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(struct amdgpu_ip_block *ip_block);
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(struct amdgpu_ip_block * ip_block)399 static int vce_v3_0_early_init(struct amdgpu_ip_block *ip_block)
400 {
401 struct amdgpu_device *adev = ip_block->adev;
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(struct amdgpu_ip_block * ip_block)418 static int vce_v3_0_sw_init(struct amdgpu_ip_block *ip_block)
419 {
420 struct amdgpu_device *adev = ip_block->adev;
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(struct amdgpu_ip_block * ip_block)456 static int vce_v3_0_sw_fini(struct amdgpu_ip_block *ip_block)
457 {
458 int r;
459 struct amdgpu_device *adev = ip_block->adev;
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(struct amdgpu_ip_block * ip_block)468 static int vce_v3_0_hw_init(struct amdgpu_ip_block *ip_block)
469 {
470 int r, i;
471 struct amdgpu_device *adev = ip_block->adev;
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(struct amdgpu_ip_block * ip_block)488 static int vce_v3_0_hw_fini(struct amdgpu_ip_block *ip_block)
489 {
490 int r;
491 struct amdgpu_device *adev = ip_block->adev;
492
493 cancel_delayed_work_sync(&adev->vce.idle_work);
494
495 r = vce_v3_0_wait_for_idle(ip_block);
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(struct amdgpu_ip_block * ip_block)503 static int vce_v3_0_suspend(struct amdgpu_ip_block *ip_block)
504 {
505 int r;
506 struct amdgpu_device *adev = ip_block->adev;
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(ip_block);
532 if (r)
533 return r;
534
535 return amdgpu_vce_suspend(adev);
536 }
537
vce_v3_0_resume(struct amdgpu_ip_block * ip_block)538 static int vce_v3_0_resume(struct amdgpu_ip_block *ip_block)
539 {
540 int r;
541
542 r = amdgpu_vce_resume(ip_block->adev);
543 if (r)
544 return r;
545
546 return vce_v3_0_hw_init(ip_block);
547 }
548
vce_v3_0_mc_resume(struct amdgpu_device * adev,int idx)549 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
550 {
551 uint32_t offset, size;
552
553 WREG32_P(mmVCE_CLOCK_GATING_A, 0, ~(1 << 16));
554 WREG32_P(mmVCE_UENC_CLOCK_GATING, 0x1FF000, ~0xFF9FF000);
555 WREG32_P(mmVCE_UENC_REG_CLOCK_GATING, 0x3F, ~0x3F);
556 WREG32(mmVCE_CLOCK_GATING_B, 0x1FF);
557
558 WREG32(mmVCE_LMI_CTRL, 0x00398000);
559 WREG32_P(mmVCE_LMI_CACHE_CTRL, 0x0, ~0x1);
560 WREG32(mmVCE_LMI_SWAP_CNTL, 0);
561 WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
562 WREG32(mmVCE_LMI_VM_CTRL, 0);
563 WREG32_OR(mmVCE_VCPU_CNTL, 0x00100000);
564
565 if (adev->asic_type >= CHIP_STONEY) {
566 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8));
567 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8));
568 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8));
569 } else
570 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
571 offset = AMDGPU_VCE_FIRMWARE_OFFSET;
572 size = VCE_V3_0_FW_SIZE;
573 WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
574 WREG32(mmVCE_VCPU_CACHE_SIZE0, size);
575
576 if (idx == 0) {
577 offset += size;
578 size = VCE_V3_0_STACK_SIZE;
579 WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0x7fffffff);
580 WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
581 offset += size;
582 size = VCE_V3_0_DATA_SIZE;
583 WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0x7fffffff);
584 WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
585 } else {
586 offset += size + VCE_V3_0_STACK_SIZE + VCE_V3_0_DATA_SIZE;
587 size = VCE_V3_0_STACK_SIZE;
588 WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0xfffffff);
