1 /*
2 * Copyright 2014 Advanced Micro Devices, Inc.
3 * All Rights Reserved.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19 * USE OR OTHER DEALINGS IN THE SOFTWARE.
20 *
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
23 * of the Software.
24 *
25 * Authors: Christian König <christian.koenig@amd.com>
26 */
27
28 #include <linux/firmware.h>
29
30 #include "amdgpu.h"
31 #include "amdgpu_vce.h"
32 #include "vid.h"
33 #include "vce/vce_3_0_d.h"
34 #include "vce/vce_3_0_sh_mask.h"
35 #include "oss/oss_3_0_d.h"
36 #include "oss/oss_3_0_sh_mask.h"
37 #include "gca/gfx_8_0_d.h"
38 #include "smu/smu_7_1_2_d.h"
39 #include "smu/smu_7_1_2_sh_mask.h"
40 #include "gca/gfx_8_0_sh_mask.h"
41 #include "ivsrcid/ivsrcid_vislands30.h"
42
43
44 #define GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT 0x04
45 #define GRBM_GFX_INDEX__VCE_INSTANCE_MASK 0x10
46 #define GRBM_GFX_INDEX__VCE_ALL_PIPE 0x07
47
48 #define mmVCE_LMI_VCPU_CACHE_40BIT_BAR0 0x8616
49 #define mmVCE_LMI_VCPU_CACHE_40BIT_BAR1 0x8617
50 #define mmVCE_LMI_VCPU_CACHE_40BIT_BAR2 0x8618
51 #define mmGRBM_GFX_INDEX_DEFAULT 0xE0000000
52
53 #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK 0x02
54
55 #define VCE_V3_0_FW_SIZE (384 * 1024)
56 #define VCE_V3_0_STACK_SIZE (64 * 1024)
57 #define VCE_V3_0_DATA_SIZE ((16 * 1024 * AMDGPU_MAX_VCE_HANDLES) + (52 * 1024))
58
59 #define FW_52_8_3 ((52 << 24) | (8 << 16) | (3 << 8))
60
61 #define GET_VCE_INSTANCE(i) ((i) << GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT \
62 | GRBM_GFX_INDEX__VCE_ALL_PIPE)
63
64 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx);
65 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev);
66 static void vce_v3_0_set_irq_funcs(struct amdgpu_device *adev);
67 static int vce_v3_0_wait_for_idle(void *handle);
68 static int vce_v3_0_set_clockgating_state(void *handle,
69 enum amd_clockgating_state state);
70 /**
71 * vce_v3_0_ring_get_rptr - get read pointer
72 *
73 * @ring: amdgpu_ring pointer
74 *
75 * Returns the current hardware read pointer
76 */
vce_v3_0_ring_get_rptr(struct amdgpu_ring * ring)77 static uint64_t vce_v3_0_ring_get_rptr(struct amdgpu_ring *ring)
78 {
79 struct amdgpu_device *adev = ring->adev;
80 u32 v;
81
82 mutex_lock(&adev->grbm_idx_mutex);
83 if (adev->vce.harvest_config == 0 ||
84 adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
85 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
86 else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
87 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
88
89 if (ring->me == 0)
90 v = RREG32(mmVCE_RB_RPTR);
91 else if (ring->me == 1)
92 v = RREG32(mmVCE_RB_RPTR2);
93 else
94 v = RREG32(mmVCE_RB_RPTR3);
95
96 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
97 mutex_unlock(&adev->grbm_idx_mutex);
98
99 return v;
100 }
101
102 /**
103 * vce_v3_0_ring_get_wptr - get write pointer
104 *
105 * @ring: amdgpu_ring pointer
106 *
107 * Returns the current hardware write pointer
108 */
vce_v3_0_ring_get_wptr(struct amdgpu_ring * ring)109 static uint64_t vce_v3_0_ring_get_wptr(struct amdgpu_ring *ring)
110 {
111 struct amdgpu_device *adev = ring->adev;
112 u32 v;
113
114 mutex_lock(&adev->grbm_idx_mutex);
115 if (adev->vce.harvest_config == 0 ||
116 adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
117 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
118 else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
119 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
120
121 if (ring->me == 0)
122 v = RREG32(mmVCE_RB_WPTR);
123 else if (ring->me == 1)
124 v = RREG32(mmVCE_RB_WPTR2);
125 else
126 v = RREG32(mmVCE_RB_WPTR3);
127
128 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
129 mutex_unlock(&adev->grbm_idx_mutex);
130
131 return v;
132 }
133
134 /**
135 * vce_v3_0_ring_set_wptr - set write pointer
136 *
137 * @ring: amdgpu_ring pointer
138 *
139 * Commits the write pointer to the hardware
140 */
vce_v3_0_ring_set_wptr(struct amdgpu_ring * ring)141 static void vce_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
142 {
143 struct amdgpu_device *adev = ring->adev;
144
145 mutex_lock(&adev->grbm_idx_mutex);
146 if (adev->vce.harvest_config == 0 ||
147 adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
148 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
149 else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
150 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
151
152 if (ring->me == 0)
153 WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
154 else if (ring->me == 1)
155 WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
156 else
157 WREG32(mmVCE_RB_WPTR3, lower_32_bits(ring->wptr));
158
159 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
160 mutex_unlock(&adev->grbm_idx_mutex);
161 }
162
vce_v3_0_override_vce_clock_gating(struct amdgpu_device * adev,bool override)163 static void vce_v3_0_override_vce_clock_gating(struct amdgpu_device *adev, bool override)
164 {
165 WREG32_FIELD(VCE_RB_ARB_CTRL, VCE_CGTT_OVERRIDE, override ? 1 : 0);
166 }
167
vce_v3_0_set_vce_sw_clock_gating(struct amdgpu_device * adev,bool gated)168 static void vce_v3_0_set_vce_sw_clock_gating(struct amdgpu_device *adev,
169 bool gated)
170 {
171 u32 data;
172
173 /* Set Override to disable Clock Gating */
174 vce_v3_0_override_vce_clock_gating(adev, true);
175
176 /* This function enables MGCG which is controlled by firmware.
