]>
Commit | Line | Data |
---|---|---|
aaa36a97 AD |
1 | /* |
2 | * Copyright 2014 Advanced Micro Devices, Inc. | |
3 | * | |
4 | * Permission is hereby granted, free of charge, to any person obtaining a | |
5 | * copy of this software and associated documentation files (the "Software"), | |
6 | * to deal in the Software without restriction, including without limitation | |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, | |
8 | * and/or sell copies of the Software, and to permit persons to whom the | |
9 | * Software is furnished to do so, subject to the following conditions: | |
10 | * | |
11 | * The above copyright notice and this permission notice shall be included in | |
12 | * all copies or substantial portions of the Software. | |
13 | * | |
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
20 | * OTHER DEALINGS IN THE SOFTWARE. | |
21 | * | |
22 | * Authors: Christian König <christian.koenig@amd.com> | |
23 | */ | |
24 | ||
25 | #include <linux/firmware.h> | |
26 | #include <drm/drmP.h> | |
27 | #include "amdgpu.h" | |
28 | #include "amdgpu_uvd.h" | |
29 | #include "vid.h" | |
30 | #include "uvd/uvd_5_0_d.h" | |
31 | #include "uvd/uvd_5_0_sh_mask.h" | |
32 | #include "oss/oss_2_0_d.h" | |
33 | #include "oss/oss_2_0_sh_mask.h" | |
d5b4e25d | 34 | #include "bif/bif_5_0_d.h" |
be3ecca7 | 35 | #include "vi.h" |
aaa36a97 AD |
36 | |
37 | static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev); | |
38 | static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev); | |
39 | static int uvd_v5_0_start(struct amdgpu_device *adev); | |
40 | static void uvd_v5_0_stop(struct amdgpu_device *adev); | |
41 | ||
42 | /** | |
43 | * uvd_v5_0_ring_get_rptr - get read pointer | |
44 | * | |
45 | * @ring: amdgpu_ring pointer | |
46 | * | |
47 | * Returns the current hardware read pointer | |
48 | */ | |
49 | static uint32_t uvd_v5_0_ring_get_rptr(struct amdgpu_ring *ring) | |
50 | { | |
51 | struct amdgpu_device *adev = ring->adev; | |
52 | ||
53 | return RREG32(mmUVD_RBC_RB_RPTR); | |
54 | } | |
55 | ||
56 | /** | |
57 | * uvd_v5_0_ring_get_wptr - get write pointer | |
58 | * | |
59 | * @ring: amdgpu_ring pointer | |
60 | * | |
61 | * Returns the current hardware write pointer | |
62 | */ | |
63 | static uint32_t uvd_v5_0_ring_get_wptr(struct amdgpu_ring *ring) | |
64 | { | |
65 | struct amdgpu_device *adev = ring->adev; | |
66 | ||
67 | return RREG32(mmUVD_RBC_RB_WPTR); | |
68 | } | |
69 | ||
70 | /** | |
71 | * uvd_v5_0_ring_set_wptr - set write pointer | |
72 | * | |
73 | * @ring: amdgpu_ring pointer | |
74 | * | |
75 | * Commits the write pointer to the hardware | |
76 | */ | |
77 | static void uvd_v5_0_ring_set_wptr(struct amdgpu_ring *ring) | |
78 | { | |
79 | struct amdgpu_device *adev = ring->adev; | |
80 | ||
81 | WREG32(mmUVD_RBC_RB_WPTR, ring->wptr); | |
82 | } | |
83 | ||
5fc3aeeb | 84 | static int uvd_v5_0_early_init(void *handle) |
aaa36a97 | 85 | { |
5fc3aeeb | 86 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
87 | ||
aaa36a97 AD |
88 | uvd_v5_0_set_ring_funcs(adev); |
89 | uvd_v5_0_set_irq_funcs(adev); | |
90 | ||
91 | return 0; | |
92 | } | |
93 | ||
5fc3aeeb | 94 | static int uvd_v5_0_sw_init(void *handle) |
aaa36a97 AD |
95 | { |
96 | struct amdgpu_ring *ring; | |
5fc3aeeb | 97 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
aaa36a97 AD |
98 | int r; |
99 | ||
100 | /* UVD TRAP */ | |
101 | r = amdgpu_irq_add_id(adev, 124, &adev->uvd.irq); | |
102 | if (r) | |
103 | return r; | |
104 | ||
105 | r = amdgpu_uvd_sw_init(adev); | |
106 | if (r) | |
107 | return r; | |
108 | ||
109 | r = amdgpu_uvd_resume(adev); | |
110 | if (r) | |
111 | return r; | |
112 | ||
113 | ring = &adev->uvd.ring; | |
114 | sprintf(ring->name, "uvd"); | |
a3f1cf35 | 115 | r = amdgpu_ring_init(adev, ring, 512, CP_PACKET2, 0xf, |
aaa36a97 AD |
116 | &adev->uvd.irq, 0, AMDGPU_RING_TYPE_UVD); |
