu_device pointer | |
442 | * | |
443 | * Disables Writeback and frees the Writeback memory (all asics). | |
444 | * Used at driver startup. | |
445 | * Returns 0 on success or an -error on failure. | |
446 | */ | |
447 | static int amdgpu_wb_init(struct amdgpu_device *adev) | |
448 | { | |
449 | int r; | |
450 | ||
451 | if (adev->wb.wb_obj == NULL) { | |
452 | r = amdgpu_bo_create(adev, AMDGPU_MAX_WB * 4, PAGE_SIZE, true, | |
72d7668b CK |
453 | AMDGPU_GEM_DOMAIN_GTT, 0, NULL, NULL, |
454 | &adev->wb.wb_obj); | |
d38ceaf9 AD |
455 | if (r) { |
456 | dev_warn(adev->dev, "(%d) create WB bo failed\n", r); | |
457 | return r; | |
458 | } | |
459 | r = amdgpu_bo_reserve(adev->wb.wb_obj, false); | |
460 | if (unlikely(r != 0)) { | |
461 | amdgpu_wb_fini(adev); | |
462 | return r; | |
463 | } | |
464 | r = amdgpu_bo_pin(adev->wb.wb_obj, AMDGPU_GEM_DOMAIN_GTT, | |
465 | &adev->wb.gpu_addr); | |
466 | if (r) { | |
467 | amdgpu_bo_unreserve(adev->wb.wb_obj); | |
468 | dev_warn(adev->dev, "(%d) pin WB bo failed\n", r); | |
469 | amdgpu_wb_fini(adev); | |
470 | return r; | |
471 | } | |
472 | r = amdgpu_bo_kmap(adev->wb.wb_obj, (void **)&adev->wb.wb); | |
473 | amdgpu_bo_unreserve(adev->wb.wb_obj); | |
474 | if (r) { | |
475 | dev_warn(adev->dev, "(%d) map WB bo failed\n", r); | |
476 | amdgpu_wb_fini(adev); | |
477 | return r; | |
478 | } | |
479 | ||
480 | adev->wb.num_wb = AMDGPU_MAX_WB; | |
481 | memset(&adev->wb.used, 0, sizeof(adev->wb.used)); | |
482 | ||
483 | /* clear wb memory */ | |
484 | memset((char *)adev->wb.wb, 0, AMDGPU_GPU_PAGE_SIZE); | |
485 | } | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
490 | /** | |
491 | * amdgpu_wb_get - Allocate a wb entry | |
492 | * | |
493 | * @adev: amdgpu_device pointer | |
494 | * @wb: wb index | |
495 | * | |
496 | * Allocate a wb slot for use by the driver (all asics). | |
497 | * Returns 0 on success or -EINVAL on failure. | |
498 | */ | |
499 | int amdgpu_wb_get(struct amdgpu_device *adev, u32 *wb) | |
500 | { | |
501 | unsigned long offset = find_first_zero_bit(adev->wb.used, adev->wb.num_wb); | |
502 | if (offset < adev->wb.num_wb) { | |
503 | __set_bit(offset, adev->wb.used); | |
504 | *wb = offset; | |
505 | return 0; | |
506 | } else { | |
507 | return -EINVAL; | |
508 | } | |
509 | } | |
510 | ||
511 | /** | |
512 | * amdgpu_wb_free - Free a wb entry | |
513 | * | |
514 | * @adev: amdgpu_device pointer | |
515 | * @wb: wb index | |
516 | * | |
517 | * Free a wb slot allocated for use by the driver (all asics) | |
518 | */ | |
519 | void amdgpu_wb_free(struct amdgpu_device *adev, u32 wb) | |
520 | { | |
521 | if (wb < adev->wb.num_wb) | |
522 | __clear_bit(wb, adev->wb.used); | |
523 | } | |
524 | ||
525 | /** | |
526 | * amdgpu_vram_location - try to find VRAM location | |
527 | * @adev: amdgpu device structure holding all necessary informations | |
528 | * @mc: memory controller structure holding memory informations | |
529 | * @base: base address at which to put VRAM | |
530 | * | |
531 | * Function will place try to place VRAM at base address provided | |
532 | * as parameter (which is so far either PCI aperture address or | |
533 | * for IGP TOM base address). | |
534 | * | |
535 | * If there is not enough space to fit the unvisible VRAM in the 32bits | |
536 | * address space then we limit the VRAM size to the aperture. | |
537 | * | |
538 | * Note: We don't explicitly enforce VRAM start to be aligned on VRAM size, | |
539 | * this shouldn't be a problem as we are using the PCI aperture as a reference. | |
540 | * Otherwise this would be needed for rv280, all r3xx, and all r4xx, but | |
541 | * not IGP. | |
542 | * | |
543 | * Note: we use mc_vram_size as on some board we need to program the mc to | |
544 | * cover the whole aperture even if VRAM size is inferior to aperture size | |
545 | * Novell bug 204882 + along with lots of ubuntu ones | |
546 | * | |
547 | * Note: when limiting vram it's safe to overwritte real_vram_size because | |
548 | * we are not in case where real_vram_size is inferior to mc_vram_size (ie | |
549 | * note afected by bogus hw of Novell bug 204882 + along with lots of ubuntu | |
550 | * ones) | |
551 | * | |
552 | * Note: IGP TOM addr should be the same as the aperture addr, we don't | |
553 | * explicitly check for that thought. | |
554 | * | |
555 | * FIXME: when reducing VRAM size align new size on power of 2. | |
556 | */ | |
557 | void amdgpu_vram_location(struct amdgpu_device *adev, struct amdgpu_mc *mc, u64 base) | |
558 | { | |
559 | uint64_t limit = (uint64_t)amdgpu_vram_limit << 20; | |
560 | ||
561 | mc->vram_start = base; | |
562 | if (mc->mc_vram_size > (adev->mc.mc_mask - base + 1)) { | |
563 | dev_warn(adev->dev, "limiting VRAM to PCI aperture size\n"); | |
564 | mc->real_vram_size = mc->aper_size; | |
565 | mc->mc_vram_size = mc->aper_size; | |
566 | } | |
567 | mc->vram_end = mc->vram_start + mc->mc_vram_size - 1; | |
568 | if (limit && limit < mc->real_vram_size) | |
569 | mc->real_vram_size = limit; | |
570 | dev_info(adev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n", | |
571 | mc->mc_vram_size >> 20, mc->vram_start, | |
572 | mc->vram_end, mc->real_vram_size >> 20); | |
573 | } | |
574 | ||
575 | /** | |
576 | * amdgpu_gtt_location - try to find GTT location | |
577 | * @adev: amdgpu device structure holding all necessary informations | |
578 | * @mc: memory controller structure holding memory informations | |
579 | * | |
580 | * Function will place try to place GTT before or after VRAM. | |
581 | * | |
582 | * If GTT size is bigger than space left then we ajust GTT size. | |
583 | * Thus function will never fails. | |
584 | * | |
585 | * FIXME: when reducing GTT size align new size on power of 2. | |
586 | */ | |
587 | void amdgpu_gtt_location(struct amdgpu_device *adev, struct amdgpu_mc *mc) | |
588 | { | |
589 | u64 size_af, size_bf; | |
590 | ||
591 | size_af = ((adev->mc.mc_mask - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align; | |
592 | size_bf = mc->vram_start & ~mc->gtt_base_align; | |
593 | if (size_bf > size_af) { | |
594 | if (mc->gtt_size > size_bf) { | |
595 | dev_warn(adev->dev, "limiting GTT\n"); | |
596 | mc->gtt_size = size_bf; | |
597 | } | |
598 | mc->gtt_start = (mc->vram_start & ~mc->gtt_base_align) - mc->gtt_size; | |
599 | } else { | |
600 | if (mc->gtt_size > size_af) { | |
601 | dev_warn(adev->dev, "limiting GTT\n"); | |
602 | mc->gtt_size = size_af; | |
603 | } | |
604 | mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align; | |
605 | } | |
606 | mc->gtt_end = mc->gtt_start + mc->gtt_size - 1; | |
607 | dev_info(adev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n", | |
608 | mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end); | |
609 | } | |
610 | ||
611 | /* | |
612 | * GPU helpers function. | |
613 | */ | |
614 | /** | |
615 | * amdgpu_card_posted - check if the hw has already been initialized | |
616 | * | |
617 | * @adev: amdgpu_device pointer | |
618 | * | |
619 | * Check if the asic has been initialized (all asics). | |
620 | * Used at driver startup. | |
