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 return r;
454 }
455
vce_v3_0_sw_fini(void * handle)456 static int vce_v3_0_sw_fini(void *handle)
457 {
458 int r;
459 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
460
461 r = amdgpu_vce_suspend(adev);
462 if (r)
463 return r;
464
465 return amdgpu_vce_sw_fini(adev);
466 }
467
vce_v3_0_hw_init(void * handle)468 static int vce_v3_0_hw_init(void *handle)
469 {
470 int r, i;
471 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
472
473 vce_v3_0_override_vce_clock_gating(adev, true);
474
475 amdgpu_asic_set_vce_clocks(adev, 10000, 10000);
476
477 for (i = 0; i < adev->vce.num_rings; i++) {
478 r = amdgpu_ring_test_helper(&adev->vce.ring[i]);
479 if (r)
480 return r;
481 }
482
483 DRM_INFO("VCE initialized successfully.\n");
484
485 return 0;
486 }
487
vce_v3_0_hw_fini(void * handle)488 static int vce_v3_0_hw_fini(void *handle)
489 {
490 int r;
491 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
492
493 cancel_delayed_work_sync(&adev->vce.idle_work);
494
495 r = vce_v3_0_wait_for_idle(handle);
496 if (r)
497 return r;
498
499 vce_v3_0_stop(adev);
500 return vce_v3_0_set_clockgating_state(adev, AMD_CG_STATE_GATE);
501 }
502
vce_v3_0_suspend(void * handle)503 static int vce_v3_0_suspend(void *handle)
504 {
505 int r;
506 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
507
508 /*
509 * Proper cleanups before halting the HW engine:
510 * - cancel the delayed idle work
511 * - enable powergating
512 * - enable clockgating
513 * - disable dpm
514 *
515 * TODO: to align with the VCN implementation, move the
516 * jobs for clockgating/powergating/dpm setting to
517 * ->set_powergating_state().
518 */
519 cancel_delayed_work_sync(&adev->vce.idle_work);
520
521 if (adev->pm.dpm_enabled) {
522 amdgpu_dpm_enable_vce(adev, false);
523 } else {
524 amdgpu_asic_set_vce_clocks(adev, 0, 0);
525 amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
526 AMD_PG_STATE_GATE);
527 amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
528 AMD_CG_STATE_GATE);
529 }
530
531 r = vce_v3_0_hw_fini(adev);
532 if (r)
533 return r;
534
535 return amdgpu_vce_suspend(adev);
536 }
537
vce_v3_0_resume(void * handle)538 static int vce_v3_0_resume(void *handle)
539 {
540 int r;
541 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
542
543 r = amdgpu_vce_resume(adev);
544 if (r)
545 return r;
546
547 return vce_v3_0_hw_init(adev);
548 }
549
vce_v3_0_mc_resume(struct amdgpu_device * adev,int idx)550 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
551 {
552 uint32_t offset, size;
553
554 WREG32_P(mmVCE_CLOCK_GATING_A, 0, ~(1 << 16));
555 WREG32_P(mmVCE_UENC_CLOCK_GATING, 0x1FF000, ~0xFF9FF000);
556 WREG32_P(mmVCE_UENC_REG_CLOCK_GATING, 0x3F, ~0x3F);
557 WREG32(mmVCE_CLOCK_GATING_B, 0x1FF);
558
559 WREG32(mmVCE_LMI_CTRL, 0x00398000);
560 WREG32_P(mmVCE_LMI_CACHE_CTRL, 0x0, ~0x1);
561 WREG32(mmVCE_LMI_SWAP_CNTL, 0);
562 WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
563 WREG32(mmVCE_LMI_VM_CTRL, 0);
564 WREG32_OR(mmVCE_VCPU_CNTL, 0x00100000);
565
566 if (adev->asic_type >= CHIP_STONEY) {
567 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8));
568 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8));
569 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8));
570 } else
571 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
572 offset = AMDGPU_VCE_FIRMWARE_OFFSET;
573 size = VCE_V3_0_FW_SIZE;
574 WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
575 WREG32(mmVCE_VCPU_CACHE_SIZE0, size);
576
577 if (idx == 0) {
578 offset += size;
579 size = VCE_V3_0_STACK_SIZE;
580 WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0x7fffffff);