117 | ||
118 | return r; | |
119 | } | |
120 | ||
5fc3aeeb | 121 | static int uvd_v5_0_sw_fini(void *handle) |
aaa36a97 AD |
122 | { |
123 | int r; | |
5fc3aeeb | 124 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
aaa36a97 AD |
125 | |
126 | r = amdgpu_uvd_suspend(adev); | |
127 | if (r) | |
128 | return r; | |
129 | ||
130 | r = amdgpu_uvd_sw_fini(adev); | |
131 | if (r) | |
132 | return r; | |
133 | ||
134 | return r; | |
135 | } | |
136 | ||
137 | /** | |
138 | * uvd_v5_0_hw_init - start and test UVD block | |
139 | * | |
140 | * @adev: amdgpu_device pointer | |
141 | * | |
142 | * Initialize the hardware, boot up the VCPU and do some testing | |
143 | */ | |
5fc3aeeb | 144 | static int uvd_v5_0_hw_init(void *handle) |
aaa36a97 | 145 | { |
5fc3aeeb | 146 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
aaa36a97 AD |
147 | struct amdgpu_ring *ring = &adev->uvd.ring; |
148 | uint32_t tmp; | |
149 | int r; | |
150 | ||
151 | /* raise clocks while booting up the VCPU */ | |
152 | amdgpu_asic_set_uvd_clocks(adev, 53300, 40000); | |
153 | ||
154 | r = uvd_v5_0_start(adev); | |
155 | if (r) | |
156 | goto done; | |
157 | ||
158 | ring->ready = true; | |
159 | r = amdgpu_ring_test_ring(ring); | |
160 | if (r) { | |
161 | ring->ready = false; | |
162 | goto done; | |
163 | } | |
164 | ||
a27de35c | 165 | r = amdgpu_ring_alloc(ring, 10); |
aaa36a97 AD |
166 | if (r) { |
167 | DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r); | |
168 | goto done; | |
169 | } | |
170 | ||
171 | tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0); | |
172 | amdgpu_ring_write(ring, tmp); | |
173 | amdgpu_ring_write(ring, 0xFFFFF); | |
174 | ||
175 | tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0); | |
176 | amdgpu_ring_write(ring, tmp); | |
177 | amdgpu_ring_write(ring, 0xFFFFF); | |
178 | ||
179 | tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0); | |
180 | amdgpu_ring_write(ring, tmp); | |
181 | amdgpu_ring_write(ring, 0xFFFFF); | |
182 | ||
183 | /* Clear timeout status bits */ | |
184 | amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0)); | |
185 | amdgpu_ring_write(ring, 0x8); | |
186 | ||
187 | amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0)); | |
188 | amdgpu_ring_write(ring, 3); | |
189 | ||
a27de35c | 190 | amdgpu_ring_commit(ring); |
aaa36a97 AD |
191 | |
192 | done: | |
193 | /* lower clocks again */ | |
194 | amdgpu_asic_set_uvd_clocks(adev, 0, 0); | |
195 | ||
196 | if (!r) | |
197 | DRM_INFO("UVD initialized successfully.\n"); | |
198 | ||
199 | return r; | |
200 | } | |
201 | ||
202 | /** | |
203 | * uvd_v5_0_hw_fini - stop the hardware block | |
204 | * | |
205 | * @adev: amdgpu_device pointer | |
206 | * | |
207 | * Stop the UVD block, mark ring as not ready any more | |
208 | */ | |
5fc3aeeb | 209 | static int uvd_v5_0_hw_fini(void *handle) |
aaa36a97 | 210 | { |
5fc3aeeb | 211 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
aaa36a97 AD |
212 | struct amdgpu_ring *ring = &adev->uvd.ring; |
213 | ||
214 | uvd_v5_0_stop(adev); | |
215 | ring->ready = false; | |
216 | ||
217 | return 0; | |
218 | } | |
219 | ||
5fc3aeeb | 220 | static int uvd_v5_0_suspend(void *handle) |
aaa36a97 AD |
221 | { |
222 | int r; | |
5fc3aeeb | 223 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
aaa36a97 | 224 | |
3f99dd81 | 225 | r = uvd_v5_0_hw_fini(adev); |
aaa36a97 AD |
226 | if (r) |
227 | return r; | |
228 | ||
3f99dd81 | 229 | r = amdgpu_uvd_suspend(adev); |
aaa36a97 AD |
230 | if (r) |
231 | return r; | |
232 | ||
233 | return r; | |
234 | } | |
235 | ||
5fc3aeeb | 236 | static int uvd_v5_0_resume(void *handle) |
aaa36a97 AD |
237 | { |
238 | int r; | |
5fc3aeeb | 239 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
aaa36a97 AD |
240 | |
241 | r = amdgpu_uvd_resume(adev); | |
242 | if (r) | |
243 | return r; | |
244 | ||
