Loading...
1/*
2 * Copyright 2019 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 */
23
24#include <linux/firmware.h>
25#include <drm/drm_drv.h>
26
27#include "amdgpu.h"
28#include "amdgpu_vcn.h"
29#include "amdgpu_pm.h"
30#include "soc15.h"
31#include "soc15d.h"
32#include "vcn_v2_0.h"
33#include "mmsch_v1_0.h"
34#include "vcn_v2_5.h"
35
36#include "vcn/vcn_2_5_offset.h"
37#include "vcn/vcn_2_5_sh_mask.h"
38#include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
39
40#define VCN_VID_SOC_ADDRESS_2_0 0x1fa00
41#define VCN1_VID_SOC_ADDRESS_3_0 0x48200
42
43#define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x27
44#define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x0f
45#define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET 0x10
46#define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET 0x11
47#define mmUVD_NO_OP_INTERNAL_OFFSET 0x29
48#define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET 0x66
49#define mmUVD_SCRATCH9_INTERNAL_OFFSET 0xc01d
50
51#define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET 0x431
52#define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET 0x3b4
53#define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 0x3b5
54#define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET 0x25c
55
56#define VCN25_MAX_HW_INSTANCES_ARCTURUS 2
57
58static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
59static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
60static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
61static int vcn_v2_5_set_powergating_state(void *handle,
62 enum amd_powergating_state state);
63static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
64 int inst_idx, struct dpg_pause_state *new_state);
65static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
66static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev);
67
68static int amdgpu_ih_clientid_vcns[] = {
69 SOC15_IH_CLIENTID_VCN,
70 SOC15_IH_CLIENTID_VCN1
71};
72
73/**
74 * vcn_v2_5_early_init - set function pointers and load microcode
75 *
76 * @handle: amdgpu_device pointer
77 *
78 * Set ring and irq function pointers
79 * Load microcode from filesystem
80 */
81static int vcn_v2_5_early_init(void *handle)
82{
83 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
84
85 if (amdgpu_sriov_vf(adev)) {
86 adev->vcn.num_vcn_inst = 2;
87 adev->vcn.harvest_config = 0;
88 adev->vcn.num_enc_rings = 1;
89 } else {
90 u32 harvest;
91 int i;
92
93 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
94 harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
95 if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
96 adev->vcn.harvest_config |= 1 << i;
97 }
98 if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
99 AMDGPU_VCN_HARVEST_VCN1))
100 /* both instances are harvested, disable the block */
101 return -ENOENT;
102
103 adev->vcn.num_enc_rings = 2;
104 }
105
106 vcn_v2_5_set_dec_ring_funcs(adev);
107 vcn_v2_5_set_enc_ring_funcs(adev);
108 vcn_v2_5_set_irq_funcs(adev);
109 vcn_v2_5_set_ras_funcs(adev);
110
111 return amdgpu_vcn_early_init(adev);
112}
113
114/**
115 * vcn_v2_5_sw_init - sw init for VCN block
116 *
117 * @handle: amdgpu_device pointer
118 *
119 * Load firmware and sw initialization
120 */
121static int vcn_v2_5_sw_init(void *handle)
122{
123 struct amdgpu_ring *ring;
124 int i, j, r;
125 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
126
127 for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
128 if (adev->vcn.harvest_config & (1 << j))
129 continue;
130 /* VCN DEC TRAP */
131 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
132 VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
133 if (r)
134 return r;
135
136 /* VCN ENC TRAP */
137 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
138 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
139 i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
140 if (r)
141 return r;
142 }
143
144 /* VCN POISON TRAP */
145 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
146 VCN_2_6__SRCID_UVD_POISON, &adev->vcn.inst[j].ras_poison_irq);
147 if (r)
148 return r;
149 }
150
151 r = amdgpu_vcn_sw_init(adev);
152 if (r)
153 return r;
154
155 amdgpu_vcn_setup_ucode(adev);
156
157 r = amdgpu_vcn_resume(adev);
158 if (r)
159 return r;
160
161 for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
162 volatile struct amdgpu_fw_shared *fw_shared;
163
164 if (adev->vcn.harvest_config & (1 << j))
165 continue;
166 adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
167 adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
168 adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
169 adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
170 adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
171 adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
172
173 adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
174 adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
175 adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
176 adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
177 adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
178 adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
179 adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
180 adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
181 adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
182 adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
183
184 ring = &adev->vcn.inst[j].ring_dec;
185 ring->use_doorbell = true;
186
187 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
188 (amdgpu_sriov_vf(adev) ? 2*j : 8*j);
189
190 if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(2, 5, 0))
191 ring->vm_hub = AMDGPU_MMHUB1(0);
192 else
193 ring->vm_hub = AMDGPU_MMHUB0(0);
194
195 sprintf(ring->name, "vcn_dec_%d", j);
196 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
197 0, AMDGPU_RING_PRIO_DEFAULT, NULL);
198 if (r)
199 return r;
200
201 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
202 enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
203
204 ring = &adev->vcn.inst[j].ring_enc[i];
205 ring->use_doorbell = true;
206
207 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
208 (amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
209
210 if (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
211 IP_VERSION(2, 5, 0))
212 ring->vm_hub = AMDGPU_MMHUB1(0);
213 else
214 ring->vm_hub = AMDGPU_MMHUB0(0);
215
216 sprintf(ring->name, "vcn_enc_%d.%d", j, i);
217 r = amdgpu_ring_init(adev, ring, 512,
218 &adev->vcn.inst[j].irq, 0,
219 hw_prio, NULL);
220 if (r)
221 return r;
222 }
223
224 fw_shared = adev->vcn.inst[j].fw_shared.cpu_addr;
225 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
226
227 if (amdgpu_vcnfw_log)
228 amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
229 }
230
231 if (amdgpu_sriov_vf(adev)) {
232 r = amdgpu_virt_alloc_mm_table(adev);
233 if (r)
234 return r;
235 }
236
237 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
238 adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
239
240 r = amdgpu_vcn_ras_sw_init(adev);
241 if (r)
242 return r;
243
244 return 0;
245}
246
247/**
248 * vcn_v2_5_sw_fini - sw fini for VCN block
249 *
250 * @handle: amdgpu_device pointer
251 *
252 * VCN suspend and free up sw allocation
253 */
254static int vcn_v2_5_sw_fini(void *handle)
255{
256 int i, r, idx;
257 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
258 volatile struct amdgpu_fw_shared *fw_shared;
259
260 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
261 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
262 if (adev->vcn.harvest_config & (1 << i))
263 continue;
264 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
265 fw_shared->present_flag_0 = 0;
266 }
267 drm_dev_exit(idx);
268 }
269
270
271 if (amdgpu_sriov_vf(adev))
272 amdgpu_virt_free_mm_table(adev);
273
274 r = amdgpu_vcn_suspend(adev);
275 if (r)
276 return r;
277
278 r = amdgpu_vcn_sw_fini(adev);
279
280 return r;
281}
282
283/**
284 * vcn_v2_5_hw_init - start and test VCN block
285 *
286 * @handle: amdgpu_device pointer
287 *
288 * Initialize the hardware, boot up the VCPU and do some testing
289 */
290static int vcn_v2_5_hw_init(void *handle)
291{
292 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
293 struct amdgpu_ring *ring;
294 int i, j, r = 0;
295
296 if (amdgpu_sriov_vf(adev))
297 r = vcn_v2_5_sriov_start(adev);
298
299 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
300 if (adev->vcn.harvest_config & (1 << j))
301 continue;
302
303 if (amdgpu_sriov_vf(adev)) {
304 adev->vcn.inst[j].ring_enc[0].sched.ready = true;
305 adev->vcn.inst[j].ring_enc[1].sched.ready = false;
306 adev->vcn.inst[j].ring_enc[2].sched.ready = false;
307 adev->vcn.inst[j].ring_dec.sched.ready = true;
308 } else {
309
310 ring = &adev->vcn.inst[j].ring_dec;
311
312 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
313 ring->doorbell_index, j);
314
315 r = amdgpu_ring_test_helper(ring);
316 if (r)
317 goto done;
318
319 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
320 ring = &adev->vcn.inst[j].ring_enc[i];
321 r = amdgpu_ring_test_helper(ring);
322 if (r)
323 goto done;
324 }
325 }
326 }
327
328done:
329 if (!r)
330 DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
331 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
332
333 return r;
334}
335
336/**
337 * vcn_v2_5_hw_fini - stop the hardware block
338 *
339 * @handle: amdgpu_device pointer
340 *
341 * Stop the VCN block, mark ring as not ready any more
342 */
343static int vcn_v2_5_hw_fini(void *handle)
344{
345 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
346 int i;
347
348 cancel_delayed_work_sync(&adev->vcn.idle_work);
349
350 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
351 if (adev->vcn.harvest_config & (1 << i))
352 continue;
353
354 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
355 (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
356 RREG32_SOC15(VCN, i, mmUVD_STATUS)))
357 vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
358
359 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
360 amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
361 }
362
363 return 0;
364}
365
366/**
367 * vcn_v2_5_suspend - suspend VCN block
368 *
369 * @handle: amdgpu_device pointer
370 *
371 * HW fini and suspend VCN block
372 */
373static int vcn_v2_5_suspend(void *handle)
374{
375 int r;
376 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
377
378 r = vcn_v2_5_hw_fini(adev);
379 if (r)
380 return r;
381
382 r = amdgpu_vcn_suspend(adev);
383
384 return r;
385}
386
387/**
388 * vcn_v2_5_resume - resume VCN block
389 *
390 * @handle: amdgpu_device pointer
391 *
392 * Resume firmware and hw init VCN block
393 */
394static int vcn_v2_5_resume(void *handle)
395{
396 int r;
397 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
398
399 r = amdgpu_vcn_resume(adev);
400 if (r)
401 return r;
402
403 r = vcn_v2_5_hw_init(adev);
404
405 return r;
406}
407
408/**
409 * vcn_v2_5_mc_resume - memory controller programming
410 *
411 * @adev: amdgpu_device pointer
412 *
413 * Let the VCN memory controller know it's offsets
414 */
415static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
416{
417 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
418 uint32_t offset;
419 int i;
420
421 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
422 if (adev->vcn.harvest_config & (1 << i))
423 continue;
424 /* cache window 0: fw */
425 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
426 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
427 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
428 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
429 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
430 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
431 offset = 0;
432 } else {
433 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
434 lower_32_bits(adev->vcn.inst[i].gpu_addr));
435 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
436 upper_32_bits(adev->vcn.inst[i].gpu_addr));
437 offset = size;
438 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
439 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
440 }
441 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
442
443 /* cache window 1: stack */
444 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
445 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
446 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
447 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
448 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
449 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
450
451 /* cache window 2: context */
452 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
453 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
454 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
455 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
456 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
457 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
458
459 /* non-cache window */
460 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
461 lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
462 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
463 upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
464 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
465 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
466 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
467 }
468}
469
470static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
471{
472 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
473 uint32_t offset;
474
475 /* cache window 0: fw */
476 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
477 if (!indirect) {
478 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
479 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
480 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
481 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
482 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
483 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
484 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
485 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
486 } else {
487 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
488 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
489 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
490 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
491 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
492 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
493 }
494 offset = 0;
495 } else {
496 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
497 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
498 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
499 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
501 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
502 offset = size;
503 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
505 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
506 }
507
508 if (!indirect)
509 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
510 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
511 else
512 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
513 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
514
515 /* cache window 1: stack */
516 if (!indirect) {
517 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
519 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
520 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
522 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
523 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
524 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
525 } else {
526 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
527 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
528 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
529 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
530 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
531 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
532 }
533 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
534 VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
535
536 /* cache window 2: context */
537 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
538 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
539 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
540 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
541 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
542 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
543 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
544 VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
545 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
546 VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
547
548 /* non-cache window */
549 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
550 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
551 lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