589 WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
590 offset += size;
591 size = VCE_V3_0_DATA_SIZE;
592 WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0xfffffff);
593 WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
594 }
595
596 WREG32_P(mmVCE_LMI_CTRL2, 0x0, ~0x100);
597 WREG32_FIELD(VCE_SYS_INT_EN, VCE_SYS_INT_TRAP_INTERRUPT_EN, 1);
598 }
599
vce_v3_0_is_idle(void * handle)600 static bool vce_v3_0_is_idle(void *handle)
601 {
602 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
603 u32 mask = 0;
604
605 mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_STATUS2__VCE0_BUSY_MASK;
606 mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_STATUS2__VCE1_BUSY_MASK;
607
608 return !(RREG32(mmSRBM_STATUS2) & mask);
609 }
610
vce_v3_0_wait_for_idle(struct amdgpu_ip_block * ip_block)611 static int vce_v3_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
612 {
613 unsigned i;
614 struct amdgpu_device *adev = ip_block->adev;
615
616 for (i = 0; i < adev->usec_timeout; i++)
617 if (vce_v3_0_is_idle(adev))
618 return 0;
619
620 return -ETIMEDOUT;
621 }
622
623 #define VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK 0x00000008L /* AUTO_BUSY */
624 #define VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK 0x00000010L /* RB0_BUSY */
625 #define VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK 0x00000020L /* RB1_BUSY */
626 #define AMDGPU_VCE_STATUS_BUSY_MASK (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK | \
627 VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK)
628
vce_v3_0_check_soft_reset(struct amdgpu_ip_block * ip_block)629 static bool vce_v3_0_check_soft_reset(struct amdgpu_ip_block *ip_block)
630 {
631 struct amdgpu_device *adev = ip_block->adev;
632 u32 srbm_soft_reset = 0;
633
634 /* According to VCE team , we should use VCE_STATUS instead
635 * SRBM_STATUS.VCE_BUSY bit for busy status checking.
636 * GRBM_GFX_INDEX.INSTANCE_INDEX is used to specify which VCE
637 * instance's registers are accessed
638 * (0 for 1st instance, 10 for 2nd instance).
639 *
640 *VCE_STATUS
641 *|UENC|ACPI|AUTO ACTIVE|RB1 |RB0 |RB2 | |FW_LOADED|JOB |
642 *|----+----+-----------+----+----+----+----------+---------+----|
643 *|bit8|bit7| bit6 |bit5|bit4|bit3| bit2 | bit1 |bit0|
644 *
645 * VCE team suggest use bit 3--bit 6 for busy status check
646 */
647 mutex_lock(&adev->grbm_idx_mutex);
648 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
649 if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
650 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
651 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
652 }
653 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
654 if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
655 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
656 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
657 }
658 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
659 mutex_unlock(&adev->grbm_idx_mutex);
660
661 if (srbm_soft_reset) {
662 adev->vce.srbm_soft_reset = srbm_soft_reset;
663 return true;
664 } else {
665 adev->vce.srbm_soft_reset = 0;
666 return false;
667 }
668 }
669
vce_v3_0_soft_reset(struct amdgpu_ip_block * ip_block)670 static int vce_v3_0_soft_reset(struct amdgpu_ip_block *ip_block)
671 {
672 struct amdgpu_device *adev = ip_block->adev;
673 u32 srbm_soft_reset;
674
675 if (!adev->vce.srbm_soft_reset)
676 return 0;
677 srbm_soft_reset = adev->vce.srbm_soft_reset;
678
679 if (srbm_soft_reset) {
680 u32 tmp;
681
682 tmp = RREG32(mmSRBM_SOFT_RESET);
683 tmp |= srbm_soft_reset;
684 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