621 | * Returns true if initialized or false if not. | |
622 | */ | |
623 | bool amdgpu_card_posted(struct amdgpu_device *adev) | |
624 | { | |
625 | uint32_t reg; | |
626 | ||
627 | /* then check MEM_SIZE, in case the crtcs are off */ | |
628 | reg = RREG32(mmCONFIG_MEMSIZE); | |
629 | ||
630 | if (reg) | |
631 | return true; | |
632 | ||
633 | return false; | |
634 | ||
635 | } | |
636 | ||
637 | /** | |
638 | * amdgpu_boot_test_post_card - check and possibly initialize the hw | |
639 | * | |
640 | * @adev: amdgpu_device pointer | |
641 | * | |
642 | * Check if the asic is initialized and if not, attempt to initialize | |
643 | * it (all asics). | |
644 | * Returns true if initialized or false if not. | |
645 | */ | |
646 | bool amdgpu_boot_test_post_card(struct amdgpu_device *adev) | |
647 | { | |
648 | if (amdgpu_card_posted(adev)) | |
649 | return true; | |
650 | ||
651 | if (adev->bios) { | |
652 | DRM_INFO("GPU not posted. posting now...\n"); | |
653 | if (adev->is_atom_bios) | |
654 | amdgpu_atom_asic_init(adev->mode_info.atom_context); | |
655 | return true; | |
656 | } else { | |
657 | dev_err(adev->dev, "Card not posted and no BIOS - ignoring\n"); | |
658 | return false; | |
659 | } | |
660 | } | |
661 | ||
662 | /** | |
663 | * amdgpu_dummy_page_init - init dummy page used by the driver | |
664 | * | |
665 | * @adev: amdgpu_device pointer | |
666 | * | |
667 | * Allocate the dummy page used by the driver (all asics). | |
668 | * This dummy page is used by the driver as a filler for gart entries | |
669 | * when pages are taken out of the GART | |
670 | * Returns 0 on sucess, -ENOMEM on failure. | |
671 | */ | |
672 | int amdgpu_dummy_page_init(struct amdgpu_device *adev) | |
673 | { | |
674 | if (adev->dummy_page.page) | |
675 | return 0; | |
676 | adev->dummy_page.page = alloc_page(GFP_DMA32 | GFP_KERNEL | __GFP_ZERO); | |
677 | if (adev->dummy_page.page == NULL) | |
678 | return -ENOMEM; | |
679 | adev->dummy_page.addr = pci_map_page(adev->pdev, adev->dummy_page.page, | |
680 | 0, PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
681 | if (pci_dma_mapping_error(adev->pdev, adev->dummy_page.addr)) { | |
682 | dev_err(&adev->pdev->dev, "Failed to DMA MAP the dummy page\n"); | |
683 | __free_page(adev->dummy_page.page); | |
684 | adev->dummy_page.page = NULL; | |
685 | return -ENOMEM; | |
686 | } | |
687 | return 0; | |
688 | } | |
689 | ||
690 | /** | |
691 | * amdgpu_dummy_page_fini - free dummy page used by the driver | |
692 | * | |
693 | * @adev: amdgpu_device pointer | |
694 | * | |
695 | * Frees the dummy page used by the driver (all asics). | |
696 | */ | |
697 | void amdgpu_dummy_page_fini(struct amdgpu_device *adev) | |
698 | { | |
699 | if (adev->dummy_page.page == NULL) | |
700 | return; | |
701 | pci_unmap_page(adev->pdev, adev->dummy_page.addr, | |
702 | PAGE_SIZE, PCI_DMA_BIDIRECTIONAL); | |
703 | __free_page(adev->dummy_page.page); | |
704 | adev->dummy_page.page = NULL; | |
705 | } | |
706 | ||
707 | ||
708 | /* ATOM accessor methods */ | |
709 | /* | |
710 | * ATOM is an interpreted byte code stored in tables in the vbios. The | |
711 | * driver registers callbacks to access registers and the interpreter | |
712 | * in the driver parses the tables and executes then to program specific | |
713 | * actions (set display modes, asic init, etc.). See amdgpu_atombios.c, | |
714 | * atombios.h, and atom.c | |
715 | */ | |
716 | ||
717 | /** | |
718 | * cail_pll_read - read PLL register | |
719 | * | |
720 | * @info: atom card_info pointer | |
721 | * @reg: PLL register offset | |
722 | * | |
723 | * Provides a PLL register accessor for the atom interpreter (r4xx+). | |
724 | * Returns the value of the PLL register. | |
725 | */ | |
726 | static uint32_t cail_pll_read(struct card_info *info, uint32_t reg) | |
727 | { | |
728 | return 0; | |
729 | } | |
730 | ||
731 | /** | |
732 | * cail_pll_write - write PLL register | |
733 | * | |
734 | * @info: atom card_info pointer | |
735 | * @reg: PLL register offset | |
736 | * @val: value to write to the pll register | |
737 | * | |
738 | * Provides a PLL register accessor for the atom interpreter (r4xx+). | |
739 | */ | |
740 | static void cail_pll_write(struct card_info *info, uint32_t reg, uint32_t val) | |
741 | { | |
742 | ||
743 | } | |
744 | ||
745 | /** | |
746 | * cail_mc_read - read MC (Memory Controller) register | |
747 | * | |
748 | * @info: atom card_info pointer | |
749 | * @reg: MC register offset | |
750 | * | |
751 | * Provides an MC register accessor for the atom interpreter (r4xx+). | |
752 | * Returns the value of the MC register. | |
753 | */ | |
754 | static uint32_t cail_mc_read(struct card_info *info, uint32_t reg) | |
755 | { | |
756 | return 0; | |
757 | } | |
758 | ||
759 | /** | |
760 | * cail_mc_write - write MC (Memory Controller) register | |
761 | * | |
762 | * @info: atom card_info pointer | |
763 | * @reg: MC register offset | |
764 | * @val: value to write to the pll register | |
765 | * | |
766 | * Provides a MC register accessor for the atom interpreter (r4xx+). | |
767 | */ | |
768 | static void cail_mc_write(struct card_info *info, uint32_t reg, uint32_t val) | |
769 | { | |
770 | ||
771 | } | |
772 | ||
773 | /** | |
774 | * cail_reg_write - write MMIO register | |
775 | * | |
776 | * @info: atom card_info pointer | |
777 | * @reg: MMIO register offset | |
778 | * @val: value to write to the pll register | |
779 | * | |
780 | * Provides a MMIO register accessor for the atom interpreter (r4xx+). | |
781 | */ | |
782 | static void cail_reg_write(struct card_info *info, uint32_t reg, uint32_t val) | |
783 | { | |
784 | struct amdgpu_device *adev = info->dev->dev_private; | |
785 | ||
786 | WREG32(reg, val); | |
787 | } | |
788 | ||
789 | /** | |
790 | * cail_reg_read - read MMIO register | |
791 | * | |
792 | * @info: atom card_info pointer | |
793 | * @reg: MMIO register offset | |
794 | * | |
795 | * Provides an MMIO register accessor for the atom interpreter (r4xx+). | |
796 | * Returns the value of the MMIO register. | |
797 | */ | |
798 | static uint32_t cail_reg_read(struct card_info *info, uint32_t reg) | |
799 | { | |
800 | struct amdgpu_device *adev = info->dev->dev_private; | |
801 | uint32_t r; | |
802 | ||
803 | r = RREG32(reg); | |
804 | return r; | |
805 | } | |
806 | ||
807 | /** | |
808 | * cail_ioreg_write - write IO register | |
809 | * | |
810 | * @info: atom card_info pointer | |
811 | * @reg: IO register offset | |
812 | * @val: value to write to the pll register | |
813 | * | |
814 | * Provides a IO register accessor for the atom interpreter (r4xx+). | |
815 | */ | |
816 | static void cail_ioreg_write(struct card_info *info, uint32_t reg, uint32_t val) | |
817 | { | |
818 | struct amdgpu_device *adev = info->dev->dev_private; | |
819 | ||
820 | WREG32_IO(reg, val); | |
821 | } | |
822 | ||
823 | /** | |
824 | * cail_ioreg_read - read IO register | |
825 | * | |
826 | * @info: atom card_info pointer | |
827 | * @reg: IO register offset | |
828 | * | |
829 | * Provides an IO register accessor for the atom interpreter (r4xx+). | |