177 With the clocks in the gated state the core is still
178 accessible but the firmware will throttle the clocks on the
179 fly as necessary.
180 */
181 if (!gated) {
182 data = RREG32(mmVCE_CLOCK_GATING_B);
183 data |= 0x1ff;
184 data &= ~0xef0000;
185 WREG32(mmVCE_CLOCK_GATING_B, data);
186
187 data = RREG32(mmVCE_UENC_CLOCK_GATING);
188 data |= 0x3ff000;
189 data &= ~0xffc00000;
190 WREG32(mmVCE_UENC_CLOCK_GATING, data);
191
192 data = RREG32(mmVCE_UENC_CLOCK_GATING_2);
193 data |= 0x2;
194 data &= ~0x00010000;
195 WREG32(mmVCE_UENC_CLOCK_GATING_2, data);
196
197 data = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
198 data |= 0x37f;
199 WREG32(mmVCE_UENC_REG_CLOCK_GATING, data);
200
201 data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL);
202 data |= VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK |
203 VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK |
204 VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK |
205 0x8;
206 WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data);
207 } else {
208 data = RREG32(mmVCE_CLOCK_GATING_B);
209 data &= ~0x80010;
210 data |= 0xe70008;
211 WREG32(mmVCE_CLOCK_GATING_B, data);
212
213 data = RREG32(mmVCE_UENC_CLOCK_GATING);
214 data |= 0xffc00000;
215 WREG32(mmVCE_UENC_CLOCK_GATING, data);
216
217 data = RREG32(mmVCE_UENC_CLOCK_GATING_2);
218 data |= 0x10000;
219 WREG32(mmVCE_UENC_CLOCK_GATING_2, data);
220
221 data = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
222 data &= ~0x3ff;
223 WREG32(mmVCE_UENC_REG_CLOCK_GATING, data);
224
225 data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL);
226 data &= ~(VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK |
227 VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK |
228 VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK |
229 0x8);
230 WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data);
231 }
232 vce_v3_0_override_vce_clock_gating(adev, false);
233 }
234
vce_v3_0_firmware_loaded(struct amdgpu_device * adev)235 static int vce_v3_0_firmware_loaded(struct amdgpu_device *adev)
236 {
237 int i, j;
238
239 for (i = 0; i < 10; ++i) {
240 for (j = 0; j < 100; ++j) {
241 uint32_t status = RREG32(mmVCE_STATUS);
242
243 if (status & VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK)
244 return 0;
245 mdelay(10);
246 }
247
248 DRM_ERROR("VCE not responding, trying to reset the ECPU!!!\n");
249 WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 1);
250 mdelay(10);
251 WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 0);
252 mdelay(10);
253 }
254
255 return -ETIMEDOUT;
256 }
257
258 /**
259 * vce_v3_0_start - start VCE block
260 *
261 * @adev: amdgpu_device pointer
262 *
263 * Setup and start the VCE block
264 */
vce_v3_0_start(struct amdgpu_device * adev)265 static int vce_v3_0_start(struct amdgpu_device *adev)
266 {
267 struct amdgpu_ring *ring;
268 int idx, r;
269
270 mutex_lock(&adev->grbm_idx_mutex);
271 for (idx = 0; idx < 2; ++idx) {
272 if (adev->vce.harvest_config & (1 << idx))
273 continue;
274
275 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(idx));
276
277 /* Program instance 0 reg space for two instances or instance 0 case
278 program instance 1 reg space for only instance 1 available case */
279 if (idx != 1 || adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0) {
280 ring = &adev->vce.ring[0];