552 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
553 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
554 upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
555 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
556 VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
557 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
558 VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
559 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
560
561 /* VCN global tiling registers */
562 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
563 VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
564}
565
566/**
567 * vcn_v2_5_disable_clock_gating - disable VCN clock gating
568 *
569 * @adev: amdgpu_device pointer
570 *
571 * Disable clock gating for VCN block
572 */
573static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
574{
575 uint32_t data;
576 int i;
577
578 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
579 if (adev->vcn.harvest_config & (1 << i))
580 continue;
581 /* UVD disable CGC */
582 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
583 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
584 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
585 else
586 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
587 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
588 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
589 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
590
591 data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
592 data &= ~(UVD_CGC_GATE__SYS_MASK
593 | UVD_CGC_GATE__UDEC_MASK
594 | UVD_CGC_GATE__MPEG2_MASK
595 | UVD_CGC_GATE__REGS_MASK
596 | UVD_CGC_GATE__RBC_MASK
597 | UVD_CGC_GATE__LMI_MC_MASK
598 | UVD_CGC_GATE__LMI_UMC_MASK
599 | UVD_CGC_GATE__IDCT_MASK
600 | UVD_CGC_GATE__MPRD_MASK
601 | UVD_CGC_GATE__MPC_MASK
602 | UVD_CGC_GATE__LBSI_MASK
603 | UVD_CGC_GATE__LRBBM_MASK
604 | UVD_CGC_GATE__UDEC_RE_MASK
605 | UVD_CGC_GATE__UDEC_CM_MASK
606 | UVD_CGC_GATE__UDEC_IT_MASK
607 | UVD_CGC_GATE__UDEC_DB_MASK
608 | UVD_CGC_GATE__UDEC_MP_MASK
609 | UVD_CGC_GATE__WCB_MASK
610 | UVD_CGC_GATE__VCPU_MASK
611 | UVD_CGC_GATE__MMSCH_MASK);
612
613 WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
614
615 SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0, 0xFFFFFFFF);
616
617 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
618 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
619 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
620 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
621 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
622 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
623 | UVD_CGC_CTRL__SYS_MODE_MASK
624 | UVD_CGC_CTRL__UDEC_MODE_MASK
625 | UVD_CGC_CTRL__MPEG2_MODE_MASK
626 | UVD_CGC_CTRL__REGS_MODE_MASK
627 | UVD_CGC_CTRL__RBC_MODE_MASK
628 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
629 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
630 | UVD_CGC_CTRL__IDCT_MODE_MASK
631 | UVD_CGC_CTRL__MPRD_MODE_MASK
632 | UVD_CGC_CTRL__MPC_MODE_MASK
633 | UVD_CGC_CTRL__LBSI_MODE_MASK
634 | UVD_CGC_CTRL__LRBBM_MODE_MASK
635 | UVD_CGC_CTRL__WCB_MODE_MASK
636 | UVD_CGC_CTRL__VCPU_MODE_MASK
637 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
638 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
639
640 /* turn on */
641 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
642 data |= (UVD_SUVD_CGC_GATE__SRE_MASK
643 | UVD_SUVD_CGC_GATE__SIT_MASK
644 | UVD_SUVD_CGC_GATE__SMP_MASK
645 | UVD_SUVD_CGC_GATE__SCM_MASK
646 | UVD_SUVD_CGC_GATE__SDB_MASK
647 | UVD_SUVD_CGC_GATE__SRE_H264_MASK
648 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
649 | UVD_SUVD_CGC_GATE__SIT_H264_MASK
650 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
651 | UVD_SUVD_CGC_GATE__SCM_H264_MASK
652 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
653 | UVD_SUVD_CGC_GATE__SDB_H264_MASK
654 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
655 | UVD_SUVD_CGC_GATE__SCLR_MASK
656 | UVD_SUVD_CGC_GATE__UVD_SC_MASK
657 | UVD_SUVD_CGC_GATE__ENT_MASK
658 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
659 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
660 | UVD_SUVD_CGC_GATE__SITE_MASK
661 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
662 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
663 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
664 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
665 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
666 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
667
668 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
669 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
670 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
671 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
672 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
673 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
674 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
675 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
676 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
677 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
678 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
679 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
680 }
681}
682
683static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
684 uint8_t sram_sel, int inst_idx, uint8_t indirect)
685{
686 uint32_t reg_data = 0;
687
688 /* enable sw clock gating control */
689 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
690 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
691 else
692 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
693 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
694 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
695 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
696 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
697 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
698 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
699 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
700 UVD_CGC_CTRL__SYS_MODE_MASK |
701 UVD_CGC_CTRL__UDEC_MODE_MASK |
702 UVD_CGC_CTRL__MPEG2_MODE_MASK |
703 UVD_CGC_CTRL__REGS_MODE_MASK |
704 UVD_CGC_CTRL__RBC_MODE_MASK |
705 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
706 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
707 UVD_CGC_CTRL__IDCT_MODE_MASK |
708 UVD_CGC_CTRL__MPRD_MODE_MASK |
709 UVD_CGC_CTRL__MPC_MODE_MASK |
710 UVD_CGC_CTRL__LBSI_MODE_MASK |
711 UVD_CGC_CTRL__LRBBM_MODE_MASK |
712 UVD_CGC_CTRL__WCB_MODE_MASK |
713 UVD_CGC_CTRL__VCPU_MODE_MASK |
714 UVD_CGC_CTRL__MMSCH_MODE_MASK);
715 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
716 VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
717
718 /* turn off clock gating */
719 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
720 VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
721
722 /* turn on SUVD clock gating */
723 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
724 VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
725
726 /* turn on sw mode in UVD_SUVD_CGC_CTRL */
727 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
728 VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
729}
730
731/**
732 * vcn_v2_5_enable_clock_gating - enable VCN clock gating
733 *
734 * @adev: amdgpu_device pointer
735 *
736 * Enable clock gating for VCN block
737 */
738static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
739{
740 uint32_t data = 0;
741 int i;
742
743 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
744 if (adev->vcn.harvest_config & (1 << i))
745 continue;
746 /* enable UVD CGC */
747 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
748 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
749 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
750 else
751 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
752 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
753 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
754 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
755
756 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
757 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
758 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
759 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
760 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
761 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
762 | UVD_CGC_CTRL__SYS_MODE_MASK
763 | UVD_CGC_CTRL__UDEC_MODE_MASK
764 | UVD_CGC_CTRL__MPEG2_MODE_MASK
765 | UVD_CGC_CTRL__REGS_MODE_MASK
766 | UVD_CGC_CTRL__RBC_MODE_MASK
767 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
768 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
769 | UVD_CGC_CTRL__IDCT_MODE_MASK
770 | UVD_CGC_CTRL__MPRD_MODE_MASK
771 | UVD_CGC_CTRL__MPC_MODE_MASK
772 | UVD_CGC_CTRL__LBSI_MODE_MASK
773 | UVD_CGC_CTRL__LRBBM_MODE_MASK
774 | UVD_CGC_CTRL__WCB_MODE_MASK
775 | UVD_CGC_CTRL__VCPU_MODE_MASK);
776 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
777
778 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
779 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
780 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
781 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
782 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
783 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
784 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
785 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
786 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
787 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
788 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
789 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
790 }
791}
792
793static void vcn_v2_6_enable_ras(struct amdgpu_device *adev, int inst_idx,
794 bool indirect)
795{
796 uint32_t tmp;
797
798 if (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(2, 6, 0))
799 return;
800
801 tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
802 VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
803 VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
804 VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
805 WREG32_SOC15_DPG_MODE(inst_idx,
806 SOC15_DPG_MODE_OFFSET(VCN, 0, mmVCN_RAS_CNTL),
807 tmp, 0, indirect);
808
809 tmp = UVD_VCPU_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
810 WREG32_SOC15_DPG_MODE(inst_idx,
811 SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_VCPU_INT_EN),
812 tmp, 0, indirect);
813
814 tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
815 WREG32_SOC15_DPG_MODE(inst_idx,
816 SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_SYS_INT_EN),
817 tmp, 0, indirect);
818}
819
820static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
821{
822 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
823 struct amdgpu_ring *ring;
824 uint32_t rb_bufsz, tmp;
825
826 /* disable register anti-hang mechanism */
827 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
828 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
829 /* enable dynamic power gating mode */
830 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
831 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
832 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
833 WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
834
835 if (indirect)
836 adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
837
838 /* enable clock gating */
839 vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
840
841 /* enable VCPU clock */
842 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
843 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
844 tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
845 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
846 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
847
848 /* disable master interupt */
849 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
850 VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
851
852 /* setup mmUVD_LMI_CTRL */
853 tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
854 UVD_LMI_CTRL__REQ_MODE_MASK |
855 UVD_LMI_CTRL__CRC_RESET_MASK |
856 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
857 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
858 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
859 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
860 0x00100000L);
861 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
862 VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
863
864 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
865 VCN, 0, mmUVD_MPC_CNTL),
866 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
867
868 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
869 VCN, 0, mmUVD_MPC_SET_MUXA0),
870 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
871 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
872 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
873 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
874
875 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
876 VCN, 0, mmUVD_MPC_SET_MUXB0),
877 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
878 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
879 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
880 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
881
882 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
883 VCN, 0, mmUVD_MPC_SET_MUX),
884 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
885 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
886 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
887
888 vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
889
890 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
891 VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
892 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
893 VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
894
895 /* enable LMI MC and UMC channels */
896 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
897 VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
898
899 vcn_v2_6_enable_ras(adev, inst_idx, indirect);
900
901 /* unblock VCPU register access */
902 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
903 VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
904
905 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
906 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
907 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
908 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
909
910 /* enable master interrupt */
911 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
912 VCN, 0, mmUVD_MASTINT_EN),
913 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
914
915 if (indirect)
916 amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
917
918 ring = &adev->vcn.inst[inst_idx].ring_dec;
919 /* force RBC into idle state */
920 rb_bufsz = order_base_2(ring->ring_size);
921 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
922 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
923 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
924 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
925 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
926 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
927
928 /* Stall DPG before WPTR/RPTR reset */
929 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
930 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
931 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
932 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
933
934 /* set the write pointer delay */
935 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
936
937 /* set the wb address */
938 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
939 (upper_32_bits(ring->gpu_addr) >> 2));
940
941 /* program the RB_BASE for ring buffer */
942 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
943 lower_32_bits(ring->gpu_addr));
944 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
945 upper_32_bits(ring->gpu_addr));
946
947 /* Initialize the ring buffer's read and write pointers */
948 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
949
950 WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
951
952 ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
953 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
954 lower_32_bits(ring->wptr));
955
956 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
957 /* Unstall DPG */
958 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
959 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
960
961 return 0;
962}
963
964static int vcn_v2_5_start(struct amdgpu_device *adev)
965{
966 struct amdgpu_ring *ring;
967 uint32_t rb_bufsz, tmp;
968 int i, j, k, r;
969
970 if (adev->pm.dpm_enabled)
971 amdgpu_dpm_enable_uvd(adev, true);
972
973 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
974 if (adev->vcn.harvest_config & (1 << i))
975 continue;
976 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