830 | * Returns the value of the IO register. | |
831 | */ | |
832 | static uint32_t cail_ioreg_read(struct card_info *info, uint32_t reg) | |
833 | { | |
834 | struct amdgpu_device *adev = info->dev->dev_private; | |
835 | uint32_t r; | |
836 | ||
837 | r = RREG32_IO(reg); | |
838 | return r; | |
839 | } | |
840 | ||
841 | /** | |
842 | * amdgpu_atombios_fini - free the driver info and callbacks for atombios | |
843 | * | |
844 | * @adev: amdgpu_device pointer | |
845 | * | |
846 | * Frees the driver info and register access callbacks for the ATOM | |
847 | * interpreter (r4xx+). | |
848 | * Called at driver shutdown. | |
849 | */ | |
850 | static void amdgpu_atombios_fini(struct amdgpu_device *adev) | |
851 | { | |
852 | if (adev->mode_info.atom_context) | |
853 | kfree(adev->mode_info.atom_context->scratch); | |
854 | kfree(adev->mode_info.atom_context); | |
855 | adev->mode_info.atom_context = NULL; | |
856 | kfree(adev->mode_info.atom_card_info); | |
857 | adev->mode_info.atom_card_info = NULL; | |
858 | } | |
859 | ||
860 | /** | |
861 | * amdgpu_atombios_init - init the driver info and callbacks for atombios | |
862 | * | |
863 | * @adev: amdgpu_device pointer | |
864 | * | |
865 | * Initializes the driver info and register access callbacks for the | |
866 | * ATOM interpreter (r4xx+). | |
867 | * Returns 0 on sucess, -ENOMEM on failure. | |
868 | * Called at driver startup. | |
869 | */ | |
870 | static int amdgpu_atombios_init(struct amdgpu_device *adev) | |
871 | { | |
872 | struct card_info *atom_card_info = | |
873 | kzalloc(sizeof(struct card_info), GFP_KERNEL); | |
874 | ||
875 | if (!atom_card_info) | |
876 | return -ENOMEM; | |
877 | ||
878 | adev->mode_info.atom_card_info = atom_card_info; | |
879 | atom_card_info->dev = adev->ddev; | |
880 | atom_card_info->reg_read = cail_reg_read; | |
881 | atom_card_info->reg_write = cail_reg_write; | |
882 | /* needed for iio ops */ | |
883 | if (adev->rio_mem) { | |
884 | atom_card_info->ioreg_read = cail_ioreg_read; | |
885 | atom_card_info->ioreg_write = cail_ioreg_write; | |
886 | } else { | |
887 | DRM_ERROR("Unable to find PCI I/O BAR; using MMIO for ATOM IIO\n"); | |
888 | atom_card_info->ioreg_read = cail_reg_read; | |
889 | atom_card_info->ioreg_write = cail_reg_write; | |
890 | } | |
891 | atom_card_info->mc_read = cail_mc_read; | |
892 | atom_card_info->mc_write = cail_mc_write; | |
893 | atom_card_info->pll_read = cail_pll_read; | |
894 | atom_card_info->pll_write = cail_pll_write; | |
895 | ||
896 | adev->mode_info.atom_context = amdgpu_atom_parse(atom_card_info, adev->bios); | |
897 | if (!adev->mode_info.atom_context) { | |
898 | amdgpu_atombios_fini(adev); | |
899 | return -ENOMEM; | |
900 | } | |
901 | ||
902 | mutex_init(&adev->mode_info.atom_context->mutex); | |
903 | amdgpu_atombios_scratch_regs_init(adev); | |
904 | amdgpu_atom_allocate_fb_scratch(adev->mode_info.atom_context); | |
905 | return 0; | |
906 | } | |
907 | ||
908 | /* if we get transitioned to only one device, take VGA back */ | |
909 | /** | |
910 | * amdgpu_vga_set_decode - enable/disable vga decode | |
911 | * | |
912 | * @cookie: amdgpu_device pointer | |
913 | * @state: enable/disable vga decode | |
914 | * | |
915 | * Enable/disable vga decode (all asics). | |
916 | * Returns VGA resource flags. | |
917 | */ | |
918 | static unsigned int amdgpu_vga_set_decode(void *cookie, bool state) | |
919 | { | |
920 | struct amdgpu_device *adev = cookie; | |
921 | amdgpu_asic_set_vga_state(adev, state); | |
922 | if (state) | |
923 | return VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM | | |
924 | VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
925 | else | |
926 | return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM; | |
927 | } | |
928 | ||
929 | /** | |
930 | * amdgpu_check_pot_argument - check that argument is a power of two | |
931 | * | |
932 | * @arg: value to check | |
933 | * | |
934 | * Validates that a certain argument is a power of two (all asics). | |
935 | * Returns true if argument is valid. | |
936 | */ | |
937 | static bool amdgpu_check_pot_argument(int arg) | |
938 | { | |
939 | return (arg & (arg - 1)) == 0; | |
940 | } | |
941 | ||
942 | /** | |
943 | * amdgpu_check_arguments - validate module params | |
944 | * | |
945 | * @adev: amdgpu_device pointer | |
946 | * | |
947 | * Validates certain module parameters and updates | |
948 | * the associated values used by the driver (all asics). | |
949 | */ | |
950 | static void amdgpu_check_arguments(struct amdgpu_device *adev) | |
951 | { | |
952 | /* vramlimit must be a power of two */ | |
953 | if (!amdgpu_check_pot_argument(amdgpu_vram_limit)) { | |
954 | dev_warn(adev->dev, "vram limit (%d) must be a power of 2\n", | |
955 | amdgpu_vram_limit); | |
956 | amdgpu_vram_limit = 0; | |
957 | } | |
958 | ||
959 | if (amdgpu_gart_size != -1) { | |
960 | /* gtt size must be power of two and greater or equal to 32M */ | |
961 | if (amdgpu_gart_size < 32) { | |
962 | dev_warn(adev->dev, "gart size (%d) too small\n", | |
963 | amdgpu_gart_size); | |
964 | amdgpu_gart_size = -1; | |
965 | } else if (!amdgpu_check_pot_argument(amdgpu_gart_size)) { | |
966 | dev_warn(adev->dev, "gart size (%d) must be a power of 2\n", | |
967 | amdgpu_gart_size); | |
968 | amdgpu_gart_size = -1; | |
969 | } | |
970 | } | |
971 | ||
972 | if (!amdgpu_check_pot_argument(amdgpu_vm_size)) { | |
973 | dev_warn(adev->dev, "VM size (%d) must be a power of 2\n", | |
974 | amdgpu_vm_size); | |
8dacc127 | 975 | amdgpu_vm_size = 8; |
d38ceaf9 AD |
976 | } |
977 | ||
978 | if (amdgpu_vm_size < 1) { | |
979 | dev_warn(adev->dev, "VM size (%d) too small, min is 1GB\n", | |
980 | amdgpu_vm_size); | |
8dacc127 | 981 | amdgpu_vm_size = 8; |
d38ceaf9 AD |
982 | } |
983 | ||
984 | /* | |
985 | * Max GPUVM size for Cayman, SI and CI are 40 bits. | |
986 | */ | |
987 | if (amdgpu_vm_size > 1024) { | |
988 | dev_warn(adev->dev, "VM size (%d) too large, max is 1TB\n", | |
989 | amdgpu_vm_size); | |
8dacc127 | 990 | amdgpu_vm_size = 8; |
d38ceaf9 AD |
991 | } |
992 | ||
993 | /* defines number of bits in page table versus page directory, | |
994 | * a page is 4KB so we have 12 bits offset, minimum 9 bits in the | |
995 | * page table and the remaining bits are in the page directory */ | |
996 | if (amdgpu_vm_block_size == -1) { | |
997 | ||
998 | /* Total bits covered by PD + PTs */ | |
999 | unsigned bits = ilog2(amdgpu_vm_size) + 18; | |
1000 | ||
1001 | /* Make sure the PD is 4K in size up to 8GB address space. | |
1002 | Above that split equal between PD and PTs */ | |
1003 | if (amdgpu_vm_size <= 8) | |
1004 | amdgpu_vm_block_size = bits - 9; | |
1005 | else | |
1006 | amdgpu_vm_block_size = (bits + 3) / 2; | |
1007 | ||
1008 | } else if (amdgpu_vm_block_size < 9) { | |
1009 | dev_warn(adev->dev, "VM page table size (%d) too small\n", | |
1010 | amdgpu_vm_block_size); | |
1011 | amdgpu_vm_block_size = 9; | |
1012 | } | |
1013 | ||
1014 | if (amdgpu_vm_block_size > 24 || | |
1015 | (amdgpu_vm_size * 1024) < (1ull << amdgpu_vm_block_size)) { | |
1016 | dev_warn(adev->dev, "VM page table size (%d) too large\n", | |