281 WREG32(mmVCE_RB_RPTR, lower_32_bits(ring->wptr));
282 WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
283 WREG32(mmVCE_RB_BASE_LO, ring->gpu_addr);
284 WREG32(mmVCE_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
285 WREG32(mmVCE_RB_SIZE, ring->ring_size / 4);
286
287 ring = &adev->vce.ring[1];
288 WREG32(mmVCE_RB_RPTR2, lower_32_bits(ring->wptr));
289 WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
290 WREG32(mmVCE_RB_BASE_LO2, ring->gpu_addr);
291 WREG32(mmVCE_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
292 WREG32(mmVCE_RB_SIZE2, ring->ring_size / 4);
293
294 ring = &adev->vce.ring[2];
295 WREG32(mmVCE_RB_RPTR3, lower_32_bits(ring->wptr));
296 WREG32(mmVCE_RB_WPTR3, lower_32_bits(ring->wptr));
297 WREG32(mmVCE_RB_BASE_LO3, ring->gpu_addr);
298 WREG32(mmVCE_RB_BASE_HI3, upper_32_bits(ring->gpu_addr));
299 WREG32(mmVCE_RB_SIZE3, ring->ring_size / 4);
300 }
301
302 vce_v3_0_mc_resume(adev, idx);
303 WREG32_FIELD(VCE_STATUS, JOB_BUSY, 1);
304
305 if (adev->asic_type >= CHIP_STONEY)
306 WREG32_P(mmVCE_VCPU_CNTL, 1, ~0x200001);
307 else
308 WREG32_FIELD(VCE_VCPU_CNTL, CLK_EN, 1);
309
310 WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 0);
311 mdelay(100);
312
313 r = vce_v3_0_firmware_loaded(adev);
314
315 /* clear BUSY flag */
316 WREG32_FIELD(VCE_STATUS, JOB_BUSY, 0);
317
318 if (r) {
319 DRM_ERROR("VCE not responding, giving up!!!\n");
320 mutex_unlock(&adev->grbm_idx_mutex);
321 return r;
322 }
323 }
324
325 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
326 mutex_unlock(&adev->grbm_idx_mutex);
327
328 return 0;
329 }
330
vce_v3_0_stop(struct amdgpu_device * adev)331 static int vce_v3_0_stop(struct amdgpu_device *adev)
332 {
333 int idx;
334
335 mutex_lock(&adev->grbm_idx_mutex);
336 for (idx = 0; idx < 2; ++idx) {
337 if (adev->vce.harvest_config & (1 << idx))
338 continue;
339
340 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(idx));
341
342 if (adev->asic_type >= CHIP_STONEY)
343 WREG32_P(mmVCE_VCPU_CNTL, 0, ~0x200001);
344 else
345 WREG32_FIELD(VCE_VCPU_CNTL, CLK_EN, 0);
346
347 /* hold on ECPU */
348 WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 1);
349
350 /* clear VCE STATUS */
351 WREG32(mmVCE_STATUS, 0);
352 }
353
354 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
355 mutex_unlock(&adev->grbm_idx_mutex);
356
357 return 0;
358 }
359
360 #define ixVCE_HARVEST_FUSE_MACRO__ADDRESS 0xC0014074
361 #define VCE_HARVEST_FUSE_MACRO__SHIFT 27
362 #define VCE_HARVEST_FUSE_MACRO__MASK 0x18000000
363
vce_v3_0_get_harvest_config(struct amdgpu_device * adev)364 static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev)
365 {
366 u32 tmp;
367
368 if ((adev->asic_type == CHIP_FIJI) ||
369 (adev->asic_type == CHIP_STONEY))
370 return AMDGPU_VCE_HARVEST_VCE1;
371
372 if (adev->flags & AMD_IS_APU)
373 tmp = (RREG32_SMC(ixVCE_HARVEST_FUSE_MACRO__ADDRESS) &
374 VCE_HARVEST_FUSE_MACRO__MASK) >>
375 VCE_HARVEST_FUSE_MACRO__SHIFT;
376 else
377 tmp = (RREG32_SMC(ixCC_HARVEST_FUSES) &
378 CC_HARVEST_FUSES__VCE_DISABLE_MASK) >>
379 CC_HARVEST_FUSES__VCE_DISABLE__SHIFT;
380
381 switch (tmp) {
382 case 1:
383 return AMDGPU_VCE_HARVEST_VCE0;
384 case 2:
385 return AMDGPU_VCE_HARVEST_VCE1;
386 case 3:
387 return AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1;
388 default:
389 if ((adev->asic_type == CHIP_POLARIS10) ||