245 | r = uvd_v5_0_hw_init(adev); | |
246 | if (r) | |
247 | return r; | |
248 | ||
249 | return r; | |
250 | } | |
251 | ||
252 | /** | |
253 | * uvd_v5_0_mc_resume - memory controller programming | |
254 | * | |
255 | * @adev: amdgpu_device pointer | |
256 | * | |
257 | * Let the UVD memory controller know it's offsets | |
258 | */ | |
259 | static void uvd_v5_0_mc_resume(struct amdgpu_device *adev) | |
260 | { | |
261 | uint64_t offset; | |
262 | uint32_t size; | |
263 | ||
264 | /* programm memory controller bits 0-27 */ | |
265 | WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW, | |
266 | lower_32_bits(adev->uvd.gpu_addr)); | |
267 | WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH, | |
268 | upper_32_bits(adev->uvd.gpu_addr)); | |
269 | ||
270 | offset = AMDGPU_UVD_FIRMWARE_OFFSET; | |
271 | size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4); | |
272 | WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3); | |
273 | WREG32(mmUVD_VCPU_CACHE_SIZE0, size); | |
274 | ||
275 | offset += size; | |
c0365541 | 276 | size = AMDGPU_UVD_HEAP_SIZE; |
aaa36a97 AD |
277 | WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3); |
278 | WREG32(mmUVD_VCPU_CACHE_SIZE1, size); | |
279 | ||
280 | offset += size; | |
c0365541 AN |
281 | size = AMDGPU_UVD_STACK_SIZE + |
282 | (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles); | |
aaa36a97 AD |
283 | WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3); |
284 | WREG32(mmUVD_VCPU_CACHE_SIZE2, size); | |
549300ce AD |
285 | |
286 | WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config); | |
287 | WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); | |
288 | WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config); | |
aaa36a97 AD |
289 | } |
290 | ||
291 | /** | |
292 | * uvd_v5_0_start - start UVD block | |
293 | * | |
294 | * @adev: amdgpu_device pointer | |
295 | * | |
296 | * Setup and start the UVD block | |
297 | */ | |
298 | static int uvd_v5_0_start(struct amdgpu_device *adev) | |
299 | { | |
300 | struct amdgpu_ring *ring = &adev->uvd.ring; | |
301 | uint32_t rb_bufsz, tmp; | |
302 | uint32_t lmi_swap_cntl; | |
303 | uint32_t mp_swap_cntl; | |
304 | int i, j, r; | |
305 | ||
306 | /*disable DPG */ | |
307 | WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2)); | |
308 | ||
309 | /* disable byte swapping */ | |
310 | lmi_swap_cntl = 0; | |
311 | mp_swap_cntl = 0; | |
312 | ||
313 | uvd_v5_0_mc_resume(adev); | |
314 | ||
315 | /* disable clock gating */ | |
316 | WREG32(mmUVD_CGC_GATE, 0); | |
317 | ||
318 | /* disable interupt */ | |
319 | WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1)); | |
320 | ||
321 | /* stall UMC and register bus before resetting VCPU */ | |
322 | WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8)); | |
323 | mdelay(1); | |
324 | ||
325 | /* put LMI, VCPU, RBC etc... into reset */ | |
326 | WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK | | |
327 | UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK | | |
328 | UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK | | |
329 | UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK | | |
330 | UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK); | |
331 | mdelay(5); | |
332 | ||
333 | /* take UVD block out of reset */ | |
334 | WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK); | |
335 | mdelay(5); | |
336 | ||
337 | /* initialize UVD memory controller */ | |
338 | WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) | | |
339 | (1 << 21) | (1 << 9) | (1 << 20)); | |
340 | ||
341 | #ifdef __BIG_ENDIAN | |
342 | /* swap (8 in 32) RB and IB */ | |
343 | lmi_swap_cntl = 0xa; | |
344 | mp_swap_cntl = 0; | |
345 | #endif | |
346 | WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl); | |
347 | WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl); | |
348 | ||