1017 | amdgpu_vm_block_size); | |
1018 | amdgpu_vm_block_size = 9; | |
1019 | } | |
1020 | } | |
1021 | ||
1022 | /** | |
1023 | * amdgpu_switcheroo_set_state - set switcheroo state | |
1024 | * | |
1025 | * @pdev: pci dev pointer | |
1694467b | 1026 | * @state: vga_switcheroo state |
d38ceaf9 AD |
1027 | * |
1028 | * Callback for the switcheroo driver. Suspends or resumes the | |
1029 | * the asics before or after it is powered up using ACPI methods. | |
1030 | */ | |
1031 | static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, enum vga_switcheroo_state state) | |
1032 | { | |
1033 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1034 | ||
1035 | if (amdgpu_device_is_px(dev) && state == VGA_SWITCHEROO_OFF) | |
1036 | return; | |
1037 | ||
1038 | if (state == VGA_SWITCHEROO_ON) { | |
1039 | unsigned d3_delay = dev->pdev->d3_delay; | |
1040 | ||
1041 | printk(KERN_INFO "amdgpu: switched on\n"); | |
1042 | /* don't suspend or resume card normally */ | |
1043 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
1044 | ||
1045 | amdgpu_resume_kms(dev, true, true); | |
1046 | ||
1047 | dev->pdev->d3_delay = d3_delay; | |
1048 | ||
1049 | dev->switch_power_state = DRM_SWITCH_POWER_ON; | |
1050 | drm_kms_helper_poll_enable(dev); | |
1051 | } else { | |
1052 | printk(KERN_INFO "amdgpu: switched off\n"); | |
1053 | drm_kms_helper_poll_disable(dev); | |
1054 | dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; | |
1055 | amdgpu_suspend_kms(dev, true, true); | |
1056 | dev->switch_power_state = DRM_SWITCH_POWER_OFF; | |
1057 | } | |
1058 | } | |
1059 | ||
1060 | /** | |
1061 | * amdgpu_switcheroo_can_switch - see if switcheroo state can change | |
1062 | * | |
1063 | * @pdev: pci dev pointer | |
1064 | * | |
1065 | * Callback for the switcheroo driver. Check of the switcheroo | |
1066 | * state can be changed. | |
1067 | * Returns true if the state can be changed, false if not. | |
1068 | */ | |
1069 | static bool amdgpu_switcheroo_can_switch(struct pci_dev *pdev) | |
1070 | { | |
1071 | struct drm_device *dev = pci_get_drvdata(pdev); | |
1072 | ||
1073 | /* | |
1074 | * FIXME: open_count is protected by drm_global_mutex but that would lead to | |
1075 | * locking inversion with the driver load path. And the access here is | |
1076 | * completely racy anyway. So don't bother with locking for now. | |
1077 | */ | |
1078 | return dev->open_count == 0; | |
1079 | } | |
1080 | ||
1081 | static const struct vga_switcheroo_client_ops amdgpu_switcheroo_ops = { | |
1082 | .set_gpu_state = amdgpu_switcheroo_set_state, | |
1083 | .reprobe = NULL, | |
1084 | .can_switch = amdgpu_switcheroo_can_switch, | |
1085 | }; | |
1086 | ||
1087 | int amdgpu_set_clockgating_state(struct amdgpu_device *adev, | |
5fc3aeeb | 1088 | enum amd_ip_block_type block_type, |
1089 | enum amd_clockgating_state state) | |
d38ceaf9 AD |
1090 | { |
1091 | int i, r = 0; | |
1092 | ||
1093 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1094 | if (adev->ip_blocks[i].type == block_type) { | |
5fc3aeeb | 1095 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, |
d38ceaf9 AD |
1096 | state); |
1097 | if (r) | |
1098 | return r; | |
1099 | } | |
1100 | } | |
1101 | return r; | |
1102 | } | |
1103 | ||
1104 | int amdgpu_set_powergating_state(struct amdgpu_device *adev, | |
5fc3aeeb | 1105 | enum amd_ip_block_type block_type, |
1106 | enum amd_powergating_state state) | |
d38ceaf9 AD |
1107 | { |
1108 | int i, r = 0; | |
1109 | ||
1110 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1111 | if (adev->ip_blocks[i].type == block_type) { | |
5fc3aeeb | 1112 | r = adev->ip_blocks[i].funcs->set_powergating_state((void *)adev, |
d38ceaf9 AD |
1113 | state); |
1114 | if (r) | |
1115 | return r; | |
1116 | } | |
1117 | } | |
1118 | return r; | |
1119 | } | |
1120 | ||
1121 | const struct amdgpu_ip_block_version * amdgpu_get_ip_block( | |
1122 | struct amdgpu_device *adev, | |
5fc3aeeb | 1123 | enum amd_ip_block_type type) |
d38ceaf9 AD |
1124 | { |
1125 | int i; | |
1126 | ||
1127 | for (i = 0; i < adev->num_ip_blocks; i++) | |
1128 | if (adev->ip_blocks[i].type == type) | |
1129 | return &adev->ip_blocks[i]; | |
1130 | ||
1131 | return NULL; | |
1132 | } | |
1133 | ||
1134 | /** | |
1135 | * amdgpu_ip_block_version_cmp | |
1136 | * | |
1137 | * @adev: amdgpu_device pointer | |
5fc3aeeb | 1138 | * @type: enum amd_ip_block_type |
d38ceaf9 AD |
1139 | * @major: major version |
1140 | * @minor: minor version | |
1141 | * | |
1142 | * return 0 if equal or greater | |
1143 | * return 1 if smaller or the ip_block doesn't exist | |
1144 | */ | |
1145 | int amdgpu_ip_block_version_cmp(struct amdgpu_device *adev, | |
5fc3aeeb | 1146 | enum amd_ip_block_type type, |
d38ceaf9 AD |
1147 | u32 major, u32 minor) |
1148 | { | |
1149 | const struct amdgpu_ip_block_version *ip_block; | |
1150 | ip_block = amdgpu_get_ip_block(adev, type); | |
1151 | ||
1152 | if (ip_block && ((ip_block->major > major) || | |
1153 | ((ip_block->major == major) && | |
1154 | (ip_block->minor >= minor)))) | |
1155 | return 0; | |
1156 | ||
1157 | return 1; | |
1158 | } | |
1159 | ||
1160 | static int amdgpu_early_init(struct amdgpu_device *adev) | |
1161 | { | |
aaa36a97 | 1162 | int i, r; |
d38ceaf9 AD |
1163 | |
1164 | switch (adev->asic_type) { | |
aaa36a97 AD |
1165 | case CHIP_TOPAZ: |
1166 | case CHIP_TONGA: | |
48299f95 | 1167 | case CHIP_FIJI: |
aaa36a97 | 1168 | case CHIP_CARRIZO: |
39bb0c92 SL |
1169 | case CHIP_STONEY: |
1170 | if (adev->asic_type == CHIP_CARRIZO || adev->asic_type == CHIP_STONEY) | |
aaa36a97 AD |
1171 | adev->family = AMDGPU_FAMILY_CZ; |
1172 | else | |
1173 | adev->family = AMDGPU_FAMILY_VI; | |
1174 | ||
1175 | r = vi_set_ip_blocks(adev); | |
1176 | if (r) | |
1177 | return r; | |
1178 | break; | |
a2e73f56 AD |
1179 | #ifdef CONFIG_DRM_AMDGPU_CIK |
1180 | case CHIP_BONAIRE: | |
1181 | case CHIP_HAWAII: | |
1182 | case CHIP_KAVERI: | |
1183 | case CHIP_KABINI: | |
1184 | case CHIP_MULLINS: | |
1185 | if ((adev->asic_type == CHIP_BONAIRE) || (adev->asic_type == CHIP_HAWAII)) | |
1186 | adev->family = AMDGPU_FAMILY_CI; | |
1187 | else | |
1188 | adev->family = AMDGPU_FAMILY_KV; | |
1189 | ||
1190 | r = cik_set_ip_blocks(adev); | |
1191 | if (r) | |
1192 | return r; | |
1193 | break; | |
1194 | #endif | |
d38ceaf9 AD |
1195 | default: |
1196 | /* FIXME: not supported yet */ | |
1197 | return -EINVAL; | |
1198 | } | |
1199 | ||
8faf0e08 AD |
1200 | adev->ip_block_status = kcalloc(adev->num_ip_blocks, |
1201 | sizeof(struct amdgpu_ip_block_status), GFP_KERNEL); | |
1202 | if (adev->ip_block_status == NULL) | |
d8d090b7 | 1203 | return -ENOMEM; |
d38ceaf9 AD |
1204 | |
1205 | if (adev->ip_blocks == NULL) { | |
1206 | DRM_ERROR("No IP blocks found!\n"); | |
1207 | return r; | |
1208 | } | |
1209 | ||
1210 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
1211 | if ((amdgpu_ip_block_mask & (1 << i)) == 0) { | |
1212 | DRM_ERROR("disabled ip block: %d\n", i); | |
8faf0e08 | 1213 | adev->ip_block_status[i].valid = false; |
d38ceaf9 AD |
1214 | } else { |
1215 | if (adev->ip_blocks[i].funcs->early_init) { | |