685 WREG32(mmSRBM_SOFT_RESET, tmp);
686 tmp = RREG32(mmSRBM_SOFT_RESET);
687
688 udelay(50);
689
690 tmp &= ~srbm_soft_reset;
691 WREG32(mmSRBM_SOFT_RESET, tmp);
692 tmp = RREG32(mmSRBM_SOFT_RESET);
693
694 /* Wait a little for things to settle down */
695 udelay(50);
696 }
697
698 return 0;
699 }
700
vce_v3_0_pre_soft_reset(struct amdgpu_ip_block * ip_block)701 static int vce_v3_0_pre_soft_reset(struct amdgpu_ip_block *ip_block)
702 {
703 struct amdgpu_device *adev = ip_block->adev;
704
705 if (!adev->vce.srbm_soft_reset)
706 return 0;
707
708 mdelay(5);
709
710 return vce_v3_0_suspend(ip_block);
711 }
712
713
vce_v3_0_post_soft_reset(struct amdgpu_ip_block * ip_block)714 static int vce_v3_0_post_soft_reset(struct amdgpu_ip_block *ip_block)
715 {
716 struct amdgpu_device *adev = ip_block->adev;
717
718 if (!adev->vce.srbm_soft_reset)
719 return 0;
720
721 mdelay(5);
722
723 return vce_v3_0_resume(ip_block);
724 }
725
vce_v3_0_set_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)726 static int vce_v3_0_set_interrupt_state(struct amdgpu_device *adev,
727 struct amdgpu_irq_src *source,
728 unsigned type,
729 enum amdgpu_interrupt_state state)
730 {
731 uint32_t val = 0;
732
733 if (state == AMDGPU_IRQ_STATE_ENABLE)
734 val |= VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK;
735
736 WREG32_P(mmVCE_SYS_INT_EN, val, ~VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK);
737 return 0;
738 }
739
vce_v3_0_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)740 static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
741 struct amdgpu_irq_src *source,
742 struct amdgpu_iv_entry *entry)
743 {
744 DRM_DEBUG("IH: VCE\n");
745
746 WREG32_FIELD(VCE_SYS_INT_STATUS, VCE_SYS_INT_TRAP_INTERRUPT_INT, 1);
747
748 switch (entry->src_data[0]) {
749 case 0:
750 case 1:
751 case 2:
752 amdgpu_fence_process(&adev->vce.ring[entry->src_data[0]]);
753 break;
754 default:
755 DRM_ERROR("Unhandled interrupt: %d %d\n",
756 entry->src_id, entry->src_data[0]);
757 break;
758 }
759
760 return 0;
761 }
762
vce_v3_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)763 static int vce_v3_0_set_clockgating_state(void *handle,
764 enum amd_clockgating_state state)
765 {
766 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
767 bool enable = (state == AMD_CG_STATE_GATE);
768 int i;
769
770 if (!(adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG))
771 return 0;
772
773 mutex_lock(&adev->grbm_idx_mutex);
774 for (i = 0; i < 2; i++) {
775 /* Program VCE Instance 0 or 1 if not harvested */
776 if (adev->vce.harvest_config & (1 << i))
777 continue;
778
779 WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(i));
780
781 if (!enable) {
782 /* initialize VCE_CLOCK_GATING_A: Clock ON/OFF delay */
783 uint32_t data = RREG32(mmVCE_CLOCK_GATING_A);
784 data &= ~(0xf | 0xff0);
785 data |= ((0x0 << 0) | (0x04 << 4));
786 WREG32(mmVCE_CLOCK_GATING_A, data);
787
788 /* initialize VCE_UENC_CLOCK_GATING: Clock ON/OFF delay */
789 data = RREG32(mmVCE_UENC_CLOCK_GATING);
790 data &= ~(0xf | 0xff0);
791 data |= ((0x0 << 0) | (0x04 << 4));
792 WREG32(mmVCE_UENC_CLOCK_GATING, data);
793 }
794
795 vce_v3_0_set_vce_sw_clock_gating(adev, enable);
796 }
797
798 WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
799 mutex_unlock(&adev->grbm_idx_mutex);
800
801 return 0;
802 }
803
vce_v3_0_set_powergating_state(void * handle,enum amd_powergating_state state)804 static int vce_v3_0_set_powergating_state(void *handle,
805 enum amd_powergating_state state)
806 {
807 /* This doesn't actually powergate the VCE block.