390 (adev->asic_type == CHIP_POLARIS11) ||
391 (adev->asic_type == CHIP_POLARIS12) ||
392 (adev->asic_type == CHIP_VEGAM))
393 return AMDGPU_VCE_HARVEST_VCE1;
394
395 return 0;
396 }
397 }
398
vce_v3_0_early_init(void * handle)399 static int vce_v3_0_early_init(void *handle)
400 {
401 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
402
403 adev->vce.harvest_config = vce_v3_0_get_harvest_config(adev);
404
405 if ((adev->vce.harvest_config &
406 (AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1)) ==
407 (AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1))
408 return -ENOENT;
409
410 adev->vce.num_rings = 3;
411
412 vce_v3_0_set_ring_funcs(adev);
413 vce_v3_0_set_irq_funcs(adev);
414
415 return 0;
416 }
417
vce_v3_0_sw_init(void * handle)418 static int vce_v3_0_sw_init(void *handle)
419 {
420 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
421 struct amdgpu_ring *ring;
422 int r, i;
423
424 /* VCE */
425 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_VCE_TRAP, &adev->vce.irq);
426 if (r)
427 return r;
428
429 r = amdgpu_vce_sw_init(adev, VCE_V3_0_FW_SIZE +
430 (VCE_V3_0_STACK_SIZE + VCE_V3_0_DATA_SIZE) * 2);
431 if (r)
432 return r;
433
434 /* 52.8.3 required for 3 ring support */
435 if (adev->vce.fw_version < FW_52_8_3)
436 adev->vce.num_rings = 2;
437
438 r = amdgpu_vce_resume(adev);
439 if (r)
440 return r;
441
442 for (i = 0; i < adev->vce.num_rings; i++) {
443 enum amdgpu_ring_priority_level hw_prio = amdgpu_vce_get_ring_prio(i);
444
445 ring = &adev->vce.ring[i];
446 sprintf(ring->name, "vce%d", i);
447 r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0,
448 hw_prio, NULL);
449 if (r)
450 return r;
451 }
452
453 r = amdgpu_vce_entity_init(adev);
454
455 return r;
456 }
457
vce_v3_0_sw_fini(void * handle)458 static int vce_v3_0_sw_fini(void *handle)
459 {
460 int r;
461 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
462
463 r = amdgpu_vce_suspend(adev);
464 if (r)
465 return r;
466
467 return amdgpu_vce_sw_fini(adev);
468 }
469
vce_v3_0_hw_init(void * handle)470 static int vce_v3_0_hw_init(void *handle)
471 {
472 int r, i;
473 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
474
475 vce_v3_0_override_vce_clock_gating(adev, true);
476
477 amdgpu_asic_set_vce_clocks(adev, 10000, 10000);
478
479 for (i = 0; i < adev->vce.num_rings; i++) {
480 r = amdgpu_ring_test_helper(&adev->vce.ring[i]);
481 if (r)
482 return r;
483 }
484
485 DRM_INFO("VCE initialized successfully.\n");
486
487 return 0;
488 }
489
vce_v3_0_hw_fini(void * handle)490 static int vce_v3_0_hw_fini(void *handle)
491 {
492 int r;
493 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
494
495 cancel_delayed_work_sync(&adev->vce.idle_work);
496
497 r = vce_v3_0_wait_for_idle(handle);
498 if (r)
499 return r;
500
501 vce_v3_0_stop(adev);
502 return vce_v3_0_set_clockgating_state(adev, AMD_CG_STATE_GATE);
503 }
504
vce_v3_0_suspend(void * handle)505 static int vce_v3_0_suspend(void *handle)
506 {
507 int r;
508 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
509
510 /*
511 * Proper cleanups before halting the HW engine:
512 * - cancel the delayed idle work
513 * - enable powergating
514 * - enable clockgating
515 * - disable dpm
516 *
517 * TODO: to align with the VCN implementation, move the
518 * jobs for clockgating/powergating/dpm setting to
519 * ->set_powergating_state().