5fc3aeeb | 1216 | r = adev->ip_blocks[i].funcs->early_init((void *)adev); |
974e6b64 | 1217 | if (r == -ENOENT) |
8faf0e08 | 1218 | adev->ip_block_status[i].valid = false; |
974e6b64 | 1219 | else if (r) |
d38ceaf9 | 1220 | return r; |
974e6b64 | 1221 | else |
8faf0e08 | 1222 | adev->ip_block_status[i].valid = true; |
974e6b64 | 1223 | } else { |
8faf0e08 | 1224 | adev->ip_block_status[i].valid = true; |
d38ceaf9 | 1225 | } |
d38ceaf9 AD |
1226 | } |
1227 | } | |
1228 | ||
1229 | return 0; | |
1230 | } | |
1231 | ||
1232 | static int amdgpu_init(struct amdgpu_device *adev) | |
1233 | { | |
1234 | int i, r; | |
1235 | ||
1236 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1237 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 | 1238 | continue; |
5fc3aeeb | 1239 | r = adev->ip_blocks[i].funcs->sw_init((void *)adev); |
d38ceaf9 AD |
1240 | if (r) |
1241 | return r; | |
8faf0e08 | 1242 | adev->ip_block_status[i].sw = true; |
d38ceaf9 | 1243 | /* need to do gmc hw init early so we can allocate gpu mem */ |
5fc3aeeb | 1244 | if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) { |
d38ceaf9 AD |
1245 | r = amdgpu_vram_scratch_init(adev); |
1246 | if (r) | |
1247 | return r; | |
5fc3aeeb | 1248 | r = adev->ip_blocks[i].funcs->hw_init((void *)adev); |
d38ceaf9 AD |
1249 | if (r) |
1250 | return r; | |
1251 | r = amdgpu_wb_init(adev); | |
1252 | if (r) | |
1253 | return r; | |
8faf0e08 | 1254 | adev->ip_block_status[i].hw = true; |
d38ceaf9 AD |
1255 | } |
1256 | } | |
1257 | ||
1258 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1259 | if (!adev->ip_block_status[i].sw) |
d38ceaf9 AD |
1260 | continue; |
1261 | /* gmc hw init is done early */ | |
5fc3aeeb | 1262 | if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) |
d38ceaf9 | 1263 | continue; |
5fc3aeeb | 1264 | r = adev->ip_blocks[i].funcs->hw_init((void *)adev); |
d38ceaf9 AD |
1265 | if (r) |
1266 | return r; | |
8faf0e08 | 1267 | adev->ip_block_status[i].hw = true; |
d38ceaf9 AD |
1268 | } |
1269 | ||
1270 | return 0; | |
1271 | } | |
1272 | ||
1273 | static int amdgpu_late_init(struct amdgpu_device *adev) | |
1274 | { | |
1275 | int i = 0, r; | |
1276 | ||
1277 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1278 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 AD |
1279 | continue; |
1280 | /* enable clockgating to save power */ | |
5fc3aeeb | 1281 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, |
1282 | AMD_CG_STATE_GATE); | |
d38ceaf9 AD |
1283 | if (r) |
1284 | return r; | |
1285 | if (adev->ip_blocks[i].funcs->late_init) { | |
5fc3aeeb | 1286 | r = adev->ip_blocks[i].funcs->late_init((void *)adev); |
d38ceaf9 AD |
1287 | if (r) |
1288 | return r; | |
1289 | } | |
1290 | } | |
1291 | ||
1292 | return 0; | |
1293 | } | |
1294 | ||
1295 | static int amdgpu_fini(struct amdgpu_device *adev) | |
1296 | { | |
1297 | int i, r; | |
1298 | ||
1299 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
8faf0e08 | 1300 | if (!adev->ip_block_status[i].hw) |
d38ceaf9 | 1301 | continue; |
5fc3aeeb | 1302 | if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) { |
d38ceaf9 AD |
1303 | amdgpu_wb_fini(adev); |
1304 | amdgpu_vram_scratch_fini(adev); | |
1305 | } | |
1306 | /* ungate blocks before hw fini so that we can shutdown the blocks safely */ | |
5fc3aeeb | 1307 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, |
1308 | AMD_CG_STATE_UNGATE); | |
d38ceaf9 AD |
1309 | if (r) |
1310 | return r; | |
5fc3aeeb | 1311 | r = adev->ip_blocks[i].funcs->hw_fini((void *)adev); |
d38ceaf9 | 1312 | /* XXX handle errors */ |
8faf0e08 | 1313 | adev->ip_block_status[i].hw = false; |
d38ceaf9 AD |
1314 | } |
1315 | ||
1316 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
8faf0e08 | 1317 | if (!adev->ip_block_status[i].sw) |
d38ceaf9 | 1318 | continue; |
5fc3aeeb | 1319 | r = adev->ip_blocks[i].funcs->sw_fini((void *)adev); |
d38ceaf9 | 1320 | /* XXX handle errors */ |
8faf0e08 AD |
1321 | adev->ip_block_status[i].sw = false; |
1322 | adev->ip_block_status[i].valid = false; | |
d38ceaf9 AD |
1323 | } |
1324 | ||
1325 | return 0; | |
1326 | } | |
1327 | ||
1328 | static int amdgpu_suspend(struct amdgpu_device *adev) | |
1329 | { | |
1330 | int i, r; | |
1331 | ||
1332 | for (i = adev->num_ip_blocks - 1; i >= 0; i--) { | |
8faf0e08 | 1333 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 AD |
1334 | continue; |
1335 | /* ungate blocks so that suspend can properly shut them down */ | |
5fc3aeeb | 1336 | r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev, |
1337 | AMD_CG_STATE_UNGATE); | |
d38ceaf9 AD |
1338 | /* XXX handle errors */ |
1339 | r = adev->ip_blocks[i].funcs->suspend(adev); | |
1340 | /* XXX handle errors */ | |
1341 | } | |
1342 | ||
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | static int amdgpu_resume(struct amdgpu_device *adev) | |
1347 | { | |
1348 | int i, r; | |
1349 | ||
1350 | for (i = 0; i < adev->num_ip_blocks; i++) { | |
8faf0e08 | 1351 | if (!adev->ip_block_status[i].valid) |
d38ceaf9 AD |
1352 | continue; |
1353 | r = adev->ip_blocks[i].funcs->resume(adev); | |
1354 | if (r) | |
1355 | return r; | |
1356 | } | |
1357 | ||
1358 | return 0; | |
1359 | } | |
1360 | ||
1361 | /** | |
1362 | * amdgpu_device_init - initialize the driver | |
1363 | * | |
1364 | * @adev: amdgpu_device pointer | |
1365 | * @pdev: drm dev pointer | |
1366 | * @pdev: pci dev pointer | |
1367 | * @flags: driver flags | |
1368 | * | |
1369 | * Initializes the driver info and hw (all asics). | |
1370 | * Returns 0 for success or an error on failure. | |
1371 | * Called at driver startup. | |
1372 | */ | |
1373 | int amdgpu_device_init(struct amdgpu_device *adev, | |
1374 | struct drm_device *ddev, | |
1375 | struct pci_dev *pdev, | |
1376 | uint32_t flags) | |
1377 | { | |
1378 | int r, i; | |
1379 | bool runtime = false; | |
1380 | ||
1381 | adev->shutdown = false; | |
1382 | adev->dev = &pdev->dev; | |
1383 | adev->ddev = ddev; | |
1384 | adev->pdev = pdev; | |
1385 | adev->flags = flags; | |
2f7d10b3 | 1386 | adev->asic_type = flags & AMD_ASIC_MASK; |
d38ceaf9 AD |
1387 | adev->is_atom_bios = false; |
1388 | adev->usec_timeout = AMDGPU_MAX_USEC_TIMEOUT; | |
1389 | adev->mc.gtt_size = 512 * 1024 * 1024; | |
1390 | adev->accel_working = false; | |
1391 | adev->num_rings = 0; | |
1392 | adev->mman.buffer_funcs = NULL; | |
1393 | adev->mman.buffer_funcs_ring = NULL; | |
1394 | adev->vm_manager.vm_pte_funcs = NULL; | |
1395 | adev->vm_manager.vm_pte_funcs_ring = NULL; | |
1396 | adev->gart.gart_funcs = NULL; | |
1397 | adev->fence_context = fence_context_alloc(AMDGPU_MAX_RINGS); | |
1398 | ||
1399 | adev->smc_rreg = &amdgpu_invalid_rreg; | |
1400 | adev->smc_wreg = &amdgpu_invalid_wreg; | |
1401 | adev->pcie_rreg = &amdgpu_invalid_rreg; | |
1402 | adev->pcie_wreg = &amdgpu_invalid_wreg; | |
1403 | adev->uvd_ctx_rreg = &amdgpu_invalid_rreg; | |
1404 | adev->uvd_ctx_wreg = &amdgpu_invalid_wreg; | |
1405 | adev->didt_rreg = &amdgpu_invalid_rreg; | |
1406 | adev->didt_wreg = &amdgpu_invalid_wreg; | |
1407 | adev->audio_endpt_rreg = &amdgpu_block_invalid_rreg; | |