977 r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
978 continue;
979 }
980
981 /* disable register anti-hang mechanism */
982 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
983 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
984
985 /* set uvd status busy */
986 tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
987 WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
988 }
989
990 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
991 return 0;
992
993 /*SW clock gating */
994 vcn_v2_5_disable_clock_gating(adev);
995
996 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
997 if (adev->vcn.harvest_config & (1 << i))
998 continue;
999 /* enable VCPU clock */
1000 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1001 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1002
1003 /* disable master interrupt */
1004 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1005 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1006
1007 /* setup mmUVD_LMI_CTRL */
1008 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1009 tmp &= ~0xff;
1010 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
1011 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1012 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1013 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1014 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1015
1016 /* setup mmUVD_MPC_CNTL */
1017 tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1018 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1019 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1020 WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1021
1022 /* setup UVD_MPC_SET_MUXA0 */
1023 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1024 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1025 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1026 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1027 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1028
1029 /* setup UVD_MPC_SET_MUXB0 */
1030 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1031 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1032 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1033 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1034 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1035
1036 /* setup mmUVD_MPC_SET_MUX */
1037 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1038 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1039 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1040 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1041 }
1042
1043 vcn_v2_5_mc_resume(adev);
1044
1045 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1046 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1047 if (adev->vcn.harvest_config & (1 << i))
1048 continue;
1049 /* VCN global tiling registers */
1050 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1051 adev->gfx.config.gb_addr_config);
1052 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1053 adev->gfx.config.gb_addr_config);
1054
1055 /* enable LMI MC and UMC channels */
1056 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1057 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1058
1059 /* unblock VCPU register access */
1060 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1061 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1062
1063 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1064 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1065
1066 for (k = 0; k < 10; ++k) {
1067 uint32_t status;
1068
1069 for (j = 0; j < 100; ++j) {
1070 status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1071 if (status & 2)
1072 break;
1073 if (amdgpu_emu_mode == 1)
1074 msleep(500);
1075 else
1076 mdelay(10);
1077 }
1078 r = 0;
1079 if (status & 2)
1080 break;
1081
1082 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1083 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1084 UVD_VCPU_CNTL__BLK_RST_MASK,
1085 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1086 mdelay(10);
1087 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1088 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1089
1090 mdelay(10);
1091 r = -1;
1092 }
1093
1094 if (r) {
1095 DRM_ERROR("VCN decode not responding, giving up!!!\n");
1096 return r;
1097 }
1098
1099 /* enable master interrupt */
1100 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1101 UVD_MASTINT_EN__VCPU_EN_MASK,
1102 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1103
1104 /* clear the busy bit of VCN_STATUS */
1105 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1106 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1107
1108 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1109
1110 ring = &adev->vcn.inst[i].ring_dec;
1111 /* force RBC into idle state */
1112 rb_bufsz = order_base_2(ring->ring_size);
1113 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1114 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1115 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1116 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1117 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1118 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1119
1120 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1121 /* program the RB_BASE for ring buffer */
1122 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1123 lower_32_bits(ring->gpu_addr));
1124 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1125 upper_32_bits(ring->gpu_addr));
1126
1127 /* Initialize the ring buffer's read and write pointers */
1128 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1129
1130 ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1131 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1132 lower_32_bits(ring->wptr));
1133 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1134
1135 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1136 ring = &adev->vcn.inst[i].ring_enc[0];
1137 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1138 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1139 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1140 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1141 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1142 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1143
1144 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1145 ring = &adev->vcn.inst[i].ring_enc[1];
1146 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1147 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1148 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1149 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1150 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1151 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1152 }
1153
1154 return 0;
1155}
1156
1157static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1158 struct amdgpu_mm_table *table)
1159{
1160 uint32_t data = 0, loop = 0, size = 0;
1161 uint64_t addr = table->gpu_addr;
1162 struct mmsch_v1_1_init_header *header = NULL;
1163
1164 header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1165 size = header->total_size;
1166
1167 /*
1168 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1169 * memory descriptor location
1170 */
1171 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1172 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1173
1174 /* 2, update vmid of descriptor */
1175 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1176 data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1177 /* use domain0 for MM scheduler */
1178 data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1179 WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1180
1181 /* 3, notify mmsch about the size of this descriptor */
1182 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1183
1184 /* 4, set resp to zero */
1185 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1186
1187 /*
1188 * 5, kick off the initialization and wait until
1189 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1190 */
1191 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1192
1193 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1194 loop = 10;
1195 while ((data & 0x10000002) != 0x10000002) {
1196 udelay(100);
1197 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1198 loop--;
1199 if (!loop)
1200 break;
1201 }
1202
1203 if (!loop) {
1204 dev_err(adev->dev,
1205 "failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1206 data);
1207 return -EBUSY;
1208 }
1209
1210 return 0;
1211}
1212
1213static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1214{
1215 struct amdgpu_ring *ring;
1216 uint32_t offset, size, tmp, i, rb_bufsz;
1217 uint32_t table_size = 0;
1218 struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1219 struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1220 struct mmsch_v1_0_cmd_end end = { { 0 } };
1221 uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1222 struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1223
1224 direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1225 direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1226 end.cmd_header.command_type = MMSCH_COMMAND__END;
1227
1228 header->version = MMSCH_VERSION;
1229 header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1230 init_table += header->total_size;
1231
1232 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1233 header->eng[i].table_offset = header->total_size;
1234 header->eng[i].init_status = 0;
1235 header->eng[i].table_size = 0;
1236
1237 table_size = 0;
1238
1239 MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1240 SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1241 ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1242
1243 size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1244 /* mc resume*/
1245 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1246 MMSCH_V1_0_INSERT_DIRECT_WT(
1247 SOC15_REG_OFFSET(VCN, i,
1248 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1249 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1250 MMSCH_V1_0_INSERT_DIRECT_WT(
1251 SOC15_REG_OFFSET(VCN, i,
1252 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1253 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1254 offset = 0;
1255 MMSCH_V1_0_INSERT_DIRECT_WT(
1256 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1257 } else {
1258 MMSCH_V1_0_INSERT_DIRECT_WT(
1259 SOC15_REG_OFFSET(VCN, i,
1260 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1261 lower_32_bits(adev->vcn.inst[i].gpu_addr));
1262 MMSCH_V1_0_INSERT_DIRECT_WT(
1263 SOC15_REG_OFFSET(VCN, i,
1264 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1265 upper_32_bits(adev->vcn.inst[i].gpu_addr));
1266 offset = size;
1267 MMSCH_V1_0_INSERT_DIRECT_WT(
1268 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1269 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1270 }
1271
1272 MMSCH_V1_0_INSERT_DIRECT_WT(
1273 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1274 size);
1275 MMSCH_V1_0_INSERT_DIRECT_WT(
1276 SOC15_REG_OFFSET(VCN, i,
1277 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1278 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1279 MMSCH_V1_0_INSERT_DIRECT_WT(
1280 SOC15_REG_OFFSET(VCN, i,
1281 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1282 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1283 MMSCH_V1_0_INSERT_DIRECT_WT(
1284 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1285 0);
1286 MMSCH_V1_0_INSERT_DIRECT_WT(
1287 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1288 AMDGPU_VCN_STACK_SIZE);
1289 MMSCH_V1_0_INSERT_DIRECT_WT(
1290 SOC15_REG_OFFSET(VCN, i,
1291 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1292 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1293 AMDGPU_VCN_STACK_SIZE));
1294 MMSCH_V1_0_INSERT_DIRECT_WT(
1295 SOC15_REG_OFFSET(VCN, i,
1296 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1297 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1298 AMDGPU_VCN_STACK_SIZE));
1299 MMSCH_V1_0_INSERT_DIRECT_WT(
1300 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1301 0);
1302 MMSCH_V1_0_INSERT_DIRECT_WT(
1303 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1304 AMDGPU_VCN_CONTEXT_SIZE);
1305
1306 ring = &adev->vcn.inst[i].ring_enc[0];
1307 ring->wptr = 0;
1308
1309 MMSCH_V1_0_INSERT_DIRECT_WT(
1310 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1311 lower_32_bits(ring->gpu_addr));
1312 MMSCH_V1_0_INSERT_DIRECT_WT(
1313 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1314 upper_32_bits(ring->gpu_addr));
1315 MMSCH_V1_0_INSERT_DIRECT_WT(
1316 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1317 ring->ring_size / 4);
1318
1319 ring = &adev->vcn.inst[i].ring_dec;
1320 ring->wptr = 0;
1321 MMSCH_V1_0_INSERT_DIRECT_WT(
1322 SOC15_REG_OFFSET(VCN, i,
1323 mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1324 lower_32_bits(ring->gpu_addr));
1325 MMSCH_V1_0_INSERT_DIRECT_WT(
1326 SOC15_REG_OFFSET(VCN, i,
1327 mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1328 upper_32_bits(ring->gpu_addr));
1329
1330 /* force RBC into idle state */
1331 rb_bufsz = order_base_2(ring->ring_size);
1332 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1333 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1334 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1335 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1336 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1337 MMSCH_V1_0_INSERT_DIRECT_WT(
1338 SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1339
1340 /* add end packet */
1341 memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1342 table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1343 init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1344
1345 /* refine header */
1346 header->eng[i].table_size = table_size;
1347 header->total_size += table_size;
1348 }
1349
1350 return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1351}
1352
1353static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1354{
1355 uint32_t tmp;
1356
1357 /* Wait for power status to be 1 */
1358 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1359 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1360
1361 /* wait for read ptr to be equal to write ptr */
1362 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1363 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1364
1365 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1366 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1367
1368 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1369 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1370
1371 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1372 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1373
1374 /* disable dynamic power gating mode */
1375 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1376 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1377
1378 return 0;
1379}
1380
1381static int vcn_v2_5_stop(struct amdgpu_device *adev)
1382{
1383 uint32_t tmp;
1384 int i, r = 0;
1385
1386 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1387 if (adev->vcn.harvest_config & (1 << i))
1388 continue;
1389 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1390 r = vcn_v2_5_stop_dpg_mode(adev, i);
1391 continue;
1392 }
1393
1394 /* wait for vcn idle */
1395 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1396 if (r)
1397 return r;
1398
1399 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1400 UVD_LMI_STATUS__READ_CLEAN_MASK |
1401 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1402 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1403 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1404 if (r)
1405 return r;
1406
1407 /* block LMI UMC channel */
1408 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1409 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1410 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1411
1412 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1413 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1414 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1415 if (r)
1416 return r;
1417
1418 /* block VCPU register access */
1419 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1420 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1421 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1422
1423 /* reset VCPU */
1424 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1425 UVD_VCPU_CNTL__BLK_RST_MASK,
1426 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1427
1428 /* disable VCPU clock */
1429 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1430 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1431
1432 /* clear status */
1433 WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1434
1435 vcn_v2_5_enable_clock_gating(adev);
1436
1437 /* enable register anti-hang mechanism */
1438 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1439 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1440 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1441 }