520 */
521 cancel_delayed_work_sync(&adev->vce.idle_work);
522
523 if (adev->pm.dpm_enabled) {
524 amdgpu_dpm_enable_vce(adev, false);
525 } else {
526 amdgpu_asic_set_vce_clocks(adev, 0, 0);
527 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
528 AMD_PG_STATE_GATE);
529 amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
530 AMD_CG_STATE_GATE);
531 }
532
533 r = vce_v3_0_hw_fini(adev);
534 if (r)
535 return r;
536
537 return amdgpu_vce_suspend(adev);
538 }
539
vce_v3_0_resume(void * handle)540 static int vce_v3_0_resume(void *handle)
541 {
542 int r;
543 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
544
545 r = amdgpu_vce_resume(adev);
546 if (r)
547 return r;
548
549 return vce_v3_0_hw_init(adev);
550 }
551
vce_v3_0_mc_resume(struct amdgpu_device * adev,int idx)552 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
553 {
554 uint32_t offset, size;
555
556 WREG32_P(mmVCE_CLOCK_GATING_A, 0, ~(1 << 16));
557 WREG32_P(mmVCE_UENC_CLOCK_GATING, 0x1FF000, ~0xFF9FF000);
558 WREG32_P(mmVCE_UENC_REG_CLOCK_GATING, 0x3F, ~0x3F);
559 WREG32(mmVCE_CLOCK_GATING_B, 0x1FF);
560
561 WREG32(mmVCE_LMI_CTRL, 0x00398000);
562 WREG32_P(mmVCE_LMI_CACHE_CTRL, 0x0, ~0x1);
563 WREG32(mmVCE_LMI_SWAP_CNTL, 0);
564 WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
565 WREG32(mmVCE_LMI_VM_CTRL, 0);
566 WREG32_OR(mmVCE_VCPU_CNTL, 0x00100000);
567
568 if (adev->asic_type >= CHIP_STONEY) {
569 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8));
570 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8));
571 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8));
572 } else
573 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
574 offset = AMDGPU_VCE_FIRMWARE_OFFSET;
575 size = VCE_V3_0_FW_SIZE;
576 WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
577 WREG32(mmVCE_VCPU_CACHE_SIZE0, size);
578
579 if (idx == 0) {
580 offset += size;
581 size = VCE_V3_0_STACK_SIZE;
582 WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0x7fffffff);
583 WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
584 offset += size;
585 size = VCE_V3_0_DATA_SIZE;
586 WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0x7fffffff);
587 WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
588 } else {
589 offset += size + VCE_V3_0_STACK_SIZE + VCE_V3_0_DATA_SIZE;
590 size = VCE_V3_0_STACK_SIZE;
591 WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0xfffffff);
592 WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
593 offset += size;
594 size = VCE_V3_0_DATA_SIZE;
595 WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0xfffffff);
596 WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
597 }
598
599 WREG32_P(mmVCE_LMI_CTRL2, 0x0, ~0x100);
600 WREG32_FIELD(VCE_SYS_INT_EN, VCE_SYS_INT_TRAP_INTERRUPT_EN, 1);
601 }
602
vce_v3_0_is_idle(void * handle)603 static bool vce_v3_0_is_idle(void *handle)
604 {
605 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
606 u32 mask = 0;
607
608 mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_STATUS2__VCE0_BUSY_MASK;
609 mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_STATUS2__VCE1_BUSY_MASK;
610
611 return !(RREG32(mmSRBM_STATUS2) & mask);
612 }
613
vce_v3_0_wait_for_idle(void * handle)614 static int vce_v3_0_wait_for_idle(void *handle)
615 {
616 unsigned i;
617 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
618
619 for (i = 0; i < adev->usec_timeout; i++)
620 if (vce_v3_0_is_idle(handle))
621 return 0;
622
623 return -ETIMEDOUT;
624 }
625
626 #define VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK 0x00000008L /* AUTO_BUSY */
627 #define VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK 0x00000010L /* RB0_BUSY */
628 #define VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK 0x00000020L /* RB1_BUSY */
629 #define AMDGPU_VCE_STATUS_BUSY_MASK (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK | \
630 VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK)
631
vce_v3_0_check_soft_reset(void * handle)632 static bool vce_v3_0_check_soft_reset(void *handle)
633 {
634 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
635 u32 srbm_soft_reset = 0;
636
637 /* According to VCE team , we should use VCE_STATUS instead
638 * SRBM_STATUS.VCE_BUSY bit for busy status checking.
639 * GRBM_GFX_INDEX.INSTANCE_INDEX is used to specify which VCE
640 * instance's registers are accessed
641 * (0 for 1st instance, 10 for 2nd instance).