1408 | adev->audio_endpt_wreg = &amdgpu_block_invalid_wreg; | |
1409 | ||
3e39ab90 AD |
1410 | DRM_INFO("initializing kernel modesetting (%s 0x%04X:0x%04X 0x%04X:0x%04X 0x%02X).\n", |
1411 | amdgpu_asic_name[adev->asic_type], pdev->vendor, pdev->device, | |
1412 | pdev->subsystem_vendor, pdev->subsystem_device, pdev->revision); | |
d38ceaf9 AD |
1413 | |
1414 | /* mutex initialization are all done here so we | |
1415 | * can recall function without having locking issues */ | |
1416 | mutex_init(&adev->ring_lock); | |
1417 | atomic_set(&adev->irq.ih.lock, 0); | |
1418 | mutex_init(&adev->gem.mutex); | |
1419 | mutex_init(&adev->pm.mutex); | |
1420 | mutex_init(&adev->gfx.gpu_clock_mutex); | |
1421 | mutex_init(&adev->srbm_mutex); | |
1422 | mutex_init(&adev->grbm_idx_mutex); | |
d38ceaf9 AD |
1423 | mutex_init(&adev->mn_lock); |
1424 | hash_init(adev->mn_hash); | |
1425 | ||
1426 | amdgpu_check_arguments(adev); | |
1427 | ||
1428 | /* Registers mapping */ | |
1429 | /* TODO: block userspace mapping of io register */ | |
1430 | spin_lock_init(&adev->mmio_idx_lock); | |
1431 | spin_lock_init(&adev->smc_idx_lock); | |
1432 | spin_lock_init(&adev->pcie_idx_lock); | |
1433 | spin_lock_init(&adev->uvd_ctx_idx_lock); | |
1434 | spin_lock_init(&adev->didt_idx_lock); | |
1435 | spin_lock_init(&adev->audio_endpt_idx_lock); | |
1436 | ||
1437 | adev->rmmio_base = pci_resource_start(adev->pdev, 5); | |
1438 | adev->rmmio_size = pci_resource_len(adev->pdev, 5); | |
1439 | adev->rmmio = ioremap(adev->rmmio_base, adev->rmmio_size); | |
1440 | if (adev->rmmio == NULL) { | |
1441 | return -ENOMEM; | |
1442 | } | |
1443 | DRM_INFO("register mmio base: 0x%08X\n", (uint32_t)adev->rmmio_base); | |
1444 | DRM_INFO("register mmio size: %u\n", (unsigned)adev->rmmio_size); | |
1445 | ||
1446 | /* doorbell bar mapping */ | |
1447 | amdgpu_doorbell_init(adev); | |
1448 | ||
1449 | /* io port mapping */ | |
1450 | for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { | |
1451 | if (pci_resource_flags(adev->pdev, i) & IORESOURCE_IO) { | |
1452 | adev->rio_mem_size = pci_resource_len(adev->pdev, i); | |
1453 | adev->rio_mem = pci_iomap(adev->pdev, i, adev->rio_mem_size); | |
1454 | break; | |
1455 | } | |
1456 | } | |
1457 | if (adev->rio_mem == NULL) | |
1458 | DRM_ERROR("Unable to find PCI I/O BAR\n"); | |
1459 | ||
1460 | /* early init functions */ | |
1461 | r = amdgpu_early_init(adev); | |
1462 | if (r) | |
1463 | return r; | |
1464 | ||
1465 | /* if we have > 1 VGA cards, then disable the amdgpu VGA resources */ | |
1466 | /* this will fail for cards that aren't VGA class devices, just | |
1467 | * ignore it */ | |
1468 | vga_client_register(adev->pdev, adev, NULL, amdgpu_vga_set_decode); | |
1469 | ||
1470 | if (amdgpu_runtime_pm == 1) | |
1471 | runtime = true; | |
1472 | if (amdgpu_device_is_px(ddev)) | |
1473 | runtime = true; | |
1474 | vga_switcheroo_register_client(adev->pdev, &amdgpu_switcheroo_ops, runtime); | |
1475 | if (runtime) | |
1476 | vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain); | |
1477 | ||
1478 | /* Read BIOS */ | |
1479 | if (!amdgpu_get_bios(adev)) | |
1480 | return -EINVAL; | |
1481 | /* Must be an ATOMBIOS */ | |
1482 | if (!adev->is_atom_bios) { | |
1483 | dev_err(adev->dev, "Expecting atombios for GPU\n"); | |
1484 | return -EINVAL; | |
1485 | } | |
1486 | r = amdgpu_atombios_init(adev); | |
1487 | if (r) | |
1488 | return r; | |
1489 | ||
1490 | /* Post card if necessary */ | |
1491 | if (!amdgpu_card_posted(adev)) { | |
1492 | if (!adev->bios) { | |
1493 | dev_err(adev->dev, "Card not posted and no BIOS - ignoring\n"); | |
1494 | return -EINVAL; | |
1495 | } | |
1496 | DRM_INFO("GPU not posted. posting now...\n"); | |
1497 | amdgpu_atom_asic_init(adev->mode_info.atom_context); | |
1498 | } | |
1499 | ||
1500 | /* Initialize clocks */ | |
1501 | r = amdgpu_atombios_get_clock_info(adev); | |
1502 | if (r) | |
1503 | return r; | |
1504 | /* init i2c buses */ | |
1505 | amdgpu_atombios_i2c_init(adev); | |
1506 | ||
1507 | /* Fence driver */ | |
1508 | r = amdgpu_fence_driver_init(adev); | |
1509 | if (r) | |
1510 | return r; | |
1511 | ||
1512 | /* init the mode config */ | |
1513 | drm_mode_config_init(adev->ddev); | |
1514 | ||
1515 | r = amdgpu_init(adev); | |
1516 | if (r) { | |
1517 | amdgpu_fini(adev); | |
1518 | return r; | |
1519 | } | |
1520 | ||
1521 | adev->accel_working = true; | |
1522 | ||
1523 | amdgpu_fbdev_init(adev); | |
1524 | ||
1525 | r = amdgpu_ib_pool_init(adev); | |
1526 | if (r) { | |
1527 | dev_err(adev->dev, "IB initialization failed (%d).\n", r); | |
1528 | return r; | |
1529 | } | |
1530 | ||
47f38501 CK |
1531 | r = amdgpu_ctx_init(adev, true, &adev->kernel_ctx); |
1532 | if (r) { | |
1533 | dev_err(adev->dev, "failed to create kernel context (%d).\n", r); | |
1534 | return r; | |
23ca0e4e | 1535 | } |
d38ceaf9 AD |
1536 | r = amdgpu_ib_ring_tests(adev); |
1537 | if (r) | |
1538 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
1539 | ||
1540 | r = amdgpu_gem_debugfs_init(adev); | |
1541 | if (r) { | |
1542 | DRM_ERROR("registering gem debugfs failed (%d).\n", r); | |
1543 | } | |
1544 | ||
1545 | r = amdgpu_debugfs_regs_init(adev); | |
1546 | if (r) { | |
1547 | DRM_ERROR("registering register debugfs failed (%d).\n", r); | |
1548 | } | |
1549 | ||
1550 | if ((amdgpu_testing & 1)) { | |
1551 | if (adev->accel_working) | |
1552 | amdgpu_test_moves(adev); | |
1553 | else | |
1554 | DRM_INFO("amdgpu: acceleration disabled, skipping move tests\n"); | |
1555 | } | |
1556 | if ((amdgpu_testing & 2)) { | |
1557 | if (adev->accel_working) | |
1558 | amdgpu_test_syncing(adev); | |
1559 | else | |
1560 | DRM_INFO("amdgpu: acceleration disabled, skipping sync tests\n"); | |
1561 | } | |
1562 | if (amdgpu_benchmarking) { | |
1563 | if (adev->accel_working) | |
1564 | amdgpu_benchmark(adev, amdgpu_benchmarking); | |
1565 | else | |
1566 | DRM_INFO("amdgpu: acceleration disabled, skipping benchmarks\n"); | |
1567 | } | |
1568 | ||
1569 | /* enable clockgating, etc. after ib tests, etc. since some blocks require | |
1570 | * explicit gating rather than handling it automatically. | |
1571 | */ | |
1572 | r = amdgpu_late_init(adev); | |
1573 | if (r) | |
1574 | return r; | |
1575 | ||
1576 | return 0; | |
1577 | } | |
1578 | ||
1579 | static void amdgpu_debugfs_remove_files(struct amdgpu_device *adev); | |
1580 | ||
1581 | /** | |
1582 | * amdgpu_device_fini - tear down the driver | |
1583 | * | |
1584 | * @adev: amdgpu_device pointer | |
1585 | * | |
1586 | * Tear down the driver info (all asics). | |
1587 | * Called at driver shutdown. | |
1588 | */ | |
1589 | void amdgpu_device_fini(struct amdgpu_device *adev) | |
1590 | { | |
1591 | int r; | |
1592 | ||
1593 | DRM_INFO("amdgpu: finishing device.\n"); | |
1594 | adev->shutdown = true; | |
1595 | /* evict vram memory */ | |
1596 | amdgpu_bo_evict_vram(adev); | |
47f38501 | 1597 | amdgpu_ctx_fini(&adev->kernel_ctx); |
d38ceaf9 AD |
1598 | amdgpu_ib_pool_fini(adev); |
1599 | amdgpu_fence_driver_fini(adev); | |
1600 | amdgpu_fbdev_fini(adev); | |