1442
1443 if (adev->pm.dpm_enabled)
1444 amdgpu_dpm_enable_uvd(adev, false);
1445
1446 return 0;
1447}
1448
1449static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1450 int inst_idx, struct dpg_pause_state *new_state)
1451{
1452 struct amdgpu_ring *ring;
1453 uint32_t reg_data = 0;
1454 int ret_code = 0;
1455
1456 /* pause/unpause if state is changed */
1457 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1458 DRM_DEBUG("dpg pause state changed %d -> %d",
1459 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based);
1460 reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1461 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1462
1463 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1464 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1465 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1466
1467 if (!ret_code) {
1468 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1469
1470 /* pause DPG */
1471 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1472 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1473
1474 /* wait for ACK */
1475 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1476 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1477 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1478
1479 /* Stall DPG before WPTR/RPTR reset */
1480 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1481 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1482 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1483
1484 /* Restore */
1485 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1486 ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1487 ring->wptr = 0;
1488 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1489 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1490 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1491 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1492 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1493 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1494
1495 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1496 ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1497 ring->wptr = 0;
1498 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1499 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1500 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1501 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1502 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1503 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1504
1505 /* Unstall DPG */
1506 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1507 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1508
1509 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1510 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1511 }
1512 } else {
1513 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1514 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1515 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1516 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1517 }
1518 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1519 }
1520
1521 return 0;
1522}
1523
1524/**
1525 * vcn_v2_5_dec_ring_get_rptr - get read pointer
1526 *
1527 * @ring: amdgpu_ring pointer
1528 *
1529 * Returns the current hardware read pointer
1530 */
1531static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1532{
1533 struct amdgpu_device *adev = ring->adev;
1534
1535 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1536}
1537
1538/**
1539 * vcn_v2_5_dec_ring_get_wptr - get write pointer
1540 *
1541 * @ring: amdgpu_ring pointer
1542 *
1543 * Returns the current hardware write pointer
1544 */
1545static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1546{
1547 struct amdgpu_device *adev = ring->adev;
1548
1549 if (ring->use_doorbell)
1550 return *ring->wptr_cpu_addr;
1551 else
1552 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1553}
1554
1555/**
1556 * vcn_v2_5_dec_ring_set_wptr - set write pointer
1557 *
1558 * @ring: amdgpu_ring pointer
1559 *
1560 * Commits the write pointer to the hardware
1561 */
1562static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1563{
1564 struct amdgpu_device *adev = ring->adev;
1565
1566 if (ring->use_doorbell) {
1567 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1568 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1569 } else {
1570 WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1571 }
1572}
1573
1574static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1575 .type = AMDGPU_RING_TYPE_VCN_DEC,
1576 .align_mask = 0xf,
1577 .secure_submission_supported = true,
1578 .get_rptr = vcn_v2_5_dec_ring_get_rptr,
1579 .get_wptr = vcn_v2_5_dec_ring_get_wptr,
1580 .set_wptr = vcn_v2_5_dec_ring_set_wptr,
1581 .emit_frame_size =
1582 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1583 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1584 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1585 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1586 6,
1587 .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1588 .emit_ib = vcn_v2_0_dec_ring_emit_ib,
1589 .emit_fence = vcn_v2_0_dec_ring_emit_fence,
1590 .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1591 .test_ring = vcn_v2_0_dec_ring_test_ring,
1592 .test_ib = amdgpu_vcn_dec_ring_test_ib,
1593 .insert_nop = vcn_v2_0_dec_ring_insert_nop,
1594 .insert_start = vcn_v2_0_dec_ring_insert_start,
1595 .insert_end = vcn_v2_0_dec_ring_insert_end,
1596 .pad_ib = amdgpu_ring_generic_pad_ib,
1597 .begin_use = amdgpu_vcn_ring_begin_use,
1598 .end_use = amdgpu_vcn_ring_end_use,
1599 .emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1600 .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1601 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1602};
1603
1604/**
1605 * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1606 *
1607 * @ring: amdgpu_ring pointer
1608 *
1609 * Returns the current hardware enc read pointer
1610 */
1611static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1612{
1613 struct amdgpu_device *adev = ring->adev;
1614
1615 if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1616 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1617 else
1618 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1619}
1620
1621/**
1622 * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1623 *
1624 * @ring: amdgpu_ring pointer
1625 *
1626 * Returns the current hardware enc write pointer
1627 */
1628static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1629{
1630 struct amdgpu_device *adev = ring->adev;
1631
1632 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1633 if (ring->use_doorbell)
1634 return *ring->wptr_cpu_addr;
1635 else
1636 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1637 } else {
1638 if (ring->use_doorbell)
1639 return *ring->wptr_cpu_addr;
1640 else
1641 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1642 }
1643}
1644
1645/**
1646 * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1647 *
1648 * @ring: amdgpu_ring pointer
1649 *
1650 * Commits the enc write pointer to the hardware
1651 */
1652static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1653{
1654 struct amdgpu_device *adev = ring->adev;
1655
1656 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1657 if (ring->use_doorbell) {
1658 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1659 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1660 } else {
1661 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1662 }
1663 } else {
1664 if (ring->use_doorbell) {
1665 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1666 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1667 } else {
1668 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1669 }
1670 }
1671}
1672
1673static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1674 .type = AMDGPU_RING_TYPE_VCN_ENC,
1675 .align_mask = 0x3f,
1676 .nop = VCN_ENC_CMD_NO_OP,
1677 .get_rptr = vcn_v2_5_enc_ring_get_rptr,
1678 .get_wptr = vcn_v2_5_enc_ring_get_wptr,
1679 .set_wptr = vcn_v2_5_enc_ring_set_wptr,
1680 .emit_frame_size =
1681 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1682 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1683 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1684 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1685 1, /* vcn_v2_0_enc_ring_insert_end */
1686 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1687 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1688 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1689 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1690 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1691 .test_ib = amdgpu_vcn_enc_ring_test_ib,
1692 .insert_nop = amdgpu_ring_insert_nop,
1693 .insert_end = vcn_v2_0_enc_ring_insert_end,
1694 .pad_ib = amdgpu_ring_generic_pad_ib,
1695 .begin_use = amdgpu_vcn_ring_begin_use,
1696 .end_use = amdgpu_vcn_ring_end_use,
1697 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1698 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1699 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1700};
1701
1702static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1703{
1704 int i;
1705
1706 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1707 if (adev->vcn.harvest_config & (1 << i))
1708 continue;
1709 adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1710 adev->vcn.inst[i].ring_dec.me = i;
1711 DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1712 }
1713}
1714
1715static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1716{
1717 int i, j;
1718
1719 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1720 if (adev->vcn.harvest_config & (1 << j))
1721 continue;
1722 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1723 adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1724 adev->vcn.inst[j].ring_enc[i].me = j;
1725 }
1726 DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1727 }
1728}
1729
1730static bool vcn_v2_5_is_idle(void *handle)
1731{
1732 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1733 int i, ret = 1;
1734
1735 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1736 if (adev->vcn.harvest_config & (1 << i))
1737 continue;
1738 ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1739 }
1740
1741 return ret;
1742}
1743
1744static int vcn_v2_5_wait_for_idle(void *handle)
1745{
1746 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1747 int i, ret = 0;
1748
1749 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1750 if (adev->vcn.harvest_config & (1 << i))
1751 continue;
1752 ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1753 UVD_STATUS__IDLE);
1754 if (ret)
1755 return ret;
1756 }
1757
1758 return ret;
1759}
1760
1761static int vcn_v2_5_set_clockgating_state(void *handle,
1762 enum amd_clockgating_state state)
1763{
1764 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1765 bool enable = (state == AMD_CG_STATE_GATE);
1766
1767 if (amdgpu_sriov_vf(adev))
1768 return 0;
1769
1770 if (enable) {
1771 if (!vcn_v2_5_is_idle(handle))
1772 return -EBUSY;
1773 vcn_v2_5_enable_clock_gating(adev);
1774 } else {
1775 vcn_v2_5_disable_clock_gating(adev);
1776 }
1777
1778 return 0;
1779}
1780
1781static int vcn_v2_5_set_powergating_state(void *handle,
1782 enum amd_powergating_state state)
1783{
1784 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1785 int ret;
1786
1787 if (amdgpu_sriov_vf(adev))
1788 return 0;
1789
1790 if(state == adev->vcn.cur_state)
1791 return 0;
1792
1793 if (state == AMD_PG_STATE_GATE)
1794 ret = vcn_v2_5_stop(adev);
1795 else
1796 ret = vcn_v2_5_start(adev);
1797
1798 if(!ret)
1799 adev->vcn.cur_state = state;
1800
1801 return ret;
1802}
1803
1804static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1805 struct amdgpu_irq_src *source,
1806 unsigned type,
1807 enum amdgpu_interrupt_state state)
1808{
1809 return 0;
1810}
1811
1812static int vcn_v2_6_set_ras_interrupt_state(struct amdgpu_device *adev,
1813 struct amdgpu_irq_src *source,
1814 unsigned int type,
1815 enum amdgpu_interrupt_state state)
1816{
1817 return 0;
1818}
1819
1820static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1821 struct amdgpu_irq_src *source,
1822 struct amdgpu_iv_entry *entry)
1823{
1824 uint32_t ip_instance;
1825
1826 switch (entry->client_id) {
1827 case SOC15_IH_CLIENTID_VCN:
1828 ip_instance = 0;
1829 break;
1830 case SOC15_IH_CLIENTID_VCN1:
1831 ip_instance = 1;
1832 break;
1833 default:
1834 DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1835 return 0;
1836 }
1837
1838 DRM_DEBUG("IH: VCN TRAP\n");
1839
1840 switch (entry->src_id) {
1841 case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1842 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1843 break;
1844 case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1845 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1846 break;
1847 case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1848 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1849 break;
1850 default:
1851 DRM_ERROR("Unhandled interrupt: %d %d\n",
1852 entry->src_id, entry->src_data[0]);
1853 break;
1854 }
1855
1856 return 0;
1857}
1858
1859static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1860 .set = vcn_v2_5_set_interrupt_state,
1861 .process = vcn_v2_5_process_interrupt,
1862};
1863
1864static const struct amdgpu_irq_src_funcs vcn_v2_6_ras_irq_funcs = {
1865 .set = vcn_v2_6_set_ras_interrupt_state,
1866 .process = amdgpu_vcn_process_poison_irq,
1867};
1868
1869static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1870{
1871 int i;
1872
1873 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1874 if (adev->vcn.harvest_config & (1 << i))
1875 continue;
1876 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1877 adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1878
1879 adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
1880 adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v2_6_ras_irq_funcs;
1881 }
1882}
1883
1884static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1885 .name = "vcn_v2_5",
1886 .early_init = vcn_v2_5_early_init,
1887 .late_init = NULL,
1888 .sw_init = vcn_v2_5_sw_init,
1889 .sw_fini = vcn_v2_5_sw_fini,
1890 .hw_init = vcn_v2_5_hw_init,
1891 .hw_fini = vcn_v2_5_hw_fini,
1892 .suspend = vcn_v2_5_suspend,
1893 .resume = vcn_v2_5_resume,
1894 .is_idle = vcn_v2_5_is_idle,
1895 .wait_for_idle = vcn_v2_5_wait_for_idle,
1896 .check_soft_reset = NULL,
1897 .pre_soft_reset = NULL,
1898 .soft_reset = NULL,
1899 .post_soft_reset = NULL,
1900 .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1901 .set_powergating_state = vcn_v2_5_set_powergating_state,
1902};
1903
1904static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1905 .name = "vcn_v2_6",
1906 .early_init = vcn_v2_5_early_init,
1907 .late_init = NULL,
1908 .sw_init = vcn_v2_5_sw_init,
1909 .sw_fini = vcn_v2_5_sw_fini,
1910 .hw_init = vcn_v2_5_hw_init,
1911 .hw_fini = vcn_v2_5_hw_fini,
1912 .suspend = vcn_v2_5_suspend,
1913 .resume = vcn_v2_5_resume,
1914 .is_idle = vcn_v2_5_is_idle,
1915 .wait_for_idle = vcn_v2_5_wait_for_idle,
1916 .check_soft_reset = NULL,
1917 .pre_soft_reset = NULL,
1918 .soft_reset = NULL,
1919 .post_soft_reset = NULL,
1920 .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1921 .set_powergating_state = vcn_v2_5_set_powergating_state,
1922};
1923
1924const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1925{
1926 .type = AMD_IP_BLOCK_TYPE_VCN,
1927 .major = 2,
1928 .minor = 5,
1929 .rev = 0,
1930 .funcs = &vcn_v2_5_ip_funcs,
1931};
1932
1933const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1934{
1935 .type = AMD_IP_BLOCK_TYPE_VCN,
1936 .major = 2,
1937 .minor = 6,
1938 .rev = 0,
1939 .funcs = &vcn_v2_6_ip_funcs,
1940};
1941
1942static uint32_t vcn_v2_6_query_poison_by_instance(struct amdgpu_device *adev,
1943 uint32_t instance, uint32_t sub_block)
1944{
1945 uint32_t poison_stat = 0, reg_value = 0;
1946
1947 switch (sub_block) {
1948 case AMDGPU_VCN_V2_6_VCPU_VCODEC:
1949 reg_value = RREG32_SOC15(VCN, instance, mmUVD_RAS_VCPU_VCODEC_STATUS);
1950 poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
1951 break;
1952 default:
1953 break;
1954 }
1955
1956 if (poison_stat)
1957 dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
1958 instance, sub_block);
1959
1960 return poison_stat;
1961}
1962
1963static bool vcn_v2_6_query_poison_status(struct amdgpu_device *adev)
1964{
1965 uint32_t inst, sub;
1966 uint32_t poison_stat = 0;
1967
1968 for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
1969 for (sub = 0; sub < AMDGPU_VCN_V2_6_MAX_SUB_BLOCK; sub++)
1970 poison_stat +=
1971 vcn_v2_6_query_poison_by_instance(adev, inst, sub);
1972
1973 return !!poison_stat;
1974}
1975
1976const struct amdgpu_ras_block_hw_ops vcn_v2_6_ras_hw_ops = {
1977 .query_poison_status = vcn_v2_6_query_poison_status,
1978};
1979
1980static struct amdgpu_vcn_ras vcn_v2_6_ras = {
1981 .ras_block = {