642 *
643 *VCE_STATUS
644 *|UENC|ACPI|AUTO ACTIVE|RB1 |RB0 |RB2 | |FW_LOADED|JOB |
645 *|----+----+-----------+----+----+----+----------+---------+----|
646 *|bit8|bit7| bit6 |bit5|bit4|bit3| bit2 | bit1 |bit0|
647 *
648 * VCE team suggest use bit 3--bit 6 for busy status check
649 */
650 mutex_lock(&adev->grbm_idx_mutex);
651 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
652 if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
653 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
654 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
655 }
656 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
657 if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
658 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
659 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
660 }
661 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
662 mutex_unlock(&adev->grbm_idx_mutex);
663
664 if (srbm_soft_reset) {
665 adev->vce.srbm_soft_reset = srbm_soft_reset;
666 return true;
667 } else {
668 adev->vce.srbm_soft_reset = 0;
669 return false;
670 }
671 }
672
vce_v3_0_soft_reset(void * handle)673 static int vce_v3_0_soft_reset(void *handle)
674 {
675 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
676 u32 srbm_soft_reset;
677
678 if (!adev->vce.srbm_soft_reset)
679 return 0;
680 srbm_soft_reset = adev->vce.srbm_soft_reset;
681
682 if (srbm_soft_reset) {
683 u32 tmp;
684
685 tmp = RREG32(mmSRBM_SOFT_RESET);
686 tmp |= srbm_soft_reset;
687 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
688 WREG32(mmSRBM_SOFT_RESET, tmp);
689 tmp = RREG32(mmSRBM_SOFT_RESET);
690
691 udelay(50);
692
693 tmp &= ~srbm_soft_reset;
694 WREG32(mmSRBM_SOFT_RESET, tmp);
695 tmp = RREG32(mmSRBM_SOFT_RESET);
696
697 /* Wait a little for things to settle down */
698 udelay(50);
699 }
700
701 return 0;
702 }
703
vce_v3_0_pre_soft_reset(void * handle)704 static int vce_v3_0_pre_soft_reset(void *handle)
705 {
706 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
707
708 if (!adev->vce.srbm_soft_reset)
709 return 0;
710
711 mdelay(5);
712
713 return vce_v3_0_suspend(adev);
714 }
715
716
vce_v3_0_post_soft_reset(void * handle)717 static int vce_v3_0_post_soft_reset(void *handle)
718 {
719 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
720
721 if (!adev->vce.srbm_soft_reset)
722 return 0;
723
724 mdelay(5);
725
726 return vce_v3_0_resume(adev);
727 }
728
vce_v3_0_set_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)729 static int vce_v3_0_set_interrupt_state(struct amdgpu_device *adev,
730 struct amdgpu_irq_src *source,
731 unsigned type,
732 enum amdgpu_interrupt_state state)
733 {
734 uint32_t val = 0;
735
736 if (state == AMDGPU_IRQ_STATE_ENABLE)
737 val |= VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK;
738
739 WREG32_P(mmVCE_SYS_INT_EN, val, ~VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK);
740 return 0;
741 }
742
vce_v3_0_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)743 static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
744 struct amdgpu_irq_src *source,
745 struct amdgpu_iv_entry *entry)
746 {
747 DRM_DEBUG("IH: VCE\n");
748
749 WREG32_FIELD(VCE_SYS_INT_STATUS, VCE_SYS_INT_TRAP_INTERRUPT_INT, 1);
750
751 switch (entry->src_data[0]) {
752 case 0:
753 case 1:
754 case 2:
755 amdgpu_fence_process(&adev->vce.ring[entry->src_data[0]]);
756 break;
757 default:
758 DRM_ERROR("Unhandled interrupt: %d %d\n",
759 entry->src_id, entry->src_data[0]);
760 break;
761 }
762
763 return 0;
764 }
765
vce_v3_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)766 static int vce_v3_0_set_clockgating_state(void *handle,
767 enum amd_clockgating_state state)
768 {
769 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
770 bool enable = (state == AMD_CG_STATE_GATE);
771 int i;
772
773 if (!(adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG))
774 return 0;
775
776 mutex_lock(&adev->grbm_idx_mutex);
777 for (i = 0; i < 2; i++) {
778 /* Program VCE Instance 0 or 1 if not harvested */
779 if (adev->vce.harvest_config & (1 << i))
780 continue;
781
782 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(i));
783
784 if (!enable) {
785 /* initialize VCE_CLOCK_GATING_A: Clock ON/OFF delay */
786 uint32_t data = RREG32(mmVCE_CLOCK_GATING_A);
787 data &= ~(0xf | 0xff0);
788 data |= ((0x0 << 0) | (0x04 << 4));
789 WREG32(mmVCE_CLOCK_GATING_A, data);
790
791 /* initialize VCE_UENC_CLOCK_GATING: Clock ON/OFF delay */
792 data = RREG32(mmVCE_UENC_CLOCK_GATING);
793 data &= ~(0xf | 0xff0);
794 data |= ((0x0 << 0) | (0x04 << 4));
795 WREG32(mmVCE_UENC_CLOCK_GATING, data);
796 }
797
798 vce_v3_0_set_vce_sw_clock_gating(adev, enable);
799 }
800
801 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
802 mutex_unlock(&adev->grbm_idx_mutex);
803
804 return 0;
805 }
806
vce_v3_0_set_powergating_state(void * handle,enum amd_powergating_state state)807 static int vce_v3_0_set_powergating_state(void *handle,
808 enum amd_powergating_state state)
809 {
810 /* This doesn't actually powergate the VCE block.