1601 | r = amdgpu_fini(adev); | |
8faf0e08 AD |
1602 | kfree(adev->ip_block_status); |
1603 | adev->ip_block_status = NULL; | |
d38ceaf9 AD |
1604 | adev->accel_working = false; |
1605 | /* free i2c buses */ | |
1606 | amdgpu_i2c_fini(adev); | |
1607 | amdgpu_atombios_fini(adev); | |
1608 | kfree(adev->bios); | |
1609 | adev->bios = NULL; | |
1610 | vga_switcheroo_unregister_client(adev->pdev); | |
1611 | vga_client_register(adev->pdev, NULL, NULL, NULL); | |
1612 | if (adev->rio_mem) | |
1613 | pci_iounmap(adev->pdev, adev->rio_mem); | |
1614 | adev->rio_mem = NULL; | |
1615 | iounmap(adev->rmmio); | |
1616 | adev->rmmio = NULL; | |
1617 | amdgpu_doorbell_fini(adev); | |
1618 | amdgpu_debugfs_regs_cleanup(adev); | |
1619 | amdgpu_debugfs_remove_files(adev); | |
1620 | } | |
1621 | ||
1622 | ||
1623 | /* | |
1624 | * Suspend & resume. | |
1625 | */ | |
1626 | /** | |
1627 | * amdgpu_suspend_kms - initiate device suspend | |
1628 | * | |
1629 | * @pdev: drm dev pointer | |
1630 | * @state: suspend state | |
1631 | * | |
1632 | * Puts the hw in the suspend state (all asics). | |
1633 | * Returns 0 for success or an error on failure. | |
1634 | * Called at driver suspend. | |
1635 | */ | |
1636 | int amdgpu_suspend_kms(struct drm_device *dev, bool suspend, bool fbcon) | |
1637 | { | |
1638 | struct amdgpu_device *adev; | |
1639 | struct drm_crtc *crtc; | |
1640 | struct drm_connector *connector; | |
5ceb54c6 | 1641 | int r; |
d38ceaf9 AD |
1642 | |
1643 | if (dev == NULL || dev->dev_private == NULL) { | |
1644 | return -ENODEV; | |
1645 | } | |
1646 | ||
1647 | adev = dev->dev_private; | |
1648 | ||
1649 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
1650 | return 0; | |
1651 | ||
1652 | drm_kms_helper_poll_disable(dev); | |
1653 | ||
1654 | /* turn off display hw */ | |
4c7fbc39 | 1655 | drm_modeset_lock_all(dev); |
d38ceaf9 AD |
1656 | list_for_each_entry(connector, &dev->mode_config.connector_list, head) { |
1657 | drm_helper_connector_dpms(connector, DRM_MODE_DPMS_OFF); | |
1658 | } | |
4c7fbc39 | 1659 | drm_modeset_unlock_all(dev); |
d38ceaf9 | 1660 | |
756e6880 | 1661 | /* unpin the front buffers and cursors */ |
d38ceaf9 | 1662 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { |
756e6880 | 1663 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); |
d38ceaf9 AD |
1664 | struct amdgpu_framebuffer *rfb = to_amdgpu_framebuffer(crtc->primary->fb); |
1665 | struct amdgpu_bo *robj; | |
1666 | ||
756e6880 AD |
1667 | if (amdgpu_crtc->cursor_bo) { |
1668 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); | |
1669 | r = amdgpu_bo_reserve(aobj, false); | |
1670 | if (r == 0) { | |
1671 | amdgpu_bo_unpin(aobj); | |
1672 | amdgpu_bo_unreserve(aobj); | |
1673 | } | |
1674 | } | |
1675 | ||
d38ceaf9 AD |
1676 | if (rfb == NULL || rfb->obj == NULL) { |
1677 | continue; | |
1678 | } | |
1679 | robj = gem_to_amdgpu_bo(rfb->obj); | |
1680 | /* don't unpin kernel fb objects */ | |
1681 | if (!amdgpu_fbdev_robj_is_fb(adev, robj)) { | |
1682 | r = amdgpu_bo_reserve(robj, false); | |
1683 | if (r == 0) { | |
1684 | amdgpu_bo_unpin(robj); | |
1685 | amdgpu_bo_unreserve(robj); | |
1686 | } | |
1687 | } | |
1688 | } | |
1689 | /* evict vram memory */ | |
1690 | amdgpu_bo_evict_vram(adev); | |
1691 | ||
5ceb54c6 | 1692 | amdgpu_fence_driver_suspend(adev); |
d38ceaf9 AD |
1693 | |
1694 | r = amdgpu_suspend(adev); | |
1695 | ||
1696 | /* evict remaining vram memory */ | |
1697 | amdgpu_bo_evict_vram(adev); | |
1698 | ||
1699 | pci_save_state(dev->pdev); | |
1700 | if (suspend) { | |
1701 | /* Shut down the device */ | |
1702 | pci_disable_device(dev->pdev); | |
1703 | pci_set_power_state(dev->pdev, PCI_D3hot); | |
1704 | } | |
1705 | ||
1706 | if (fbcon) { | |
1707 | console_lock(); | |
1708 | amdgpu_fbdev_set_suspend(adev, 1); | |
1709 | console_unlock(); | |
1710 | } | |
1711 | return 0; | |
1712 | } | |
1713 | ||
1714 | /** | |
1715 | * amdgpu_resume_kms - initiate device resume | |
1716 | * | |
1717 | * @pdev: drm dev pointer | |
1718 | * | |
1719 | * Bring the hw back to operating state (all asics). | |
1720 | * Returns 0 for success or an error on failure. | |
1721 | * Called at driver resume. | |
1722 | */ | |
1723 | int amdgpu_resume_kms(struct drm_device *dev, bool resume, bool fbcon) | |
1724 | { | |
1725 | struct drm_connector *connector; | |
1726 | struct amdgpu_device *adev = dev->dev_private; | |
756e6880 | 1727 | struct drm_crtc *crtc; |
d38ceaf9 AD |
1728 | int r; |
1729 | ||
1730 | if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) | |
1731 | return 0; | |
1732 | ||
1733 | if (fbcon) { | |
1734 | console_lock(); | |
1735 | } | |
1736 | if (resume) { | |
1737 | pci_set_power_state(dev->pdev, PCI_D0); | |
1738 | pci_restore_state(dev->pdev); | |
1739 | if (pci_enable_device(dev->pdev)) { | |
1740 | if (fbcon) | |
1741 | console_unlock(); | |
1742 | return -1; | |
1743 | } | |
1744 | } | |
1745 | ||
1746 | /* post card */ | |
1747 | amdgpu_atom_asic_init(adev->mode_info.atom_context); | |
1748 | ||
1749 | r = amdgpu_resume(adev); | |
1750 | ||
5ceb54c6 AD |
1751 | amdgpu_fence_driver_resume(adev); |
1752 | ||
d38ceaf9 AD |
1753 | r = amdgpu_ib_ring_tests(adev); |
1754 | if (r) | |
1755 | DRM_ERROR("ib ring test failed (%d).\n", r); | |
1756 | ||
1757 | r = amdgpu_late_init(adev); | |
1758 | if (r) | |
1759 | return r; | |
1760 | ||
756e6880 AD |
1761 | /* pin cursors */ |
1762 | list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) { | |
1763 | struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); | |
1764 | ||
1765 | if (amdgpu_crtc->cursor_bo) { | |
1766 | struct amdgpu_bo *aobj = gem_to_amdgpu_bo(amdgpu_crtc->cursor_bo); | |
1767 | r = amdgpu_bo_reserve(aobj, false); | |
1768 | if (r == 0) { | |
1769 | r = amdgpu_bo_pin(aobj, | |
1770 | AMDGPU_GEM_DOMAIN_VRAM, | |
1771 | &amdgpu_crtc->cursor_addr); | |
1772 | if (r != 0) | |
1773 | DRM_ERROR("Failed to pin cursor BO (%d)\n", r); | |
1774 | amdgpu_bo_unreserve(aobj); | |
1775 | } | |
1776 | } | |
1777 | } | |
1778 | ||
d38ceaf9 AD |
1779 | /* blat the mode back in */ |
1780 | if (fbcon) { | |
1781 | drm_helper_resume_force_mode(dev); | |
1782 | /* turn on display hw */ | |
4c7fbc39 | 1783 | drm_modeset_lock_all(dev); |
d38ceaf9 AD |
1784 | list_for_each_entry(connector, &dev->mode_config.connector_list, head) { |
1785 | drm_helper_connector_dpms(connector, DRM_MODE_DPMS_ON); | |
1786 | } | |
4c7fbc39 | 1787 | drm_modeset_unlock_all(dev); |
d38ceaf9 AD |
1788 | } |
1789 | ||
1790 | drm_kms_helper_poll_enable(dev); | |
1791 | ||
1792 | if (fbcon) { | |
1793 | amdgpu_fbdev_set_suspend(adev, 0); | |
1794 | console_unlock(); | |
1795 | } | |
1796 | ||
1797 | return 0; | |
1798 | } | |
1799 | ||
1800 | /** | |
1801 | * amdgpu_gpu_reset - reset the asic | |
1802 | * | |
1803 | * @adev: amdgpu device pointer | |
1804 | * | |
1805 | * Attempt the reset the GPU if it has hung (all asics). | |
1806 | * Returns 0 for success or an error on failure. | |
1807 | */ | |
1808 | int amdgpu_gpu_reset(struct amdgpu_device *adev) | |
1809 | { | |
1810 | unsigned ring_sizes[AMDGPU_MAX_RINGS]; | |