1982 .hw_ops = &vcn_v2_6_ras_hw_ops,
1983 .ras_late_init = amdgpu_vcn_ras_late_init,
1984 },
1985};
1986
1987static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev)
1988{
1989 switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
1990 case IP_VERSION(2, 6, 0):
1991 adev->vcn.ras = &vcn_v2_6_ras;
1992 break;
1993 default:
1994 break;
1995 }
1996}
1/*
2 * Copyright 2019 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 */
23
24#include <linux/firmware.h>
25
26#include "amdgpu.h"
27#include "amdgpu_vcn.h"
28#include "amdgpu_pm.h"
29#include "soc15.h"
30#include "soc15d.h"
31#include "vcn_v2_0.h"
32#include "mmsch_v1_0.h"
33
34#include "vcn/vcn_2_5_offset.h"
35#include "vcn/vcn_2_5_sh_mask.h"
36#include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
37
38#define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x27
39#define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x0f
40#define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET 0x10
41#define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET 0x11
42#define mmUVD_NO_OP_INTERNAL_OFFSET 0x29
43#define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET 0x66
44#define mmUVD_SCRATCH9_INTERNAL_OFFSET 0xc01d
45
46#define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET 0x431
47#define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET 0x3b4
48#define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 0x3b5
49#define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET 0x25c
50
51#define VCN25_MAX_HW_INSTANCES_ARCTURUS 2
52
53static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
54static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
55static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
56static int vcn_v2_5_set_powergating_state(void *handle,
57 enum amd_powergating_state state);
58static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
59 int inst_idx, struct dpg_pause_state *new_state);
60static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
61
62static int amdgpu_ih_clientid_vcns[] = {
63 SOC15_IH_CLIENTID_VCN,
64 SOC15_IH_CLIENTID_VCN1
65};
66
67/**
68 * vcn_v2_5_early_init - set function pointers
69 *
70 * @handle: amdgpu_device pointer
71 *
72 * Set ring and irq function pointers
73 */
74static int vcn_v2_5_early_init(void *handle)
75{
76 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
77
78 if (amdgpu_sriov_vf(adev)) {
79 adev->vcn.num_vcn_inst = 2;
80 adev->vcn.harvest_config = 0;
81 adev->vcn.num_enc_rings = 1;
82 } else {
83 u32 harvest;
84 int i;
85 adev->vcn.num_vcn_inst = VCN25_MAX_HW_INSTANCES_ARCTURUS;
86 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
87 harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
88 if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
89 adev->vcn.harvest_config |= 1 << i;
90 }
91 if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
92 AMDGPU_VCN_HARVEST_VCN1))
93 /* both instances are harvested, disable the block */
94 return -ENOENT;
95
96 adev->vcn.num_enc_rings = 2;
97 }
98
99 vcn_v2_5_set_dec_ring_funcs(adev);
100 vcn_v2_5_set_enc_ring_funcs(adev);
101 vcn_v2_5_set_irq_funcs(adev);
102
103 return 0;
104}
105
106/**
107 * vcn_v2_5_sw_init - sw init for VCN block
108 *
109 * @handle: amdgpu_device pointer
110 *
111 * Load firmware and sw initialization
112 */
113static int vcn_v2_5_sw_init(void *handle)
114{
115 struct amdgpu_ring *ring;
116 int i, j, r;
117 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
118
119 for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
120 if (adev->vcn.harvest_config & (1 << j))
121 continue;
122 /* VCN DEC TRAP */
123 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
124 VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
125 if (r)
126 return r;
127
128 /* VCN ENC TRAP */
129 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
130 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
131 i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
132 if (r)
133 return r;
134 }
135 }
136
137 r = amdgpu_vcn_sw_init(adev);
138 if (r)
139 return r;
140
141 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
142 const struct common_firmware_header *hdr;
143 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
144 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
145 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
146 adev->firmware.fw_size +=
147 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
148
149 if (adev->vcn.num_vcn_inst == VCN25_MAX_HW_INSTANCES_ARCTURUS) {
150 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].ucode_id = AMDGPU_UCODE_ID_VCN1;
151 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].fw = adev->vcn.fw;
152 adev->firmware.fw_size +=
153 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
154 }
155 dev_info(adev->dev, "Will use PSP to load VCN firmware\n");
156 }
157
158 r = amdgpu_vcn_resume(adev);
159 if (r)
160 return r;
161
162 for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
163 volatile struct amdgpu_fw_shared *fw_shared;
164
165 if (adev->vcn.harvest_config & (1 << j))
166 continue;
167 adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
168 adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
169 adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
170 adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
171 adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
172 adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
173
174 adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
175 adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
176 adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
177 adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
178 adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
179 adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
180 adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
181 adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
182 adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
183 adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
184
185 ring = &adev->vcn.inst[j].ring_dec;
186 ring->use_doorbell = true;
187
188 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
189 (amdgpu_sriov_vf(adev) ? 2*j : 8*j);
190 sprintf(ring->name, "vcn_dec_%d", j);
191 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
192 0, AMDGPU_RING_PRIO_DEFAULT, NULL);
193 if (r)
194 return r;
195
196 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
197 ring = &adev->vcn.inst[j].ring_enc[i];
198 ring->use_doorbell = true;
199
200 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
201 (amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
202
203 sprintf(ring->name, "vcn_enc_%d.%d", j, i);
204 r = amdgpu_ring_init(adev, ring, 512,
205 &adev->vcn.inst[j].irq, 0,
206 AMDGPU_RING_PRIO_DEFAULT, NULL);
207 if (r)
208 return r;
209 }
210
211 fw_shared = adev->vcn.inst[j].fw_shared_cpu_addr;
212 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
213 }
214
215 if (amdgpu_sriov_vf(adev)) {
216 r = amdgpu_virt_alloc_mm_table(adev);
217 if (r)
218 return r;
219 }
220
221 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
222 adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
223
224 return 0;
225}
226
227/**
228 * vcn_v2_5_sw_fini - sw fini for VCN block
229 *
230 * @handle: amdgpu_device pointer
231 *
232 * VCN suspend and free up sw allocation
233 */
234static int vcn_v2_5_sw_fini(void *handle)
235{
236 int i, r;
237 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
238 volatile struct amdgpu_fw_shared *fw_shared;
239
240 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
241 if (adev->vcn.harvest_config & (1 << i))
242 continue;
243 fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
244 fw_shared->present_flag_0 = 0;
245 }
246
247 if (amdgpu_sriov_vf(adev))
248 amdgpu_virt_free_mm_table(adev);
249
250 r = amdgpu_vcn_suspend(adev);
251 if (r)
252 return r;
253
254 r = amdgpu_vcn_sw_fini(adev);
255
256 return r;
257}
258
259/**
260 * vcn_v2_5_hw_init - start and test VCN block
261 *
262 * @handle: amdgpu_device pointer
263 *
264 * Initialize the hardware, boot up the VCPU and do some testing
265 */
266static int vcn_v2_5_hw_init(void *handle)
267{
268 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
269 struct amdgpu_ring *ring;
270 int i, j, r = 0;
271
272 if (amdgpu_sriov_vf(adev))
273 r = vcn_v2_5_sriov_start(adev);
274
275 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
276 if (adev->vcn.harvest_config & (1 << j))
277 continue;
278
279 if (amdgpu_sriov_vf(adev)) {
280 adev->vcn.inst[j].ring_enc[0].sched.ready = true;
281 adev->vcn.inst[j].ring_enc[1].sched.ready = false;
282 adev->vcn.inst[j].ring_enc[2].sched.ready = false;
283 adev->vcn.inst[j].ring_dec.sched.ready = true;
284 } else {
285
286 ring = &adev->vcn.inst[j].ring_dec;
287
288 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
289 ring->doorbell_index, j);
290
291 r = amdgpu_ring_test_helper(ring);
292 if (r)
293 goto done;
294
295 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
296 ring = &adev->vcn.inst[j].ring_enc[i];
297 r = amdgpu_ring_test_helper(ring);
298 if (r)
299 goto done;
300 }
301 }
302 }
303
304done:
305 if (!r)
306 DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
307 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
308
309 return r;
310}
311
312/**
313 * vcn_v2_5_hw_fini - stop the hardware block
314 *
315 * @handle: amdgpu_device pointer
316 *
317 * Stop the VCN block, mark ring as not ready any more
318 */
319static int vcn_v2_5_hw_fini(void *handle)
320{
321 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
322 int i;
323
324 cancel_delayed_work_sync(&adev->vcn.idle_work);
325
326 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
327 if (adev->vcn.harvest_config & (1 << i))
328 continue;
329
330 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
331 (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
332 RREG32_SOC15(VCN, i, mmUVD_STATUS)))
333 vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
334 }
335
336 return 0;
337}
338
339/**
340 * vcn_v2_5_suspend - suspend VCN block
341 *
342 * @handle: amdgpu_device pointer
343 *
344 * HW fini and suspend VCN block
345 */
346static int vcn_v2_5_suspend(void *handle)
347{
348 int r;
349 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
350
351 r = vcn_v2_5_hw_fini(adev);
352 if (r)
353 return r;
354
355 r = amdgpu_vcn_suspend(adev);
356
357 return r;
358}
359
360/**
361 * vcn_v2_5_resume - resume VCN block
362 *
363 * @handle: amdgpu_device pointer
364 *
365 * Resume firmware and hw init VCN block
366 */
367static int vcn_v2_5_resume(void *handle)
368{
369 int r;
370 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
371
372 r = amdgpu_vcn_resume(adev);
373 if (r)
374 return r;
375
376 r = vcn_v2_5_hw_init(adev);
377
378 return r;
379}
380
381/**
382 * vcn_v2_5_mc_resume - memory controller programming
383 *
384 * @adev: amdgpu_device pointer
385 *
386 * Let the VCN memory controller know it's offsets
387 */
388static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
389{
390 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
391 uint32_t offset;
392 int i;
393
394 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
395 if (adev->vcn.harvest_config & (1 << i))
396 continue;
397 /* cache window 0: fw */
398 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
399 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
400 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
401 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
402 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
403 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
404 offset = 0;
405 } else {
406 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
407 lower_32_bits(adev->vcn.inst[i].gpu_addr));
408 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
409 upper_32_bits(adev->vcn.inst[i].gpu_addr));
410 offset = size;
411 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
412 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
413 }
414 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
415
416 /* cache window 1: stack */
417 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
418 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
419 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
420 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
421 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
422 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
423
424 /* cache window 2: context */
425 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
426 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
427 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
428 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
429 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
430 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
431
432 /* non-cache window */
433 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
434 lower_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
435 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
436 upper_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
437 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
438 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
439 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
440 }
441}
442
443static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
444{
445 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
446 uint32_t offset;
447
448 /* cache window 0: fw */
449 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
450 if (!indirect) {
451 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
452 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
453 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
454 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
455 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
456 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
457 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
458 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
459 } else {
460 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
461 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
462 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
463 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
464 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
466 }
467 offset = 0;
468 } else {
469 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
470 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
471 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
472 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
473 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
474 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
475 offset = size;
476 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
478 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
479 }
480
481 if (!indirect)
482 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
483 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
484 else
485 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
486 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
487
488 /* cache window 1: stack */
489 if (!indirect) {
490 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
492 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
493 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
494 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
495 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
496 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
497 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
498 } else {
499 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
501 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
502 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
503 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
505 }
506 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
508
509 /* cache window 2: context */
510 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
512 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
513 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
515 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
516 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
517 VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
518 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
519 VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
520
521 /* non-cache window */
522 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
523 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
524 lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
525 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
527 upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
528 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
529 VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
530 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
531 VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