811 * That's done in the dpm code via the SMC. This
812 * just re-inits the block as necessary. The actual
813 * gating still happens in the dpm code. We should
814 * revisit this when there is a cleaner line between
815 * the smc and the hw blocks
816 */
817 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
818 int ret = 0;
819
820 if (state == AMD_PG_STATE_GATE) {
821 ret = vce_v3_0_stop(adev);
822 if (ret)
823 goto out;
824 } else {
825 ret = vce_v3_0_start(adev);
826 if (ret)
827 goto out;
828 }
829
830 out:
831 return ret;
832 }
833
vce_v3_0_get_clockgating_state(void * handle,u64 * flags)834 static void vce_v3_0_get_clockgating_state(void *handle, u64 *flags)
835 {
836 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
837 int data;
838
839 mutex_lock(&adev->pm.mutex);
840
841 if (adev->flags & AMD_IS_APU)
842 data = RREG32_SMC(ixCURRENT_PG_STATUS_APU);
843 else
844 data = RREG32_SMC(ixCURRENT_PG_STATUS);
845
846 if (data & CURRENT_PG_STATUS__VCE_PG_STATUS_MASK) {
847 DRM_INFO("Cannot get clockgating state when VCE is powergated.\n");
848 goto out;
849 }
850
851 WREG32_FIELD(GRBM_GFX_INDEX, VCE_INSTANCE, 0);
852
853 /* AMD_CG_SUPPORT_VCE_MGCG */
854 data = RREG32(mmVCE_CLOCK_GATING_A);
855 if (data & (0x04 << 4))
856 *flags |= AMD_CG_SUPPORT_VCE_MGCG;
857
858 out:
859 mutex_unlock(&adev->pm.mutex);
860 }
861
vce_v3_0_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)862 static void vce_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
863 struct amdgpu_job *job,
864 struct amdgpu_ib *ib,
865 uint32_t flags)
866 {
867 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
868
869 amdgpu_ring_write(ring, VCE_CMD_IB_VM);
870 amdgpu_ring_write(ring, vmid);
871 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
872 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
873 amdgpu_ring_write(ring, ib->length_dw);
874 }
875
vce_v3_0_emit_vm_flush(struct amdgpu_ring * ring,unsigned int vmid,uint64_t pd_addr)876 static void vce_v3_0_emit_vm_flush(struct amdgpu_ring *ring,
877 unsigned int vmid, uint64_t pd_addr)
878 {
879 amdgpu_ring_write(ring, VCE_CMD_UPDATE_PTB);
880 amdgpu_ring_write(ring, vmid);
881 amdgpu_ring_write(ring, pd_addr >> 12);
882
883 amdgpu_ring_write(ring, VCE_CMD_FLUSH_TLB);
884 amdgpu_ring_write(ring, vmid);
885 amdgpu_ring_write(ring, VCE_CMD_END);
886 }
887
vce_v3_0_emit_pipeline_sync(struct amdgpu_ring * ring)888 static void vce_v3_0_emit_pipeline_sync(struct amdgpu_ring *ring)
889 {
890 uint32_t seq = ring->fence_drv.sync_seq;
891 uint64_t addr = ring->fence_drv.gpu_addr;
892
893 amdgpu_ring_write(ring, VCE_CMD_WAIT_GE);
894 amdgpu_ring_write(ring, lower_32_bits(addr));
895 amdgpu_ring_write(ring, upper_32_bits(addr));
896 amdgpu_ring_write(ring, seq);
897 }
898
899 static const struct amd_ip_funcs vce_v3_0_ip_funcs = {
900 .name = "vce_v3_0",
901 .early_init = vce_v3_0_early_init,
902 .late_init = NULL,
903 .sw_init = vce_v3_0_sw_init,
904 .sw_fini = vce_v3_0_sw_fini,
905 .hw_init = vce_v3_0_hw_init,
906 .hw_fini = vce_v3_0_hw_fini,
907 .suspend = vce_v3_0_suspend,
908 .resume = vce_v3_0_resume,
909 .is_idle = vce_v3_0_is_idle,
910 .wait_for_idle = vce_v3_0_wait_for_idle,
911 .check_soft_reset = vce_v3_0_check_soft_reset,
912 .pre_soft_reset = vce_v3_0_pre_soft_reset,
913 .soft_reset = vce_v3_0_soft_reset,
914 .post_soft_reset = vce_v3_0_post_soft_reset,
915 .set_clockgating_state = vce_v3_0_set_clockgating_state,
916 .set_powergating_state = vce_v3_0_set_powergating_state,
917 .get_clockgating_state = vce_v3_0_get_clockgating_state,