581 WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
582 offset += size;
583 size = VCE_V3_0_DATA_SIZE;
584 WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0x7fffffff);
585 WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
586 } else {
587 offset += size + VCE_V3_0_STACK_SIZE + VCE_V3_0_DATA_SIZE;
588 size = VCE_V3_0_STACK_SIZE;
589 WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0xfffffff);
590 WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
591 offset += size;
592 size = VCE_V3_0_DATA_SIZE;
593 WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0xfffffff);
594 WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
595 }
596
597 WREG32_P(mmVCE_LMI_CTRL2, 0x0, ~0x100);
598 WREG32_FIELD(VCE_SYS_INT_EN, VCE_SYS_INT_TRAP_INTERRUPT_EN, 1);
599 }
600
vce_v3_0_is_idle(void * handle)601 static bool vce_v3_0_is_idle(void *handle)
602 {
603 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
604 u32 mask = 0;
605
606 mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_STATUS2__VCE0_BUSY_MASK;
607 mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_STATUS2__VCE1_BUSY_MASK;
608
609 return !(RREG32(mmSRBM_STATUS2) & mask);
610 }
611
vce_v3_0_wait_for_idle(void * handle)612 static int vce_v3_0_wait_for_idle(void *handle)
613 {
614 unsigned i;
615 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
616
617 for (i = 0; i < adev->usec_timeout; i++)
618 if (vce_v3_0_is_idle(handle))
619 return 0;
620
621 return -ETIMEDOUT;
622 }
623
624 #define VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK 0x00000008L /* AUTO_BUSY */
625 #define VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK 0x00000010L /* RB0_BUSY */
626 #define VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK 0x00000020L /* RB1_BUSY */
627 #define AMDGPU_VCE_STATUS_BUSY_MASK (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK | \
628 VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK)
629
vce_v3_0_check_soft_reset(void * handle)630 static bool vce_v3_0_check_soft_reset(void *handle)
631 {
632 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
633 u32 srbm_soft_reset = 0;
634
635 /* According to VCE team , we should use VCE_STATUS instead
636 * SRBM_STATUS.VCE_BUSY bit for busy status checking.
637 * GRBM_GFX_INDEX.INSTANCE_INDEX is used to specify which VCE
638 * instance's registers are accessed
639 * (0 for 1st instance, 10 for 2nd instance).
640 *
641 *VCE_STATUS
642 *|UENC|ACPI|AUTO ACTIVE|RB1 |RB0 |RB2 | |FW_LOADED|JOB |
643 *|----+----+-----------+----+----+----+----------+---------+----|
644 *|bit8|bit7| bit6 |bit5|bit4|bit3| bit2 | bit1 |bit0|
645 *
646 * VCE team suggest use bit 3--bit 6 for busy status check
647 */
648 mutex_lock(&adev->grbm_idx_mutex);
649 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
650 if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
651 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
652 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
653 }
654 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
655 if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
656 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
657 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
658 }
659 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
660 mutex_unlock(&adev->grbm_idx_mutex);
661
662 if (srbm_soft_reset) {
663 adev->vce.srbm_soft_reset = srbm_soft_reset;
664 return true;
665 } else {
666 adev->vce.srbm_soft_reset = 0;
667 return false;
668 }
669 }
670
vce_v3_0_soft_reset(void * handle)671 static int vce_v3_0_soft_reset(void *handle)
672 {
673 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
674 u32 srbm_soft_reset;
675
676 if (!adev->vce.srbm_soft_reset)
677 return 0;
