This source file includes following definitions.
- vce_v2_0_ring_get_rptr
- vce_v2_0_ring_get_wptr
- vce_v2_0_ring_set_wptr
- vce_v2_0_lmi_clean
- vce_v2_0_firmware_loaded
- vce_v2_0_disable_cg
- vce_v2_0_init_cg
- vce_v2_0_mc_resume
- vce_v2_0_is_idle
- vce_v2_0_wait_for_idle
- vce_v2_0_start
- vce_v2_0_stop
- vce_v2_0_set_sw_cg
- vce_v2_0_set_dyn_cg
- vce_v2_0_enable_mgcg
- vce_v2_0_early_init
- vce_v2_0_sw_init
- vce_v2_0_sw_fini
- vce_v2_0_hw_init
- vce_v2_0_hw_fini
- vce_v2_0_suspend
- vce_v2_0_resume
- vce_v2_0_soft_reset
- vce_v2_0_set_interrupt_state
- vce_v2_0_process_interrupt
- vce_v2_0_set_clockgating_state
- vce_v2_0_set_powergating_state
- vce_v2_0_set_ring_funcs
- vce_v2_0_set_irq_funcs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28 #include <linux/firmware.h>
29
30 #include "amdgpu.h"
31 #include "amdgpu_vce.h"
32 #include "cikd.h"
33 #include "vce/vce_2_0_d.h"
34 #include "vce/vce_2_0_sh_mask.h"
35 #include "smu/smu_7_0_1_d.h"
36 #include "smu/smu_7_0_1_sh_mask.h"
37 #include "oss/oss_2_0_d.h"
38 #include "oss/oss_2_0_sh_mask.h"
39
40 #define VCE_V2_0_FW_SIZE (256 * 1024)
41 #define VCE_V2_0_STACK_SIZE (64 * 1024)
42 #define VCE_V2_0_DATA_SIZE (23552 * AMDGPU_MAX_VCE_HANDLES)
43 #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK 0x02
44
45 static void vce_v2_0_set_ring_funcs(struct amdgpu_device *adev);
46 static void vce_v2_0_set_irq_funcs(struct amdgpu_device *adev);
47
48
49
50
51
52
53
54
55 static uint64_t vce_v2_0_ring_get_rptr(struct amdgpu_ring *ring)
56 {
57 struct amdgpu_device *adev = ring->adev;
58
59 if (ring->me == 0)
60 return RREG32(mmVCE_RB_RPTR);
61 else
62 return RREG32(mmVCE_RB_RPTR2);
63 }
64
65
66
67
68
69
70
71
72 static uint64_t vce_v2_0_ring_get_wptr(struct amdgpu_ring *ring)
73 {
74 struct amdgpu_device *adev = ring->adev;
75
76 if (ring->me == 0)
77 return RREG32(mmVCE_RB_WPTR);
78 else
79 return RREG32(mmVCE_RB_WPTR2);
80 }
81
82
83
84
85
86
87
88
89 static void vce_v2_0_ring_set_wptr(struct amdgpu_ring *ring)
90 {
91 struct amdgpu_device *adev = ring->adev;
92
93 if (ring->me == 0)
94 WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
95 else
96 WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
97 }
98
99 static int vce_v2_0_lmi_clean(struct amdgpu_device *adev)
100 {
101 int i, j;
102
103 for (i = 0; i < 10; ++i) {
104 for (j = 0; j < 100; ++j) {
105 uint32_t status = RREG32(mmVCE_LMI_STATUS);
106
107 if (status & 0x337f)
108 return 0;
109 mdelay(10);
110 }
111 }
112
113 return -ETIMEDOUT;
114 }
115
116 static int vce_v2_0_firmware_loaded(struct amdgpu_device *adev)
117 {
118 int i, j;
119
120 for (i = 0; i < 10; ++i) {
121 for (j = 0; j < 100; ++j) {
122 uint32_t status = RREG32(mmVCE_STATUS);
123
124 if (status & VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK)
125 return 0;
126 mdelay(10);
127 }
128
129 DRM_ERROR("VCE not responding, trying to reset the ECPU!!!\n");
130 WREG32_P(mmVCE_SOFT_RESET,
131 VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK,
132 ~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
133 mdelay(10);
134 WREG32_P(mmVCE_SOFT_RESET, 0,
135 ~VCE_SOFT_RESET__ECPU_SOFT_RESET_MASK);
136 mdelay(10);