349 | WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040); | |
350 | WREG32(mmUVD_MPC_SET_MUXA1, 0x0); | |
351 | WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040); | |
352 | WREG32(mmUVD_MPC_SET_MUXB1, 0x0); | |
353 | WREG32(mmUVD_MPC_SET_ALU, 0); | |
354 | WREG32(mmUVD_MPC_SET_MUX, 0x88); | |
355 | ||
356 | /* take all subblocks out of reset, except VCPU */ | |
357 | WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); | |
358 | mdelay(5); | |
359 | ||
360 | /* enable VCPU clock */ | |
361 | WREG32(mmUVD_VCPU_CNTL, 1 << 9); | |
362 | ||
363 | /* enable UMC */ | |
364 | WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8)); | |
365 | ||
366 | /* boot up the VCPU */ | |
367 | WREG32(mmUVD_SOFT_RESET, 0); | |
368 | mdelay(10); | |
369 | ||
370 | for (i = 0; i < 10; ++i) { | |
371 | uint32_t status; | |
372 | for (j = 0; j < 100; ++j) { | |
373 | status = RREG32(mmUVD_STATUS); | |
374 | if (status & 2) | |
375 | break; | |
376 | mdelay(10); | |
377 | } | |
378 | r = 0; | |
379 | if (status & 2) | |
380 | break; | |
381 | ||
382 | DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n"); | |
383 | WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK, | |
384 | ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); | |
385 | mdelay(10); | |
386 | WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); | |
387 | mdelay(10); | |
388 | r = -1; | |
389 | } | |
390 | ||
391 | if (r) { | |
392 | DRM_ERROR("UVD not responding, giving up!!!\n"); | |
393 | return r; | |
394 | } | |
395 | /* enable master interrupt */ | |
396 | WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1)); | |
397 | ||
398 | /* clear the bit 4 of UVD_STATUS */ | |
399 | WREG32_P(mmUVD_STATUS, 0, ~(2 << 1)); | |
400 | ||
401 | rb_bufsz = order_base_2(ring->ring_size); | |
402 | tmp = 0; | |
403 | tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz); | |
404 | tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1); | |
405 | tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1); | |
406 | tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0); | |
407 | tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1); | |
408 | tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1); | |
409 | /* force RBC into idle state */ | |
410 | WREG32(mmUVD_RBC_RB_CNTL, tmp); | |
411 | ||
412 | /* set the write pointer delay */ | |
413 | WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0); | |
414 | ||
415 | /* set the wb address */ | |
416 | WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2)); | |
417 | ||
418 | /* programm the RB_BASE for ring buffer */ | |
419 | WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW, | |
420 | lower_32_bits(ring->gpu_addr)); | |
421 | WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH, | |
422 | upper_32_bits(ring->gpu_addr)); | |
423 | ||
424 | /* Initialize the ring buffer's read and write pointers */ | |
425 | WREG32(mmUVD_RBC_RB_RPTR, 0); | |
426 | ||
427 | ring->wptr = RREG32(mmUVD_RBC_RB_RPTR); | |
428 | WREG32(mmUVD_RBC_RB_WPTR, ring->wptr); | |
429 | ||
430 | WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK); | |
431 | ||
432 | return 0; | |
433 | } | |
434 | ||
435 | /** | |
436 | * uvd_v5_0_stop - stop UVD block | |
437 | * | |
438 | * @adev: amdgpu_device pointer | |
439 | * | |
440 | * stop the UVD block | |
441 | */ | |
442 | static void uvd_v5_0_stop(struct amdgpu_device *adev) | |
443 | { | |
444 | /* force RBC into idle state */ | |
445 | WREG32(mmUVD_RBC_RB_CNTL, 0x11010101); | |
446 | ||
447 | /* Stall UMC and register bus before resetting VCPU */ | |
448 | WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8)); | |
449 | mdelay(1); | |
450 | ||
451 | /* put VCPU into reset */ | |
452 | WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK); | |
453 | mdelay(5); | |
454 | ||
455 | /* disable VCPU clock */ | |
456 | WREG32(mmUVD_VCPU_CNTL, 0x0); | |