918 };
919
920 static const struct amdgpu_ring_funcs vce_v3_0_ring_phys_funcs = {
921 .type = AMDGPU_RING_TYPE_VCE,
922 .align_mask = 0xf,
923 .nop = VCE_CMD_NO_OP,
924 .support_64bit_ptrs = false,
925 .no_user_fence = true,
926 .get_rptr = vce_v3_0_ring_get_rptr,
927 .get_wptr = vce_v3_0_ring_get_wptr,
928 .set_wptr = vce_v3_0_ring_set_wptr,
929 .parse_cs = amdgpu_vce_ring_parse_cs,
930 .emit_frame_size =
931 4 + /* vce_v3_0_emit_pipeline_sync */
932 6, /* amdgpu_vce_ring_emit_fence x1 no user fence */
933 .emit_ib_size = 4, /* amdgpu_vce_ring_emit_ib */
934 .emit_ib = amdgpu_vce_ring_emit_ib,
935 .emit_fence = amdgpu_vce_ring_emit_fence,
936 .test_ring = amdgpu_vce_ring_test_ring,
937 .test_ib = amdgpu_vce_ring_test_ib,
938 .insert_nop = amdgpu_ring_insert_nop,
939 .pad_ib = amdgpu_ring_generic_pad_ib,
940 .begin_use = amdgpu_vce_ring_begin_use,
941 .end_use = amdgpu_vce_ring_end_use,
942 };
943
944 static const struct amdgpu_ring_funcs vce_v3_0_ring_vm_funcs = {
945 .type = AMDGPU_RING_TYPE_VCE,
946 .align_mask = 0xf,
947 .nop = VCE_CMD_NO_OP,
948 .support_64bit_ptrs = false,
949 .no_user_fence = true,
950 .get_rptr = vce_v3_0_ring_get_rptr,
951 .get_wptr = vce_v3_0_ring_get_wptr,
952 .set_wptr = vce_v3_0_ring_set_wptr,
953 .parse_cs = amdgpu_vce_ring_parse_cs_vm,
954 .emit_frame_size =
955 6 + /* vce_v3_0_emit_vm_flush */
956 4 + /* vce_v3_0_emit_pipeline_sync */
957 6 + 6, /* amdgpu_vce_ring_emit_fence x2 vm fence */
958 .emit_ib_size = 5, /* vce_v3_0_ring_emit_ib */
959 .emit_ib = vce_v3_0_ring_emit_ib,
960 .emit_vm_flush = vce_v3_0_emit_vm_flush,
961 .emit_pipeline_sync = vce_v3_0_emit_pipeline_sync,
962 .emit_fence = amdgpu_vce_ring_emit_fence,
963 .test_ring = amdgpu_vce_ring_test_ring,
964 .test_ib = amdgpu_vce_ring_test_ib,
965 .insert_nop = amdgpu_ring_insert_nop,
966 .pad_ib = amdgpu_ring_generic_pad_ib,
967 .begin_use = amdgpu_vce_ring_begin_use,
968 .end_use = amdgpu_vce_ring_end_use,
969 };
970
vce_v3_0_set_ring_funcs(struct amdgpu_device * adev)971 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev)
972 {
973 int i;
974
975 if (adev->asic_type >= CHIP_STONEY) {
976 for (i = 0; i < adev->vce.num_rings; i++) {
977 adev->vce.ring[i].funcs = &vce_v3_0_ring_vm_funcs;
978 adev->vce.ring[i].me = i;
979 }
980 DRM_INFO("VCE enabled in VM mode\n");
981 } else {
982 for (i = 0; i < adev->vce.num_rings; i++) {
983 adev->vce.ring[i].funcs = &vce_v3_0_ring_phys_funcs;
984 adev->vce.ring[i].me = i;
985 }
986 DRM_INFO("VCE enabled in physical mode\n");
987 }
988 }
989
990 static const struct amdgpu_irq_src_funcs vce_v3_0_irq_funcs = {
991 .set = vce_v3_0_set_interrupt_state,
992 .process = vce_v3_0_process_interrupt,
993 };
994
vce_v3_0_set_irq_funcs(struct amdgpu_device * adev)995 static void vce_v3_0_set_irq_funcs(struct amdgpu_device *adev)
996 {
997 adev->vce.irq.num_types = 1;
998 adev->vce.irq.funcs = &vce_v3_0_irq_funcs;
999 };
1000
1001 const struct amdgpu_ip_block_version vce_v3_0_ip_block =
1002 {
1003 .type = AMD_IP_BLOCK_TYPE_VCE,
1004 .major = 3,
1005 .minor = 0,
1006 .rev = 0,
1007 .funcs = &vce_v3_0_ip_funcs,
1008 };
1009
1010 const struct amdgpu_ip_block_version vce_v3_1_ip_block =
1011 {
1012 .type = AMD_IP_BLOCK_TYPE_VCE,
1013 .major = 3,
1014 .minor = 1,
1015 .rev = 0,
1016 .funcs = &vce_v3_0_ip_funcs,
1017 };
1018
1019 const struct amdgpu_ip_block_version vce_v3_4_ip_block =
1020 {
1021 .type = AMD_IP_BLOCK_TYPE_VCE,
1022 .major = 3,
1023 .minor = 4,
1024 .rev = 0,
1025 .funcs = &vce_v3_0_ip_funcs,
1026 };
1027