532 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
533
534 /* VCN global tiling registers */
535 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
536 VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
537}
538
539/**
540 * vcn_v2_5_disable_clock_gating - disable VCN clock gating
541 *
542 * @adev: amdgpu_device pointer
543 *
544 * Disable clock gating for VCN block
545 */
546static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
547{
548 uint32_t data;
549 int i;
550
551 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
552 if (adev->vcn.harvest_config & (1 << i))
553 continue;
554 /* UVD disable CGC */
555 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
556 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
557 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
558 else
559 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
560 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
561 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
562 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
563
564 data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
565 data &= ~(UVD_CGC_GATE__SYS_MASK
566 | UVD_CGC_GATE__UDEC_MASK
567 | UVD_CGC_GATE__MPEG2_MASK
568 | UVD_CGC_GATE__REGS_MASK
569 | UVD_CGC_GATE__RBC_MASK
570 | UVD_CGC_GATE__LMI_MC_MASK
571 | UVD_CGC_GATE__LMI_UMC_MASK
572 | UVD_CGC_GATE__IDCT_MASK
573 | UVD_CGC_GATE__MPRD_MASK
574 | UVD_CGC_GATE__MPC_MASK
575 | UVD_CGC_GATE__LBSI_MASK
576 | UVD_CGC_GATE__LRBBM_MASK
577 | UVD_CGC_GATE__UDEC_RE_MASK
578 | UVD_CGC_GATE__UDEC_CM_MASK
579 | UVD_CGC_GATE__UDEC_IT_MASK
580 | UVD_CGC_GATE__UDEC_DB_MASK
581 | UVD_CGC_GATE__UDEC_MP_MASK
582 | UVD_CGC_GATE__WCB_MASK
583 | UVD_CGC_GATE__VCPU_MASK
584 | UVD_CGC_GATE__MMSCH_MASK);
585
586 WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
587
588 SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0, 0xFFFFFFFF);
589
590 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
591 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
592 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
593 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
594 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
595 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
596 | UVD_CGC_CTRL__SYS_MODE_MASK
597 | UVD_CGC_CTRL__UDEC_MODE_MASK
598 | UVD_CGC_CTRL__MPEG2_MODE_MASK
599 | UVD_CGC_CTRL__REGS_MODE_MASK
600 | UVD_CGC_CTRL__RBC_MODE_MASK
601 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
602 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
603 | UVD_CGC_CTRL__IDCT_MODE_MASK
604 | UVD_CGC_CTRL__MPRD_MODE_MASK
605 | UVD_CGC_CTRL__MPC_MODE_MASK
606 | UVD_CGC_CTRL__LBSI_MODE_MASK
607 | UVD_CGC_CTRL__LRBBM_MODE_MASK
608 | UVD_CGC_CTRL__WCB_MODE_MASK
609 | UVD_CGC_CTRL__VCPU_MODE_MASK
610 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
611 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
612
613 /* turn on */
614 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
615 data |= (UVD_SUVD_CGC_GATE__SRE_MASK
616 | UVD_SUVD_CGC_GATE__SIT_MASK
617 | UVD_SUVD_CGC_GATE__SMP_MASK
618 | UVD_SUVD_CGC_GATE__SCM_MASK
619 | UVD_SUVD_CGC_GATE__SDB_MASK
620 | UVD_SUVD_CGC_GATE__SRE_H264_MASK
621 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
622 | UVD_SUVD_CGC_GATE__SIT_H264_MASK
623 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
624 | UVD_SUVD_CGC_GATE__SCM_H264_MASK
625 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
626 | UVD_SUVD_CGC_GATE__SDB_H264_MASK
627 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
628 | UVD_SUVD_CGC_GATE__SCLR_MASK
629 | UVD_SUVD_CGC_GATE__UVD_SC_MASK
630 | UVD_SUVD_CGC_GATE__ENT_MASK
631 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
632 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
633 | UVD_SUVD_CGC_GATE__SITE_MASK
634 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
635 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
636 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
637 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
638 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
639 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
640
641 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
642 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
643 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
644 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
645 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
646 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
647 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
648 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
649 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
650 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
651 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
652 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
653 }
654}
655
656static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
657 uint8_t sram_sel, int inst_idx, uint8_t indirect)
658{
659 uint32_t reg_data = 0;
660
661 /* enable sw clock gating control */
662 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
663 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
664 else
665 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
666 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
667 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
668 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
669 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
670 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
671 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
672 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
673 UVD_CGC_CTRL__SYS_MODE_MASK |
674 UVD_CGC_CTRL__UDEC_MODE_MASK |
675 UVD_CGC_CTRL__MPEG2_MODE_MASK |
676 UVD_CGC_CTRL__REGS_MODE_MASK |
677 UVD_CGC_CTRL__RBC_MODE_MASK |
678 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
679 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
680 UVD_CGC_CTRL__IDCT_MODE_MASK |
681 UVD_CGC_CTRL__MPRD_MODE_MASK |
682 UVD_CGC_CTRL__MPC_MODE_MASK |
683 UVD_CGC_CTRL__LBSI_MODE_MASK |
684 UVD_CGC_CTRL__LRBBM_MODE_MASK |
685 UVD_CGC_CTRL__WCB_MODE_MASK |
686 UVD_CGC_CTRL__VCPU_MODE_MASK |
687 UVD_CGC_CTRL__MMSCH_MODE_MASK);
688 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
689 VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
690
691 /* turn off clock gating */
692 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
693 VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
694
695 /* turn on SUVD clock gating */
696 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
697 VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
698
699 /* turn on sw mode in UVD_SUVD_CGC_CTRL */
700 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
701 VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
702}
703
704/**
705 * vcn_v2_5_enable_clock_gating - enable VCN clock gating
706 *
707 * @adev: amdgpu_device pointer
708 *
709 * Enable clock gating for VCN block
710 */
711static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
712{
713 uint32_t data = 0;
714 int i;
715
716 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
717 if (adev->vcn.harvest_config & (1 << i))
718 continue;
719 /* enable UVD CGC */
720 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
721 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
722 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
723 else
724 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
725 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
726 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
727 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
728
729 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
730 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
731 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
732 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
733 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
734 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
735 | UVD_CGC_CTRL__SYS_MODE_MASK
736 | UVD_CGC_CTRL__UDEC_MODE_MASK
737 | UVD_CGC_CTRL__MPEG2_MODE_MASK
738 | UVD_CGC_CTRL__REGS_MODE_MASK
739 | UVD_CGC_CTRL__RBC_MODE_MASK
740 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
741 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
742 | UVD_CGC_CTRL__IDCT_MODE_MASK
743 | UVD_CGC_CTRL__MPRD_MODE_MASK
744 | UVD_CGC_CTRL__MPC_MODE_MASK
745 | UVD_CGC_CTRL__LBSI_MODE_MASK
746 | UVD_CGC_CTRL__LRBBM_MODE_MASK
747 | UVD_CGC_CTRL__WCB_MODE_MASK
748 | UVD_CGC_CTRL__VCPU_MODE_MASK);
749 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
750
751 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
752 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
753 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
754 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
755 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
756 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
757 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
758 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
759 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
760 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
761 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
762 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
763 }
764}
765
766static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
767{
768 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
769 struct amdgpu_ring *ring;
770 uint32_t rb_bufsz, tmp;
771
772 /* disable register anti-hang mechanism */
773 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
774 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
775 /* enable dynamic power gating mode */
776 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
777 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
778 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
779 WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
780
781 if (indirect)
782 adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
783
784 /* enable clock gating */
785 vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
786
787 /* enable VCPU clock */
788 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
789 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
790 tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
791 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
792 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
793
794 /* disable master interupt */
795 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
796 VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
797
798 /* setup mmUVD_LMI_CTRL */
799 tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
800 UVD_LMI_CTRL__REQ_MODE_MASK |
801 UVD_LMI_CTRL__CRC_RESET_MASK |
802 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
803 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
804 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
805 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
806 0x00100000L);
807 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
808 VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
809
810 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
811 VCN, 0, mmUVD_MPC_CNTL),
812 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
813
814 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
815 VCN, 0, mmUVD_MPC_SET_MUXA0),
816 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
817 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
818 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
819 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
820
821 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
822 VCN, 0, mmUVD_MPC_SET_MUXB0),
823 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
824 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
825 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
826 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
827
828 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
829 VCN, 0, mmUVD_MPC_SET_MUX),
830 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
831 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
832 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
833
834 vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
835
836 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
837 VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
838 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
839 VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
840
841 /* enable LMI MC and UMC channels */
842 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
843 VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
844
845 /* unblock VCPU register access */
846 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
847 VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
848
849 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
850 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
851 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
852 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
853
854 /* enable master interrupt */
855 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
856 VCN, 0, mmUVD_MASTINT_EN),
857 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
858
859 if (indirect)
860 psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
861 (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
862 (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
863
864 ring = &adev->vcn.inst[inst_idx].ring_dec;
865 /* force RBC into idle state */
866 rb_bufsz = order_base_2(ring->ring_size);
867 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
868 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
869 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
870 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
871 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
872 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
873
874 /* Stall DPG before WPTR/RPTR reset */
875 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
876 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
877 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
878 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
879
880 /* set the write pointer delay */
881 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
882
883 /* set the wb address */
884 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
885 (upper_32_bits(ring->gpu_addr) >> 2));
886
887 /* program the RB_BASE for ring buffer */
888 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
889 lower_32_bits(ring->gpu_addr));
890 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
891 upper_32_bits(ring->gpu_addr));
892
893 /* Initialize the ring buffer's read and write pointers */
894 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
895
896 WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
897
898 ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
899 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
900 lower_32_bits(ring->wptr));
901
902 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
903 /* Unstall DPG */
904 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
905 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
906
907 return 0;
908}
909
910static int vcn_v2_5_start(struct amdgpu_device *adev)
911{
912 struct amdgpu_ring *ring;
913 uint32_t rb_bufsz, tmp;
914 int i, j, k, r;
915
916 if (adev->pm.dpm_enabled)
917 amdgpu_dpm_enable_uvd(adev, true);
918
919 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
920 if (adev->vcn.harvest_config & (1 << i))
921 continue;
922 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
923 r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
924 continue;
925 }
926
927 /* disable register anti-hang mechanism */
928 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
929 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
930
931 /* set uvd status busy */
932 tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
933 WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
934 }
935
936 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
937 return 0;
938
939 /*SW clock gating */
940 vcn_v2_5_disable_clock_gating(adev);
941
942 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
943 if (adev->vcn.harvest_config & (1 << i))
944 continue;
945 /* enable VCPU clock */
946 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
947 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
948
949 /* disable master interrupt */
950 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
951 ~UVD_MASTINT_EN__VCPU_EN_MASK);
952
953 /* setup mmUVD_LMI_CTRL */
954 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
955 tmp &= ~0xff;
956 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
957 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
958 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
959 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