678 srbm_soft_reset = adev->vce.srbm_soft_reset;
679
680 if (srbm_soft_reset) {
681 u32 tmp;
682
683 tmp = RREG32(mmSRBM_SOFT_RESET);
684 tmp |= srbm_soft_reset;
685 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
686 WREG32(mmSRBM_SOFT_RESET, tmp);
687 tmp = RREG32(mmSRBM_SOFT_RESET);
688
689 udelay(50);
690
691 tmp &= ~srbm_soft_reset;
692 WREG32(mmSRBM_SOFT_RESET, tmp);
693 tmp = RREG32(mmSRBM_SOFT_RESET);
694
695 /* Wait a little for things to settle down */
696 udelay(50);
697 }
698
699 return 0;
700 }
701
vce_v3_0_pre_soft_reset(void * handle)702 static int vce_v3_0_pre_soft_reset(void *handle)
703 {
704 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
705
706 if (!adev->vce.srbm_soft_reset)
707 return 0;
708
709 mdelay(5);
710
711 return vce_v3_0_suspend(adev);
712 }
713
714
vce_v3_0_post_soft_reset(void * handle)715 static int vce_v3_0_post_soft_reset(void *handle)
716 {
717 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
718
719 if (!adev->vce.srbm_soft_reset)
720 return 0;
721
722 mdelay(5);
723
724 return vce_v3_0_resume(adev);
725 }
726
vce_v3_0_set_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)727 static int vce_v3_0_set_interrupt_state(struct amdgpu_device *adev,
728 struct amdgpu_irq_src *source,
729 unsigned type,
730 enum amdgpu_interrupt_state state)
731 {
732 uint32_t val = 0;
733
734 if (state == AMDGPU_IRQ_STATE_ENABLE)
735 val |= VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK;
736
737 WREG32_P(mmVCE_SYS_INT_EN, val, ~VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK);
738 return 0;
739 }
740
vce_v3_0_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)741 static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
742 struct amdgpu_irq_src *source,
743 struct amdgpu_iv_entry *entry)
744 {
745 DRM_DEBUG("IH: VCE\n");
746
747 WREG32_FIELD(VCE_SYS_INT_STATUS, VCE_SYS_INT_TRAP_INTERRUPT_INT, 1);
748
749 switch (entry->src_data[0]) {
750 case 0:
751 case 1:
752 case 2:
753 amdgpu_fence_process(&adev->vce.ring[entry->src_data[0]]);
754 break;
755 default:
756 DRM_ERROR("Unhandled interrupt: %d %d\n",
757 entry->src_id, entry->src_data[0]);
758 break;
759 }
760
761 return 0;
762 }
763
vce_v3_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)764 static int vce_v3_0_set_clockgating_state(void *handle,
765 enum amd_clockgating_state state)
766 {
767 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
768 bool enable = (state == AMD_CG_STATE_GATE);
769 int i;
770
771 if (!(adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG))
772 return 0;
773
774 mutex_lock(&adev->grbm_idx_mutex);
775 for (i = 0; i < 2; i++) {
776 /* Program VCE Instance 0 or 1 if not harvested */
777 if (adev->vce.harvest_config & (1 << i))
778 continue;
779
780 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(i));
781
782 if (!enable) {
783 /* initialize VCE_CLOCK_GATING_A: Clock ON/OFF delay */
784 uint32_t data = RREG32(mmVCE_CLOCK_GATING_A);
785 data &= ~(0xf | 0xff0);
786 data |= ((0x0 << 0) | (0x04 << 4));
787 WREG32(mmVCE_CLOCK_GATING_A, data);
788
789 /* initialize VCE_UENC_CLOCK_GATING: Clock ON/OFF delay */
790 data = RREG32(mmVCE_UENC_CLOCK_GATING);
791 data &= ~(0xf | 0xff0);
792 data |= ((0x0 << 0) | (0x04 << 4));
793 WREG32(mmVCE_UENC_CLOCK_GATING, data);
794 }
795
796 vce_v3_0_set_vce_sw_clock_gating(adev, enable);
797 }
798
799 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
800 mutex_unlock(&adev->grbm_idx_mutex);
801
802 return 0;
803 }
804
vce_v3_0_set_powergating_state(void * handle,enum amd_powergating_state state)805 static int vce_v3_0_set_powergating_state(void *handle,
806 enum amd_powergating_state state)
807 {
808 /* This doesn't actually powergate the VCE block.