1811 | uint32_t *ring_data[AMDGPU_MAX_RINGS]; | |
1812 | ||
1813 | bool saved = false; | |
1814 | ||
1815 | int i, r; | |
1816 | int resched; | |
1817 | ||
d94aed5a | 1818 | atomic_inc(&adev->gpu_reset_counter); |
d38ceaf9 AD |
1819 | |
1820 | /* block TTM */ | |
1821 | resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev); | |
1822 | ||
1823 | r = amdgpu_suspend(adev); | |
1824 | ||
1825 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
1826 | struct amdgpu_ring *ring = adev->rings[i]; | |
1827 | if (!ring) | |
1828 | continue; | |
1829 | ||
1830 | ring_sizes[i] = amdgpu_ring_backup(ring, &ring_data[i]); | |
1831 | if (ring_sizes[i]) { | |
1832 | saved = true; | |
1833 | dev_info(adev->dev, "Saved %d dwords of commands " | |
1834 | "on ring %d.\n", ring_sizes[i], i); | |
1835 | } | |
1836 | } | |
1837 | ||
1838 | retry: | |
1839 | r = amdgpu_asic_reset(adev); | |
1840 | if (!r) { | |
1841 | dev_info(adev->dev, "GPU reset succeeded, trying to resume\n"); | |
1842 | r = amdgpu_resume(adev); | |
1843 | } | |
1844 | ||
1845 | if (!r) { | |
1846 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
1847 | struct amdgpu_ring *ring = adev->rings[i]; | |
1848 | if (!ring) | |
1849 | continue; | |
1850 | ||
1851 | amdgpu_ring_restore(ring, ring_sizes[i], ring_data[i]); | |
1852 | ring_sizes[i] = 0; | |
1853 | ring_data[i] = NULL; | |
1854 | } | |
1855 | ||
1856 | r = amdgpu_ib_ring_tests(adev); | |
1857 | if (r) { | |
1858 | dev_err(adev->dev, "ib ring test failed (%d).\n", r); | |
1859 | if (saved) { | |
1860 | saved = false; | |
1861 | r = amdgpu_suspend(adev); | |
1862 | goto retry; | |
1863 | } | |
1864 | } | |
1865 | } else { | |
1866 | amdgpu_fence_driver_force_completion(adev); | |
1867 | for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { | |
1868 | if (adev->rings[i]) | |
1869 | kfree(ring_data[i]); | |
1870 | } | |
1871 | } | |
1872 | ||
1873 | drm_helper_resume_force_mode(adev->ddev); | |
1874 | ||
1875 | ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched); | |
1876 | if (r) { | |
1877 | /* bad news, how to tell it to userspace ? */ | |
1878 | dev_info(adev->dev, "GPU reset failed\n"); | |
1879 | } | |
1880 | ||
d38ceaf9 AD |
1881 | return r; |
1882 | } | |
1883 | ||
1884 | ||
1885 | /* | |
1886 | * Debugfs | |
1887 | */ | |
1888 | int amdgpu_debugfs_add_files(struct amdgpu_device *adev, | |
1889 | struct drm_info_list *files, | |
1890 | unsigned nfiles) | |
1891 | { | |
1892 | unsigned i; | |
1893 | ||
1894 | for (i = 0; i < adev->debugfs_count; i++) { | |
1895 | if (adev->debugfs[i].files == files) { | |
1896 | /* Already registered */ | |
1897 | return 0; | |
1898 | } | |
1899 | } | |
1900 | ||
1901 | i = adev->debugfs_count + 1; | |
1902 | if (i > AMDGPU_DEBUGFS_MAX_COMPONENTS) { | |
1903 | DRM_ERROR("Reached maximum number of debugfs components.\n"); | |
1904 | DRM_ERROR("Report so we increase " | |
1905 | "AMDGPU_DEBUGFS_MAX_COMPONENTS.\n"); | |
1906 | return -EINVAL; | |
1907 | } | |
1908 | adev->debugfs[adev->debugfs_count].files = files; | |
1909 | adev->debugfs[adev->debugfs_count].num_files = nfiles; | |
1910 | adev->debugfs_count = i; | |
1911 | #if defined(CONFIG_DEBUG_FS) | |
1912 | drm_debugfs_create_files(files, nfiles, | |
1913 | adev->ddev->control->debugfs_root, | |
1914 | adev->ddev->control); | |
1915 | drm_debugfs_create_files(files, nfiles, | |
1916 | adev->ddev->primary->debugfs_root, | |
1917 | adev->ddev->primary); | |
1918 | #endif | |
1919 | return 0; | |
1920 | } | |
1921 | ||
1922 | static void amdgpu_debugfs_remove_files(struct amdgpu_device *adev) | |
1923 | { | |
1924 | #if defined(CONFIG_DEBUG_FS) | |
1925 | unsigned i; | |
1926 | ||
1927 | for (i = 0; i < adev->debugfs_count; i++) { | |
1928 | drm_debugfs_remove_files(adev->debugfs[i].files, | |
1929 | adev->debugfs[i].num_files, | |
1930 | adev->ddev->control); | |
1931 | drm_debugfs_remove_files(adev->debugfs[i].files, | |
1932 | adev->debugfs[i].num_files, | |
1933 | adev->ddev->primary); | |
1934 | } | |
1935 | #endif | |
1936 | } | |
1937 | ||
1938 | #if defined(CONFIG_DEBUG_FS) | |
1939 | ||
1940 | static ssize_t amdgpu_debugfs_regs_read(struct file *f, char __user *buf, | |
1941 | size_t size, loff_t *pos) | |
1942 | { | |
1943 | struct amdgpu_device *adev = f->f_inode->i_private; | |
1944 | ssize_t result = 0; | |
1945 | int r; | |
1946 | ||
1947 | if (size & 0x3 || *pos & 0x3) | |
1948 | return -EINVAL; | |
1949 | ||
1950 | while (size) { | |
1951 | uint32_t value; | |
1952 | ||
1953 | if (*pos > adev->rmmio_size) | |
1954 | return result; | |
1955 | ||
1956 | value = RREG32(*pos >> 2); | |
1957 | r = put_user(value, (uint32_t *)buf); | |
1958 | if (r) | |
1959 | return r; | |
1960 | ||
1961 | result += 4; | |
1962 | buf += 4; | |
1963 | *pos += 4; | |
1964 | size -= 4; | |
1965 | } | |
1966 | ||
1967 | return result; | |
1968 | } | |
1969 | ||
1970 | static ssize_t amdgpu_debugfs_regs_write(struct file *f, const char __user *buf, | |
1971 | size_t size, loff_t *pos) | |
1972 | { | |
1973 | struct amdgpu_device *adev = f->f_inode->i_private; | |
1974 | ssize_t result = 0; | |
1975 | int r; | |
1976 | ||
1977 | if (size & 0x3 || *pos & 0x3) | |
1978 | return -EINVAL; | |
1979 | ||
1980 | while (size) { | |
1981 | uint32_t value; | |
1982 | ||
1983 | if (*pos > adev->rmmio_size) | |
1984 | return result; | |
1985 | ||
1986 | r = get_user(value, (uint32_t *)buf); | |
1987 | if (r) | |
1988 | return r; | |
1989 | ||
1990 | WREG32(*pos >> 2, value); | |
1991 | ||
1992 | result += 4; | |
1993 | buf += 4; | |
1994 | *pos += 4; | |
1995 | size -= 4; | |
1996 | } | |
1997 | ||
1998 | return result; | |
1999 | } | |
2000 | ||
2001 | static const struct file_operations amdgpu_debugfs_regs_fops = { | |
2002 | .owner = THIS_MODULE, | |
2003 | .read = amdgpu_debugfs_regs_read, | |
2004 | .write = amdgpu_debugfs_regs_write, | |
2005 | .llseek = default_llseek | |
2006 | }; | |
2007 | ||
2008 | static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev) | |
2009 | { | |
2010 | struct drm_minor *minor = adev->ddev->primary; | |
2011 | struct dentry *ent, *root = minor->debugfs_root; | |
2012 | ||
2013 | ent = debugfs_create_file("amdgpu_regs", S_IFREG | S_IRUGO, root, | |
2014 | adev, &amdgpu_debugfs_regs_fops); | |
2015 | if (IS_ERR(ent)) | |
2016 | return PTR_ERR(ent); | |
2017 | i_size_write(ent->d_inode, adev->rmmio_size); | |
2018 | adev->debugfs_regs = ent; | |
2019 | ||
2020 | return 0; | |
2021 | } | |
2022 | ||
2023 | static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev) | |
2024 | { | |
2025 | debugfs_remove(adev->debugfs_regs); | |
2026 | adev->debugfs_regs = NULL; | |
2027 | } | |
2028 | ||
2029 | int amdgpu_debugfs_init(struct drm_minor *minor) | |
2030 | { | |
2031 | return 0; | |
2032 | } | |
2033 | ||
2034 | void amdgpu_debugfs_cleanup(struct drm_minor *minor) | |
2035 | { | |
2036 | } | |
7cebc728 AK |
2037 | #else |
2038 | static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev) | |
2039 | { | |
2040 | return 0; | |
2041 | } | |
2042 | static void amdgpu_debugfs_regs_cleanup(struct amdgpu_device *adev) { } | |
d38ceaf9 | 2043 | #endif |