960 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
961
962 /* setup mmUVD_MPC_CNTL */
963 tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
964 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
965 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
966 WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
967
968 /* setup UVD_MPC_SET_MUXA0 */
969 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
970 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
971 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
972 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
973 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
974
975 /* setup UVD_MPC_SET_MUXB0 */
976 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
977 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
978 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
979 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
980 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
981
982 /* setup mmUVD_MPC_SET_MUX */
983 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
984 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
985 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
986 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
987 }
988
989 vcn_v2_5_mc_resume(adev);
990
991 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
992 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
993 if (adev->vcn.harvest_config & (1 << i))
994 continue;
995 /* VCN global tiling registers */
996 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
997 adev->gfx.config.gb_addr_config);
998 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
999 adev->gfx.config.gb_addr_config);
1000
1001 /* enable LMI MC and UMC channels */
1002 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1003 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1004
1005 /* unblock VCPU register access */
1006 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1007 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1008
1009 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1010 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1011
1012 for (k = 0; k < 10; ++k) {
1013 uint32_t status;
1014
1015 for (j = 0; j < 100; ++j) {
1016 status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1017 if (status & 2)
1018 break;
1019 if (amdgpu_emu_mode == 1)
1020 msleep(500);
1021 else
1022 mdelay(10);
1023 }
1024 r = 0;
1025 if (status & 2)
1026 break;
1027
1028 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1029 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1030 UVD_VCPU_CNTL__BLK_RST_MASK,
1031 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1032 mdelay(10);
1033 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1034 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1035
1036 mdelay(10);
1037 r = -1;
1038 }
1039
1040 if (r) {
1041 DRM_ERROR("VCN decode not responding, giving up!!!\n");
1042 return r;
1043 }
1044
1045 /* enable master interrupt */
1046 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1047 UVD_MASTINT_EN__VCPU_EN_MASK,
1048 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1049
1050 /* clear the busy bit of VCN_STATUS */
1051 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1052 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1053
1054 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1055
1056 ring = &adev->vcn.inst[i].ring_dec;
1057 /* force RBC into idle state */
1058 rb_bufsz = order_base_2(ring->ring_size);
1059 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1060 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1061 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1062 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1063 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1064 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1065
1066 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1067 /* program the RB_BASE for ring buffer */
1068 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1069 lower_32_bits(ring->gpu_addr));
1070 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1071 upper_32_bits(ring->gpu_addr));
1072
1073 /* Initialize the ring buffer's read and write pointers */
1074 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1075
1076 ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1077 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1078 lower_32_bits(ring->wptr));
1079 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1080
1081 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1082 ring = &adev->vcn.inst[i].ring_enc[0];
1083 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1084 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1085 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1086 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1087 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1088 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1089
1090 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1091 ring = &adev->vcn.inst[i].ring_enc[1];
1092 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1093 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1094 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1095 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1096 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1097 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1098 }
1099
1100 return 0;
1101}
1102
1103static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1104 struct amdgpu_mm_table *table)
1105{
1106 uint32_t data = 0, loop = 0, size = 0;
1107 uint64_t addr = table->gpu_addr;
1108 struct mmsch_v1_1_init_header *header = NULL;
1109
1110 header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1111 size = header->total_size;
1112
1113 /*
1114 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1115 * memory descriptor location
1116 */
1117 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1118 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1119
1120 /* 2, update vmid of descriptor */
1121 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1122 data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1123 /* use domain0 for MM scheduler */
1124 data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1125 WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1126
1127 /* 3, notify mmsch about the size of this descriptor */
1128 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1129
1130 /* 4, set resp to zero */
1131 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1132
1133 /*
1134 * 5, kick off the initialization and wait until
1135 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1136 */
1137 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1138
1139 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1140 loop = 10;
1141 while ((data & 0x10000002) != 0x10000002) {
1142 udelay(100);
1143 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1144 loop--;
1145 if (!loop)
1146 break;
1147 }
1148
1149 if (!loop) {
1150 dev_err(adev->dev,
1151 "failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1152 data);
1153 return -EBUSY;
1154 }
1155
1156 return 0;
1157}
1158
1159static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1160{
1161 struct amdgpu_ring *ring;
1162 uint32_t offset, size, tmp, i, rb_bufsz;
1163 uint32_t table_size = 0;
1164 struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1165 struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1166 struct mmsch_v1_0_cmd_end end = { { 0 } };
1167 uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1168 struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1169
1170 direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1171 direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1172 end.cmd_header.command_type = MMSCH_COMMAND__END;
1173
1174 header->version = MMSCH_VERSION;
1175 header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1176 init_table += header->total_size;
1177
1178 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1179 header->eng[i].table_offset = header->total_size;
1180 header->eng[i].init_status = 0;
1181 header->eng[i].table_size = 0;
1182
1183 table_size = 0;
1184
1185 MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1186 SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1187 ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1188
1189 size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1190 /* mc resume*/
1191 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1192 MMSCH_V1_0_INSERT_DIRECT_WT(
1193 SOC15_REG_OFFSET(VCN, i,
1194 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1195 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1196 MMSCH_V1_0_INSERT_DIRECT_WT(
1197 SOC15_REG_OFFSET(VCN, i,
1198 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1199 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1200 offset = 0;
1201 MMSCH_V1_0_INSERT_DIRECT_WT(
1202 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1203 } else {
1204 MMSCH_V1_0_INSERT_DIRECT_WT(
1205 SOC15_REG_OFFSET(VCN, i,
1206 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1207 lower_32_bits(adev->vcn.inst[i].gpu_addr));
1208 MMSCH_V1_0_INSERT_DIRECT_WT(
1209 SOC15_REG_OFFSET(VCN, i,
1210 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1211 upper_32_bits(adev->vcn.inst[i].gpu_addr));
1212 offset = size;
1213 MMSCH_V1_0_INSERT_DIRECT_WT(
1214 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1215 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1216 }
1217
1218 MMSCH_V1_0_INSERT_DIRECT_WT(
1219 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1220 size);
1221 MMSCH_V1_0_INSERT_DIRECT_WT(
1222 SOC15_REG_OFFSET(VCN, i,
1223 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1224 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1225 MMSCH_V1_0_INSERT_DIRECT_WT(
1226 SOC15_REG_OFFSET(VCN, i,
1227 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1228 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1229 MMSCH_V1_0_INSERT_DIRECT_WT(
1230 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1231 0);
1232 MMSCH_V1_0_INSERT_DIRECT_WT(
1233 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1234 AMDGPU_VCN_STACK_SIZE);
1235 MMSCH_V1_0_INSERT_DIRECT_WT(
1236 SOC15_REG_OFFSET(VCN, i,
1237 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1238 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1239 AMDGPU_VCN_STACK_SIZE));
1240 MMSCH_V1_0_INSERT_DIRECT_WT(
1241 SOC15_REG_OFFSET(VCN, i,
1242 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1243 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1244 AMDGPU_VCN_STACK_SIZE));
1245 MMSCH_V1_0_INSERT_DIRECT_WT(
1246 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1247 0);
1248 MMSCH_V1_0_INSERT_DIRECT_WT(
1249 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1250 AMDGPU_VCN_CONTEXT_SIZE);
1251
1252 ring = &adev->vcn.inst[i].ring_enc[0];
1253 ring->wptr = 0;
1254
1255 MMSCH_V1_0_INSERT_DIRECT_WT(
1256 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1257 lower_32_bits(ring->gpu_addr));
1258 MMSCH_V1_0_INSERT_DIRECT_WT(
1259 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1260 upper_32_bits(ring->gpu_addr));
1261 MMSCH_V1_0_INSERT_DIRECT_WT(
1262 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1263 ring->ring_size / 4);
1264
1265 ring = &adev->vcn.inst[i].ring_dec;
1266 ring->wptr = 0;
1267 MMSCH_V1_0_INSERT_DIRECT_WT(
1268 SOC15_REG_OFFSET(VCN, i,
1269 mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1270 lower_32_bits(ring->gpu_addr));
1271 MMSCH_V1_0_INSERT_DIRECT_WT(
1272 SOC15_REG_OFFSET(VCN, i,
1273 mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1274 upper_32_bits(ring->gpu_addr));
1275
1276 /* force RBC into idle state */
1277 rb_bufsz = order_base_2(ring->ring_size);
1278 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1279 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1280 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1281 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1282 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1283 MMSCH_V1_0_INSERT_DIRECT_WT(
1284 SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1285
1286 /* add end packet */
1287 memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1288 table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1289 init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1290
1291 /* refine header */
1292 header->eng[i].table_size = table_size;
1293 header->total_size += table_size;
1294 }
1295
1296 return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1297}
1298
1299static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1300{
1301 uint32_t tmp;
1302
1303 /* Wait for power status to be 1 */
1304 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1305 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1306
1307 /* wait for read ptr to be equal to write ptr */
1308 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1309 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1310
1311 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1312 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1313
1314 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1315 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1316
1317 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1318 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1319
1320 /* disable dynamic power gating mode */
1321 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1322 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1323
1324 return 0;
1325}
1326
1327static int vcn_v2_5_stop(struct amdgpu_device *adev)
1328{
1329 uint32_t tmp;
1330 int i, r = 0;
1331
1332 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1333 if (adev->vcn.harvest_config & (1 << i))
1334 continue;
1335 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1336 r = vcn_v2_5_stop_dpg_mode(adev, i);
1337 continue;
1338 }
1339
1340 /* wait for vcn idle */
1341 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1342 if (r)
1343 return r;
1344
1345 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1346 UVD_LMI_STATUS__READ_CLEAN_MASK |
1347 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1348 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1349 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1350 if (r)
1351 return r;
1352
1353 /* block LMI UMC channel */
1354 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1355 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1356 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1357
1358 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1359 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1360 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1361 if (r)
1362 return r;
1363
1364 /* block VCPU register access */
1365 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1366 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1367 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1368
1369 /* reset VCPU */
1370 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1371 UVD_VCPU_CNTL__BLK_RST_MASK,
1372 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1373
1374 /* disable VCPU clock */
1375 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1376 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1377
1378 /* clear status */
1379 WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1380
1381 vcn_v2_5_enable_clock_gating(adev);
1382
1383 /* enable register anti-hang mechanism */
1384 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1385 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1386 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1387 }
1388
1389 if (adev->pm.dpm_enabled)
1390 amdgpu_dpm_enable_uvd(adev, false);
1391
1392 return 0;
1393}
1394
1395static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1396 int inst_idx, struct dpg_pause_state *new_state)
1397{
1398 struct amdgpu_ring *ring;
1399 uint32_t reg_data = 0;
1400 int ret_code = 0;
1401
1402 /* pause/unpause if state is changed */
1403 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1404 DRM_DEBUG("dpg pause state changed %d -> %d",
1405 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based);
1406 reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1407 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1408
1409 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1410 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1411 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1412
1413 if (!ret_code) {
1414 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
1415
1416 /* pause DPG */
1417 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1418 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1419
1420 /* wait for ACK */
1421 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1422 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1423 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1424