808 * That's done in the dpm code via the SMC. This
809 * just re-inits the block as necessary. The actual
810 * gating still happens in the dpm code. We should
811 * revisit this when there is a cleaner line between
812 * the smc and the hw blocks
813 */
814 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
815 int ret = 0;
816
817 if (state == AMD_PG_STATE_GATE) {
818 ret = vce_v3_0_stop(adev);
819 if (ret)
820 goto out;
821 } else {
822 ret = vce_v3_0_start(adev);
823 if (ret)
824 goto out;
825 }
826
827 out:
828 return ret;
829 }
830
vce_v3_0_get_clockgating_state(void * handle,u64 * flags)831 static void vce_v3_0_get_clockgating_state(void *handle, u64 *flags)
832 {
833 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
834 int data;
835
836 mutex_lock(&adev->pm.mutex);
837
838 if (adev->flags & AMD_IS_APU)
839 data = RREG32_SMC(ixCURRENT_PG_STATUS_APU);
840 else
841 data = RREG32_SMC(ixCURRENT_PG_STATUS);
842
843 if (data & CURRENT_PG_STATUS__VCE_PG_STATUS_MASK) {
844 DRM_INFO("Cannot get clockgating state when VCE is powergated.\n");
845 goto out;
846 }
847
848 WREG32_FIELD(GRBM_GFX_INDEX, VCE_INSTANCE, 0);
849
850 /* AMD_CG_SUPPORT_VCE_MGCG */
851 data = RREG32(mmVCE_CLOCK_GATING_A);
852 if (data & (0x04 << 4))
853 *flags |= AMD_CG_SUPPORT_VCE_MGCG;
854
855 out:
856 mutex_unlock(&adev->pm.mutex);
857 }
858
vce_v3_0_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)859 static void vce_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
860 struct amdgpu_job *job,
861 struct amdgpu_ib *ib,
862 uint32_t flags)
863 {
864 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
865
866 amdgpu_ring_write(ring, VCE_CMD_IB_VM);
867 amdgpu_ring_write(ring, vmid);
868 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
869 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
870 amdgpu_ring_write(ring, ib->length_dw);
871 }
872
vce_v3_0_emit_vm_flush(struct amdgpu_ring * ring,unsigned int vmid,uint64_t pd_addr)873 static void vce_v3_0_emit_vm_flush(struct amdgpu_ring *ring,
874 unsigned int vmid, uint64_t pd_addr)
875 {
876 amdgpu_ring_write(ring, VCE_CMD_UPDATE_PTB);
877 amdgpu_ring_write(ring, vmid);
878 amdgpu_ring_write(ring, pd_addr >> 12);
879
880 amdgpu_ring_write(ring, VCE_CMD_FLUSH_TLB);
881 amdgpu_ring_write(ring, vmid);
882 amdgpu_ring_write(ring, VCE_CMD_END);
883 }
884
vce_v3_0_emit_pipeline_sync(struct amdgpu_ring * ring)885 static void vce_v3_0_emit_pipeline_sync(struct amdgpu_ring *ring)
886 {
887 uint32_t seq = ring->fence_drv.sync_seq;
888 uint64_t addr = ring->fence_drv.gpu_addr;
889
890 amdgpu_ring_write(ring, VCE_CMD_WAIT_GE);
891 amdgpu_ring_write(ring, lower_32_bits(addr));
892 amdgpu_ring_write(ring, upper_32_bits(addr));
893 amdgpu_ring_write(ring, seq);
894 }
895
896 static const struct amd_ip_funcs vce_v3_0_ip_funcs = {
897 .name = "vce_v3_0",
898 .early_init = vce_v3_0_early_init,
899 .sw_init = vce_v3_0_sw_init,
900 .sw_fini = vce_v3_0_sw_fini,
901 .hw_init = vce_v3_0_hw_init,
902 .hw_fini = vce_v3_0_hw_fini,
903 .suspend = vce_v3_0_suspend,
904 .resume = vce_v3_0_resume,
905 .is_idle = vce_v3_0_is_idle,
906 .wait_for_idle = vce_v3_0_wait_for_idle,
907 .check_soft_reset = vce_v3_0_check_soft_reset,
908 .pre_soft_reset = vce_v3_0_pre_soft_reset,
909 .soft_reset = vce_v3_0_soft_reset,
910 .post_soft_reset = vce_v3_0_post_soft_reset,
911 .set_clockgating_state = vce_v3_0_set_clockgating_state,
912 .set_powergating_state = vce_v3_0_set_powergating_state,
913 .get_clockgating_state = vce_v3_0_get_clockgating_state,