457 | ||
458 | /* Unstall UMC and register bus */ | |
459 | WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8)); | |
460 | } | |
461 | ||
462 | /** | |
463 | * uvd_v5_0_ring_emit_fence - emit an fence & trap command | |
464 | * | |
465 | * @ring: amdgpu_ring pointer | |
466 | * @fence: fence to emit | |
467 | * | |
468 | * Write a fence and a trap command to the ring. | |
469 | */ | |
470 | static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, | |
890ee23f | 471 | unsigned flags) |
aaa36a97 | 472 | { |
890ee23f | 473 | WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); |
aaa36a97 AD |
474 | |
475 | amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); | |
476 | amdgpu_ring_write(ring, seq); | |
477 | amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0)); | |
478 | amdgpu_ring_write(ring, addr & 0xffffffff); | |
479 | amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0)); | |
480 | amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff); | |
481 | amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0)); | |
482 | amdgpu_ring_write(ring, 0); | |
483 | ||
484 | amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0)); | |
485 | amdgpu_ring_write(ring, 0); | |
486 | amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0)); | |
487 | amdgpu_ring_write(ring, 0); | |
488 | amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0)); | |
489 | amdgpu_ring_write(ring, 2); | |
490 | } | |
491 | ||
d5b4e25d CK |
492 | /** |
493 | * uvd_v5_0_ring_emit_hdp_flush - emit an hdp flush | |
494 | * | |
495 | * @ring: amdgpu_ring pointer | |
496 | * | |
497 | * Emits an hdp flush. | |
498 | */ | |
499 | static void uvd_v5_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) | |
500 | { | |
501 | amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0)); | |
502 | amdgpu_ring_write(ring, 0); | |
503 | } | |
504 | ||
505 | /** | |
506 | * uvd_v5_0_ring_hdp_invalidate - emit an hdp invalidate | |
507 | * | |
508 | * @ring: amdgpu_ring pointer | |
509 | * | |
510 | * Emits an hdp invalidate. | |
511 | */ | |
512 | static void uvd_v5_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) | |
513 | { | |
514 | amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0)); | |
515 | amdgpu_ring_write(ring, 1); | |
516 | } | |
517 | ||
aaa36a97 AD |
518 | /** |
519 | * uvd_v5_0_ring_test_ring - register write test | |
520 | * | |
521 | * @ring: amdgpu_ring pointer | |
522 | * | |
523 | * Test if we can successfully write to the context register | |
524 | */ | |
525 | static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring) | |
526 | { | |
527 | struct amdgpu_device *adev = ring->adev; | |
528 | uint32_t tmp = 0; | |
529 | unsigned i; | |
530 | int r; | |
531 | ||
532 | WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD); | |
a27de35c | 533 | r = amdgpu_ring_alloc(ring, 3); |
aaa36a97 AD |
534 | if (r) { |
535 | DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", | |
536 | ring->idx, r); | |
537 | return r; | |
538 | } | |
539 | amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0)); | |
540 | amdgpu_ring_write(ring, 0xDEADBEEF); | |
a27de35c | 541 | amdgpu_ring_commit(ring); |
aaa36a97 AD |
542 | for (i = 0; i < adev->usec_timeout; i++) { |
543 | tmp = RREG32(mmUVD_CONTEXT_ID); | |
544 | if (tmp == 0xDEADBEEF) | |
545 | break; | |
546 | DRM_UDELAY(1); | |
547 | } | |
548 | ||
549 | if (i < adev->usec_timeout) { | |
550 | DRM_INFO("ring test on %d succeeded in %d usecs\n", | |
551 | ring->idx, i); | |
552 | } else { | |
553 | DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n", | |
554 | ring->idx, tmp); | |
555 | r = -EINVAL; | |
556 | } | |
557 | return r; | |
558 | } | |
559 | ||
560 | /** | |
561 | * uvd_v5_0_ring_emit_ib - execute indirect buffer | |
562 | * | |
563 | * @ring: amdgpu_ring pointer | |
564 | * @ib: indirect buffer to execute | |