137 }
138
139 return -ETIMEDOUT;
140 }
141
142 static void vce_v2_0_disable_cg(struct amdgpu_device *adev)
143 {
144 WREG32(mmVCE_CGTT_CLK_OVERRIDE, 7);
145 }
146
147 static void vce_v2_0_init_cg(struct amdgpu_device *adev)
148 {
149 u32 tmp;
150
151 tmp = RREG32(mmVCE_CLOCK_GATING_A);
152 tmp &= ~0xfff;
153 tmp |= ((0 << 0) | (4 << 4));
154 tmp |= 0x40000;
155 WREG32(mmVCE_CLOCK_GATING_A, tmp);
156
157 tmp = RREG32(mmVCE_UENC_CLOCK_GATING);
158 tmp &= ~0xfff;
159 tmp |= ((0 << 0) | (4 << 4));
160 WREG32(mmVCE_UENC_CLOCK_GATING, tmp);
161
162 tmp = RREG32(mmVCE_CLOCK_GATING_B);
163 tmp |= 0x10;
164 tmp &= ~0x100000;
165 WREG32(mmVCE_CLOCK_GATING_B, tmp);
166 }
167
168 static void vce_v2_0_mc_resume(struct amdgpu_device *adev)
169 {
170 uint32_t size, offset;
171
172 WREG32_P(mmVCE_CLOCK_GATING_A, 0, ~(1 << 16));
173 WREG32_P(mmVCE_UENC_CLOCK_GATING, 0x1FF000, ~0xFF9FF000);
174 WREG32_P(mmVCE_UENC_REG_CLOCK_GATING, 0x3F, ~0x3F);
175 WREG32(mmVCE_CLOCK_GATING_B, 0xf7);
176
177 WREG32(mmVCE_LMI_CTRL, 0x00398000);
178 WREG32_P(mmVCE_LMI_CACHE_CTRL, 0x0, ~0x1);
179 WREG32(mmVCE_LMI_SWAP_CNTL, 0);
180 WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
181 WREG32(mmVCE_LMI_VM_CTRL, 0);
182
183 WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
184
185 offset = AMDGPU_VCE_FIRMWARE_OFFSET;
186 size = VCE_V2_0_FW_SIZE;
187 WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
188 WREG32(mmVCE_VCPU_CACHE_SIZE0, size);
189
190 offset += size;
191 size = VCE_V2_0_STACK_SIZE;
192 WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0x7fffffff);
193 WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
194
195 offset += size;
196 size = VCE_V2_0_DATA_SIZE;
197 WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0x7fffffff);
198 WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
199
200 WREG32_P(mmVCE_LMI_CTRL2, 0x0, ~0x100);
201 WREG32_FIELD(VCE_SYS_INT_EN, VCE_SYS_INT_TRAP_INTERRUPT_EN, 1);
202 }
203
204 static bool vce_v2_0_is_idle(void *handle)
205 {
206 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
207
208 return !(RREG32(mmSRBM_STATUS2) & SRBM_STATUS2__VCE_BUSY_MASK);
209 }
210
211 static int vce_v2_0_wait_for_idle(void *handle)
212 {
213 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
214 unsigned i;
215
216 for (i = 0; i < adev->usec_timeout; i++) {
217 if (vce_v2_0_is_idle(handle))
218 return 0;
219 }
220 return -ETIMEDOUT;
221 }
222
223
224
225
226
227
228
229
230 static int vce_v2_0_start(struct amdgpu_device *adev)
231 {
232 struct amdgpu_ring *ring;
233 int r;
234
235
236 WREG32_P(mmVCE_STATUS, 1, ~1);
237
238 vce_v2_0_init_cg(adev);
239 vce_v2_0_disable_cg(adev);
240
241 vce_v2_0_mc_resume(adev);
242
243 ring = &adev->vce.ring[0];
244 WREG32(mmVCE_RB_RPTR, lower_32_bits(ring->wptr));
245 WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
246 WREG32(mmVCE_RB_BASE_LO, ring->gpu_addr);
247 WREG32(mmVCE_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
248 WREG32(mmVCE_RB_SIZE, ring->ring_size / 4);
249
250 ring = &adev->vce.ring[1];
251 WREG32(mmVCE_RB_RPTR2, lower_32_bits(ring->wptr));
252 WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