809 * That's done in the dpm code via the SMC. This
810 * just re-inits the block as necessary. The actual
811 * gating still happens in the dpm code. We should
812 * revisit this when there is a cleaner line between
813 * the smc and the hw blocks
814 */
815 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
816 int ret = 0;
817
818 if (state == AMD_PG_STATE_GATE) {
819 ret = vce_v3_0_stop(adev);
820 if (ret)
821 goto out;
822 } else {
823 ret = vce_v3_0_start(adev);
824 if (ret)
825 goto out;
826 }
827
828 out:
829 return ret;
830 }
831
vce_v3_0_get_clockgating_state(void * handle,u64 * flags)832 static void vce_v3_0_get_clockgating_state(void *handle, u64 *flags)
833 {
834 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
835 int data;
836
837 mutex_lock(&adev->pm.mutex);
838
839 if (adev->flags & AMD_IS_APU)
840 data = RREG32_SMC(ixCURRENT_PG_STATUS_APU);
841 else
842 data = RREG32_SMC(ixCURRENT_PG_STATUS);
843
844 if (data & CURRENT_PG_STATUS__VCE_PG_STATUS_MASK) {
845 DRM_INFO("Cannot get clockgating state when VCE is powergated.\n");
846 goto out;
847 }
848
849 WREG32_FIELD(GRBM_GFX_INDEX, VCE_INSTANCE, 0);
850
851 /* AMD_CG_SUPPORT_VCE_MGCG */
852 data = RREG32(mmVCE_CLOCK_GATING_A);
853 if (data & (0x04 << 4))
854 *flags |= AMD_CG_SUPPORT_VCE_MGCG;
855
856 out:
857 mutex_unlock(&adev->pm.mutex);
858 }
859
vce_v3_0_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)860 static void vce_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
861 struct amdgpu_job *job,
862 struct amdgpu_ib *ib,
863 uint32_t flags)
864 {
865 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
866
867 amdgpu_ring_write(ring, VCE_CMD_IB_VM);
868 amdgpu_ring_write(ring, vmid);
869 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
870 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
871 amdgpu_ring_write(ring, ib->length_dw);
872 }
873
vce_v3_0_emit_vm_flush(struct amdgpu_ring * ring,unsigned int vmid,uint64_t pd_addr)874 static void vce_v3_0_emit_vm_flush(struct amdgpu_ring *ring,
875 unsigned int vmid, uint64_t pd_addr)
876 {
877 amdgpu_ring_write(ring, VCE_CMD_UPDATE_PTB);
878 amdgpu_ring_write(ring, vmid);
879 amdgpu_ring_write(ring, pd_addr >> 12);
880
881 amdgpu_ring_write(ring, VCE_CMD_FLUSH_TLB);
882 amdgpu_ring_write(ring, vmid);
883 amdgpu_ring_write(ring, VCE_CMD_END);
884 }
885
vce_v3_0_emit_pipeline_sync(struct amdgpu_ring * ring)886 static void vce_v3_0_emit_pipeline_sync(struct amdgpu_ring *ring)
887 {
888 uint32_t seq = ring->fence_drv.sync_seq;
889 uint64_t addr = ring->fence_drv.gpu_addr;
890
891 amdgpu_ring_write(ring, VCE_CMD_WAIT_GE);
892 amdgpu_ring_write(ring, lower_32_bits(addr));
893 amdgpu_ring_write(ring, upper_32_bits(addr));
894 amdgpu_ring_write(ring, seq);
895 }
896
897 static const struct amd_ip_funcs vce_v3_0_ip_funcs = {
898 .name = "vce_v3_0",
899 .early_init = vce_v3_0_early_init,
900 .late_init = NULL,
901 .sw_init = vce_v3_0_sw_init,
902 .sw_fini = vce_v3_0_sw_fini,
903 .hw_init = vce_v3_0_hw_init,
904 .hw_fini = vce_v3_0_hw_fini,
905 .suspend = vce_v3_0_suspend,
906 .resume = vce_v3_0_resume,
907 .is_idle = vce_v3_0_is_idle,
908 .wait_for_idle = vce_v3_0_wait_for_idle,
909 .check_soft_reset = vce_v3_0_check_soft_reset,
910 .pre_soft_reset = vce_v3_0_pre_soft_reset,
911 .soft_reset = vce_v3_0_soft_reset,
912 .post_soft_reset = vce_v3_0_post_soft_reset,
913 .set_clockgating_state = vce_v3_0_set_clockgating_state,
914 .set_powergating_state = vce_v3_0_set_powergating_state,
915 .get_clockgating_state = vce_v3_0_get_clockgating_state,
916 .dump_ip_state = NULL,
917 .print_ip_state = NULL,
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 .patch_cs_in_place = 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 .type = AMD_IP_BLOCK_TYPE_VCE,
1003 .major = 3,
1004 .minor = 0,
1005 .rev = 0,
1006 .funcs = &vce_v3_0_ip_funcs,
1007 };
1008
1009 const struct amdgpu_ip_block_version vce_v3_1_ip_block = {
1010 .type = AMD_IP_BLOCK_TYPE_VCE,
1011 .major = 3,
1012 .minor = 1,
1013 .rev = 0,
1014 .funcs = &vce_v3_0_ip_funcs,
1015 };
1016
1017 const struct amdgpu_ip_block_version vce_v3_4_ip_block = {
1018 .type = AMD_IP_BLOCK_TYPE_VCE,
1019 .major = 3,
1020 .minor = 4,
1021 .rev = 0,
1022 .funcs = &vce_v3_0_ip_funcs,
1023 };
1024