565 | * | |
566 | * Write ring commands to execute the indirect buffer | |
567 | */ | |
568 | static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring, | |
d88bf583 CK |
569 | struct amdgpu_ib *ib, |
570 | unsigned vm_id, bool ctx_switch) | |
aaa36a97 AD |
571 | { |
572 | amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0)); | |
573 | amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); | |
574 | amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0)); | |
575 | amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); | |
576 | amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0)); | |
577 | amdgpu_ring_write(ring, ib->length_dw); | |
578 | } | |
579 | ||
580 | /** | |
581 | * uvd_v5_0_ring_test_ib - test ib execution | |
582 | * | |
583 | * @ring: amdgpu_ring pointer | |
584 | * | |
585 | * Test if we can successfully execute an IB | |
586 | */ | |
587 | static int uvd_v5_0_ring_test_ib(struct amdgpu_ring *ring) | |
588 | { | |
589 | struct amdgpu_device *adev = ring->adev; | |
0e3f154a | 590 | struct fence *fence = NULL; |
aaa36a97 AD |
591 | int r; |
592 | ||
593 | r = amdgpu_asic_set_uvd_clocks(adev, 53300, 40000); | |
594 | if (r) { | |
595 | DRM_ERROR("amdgpu: failed to raise UVD clocks (%d).\n", r); | |
596 | return r; | |
597 | } | |
598 | ||
599 | r = amdgpu_uvd_get_create_msg(ring, 1, NULL); | |
600 | if (r) { | |
601 | DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r); | |
602 | goto error; | |
603 | } | |
604 | ||
d7af97db | 605 | r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence); |
aaa36a97 AD |
606 | if (r) { |
607 | DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r); | |
608 | goto error; | |
609 | } | |
610 | ||
0e3f154a | 611 | r = fence_wait(fence, false); |
aaa36a97 AD |
612 | if (r) { |
613 | DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); | |
614 | goto error; | |
615 | } | |
616 | DRM_INFO("ib test on ring %d succeeded\n", ring->idx); | |
617 | error: | |
0e3f154a | 618 | fence_put(fence); |
aaa36a97 AD |
619 | amdgpu_asic_set_uvd_clocks(adev, 0, 0); |
620 | return r; | |
621 | } | |
622 | ||
5fc3aeeb | 623 | static bool uvd_v5_0_is_idle(void *handle) |
aaa36a97 | 624 | { |
5fc3aeeb | 625 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
626 | ||
aaa36a97 AD |
627 | return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK); |
628 | } | |
629 | ||
5fc3aeeb | 630 | static int uvd_v5_0_wait_for_idle(void *handle) |
aaa36a97 AD |
631 | { |
632 | unsigned i; | |
5fc3aeeb | 633 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
aaa36a97 AD |
634 | |
635 | for (i = 0; i < adev->usec_timeout; i++) { | |
636 | if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK)) | |
637 | return 0; | |
638 | } | |
639 | return -ETIMEDOUT; | |
640 | } | |
641 | ||
5fc3aeeb | 642 | static int uvd_v5_0_soft_reset(void *handle) |
aaa36a97 | 643 | { |
5fc3aeeb | 644 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
645 | ||
aaa36a97 AD |
646 | uvd_v5_0_stop(adev); |
647 | ||
648 | WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK, | |
649 | ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK); | |
650 | mdelay(5); | |
651 | ||
652 | return uvd_v5_0_start(adev); | |
653 | } | |
654 | ||
aaa36a97 AD |
655 | static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev, |
656 | struct amdgpu_irq_src *source, | |
657 | unsigned type, | |
658 | enum amdgpu_interrupt_state state) | |
659 | { | |
660 | // TODO | |
661 | return 0; | |
662 | } | |
663 | ||
664 | static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev, | |
665 | struct amdgpu_irq_src *source, | |
666 | struct amdgpu_iv_entry *entry) | |
667 | { | |
668 | DRM_DEBUG("IH: UVD TRAP\n"); | |
669 | amdgpu_fence_process(&adev->uvd.ring); | |
670 | return 0; | |
671 | } | |
672 | ||
be3ecca7 TSD |
673 | static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev) |