253 WREG32(mmVCE_RB_BASE_LO2, ring->gpu_addr);
254 WREG32(mmVCE_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
255 WREG32(mmVCE_RB_SIZE2, ring->ring_size / 4);
256
257 WREG32_FIELD(VCE_VCPU_CNTL, CLK_EN, 1);
258 WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 1);
259 mdelay(100);
260 WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 0);
261
262 r = vce_v2_0_firmware_loaded(adev);
263
264
265 WREG32_P(mmVCE_STATUS, 0, ~1);
266
267 if (r) {
268 DRM_ERROR("VCE not responding, giving up!!!\n");
269 return r;
270 }
271
272 return 0;
273 }
274
275 static int vce_v2_0_stop(struct amdgpu_device *adev)
276 {
277 int i;
278 int status;
279
280 if (vce_v2_0_lmi_clean(adev)) {
281 DRM_INFO("vce is not idle \n");
282 return 0;
283 }
284
285 if (vce_v2_0_wait_for_idle(adev)) {
286 DRM_INFO("VCE is busy, Can't set clock gating");
287 return 0;
288 }
289
290
291 WREG32_P(mmVCE_LMI_CTRL2, 1 << 8, ~(1 << 8));
292
293 for (i = 0; i < 100; ++i) {
294 status = RREG32(mmVCE_LMI_STATUS);
295 if (status & 0x240)
296 break;
297 mdelay(1);
298 }
299
300 WREG32_P(mmVCE_VCPU_CNTL, 0, ~0x80001);
301
302
303 WREG32_P(mmVCE_SOFT_RESET, 1, ~0x1);
304
305 WREG32(mmVCE_STATUS, 0);
306
307 return 0;
308 }
309
310 static void vce_v2_0_set_sw_cg(struct amdgpu_device *adev, bool gated)
311 {
312 u32 tmp;
313
314 if (gated) {
315 tmp = RREG32(mmVCE_CLOCK_GATING_B);
316 tmp |= 0xe70000;
317 WREG32(mmVCE_CLOCK_GATING_B, tmp);
318
319 tmp = RREG32(mmVCE_UENC_CLOCK_GATING);
320 tmp |= 0xff000000;
321 WREG32(mmVCE_UENC_CLOCK_GATING, tmp);
322
323 tmp = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
324 tmp &= ~0x3fc;
325 WREG32(mmVCE_UENC_REG_CLOCK_GATING, tmp);
326
327 WREG32(mmVCE_CGTT_CLK_OVERRIDE, 0);
328 } else {
329 tmp = RREG32(mmVCE_CLOCK_GATING_B);
330 tmp |= 0xe7;
331 tmp &= ~0xe70000;
332 WREG32(mmVCE_CLOCK_GATING_B, tmp);
333
334 tmp = RREG32(mmVCE_UENC_CLOCK_GATING);
335 tmp |= 0x1fe000;
336 tmp &= ~0xff000000;
337 WREG32(mmVCE_UENC_CLOCK_GATING, tmp);
338
339 tmp = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
340 tmp |= 0x3fc;
341 WREG32(mmVCE_UENC_REG_CLOCK_GATING, tmp);
342 }
343 }
344
345 static void vce_v2_0_set_dyn_cg(struct amdgpu_device *adev, bool gated)
346 {
347 u32 orig, tmp;
348
349
350
351
352 tmp = RREG32(mmVCE_CLOCK_GATING_B);
353 tmp &= ~0x00060006;
354
355
356 if (gated) {
357 tmp |= 0xe10000;
358 WREG32(mmVCE_CLOCK_GATING_B, tmp);
359 } else {
360 tmp |= 0xe1;
361 tmp &= ~0xe10000;
362 WREG32(mmVCE_CLOCK_GATING_B, tmp);
363 }
364
365 orig = tmp = RREG32(mmVCE_UENC_CLOCK_GATING);
366 tmp &= ~0x1fe000;
367 tmp &= ~0xff000000;
368 if (tmp != orig)
369 WREG32(mmVCE_UENC_CLOCK_GATING, tmp);
370
371 orig = tmp = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
372 tmp &= ~0x3fc;
373 if (tmp != orig)
374 WREG32(mmVCE_UENC_REG_CLOCK_GATING, tmp);
375
376
377 WREG32(mmVCE_UENC_REG_CLOCK_GATING, 0x00);
378
379 if(gated)
380 WREG32(mmVCE_CGTT_CLK_OVERRIDE, 0);
381 }
382
383 static void vce_v2_0_enable_mgcg(struct amdgpu_device *adev, bool enable,
384 bool sw_cg)
385 {
386 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG)) {
387 if (sw_cg)
388 vce_v2_0_set_sw_cg(adev, true);
389 else