914 };
915
916 static const struct amdgpu_ring_funcs vce_v3_0_ring_phys_funcs = {
917 .type = AMDGPU_RING_TYPE_VCE,
918 .align_mask = 0xf,
919 .nop = VCE_CMD_NO_OP,
920 .support_64bit_ptrs = false,
921 .no_user_fence = true,
922 .get_rptr = vce_v3_0_ring_get_rptr,
923 .get_wptr = vce_v3_0_ring_get_wptr,
924 .set_wptr = vce_v3_0_ring_set_wptr,
925 .parse_cs = amdgpu_vce_ring_parse_cs,
926 .emit_frame_size =
927 4 + /* vce_v3_0_emit_pipeline_sync */
928 6, /* amdgpu_vce_ring_emit_fence x1 no user fence */
929 .emit_ib_size = 4, /* amdgpu_vce_ring_emit_ib */
930 .emit_ib = amdgpu_vce_ring_emit_ib,
931 .emit_fence = amdgpu_vce_ring_emit_fence,
932 .test_ring = amdgpu_vce_ring_test_ring,
933 .test_ib = amdgpu_vce_ring_test_ib,
934 .insert_nop = amdgpu_ring_insert_nop,
935 .pad_ib = amdgpu_ring_generic_pad_ib,
936 .begin_use = amdgpu_vce_ring_begin_use,
937 .end_use = amdgpu_vce_ring_end_use,
938 };
939
940 static const struct amdgpu_ring_funcs vce_v3_0_ring_vm_funcs = {
941 .type = AMDGPU_RING_TYPE_VCE,
942 .align_mask = 0xf,
943 .nop = VCE_CMD_NO_OP,
944 .support_64bit_ptrs = false,
945 .no_user_fence = true,
946 .get_rptr = vce_v3_0_ring_get_rptr,
947 .get_wptr = vce_v3_0_ring_get_wptr,
948 .set_wptr = vce_v3_0_ring_set_wptr,
949 .patch_cs_in_place = amdgpu_vce_ring_parse_cs_vm,
950 .emit_frame_size =
951 6 + /* vce_v3_0_emit_vm_flush */
952 4 + /* vce_v3_0_emit_pipeline_sync */
953 6 + 6, /* amdgpu_vce_ring_emit_fence x2 vm fence */
954 .emit_ib_size = 5, /* vce_v3_0_ring_emit_ib */
955 .emit_ib = vce_v3_0_ring_emit_ib,
956 .emit_vm_flush = vce_v3_0_emit_vm_flush,
957 .emit_pipeline_sync = vce_v3_0_emit_pipeline_sync,
958 .emit_fence = amdgpu_vce_ring_emit_fence,
959 .test_ring = amdgpu_vce_ring_test_ring,
960 .test_ib = amdgpu_vce_ring_test_ib,
961 .insert_nop = amdgpu_ring_insert_nop,
962 .pad_ib = amdgpu_ring_generic_pad_ib,
963 .begin_use = amdgpu_vce_ring_begin_use,
964 .end_use = amdgpu_vce_ring_end_use,
965 };
966
vce_v3_0_set_ring_funcs(struct amdgpu_device * adev)967 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev)
968 {
969 int i;
970
971 if (adev->asic_type >= CHIP_STONEY) {
972 for (i = 0; i < adev->vce.num_rings; i++) {
973 adev->vce.ring[i].funcs = &vce_v3_0_ring_vm_funcs;
974 adev->vce.ring[i].me = i;
975 }
976 DRM_INFO("VCE enabled in VM mode\n");
977 } else {
978 for (i = 0; i < adev->vce.num_rings; i++) {
979 adev->vce.ring[i].funcs = &vce_v3_0_ring_phys_funcs;
980 adev->vce.ring[i].me = i;
981 }
982 DRM_INFO("VCE enabled in physical mode\n");
983 }
984 }
985
986 static const struct amdgpu_irq_src_funcs vce_v3_0_irq_funcs = {
987 .set = vce_v3_0_set_interrupt_state,
988 .process = vce_v3_0_process_interrupt,
989 };
990
vce_v3_0_set_irq_funcs(struct amdgpu_device * adev)991 static void vce_v3_0_set_irq_funcs(struct amdgpu_device *adev)
992 {
993 adev->vce.irq.num_types = 1;
994 adev->vce.irq.funcs = &vce_v3_0_irq_funcs;
995 };
996
997 const struct amdgpu_ip_block_version vce_v3_0_ip_block = {
998 .type = AMD_IP_BLOCK_TYPE_VCE,
999 .major = 3,
1000 .minor = 0,
1001 .rev = 0,
1002 .funcs = &vce_v3_0_ip_funcs,
1003 };
1004
1005 const struct amdgpu_ip_block_version vce_v3_1_ip_block = {
1006 .type = AMD_IP_BLOCK_TYPE_VCE,
1007 .major = 3,
1008 .minor = 1,
1009 .rev = 0,
1010 .funcs = &vce_v3_0_ip_funcs,
1011 };
1012
1013 const struct amdgpu_ip_block_version vce_v3_4_ip_block = {
1014 .type = AMD_IP_BLOCK_TYPE_VCE,
1015 .major = 3,
1016 .minor = 4,
1017 .rev = 0,
1018 .funcs = &vce_v3_0_ip_funcs,
1019 };
1020