674 | { | |
675 | uint32_t data, data1, data2, suvd_flags; | |
676 | ||
677 | data = RREG32(mmUVD_CGC_CTRL); | |
678 | data1 = RREG32(mmUVD_SUVD_CGC_GATE); | |
679 | data2 = RREG32(mmUVD_SUVD_CGC_CTRL); | |
680 | ||
681 | data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | | |
682 | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK); | |
683 | ||
684 | suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK | | |
685 | UVD_SUVD_CGC_GATE__SIT_MASK | | |
686 | UVD_SUVD_CGC_GATE__SMP_MASK | | |
687 | UVD_SUVD_CGC_GATE__SCM_MASK | | |
688 | UVD_SUVD_CGC_GATE__SDB_MASK; | |
689 | ||
690 | data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK | | |
691 | (1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) | | |
692 | (4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY)); | |
693 | ||
694 | data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK | | |
695 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK | | |
696 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK | | |
697 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK | | |
698 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK | | |
699 | UVD_CGC_CTRL__SYS_MODE_MASK | | |
700 | UVD_CGC_CTRL__UDEC_MODE_MASK | | |
701 | UVD_CGC_CTRL__MPEG2_MODE_MASK | | |
702 | UVD_CGC_CTRL__REGS_MODE_MASK | | |
703 | UVD_CGC_CTRL__RBC_MODE_MASK | | |
704 | UVD_CGC_CTRL__LMI_MC_MODE_MASK | | |
705 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK | | |
706 | UVD_CGC_CTRL__IDCT_MODE_MASK | | |
707 | UVD_CGC_CTRL__MPRD_MODE_MASK | | |
708 | UVD_CGC_CTRL__MPC_MODE_MASK | | |
709 | UVD_CGC_CTRL__LBSI_MODE_MASK | | |
710 | UVD_CGC_CTRL__LRBBM_MODE_MASK | | |
711 | UVD_CGC_CTRL__WCB_MODE_MASK | | |
712 | UVD_CGC_CTRL__VCPU_MODE_MASK | | |
713 | UVD_CGC_CTRL__JPEG_MODE_MASK | | |
714 | UVD_CGC_CTRL__SCPU_MODE_MASK); | |
715 | data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK | | |
716 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK | | |
717 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK | | |
718 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK | | |
719 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK); | |
720 | data1 |= suvd_flags; | |
721 | ||
722 | WREG32(mmUVD_CGC_CTRL, data); | |
723 | WREG32(mmUVD_CGC_GATE, 0); | |
724 | WREG32(mmUVD_SUVD_CGC_GATE, data1); | |
725 | WREG32(mmUVD_SUVD_CGC_CTRL, data2); | |
726 | } | |
727 | ||
728 | #if 0 | |
729 | static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev) | |
730 | { | |
731 | uint32_t data, data1, cgc_flags, suvd_flags; | |
732 | ||
733 | data = RREG32(mmUVD_CGC_GATE); | |
734 | data1 = RREG32(mmUVD_SUVD_CGC_GATE); | |
735 | ||
736 | cgc_flags = UVD_CGC_GATE__SYS_MASK | | |
737 | UVD_CGC_GATE__UDEC_MASK | | |
738 | UVD_CGC_GATE__MPEG2_MASK | | |
739 | UVD_CGC_GATE__RBC_MASK | | |
740 | UVD_CGC_GATE__LMI_MC_MASK | | |
741 | UVD_CGC_GATE__IDCT_MASK | | |
742 | UVD_CGC_GATE__MPRD_MASK | | |
743 | UVD_CGC_GATE__MPC_MASK | | |
744 | UVD_CGC_GATE__LBSI_MASK | | |
745 | UVD_CGC_GATE__LRBBM_MASK | | |
746 | UVD_CGC_GATE__UDEC_RE_MASK | | |
747 | UVD_CGC_GATE__UDEC_CM_MASK | | |
748 | UVD_CGC_GATE__UDEC_IT_MASK | | |
749 | UVD_CGC_GATE__UDEC_DB_MASK | | |
750 | UVD_CGC_GATE__UDEC_MP_MASK | | |
751 | UVD_CGC_GATE__WCB_MASK | | |
752 | UVD_CGC_GATE__VCPU_MASK | | |
753 | UVD_CGC_GATE__SCPU_MASK; | |
754 | ||
755 | suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK | | |
756 | UVD_SUVD_CGC_GATE__SIT_MASK | | |
757 | UVD_SUVD_CGC_GATE__SMP_MASK | | |
758 | UVD_SUVD_CGC_GATE__SCM_MASK | | |
759 | UVD_SUVD_CGC_GATE__SDB_MASK; | |
760 | ||
761 | data |= cgc_flags; | |
762 | data1 |= suvd_flags; | |
763 | ||
764 | WREG32(mmUVD_CGC_GATE, data); | |
765 | WREG32(mmUVD_SUVD_CGC_GATE, data1); | |
766 | } | |
767 | #endif | |
768 | ||
5fc3aeeb | 769 | static int uvd_v5_0_set_clockgating_state(void *handle, |