1425 /* Stall DPG before WPTR/RPTR reset */
1426 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1427 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1428 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1429
1430 /* Restore */
1431 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1432 ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1433 ring->wptr = 0;
1434 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1435 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1436 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1437 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1438 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1439 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1440
1441 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1442 ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1443 ring->wptr = 0;
1444 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1445 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1446 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1447 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1448 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1449 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1450
1451 /* Unstall DPG */
1452 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1453 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1454
1455 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1456 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1457 }
1458 } else {
1459 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1460 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1461 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1462 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1463 }
1464 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1465 }
1466
1467 return 0;
1468}
1469
1470/**
1471 * vcn_v2_5_dec_ring_get_rptr - get read pointer
1472 *
1473 * @ring: amdgpu_ring pointer
1474 *
1475 * Returns the current hardware read pointer
1476 */
1477static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1478{
1479 struct amdgpu_device *adev = ring->adev;
1480
1481 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1482}
1483
1484/**
1485 * vcn_v2_5_dec_ring_get_wptr - get write pointer
1486 *
1487 * @ring: amdgpu_ring pointer
1488 *
1489 * Returns the current hardware write pointer
1490 */
1491static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1492{
1493 struct amdgpu_device *adev = ring->adev;
1494
1495 if (ring->use_doorbell)
1496 return adev->wb.wb[ring->wptr_offs];
1497 else
1498 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1499}
1500
1501/**
1502 * vcn_v2_5_dec_ring_set_wptr - set write pointer
1503 *
1504 * @ring: amdgpu_ring pointer
1505 *
1506 * Commits the write pointer to the hardware
1507 */
1508static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1509{
1510 struct amdgpu_device *adev = ring->adev;
1511
1512 if (ring->use_doorbell) {
1513 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1514 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1515 } else {
1516 WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1517 }
1518}
1519
1520static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1521 .type = AMDGPU_RING_TYPE_VCN_DEC,
1522 .align_mask = 0xf,
1523 .vmhub = AMDGPU_MMHUB_1,
1524 .get_rptr = vcn_v2_5_dec_ring_get_rptr,
1525 .get_wptr = vcn_v2_5_dec_ring_get_wptr,
1526 .set_wptr = vcn_v2_5_dec_ring_set_wptr,
1527 .emit_frame_size =
1528 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1529 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1530 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1531 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1532 6,
1533 .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1534 .emit_ib = vcn_v2_0_dec_ring_emit_ib,
1535 .emit_fence = vcn_v2_0_dec_ring_emit_fence,
1536 .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1537 .test_ring = vcn_v2_0_dec_ring_test_ring,
1538 .test_ib = amdgpu_vcn_dec_ring_test_ib,
1539 .insert_nop = vcn_v2_0_dec_ring_insert_nop,
1540 .insert_start = vcn_v2_0_dec_ring_insert_start,
1541 .insert_end = vcn_v2_0_dec_ring_insert_end,
1542 .pad_ib = amdgpu_ring_generic_pad_ib,
1543 .begin_use = amdgpu_vcn_ring_begin_use,
1544 .end_use = amdgpu_vcn_ring_end_use,
1545 .emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1546 .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1547 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1548};
1549
1550static const struct amdgpu_ring_funcs vcn_v2_6_dec_ring_vm_funcs = {
1551 .type = AMDGPU_RING_TYPE_VCN_DEC,
1552 .align_mask = 0xf,
1553 .vmhub = AMDGPU_MMHUB_0,
1554 .get_rptr = vcn_v2_5_dec_ring_get_rptr,
1555 .get_wptr = vcn_v2_5_dec_ring_get_wptr,
1556 .set_wptr = vcn_v2_5_dec_ring_set_wptr,
1557 .emit_frame_size =
1558 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1559 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1560 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1561 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1562 6,
1563 .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1564 .emit_ib = vcn_v2_0_dec_ring_emit_ib,
1565 .emit_fence = vcn_v2_0_dec_ring_emit_fence,
1566 .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1567 .test_ring = vcn_v2_0_dec_ring_test_ring,
1568 .test_ib = amdgpu_vcn_dec_ring_test_ib,
1569 .insert_nop = vcn_v2_0_dec_ring_insert_nop,
1570 .insert_start = vcn_v2_0_dec_ring_insert_start,
1571 .insert_end = vcn_v2_0_dec_ring_insert_end,
1572 .pad_ib = amdgpu_ring_generic_pad_ib,
1573 .begin_use = amdgpu_vcn_ring_begin_use,
1574 .end_use = amdgpu_vcn_ring_end_use,
1575 .emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1576 .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1577 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1578};
1579
1580/**
1581 * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1582 *
1583 * @ring: amdgpu_ring pointer
1584 *
1585 * Returns the current hardware enc read pointer
1586 */
1587static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1588{
1589 struct amdgpu_device *adev = ring->adev;
1590
1591 if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1592 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1593 else
1594 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1595}
1596
1597/**
1598 * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1599 *
1600 * @ring: amdgpu_ring pointer
1601 *
1602 * Returns the current hardware enc write pointer
1603 */
1604static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1605{
1606 struct amdgpu_device *adev = ring->adev;
1607
1608 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1609 if (ring->use_doorbell)
1610 return adev->wb.wb[ring->wptr_offs];
1611 else
1612 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1613 } else {
1614 if (ring->use_doorbell)
1615 return adev->wb.wb[ring->wptr_offs];
1616 else
1617 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1618 }
1619}
1620
1621/**
1622 * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1623 *
1624 * @ring: amdgpu_ring pointer
1625 *
1626 * Commits the enc write pointer to the hardware
1627 */
1628static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1629{
1630 struct amdgpu_device *adev = ring->adev;
1631
1632 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1633 if (ring->use_doorbell) {
1634 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1635 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1636 } else {
1637 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1638 }
1639 } else {
1640 if (ring->use_doorbell) {
1641 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1642 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1643 } else {
1644 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1645 }
1646 }
1647}
1648
1649static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1650 .type = AMDGPU_RING_TYPE_VCN_ENC,
1651 .align_mask = 0x3f,
1652 .nop = VCN_ENC_CMD_NO_OP,
1653 .vmhub = AMDGPU_MMHUB_1,
1654 .get_rptr = vcn_v2_5_enc_ring_get_rptr,
1655 .get_wptr = vcn_v2_5_enc_ring_get_wptr,
1656 .set_wptr = vcn_v2_5_enc_ring_set_wptr,
1657 .emit_frame_size =
1658 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1659 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1660 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1661 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1662 1, /* vcn_v2_0_enc_ring_insert_end */
1663 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1664 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1665 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1666 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1667 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1668 .test_ib = amdgpu_vcn_enc_ring_test_ib,
1669 .insert_nop = amdgpu_ring_insert_nop,
1670 .insert_end = vcn_v2_0_enc_ring_insert_end,
1671 .pad_ib = amdgpu_ring_generic_pad_ib,
1672 .begin_use = amdgpu_vcn_ring_begin_use,
1673 .end_use = amdgpu_vcn_ring_end_use,
1674 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1675 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1676 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1677};
1678
1679static const struct amdgpu_ring_funcs vcn_v2_6_enc_ring_vm_funcs = {
1680 .type = AMDGPU_RING_TYPE_VCN_ENC,
1681 .align_mask = 0x3f,
1682 .nop = VCN_ENC_CMD_NO_OP,
1683 .vmhub = AMDGPU_MMHUB_0,
1684 .get_rptr = vcn_v2_5_enc_ring_get_rptr,
1685 .get_wptr = vcn_v2_5_enc_ring_get_wptr,
1686 .set_wptr = vcn_v2_5_enc_ring_set_wptr,
1687 .emit_frame_size =
1688 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1689 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1690 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1691 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1692 1, /* vcn_v2_0_enc_ring_insert_end */
1693 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1694 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1695 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1696 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1697 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1698 .test_ib = amdgpu_vcn_enc_ring_test_ib,
1699 .insert_nop = amdgpu_ring_insert_nop,
1700 .insert_end = vcn_v2_0_enc_ring_insert_end,
1701 .pad_ib = amdgpu_ring_generic_pad_ib,
1702 .begin_use = amdgpu_vcn_ring_begin_use,
1703 .end_use = amdgpu_vcn_ring_end_use,
1704 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1705 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1706 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1707};
1708
1709static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1710{
1711 int i;
1712
1713 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1714 if (adev->vcn.harvest_config & (1 << i))
1715 continue;
1716 if (adev->asic_type == CHIP_ARCTURUS)
1717 adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1718 else /* CHIP_ALDEBARAN */
1719 adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_6_dec_ring_vm_funcs;
1720 adev->vcn.inst[i].ring_dec.me = i;
1721 DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1722 }
1723}
1724
1725static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1726{
1727 int i, j;
1728
1729 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1730 if (adev->vcn.harvest_config & (1 << j))
1731 continue;
1732 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1733 if (adev->asic_type == CHIP_ARCTURUS)
1734 adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1735 else /* CHIP_ALDEBARAN */
1736 adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_6_enc_ring_vm_funcs;
1737 adev->vcn.inst[j].ring_enc[i].me = j;
1738 }
1739 DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1740 }
1741}
1742
1743static bool vcn_v2_5_is_idle(void *handle)
1744{
1745 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1746 int i, ret = 1;
1747
1748 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1749 if (adev->vcn.harvest_config & (1 << i))
1750 continue;
1751 ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1752 }
1753
1754 return ret;
1755}
1756
1757static int vcn_v2_5_wait_for_idle(void *handle)
1758{
1759 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1760 int i, ret = 0;
1761
1762 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1763 if (adev->vcn.harvest_config & (1 << i))
1764 continue;
1765 ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1766 UVD_STATUS__IDLE);
1767 if (ret)
1768 return ret;
1769 }
1770
1771 return ret;
1772}
1773
1774static int vcn_v2_5_set_clockgating_state(void *handle,
1775 enum amd_clockgating_state state)
1776{
1777 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1778 bool enable = (state == AMD_CG_STATE_GATE);
1779
1780 if (amdgpu_sriov_vf(adev))
1781 return 0;
1782
1783 if (enable) {
1784 if (!vcn_v2_5_is_idle(handle))
1785 return -EBUSY;
1786 vcn_v2_5_enable_clock_gating(adev);
1787 } else {
1788 vcn_v2_5_disable_clock_gating(adev);
1789 }
1790
1791 return 0;
1792}
1793
1794static int vcn_v2_5_set_powergating_state(void *handle,
1795 enum amd_powergating_state state)
1796{
1797 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1798 int ret;
1799
1800 if (amdgpu_sriov_vf(adev))
1801 return 0;
1802
1803 if(state == adev->vcn.cur_state)
1804 return 0;
1805
1806 if (state == AMD_PG_STATE_GATE)
1807 ret = vcn_v2_5_stop(adev);
1808 else
1809 ret = vcn_v2_5_start(adev);
1810
1811 if(!ret)
1812 adev->vcn.cur_state = state;
1813
1814 return ret;
1815}
1816
1817static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1818 struct amdgpu_irq_src *source,
1819 unsigned type,
1820 enum amdgpu_interrupt_state state)
1821{
1822 return 0;
1823}
1824
1825static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1826 struct amdgpu_irq_src *source,
1827 struct amdgpu_iv_entry *entry)
1828{
1829 uint32_t ip_instance;
1830
1831 switch (entry->client_id) {
1832 case SOC15_IH_CLIENTID_VCN:
1833 ip_instance = 0;
1834 break;
1835 case SOC15_IH_CLIENTID_VCN1:
1836 ip_instance = 1;
1837 break;
1838 default:
1839 DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1840 return 0;
1841 }
1842
1843 DRM_DEBUG("IH: VCN TRAP\n");
1844
1845 switch (entry->src_id) {
1846 case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1847 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1848 break;
1849 case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1850 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1851 break;
1852 case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1853 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1854 break;
1855 default:
1856 DRM_ERROR("Unhandled interrupt: %d %d\n",
1857 entry->src_id, entry->src_data[0]);
1858 break;
1859 }
1860
1861 return 0;
1862}
1863
1864static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1865 .set = vcn_v2_5_set_interrupt_state,
1866 .process = vcn_v2_5_process_interrupt,
1867};
1868
1869static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1870{
1871 int i;
1872
1873 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1874 if (adev->vcn.harvest_config & (1 << i))
1875 continue;
1876 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1877 adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1878 }
1879}
1880
1881static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1882 .name = "vcn_v2_5",
1883 .early_init = vcn_v2_5_early_init,
1884 .late_init = NULL,
1885 .sw_init = vcn_v2_5_sw_init,
1886 .sw_fini = vcn_v2_5_sw_fini,
1887 .hw_init = vcn_v2_5_hw_init,
1888 .hw_fini = vcn_v2_5_hw_fini,
1889 .suspend = vcn_v2_5_suspend,
1890 .resume = vcn_v2_5_resume,
1891 .is_idle = vcn_v2_5_is_idle,
1892 .wait_for_idle = vcn_v2_5_wait_for_idle,
1893 .check_soft_reset = NULL,
1894 .pre_soft_reset = NULL,
1895 .soft_reset = NULL,
1896 .post_soft_reset = NULL,
1897 .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1898 .set_powergating_state = vcn_v2_5_set_powergating_state,
1899};
1900
1901static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1902 .name = "vcn_v2_6",
1903 .early_init = vcn_v2_5_early_init,
1904 .late_init = NULL,
1905 .sw_init = vcn_v2_5_sw_init,
1906 .sw_fini = vcn_v2_5_sw_fini,
1907 .hw_init = vcn_v2_5_hw_init,
1908 .hw_fini = vcn_v2_5_hw_fini,
1909 .suspend = vcn_v2_5_suspend,
1910 .resume = vcn_v2_5_resume,
1911 .is_idle = vcn_v2_5_is_idle,
1912 .wait_for_idle = vcn_v2_5_wait_for_idle,
1913 .check_soft_reset = NULL,
1914 .pre_soft_reset = NULL,
1915 .soft_reset = NULL,
1916 .post_soft_reset = NULL,
1917 .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1918 .set_powergating_state = vcn_v2_5_set_powergating_state,
1919};
1920
1921const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1922{
1923 .type = AMD_IP_BLOCK_TYPE_VCN,
1924 .major = 2,
1925 .minor = 5,
1926 .rev = 0,
1927 .funcs = &vcn_v2_5_ip_funcs,
1928};
1929
1930const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1931{
1932 .type = AMD_IP_BLOCK_TYPE_VCN,
1933 .major = 2,
1934 .minor = 6,
1935 .rev = 0,
1936 .funcs = &vcn_v2_6_ip_funcs,
1937};