390 vce_v2_0_set_dyn_cg(adev, true);
391 } else {
392 vce_v2_0_disable_cg(adev);
393
394 if (sw_cg)
395 vce_v2_0_set_sw_cg(adev, false);
396 else
397 vce_v2_0_set_dyn_cg(adev, false);
398 }
399 }
400
401 static int vce_v2_0_early_init(void *handle)
402 {
403 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
404
405 adev->vce.num_rings = 2;
406
407 vce_v2_0_set_ring_funcs(adev);
408 vce_v2_0_set_irq_funcs(adev);
409
410 return 0;
411 }
412
413 static int vce_v2_0_sw_init(void *handle)
414 {
415 struct amdgpu_ring *ring;
416 int r, i;
417 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
418
419
420 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 167, &adev->vce.irq);
421 if (r)
422 return r;
423
424 r = amdgpu_vce_sw_init(adev, VCE_V2_0_FW_SIZE +
425 VCE_V2_0_STACK_SIZE + VCE_V2_0_DATA_SIZE);
426 if (r)
427 return r;
428
429 r = amdgpu_vce_resume(adev);
430 if (r)
431 return r;
432
433 for (i = 0; i < adev->vce.num_rings; i++) {
434 ring = &adev->vce.ring[i];
435 sprintf(ring->name, "vce%d", i);
436 r = amdgpu_ring_init(adev, ring, 512,
437 &adev->vce.irq, 0);
438 if (r)
439 return r;
440 }
441
442 r = amdgpu_vce_entity_init(adev);
443
444 return r;
445 }
446
447 static int vce_v2_0_sw_fini(void *handle)
448 {
449 int r;
450 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
451
452 r = amdgpu_vce_suspend(adev);
453 if (r)
454 return r;
455
456 return amdgpu_vce_sw_fini(adev);
457 }
458
459 static int vce_v2_0_hw_init(void *handle)
460 {
461 int r, i;
462 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
463
464 amdgpu_asic_set_vce_clocks(adev, 10000, 10000);
465 vce_v2_0_enable_mgcg(adev, true, false);
466
467 for (i = 0; i < adev->vce.num_rings; i++) {
468 r = amdgpu_ring_test_helper(&adev->vce.ring[i]);
469 if (r)
470 return r;
471 }
472
473 DRM_INFO("VCE initialized successfully.\n");
474
475 return 0;
476 }
477
478 static int vce_v2_0_hw_fini(void *handle)
479 {
480 return 0;
481 }
482
483 static int vce_v2_0_suspend(void *handle)
484 {
485 int r;
486 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
487
488 r = vce_v2_0_hw_fini(adev);
489 if (r)
490 return r;
491
492 return amdgpu_vce_suspend(adev);
493 }
494
495 static int vce_v2_0_resume(void *handle)
496 {
497 int r;
498 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
499
500 r = amdgpu_vce_resume(adev);
501 if (r)
502 return r;
503
504 return vce_v2_0_hw_init(adev);
505 }
506
507 static int vce_v2_0_soft_reset(void *handle)
508 {
509 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
510
511 WREG32_FIELD(SRBM_SOFT_RESET, SOFT_RESET_VCE, 1);
512 mdelay(5);
513
514 return vce_v2_0_start(adev);
515 }
516
517 static int vce_v2_0_set_interrupt_state(struct amdgpu_device *adev,
518 struct amdgpu_irq_src *source,
519 unsigned type,
520 enum amdgpu_interrupt_state state)
521 {
522 uint32_t val = 0;
523
524 if (state == AMDGPU_IRQ_STATE_ENABLE)
525 val |= VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK;
526
527 WREG32_P(mmVCE_SYS_INT_EN, val, ~VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK);
528 return 0;
529 }
530
531 static int vce_v2_0_process_interrupt(struct amdgpu_device *adev,
532 struct amdgpu_irq_src *source,