770 | enum amd_clockgating_state state) | |
aaa36a97 | 771 | { |
35e5912d | 772 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
be3ecca7 TSD |
773 | bool enable = (state == AMD_CG_STATE_GATE) ? true : false; |
774 | static int curstate = -1; | |
35e5912d | 775 | |
e3b04bc7 | 776 | if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) |
35e5912d AD |
777 | return 0; |
778 | ||
be3ecca7 TSD |
779 | if (curstate == state) |
780 | return 0; | |
781 | ||
782 | curstate = state; | |
783 | if (enable) { | |
784 | /* disable HW gating and enable Sw gating */ | |
785 | uvd_v5_0_set_sw_clock_gating(adev); | |
786 | } else { | |
787 | /* wait for STATUS to clear */ | |
788 | if (uvd_v5_0_wait_for_idle(handle)) | |
789 | return -EBUSY; | |
790 | ||
791 | /* enable HW gates because UVD is idle */ | |
792 | /* uvd_v5_0_set_hw_clock_gating(adev); */ | |
793 | } | |
794 | ||
aaa36a97 AD |
795 | return 0; |
796 | } | |
797 | ||
5fc3aeeb | 798 | static int uvd_v5_0_set_powergating_state(void *handle, |
799 | enum amd_powergating_state state) | |
aaa36a97 AD |
800 | { |
801 | /* This doesn't actually powergate the UVD block. | |
802 | * That's done in the dpm code via the SMC. This | |
803 | * just re-inits the block as necessary. The actual | |
804 | * gating still happens in the dpm code. We should | |
805 | * revisit this when there is a cleaner line between | |
806 | * the smc and the hw blocks | |
807 | */ | |
5fc3aeeb | 808 | struct amdgpu_device *adev = (struct amdgpu_device *)handle; |
809 | ||
e3b04bc7 | 810 | if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD)) |
b6df77fc AD |
811 | return 0; |
812 | ||
5fc3aeeb | 813 | if (state == AMD_PG_STATE_GATE) { |
aaa36a97 AD |
814 | uvd_v5_0_stop(adev); |
815 | return 0; | |
816 | } else { | |
817 | return uvd_v5_0_start(adev); | |
818 | } | |
819 | } | |
820 | ||
5fc3aeeb | 821 | const struct amd_ip_funcs uvd_v5_0_ip_funcs = { |
88a907d6 | 822 | .name = "uvd_v5_0", |
aaa36a97 AD |
823 | .early_init = uvd_v5_0_early_init, |
824 | .late_init = NULL, | |
825 | .sw_init = uvd_v5_0_sw_init, | |
826 | .sw_fini = uvd_v5_0_sw_fini, | |
827 | .hw_init = uvd_v5_0_hw_init, | |
828 | .hw_fini = uvd_v5_0_hw_fini, | |
829 | .suspend = uvd_v5_0_suspend, | |
830 | .resume = uvd_v5_0_resume, | |
831 | .is_idle = uvd_v5_0_is_idle, | |
832 | .wait_for_idle = uvd_v5_0_wait_for_idle, | |
833 | .soft_reset = uvd_v5_0_soft_reset, | |
aaa36a97 AD |
834 | .set_clockgating_state = uvd_v5_0_set_clockgating_state, |
835 | .set_powergating_state = uvd_v5_0_set_powergating_state, | |
836 | }; | |
837 | ||
838 | static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = { | |
839 | .get_rptr = uvd_v5_0_ring_get_rptr, | |
840 | .get_wptr = uvd_v5_0_ring_get_wptr, | |
841 | .set_wptr = uvd_v5_0_ring_set_wptr, | |
842 | .parse_cs = amdgpu_uvd_ring_parse_cs, | |
843 | .emit_ib = uvd_v5_0_ring_emit_ib, | |
844 | .emit_fence = uvd_v5_0_ring_emit_fence, | |
d5b4e25d CK |
845 | .emit_hdp_flush = uvd_v5_0_ring_emit_hdp_flush, |
846 | .emit_hdp_invalidate = uvd_v5_0_ring_emit_hdp_invalidate, | |
aaa36a97 AD |
847 | .test_ring = uvd_v5_0_ring_test_ring, |
848 | .test_ib = uvd_v5_0_ring_test_ib, | |
edff0e28 | 849 | .insert_nop = amdgpu_ring_insert_nop, |
9e5d5309 | 850 | .pad_ib = amdgpu_ring_generic_pad_ib, |
aaa36a97 AD |
851 | }; |
852 | ||
853 | static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev) | |
854 | { | |
855 | adev->uvd.ring.funcs = &uvd_v5_0_ring_funcs; | |
856 | } | |
857 | ||
858 | static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = { | |
859 | .set = uvd_v5_0_set_interrupt_state, | |
860 | .process = uvd_v5_0_process_interrupt, | |
861 | }; | |
862 | ||
863 | static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev) | |
864 | { | |
865 | adev->uvd.irq.num_types = 1; | |
866 | adev->uvd.irq.funcs = &uvd_v5_0_irq_funcs; | |
867 | } |