533 struct amdgpu_iv_entry *entry)
534 {
535 DRM_DEBUG("IH: VCE\n");
536 switch (entry->src_data[0]) {
537 case 0:
538 case 1:
539 amdgpu_fence_process(&adev->vce.ring[entry->src_data[0]]);
540 break;
541 default:
542 DRM_ERROR("Unhandled interrupt: %d %d\n",
543 entry->src_id, entry->src_data[0]);
544 break;
545 }
546
547 return 0;
548 }
549
550 static int vce_v2_0_set_clockgating_state(void *handle,
551 enum amd_clockgating_state state)
552 {
553 bool gate = false;
554 bool sw_cg = false;
555
556 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
557
558 if (state == AMD_CG_STATE_GATE) {
559 gate = true;
560 sw_cg = true;
561 }
562
563 vce_v2_0_enable_mgcg(adev, gate, sw_cg);
564
565 return 0;
566 }
567
568 static int vce_v2_0_set_powergating_state(void *handle,
569 enum amd_powergating_state state)
570 {
571
572
573
574
575
576
577
578 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
579
580 if (state == AMD_PG_STATE_GATE)
581 return vce_v2_0_stop(adev);
582 else
583 return vce_v2_0_start(adev);
584 }
585
586 static const struct amd_ip_funcs vce_v2_0_ip_funcs = {
587 .name = "vce_v2_0",
588 .early_init = vce_v2_0_early_init,
589 .late_init = NULL,
590 .sw_init = vce_v2_0_sw_init,
591 .sw_fini = vce_v2_0_sw_fini,
592 .hw_init = vce_v2_0_hw_init,
593 .hw_fini = vce_v2_0_hw_fini,
594 .suspend = vce_v2_0_suspend,
595 .resume = vce_v2_0_resume,
596 .is_idle = vce_v2_0_is_idle,
597 .wait_for_idle = vce_v2_0_wait_for_idle,
598 .soft_reset = vce_v2_0_soft_reset,
599 .set_clockgating_state = vce_v2_0_set_clockgating_state,
600 .set_powergating_state = vce_v2_0_set_powergating_state,
601 };
602
603 static const struct amdgpu_ring_funcs vce_v2_0_ring_funcs = {
604 .type = AMDGPU_RING_TYPE_VCE,
605 .align_mask = 0xf,
606 .nop = VCE_CMD_NO_OP,
607 .support_64bit_ptrs = false,
608 .no_user_fence = true,
609 .get_rptr = vce_v2_0_ring_get_rptr,
610 .get_wptr = vce_v2_0_ring_get_wptr,
611 .set_wptr = vce_v2_0_ring_set_wptr,
612 .parse_cs = amdgpu_vce_ring_parse_cs,
613 .emit_frame_size = 6,
614 .emit_ib_size = 4,
615 .emit_ib = amdgpu_vce_ring_emit_ib,
616 .emit_fence = amdgpu_vce_ring_emit_fence,
617 .test_ring = amdgpu_vce_ring_test_ring,
618 .test_ib = amdgpu_vce_ring_test_ib,
619 .insert_nop = amdgpu_ring_insert_nop,
620 .pad_ib = amdgpu_ring_generic_pad_ib,
621 .begin_use = amdgpu_vce_ring_begin_use,
622 .end_use = amdgpu_vce_ring_end_use,
623 };
624
625 static void vce_v2_0_set_ring_funcs(struct amdgpu_device *adev)
626 {
627 int i;
628
629 for (i = 0; i < adev->vce.num_rings; i++) {
630 adev->vce.ring[i].funcs = &vce_v2_0_ring_funcs;
631 adev->vce.ring[i].me = i;
632 }
633 }
634
635 static const struct amdgpu_irq_src_funcs vce_v2_0_irq_funcs = {
636 .set = vce_v2_0_set_interrupt_state,
637 .process = vce_v2_0_process_interrupt,
638 };
639
640 static void vce_v2_0_set_irq_funcs(struct amdgpu_device *adev)
641 {
642 adev->vce.irq.num_types = 1;
643 adev->vce.irq.funcs = &vce_v2_0_irq_funcs;
644 };
645
646 const struct amdgpu_ip_block_version vce_v2_0_ip_block =
647 {
648 .type = AMD_IP_BLOCK_TYPE_VCE,
649 .major = 2,
650 .minor = 0,
651 .rev = 0,
652 .funcs = &vce_v2_0_ip_funcs,
653 };