Loading...
1/*
2 * Copyright 2021 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 "amdgpu.h"
26#include "amdgpu_vcn.h"
27#include "amdgpu_pm.h"
28#include "amdgpu_cs.h"
29#include "soc15.h"
30#include "soc15d.h"
31#include "soc15_hw_ip.h"
32#include "vcn_v2_0.h"
33#include "mmsch_v4_0.h"
34#include "vcn_v4_0.h"
35
36#include "vcn/vcn_4_0_0_offset.h"
37#include "vcn/vcn_4_0_0_sh_mask.h"
38#include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
39
40#include <drm/drm_drv.h>
41
42#define mmUVD_DPG_LMA_CTL regUVD_DPG_LMA_CTL
43#define mmUVD_DPG_LMA_CTL_BASE_IDX regUVD_DPG_LMA_CTL_BASE_IDX
44#define mmUVD_DPG_LMA_DATA regUVD_DPG_LMA_DATA
45#define mmUVD_DPG_LMA_DATA_BASE_IDX regUVD_DPG_LMA_DATA_BASE_IDX
46
47#define VCN_VID_SOC_ADDRESS_2_0 0x1fb00
48#define VCN1_VID_SOC_ADDRESS_3_0 0x48300
49
50#define VCN_HARVEST_MMSCH 0
51
52#define RDECODE_MSG_CREATE 0x00000000
53#define RDECODE_MESSAGE_CREATE 0x00000001
54
55static int amdgpu_ih_clientid_vcns[] = {
56 SOC15_IH_CLIENTID_VCN,
57 SOC15_IH_CLIENTID_VCN1
58};
59
60static int vcn_v4_0_start_sriov(struct amdgpu_device *adev);
61static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev);
62static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
63static int vcn_v4_0_set_powergating_state(void *handle,
64 enum amd_powergating_state state);
65static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
66 int inst_idx, struct dpg_pause_state *new_state);
67static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring);
68static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev);
69
70/**
71 * vcn_v4_0_early_init - set function pointers and load microcode
72 *
73 * @handle: amdgpu_device pointer
74 *
75 * Set ring and irq function pointers
76 * Load microcode from filesystem
77 */
78static int vcn_v4_0_early_init(void *handle)
79{
80 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
81 int i;
82
83 if (amdgpu_sriov_vf(adev)) {
84 adev->vcn.harvest_config = VCN_HARVEST_MMSCH;
85 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
86 if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
87 adev->vcn.harvest_config |= 1 << i;
88 dev_info(adev->dev, "VCN%d is disabled by hypervisor\n", i);
89 }
90 }
91 }
92
93 /* re-use enc ring as unified ring */
94 adev->vcn.num_enc_rings = 1;
95
96 vcn_v4_0_set_unified_ring_funcs(adev);
97 vcn_v4_0_set_irq_funcs(adev);
98 vcn_v4_0_set_ras_funcs(adev);
99
100 return amdgpu_vcn_early_init(adev);
101}
102
103static int vcn_v4_0_fw_shared_init(struct amdgpu_device *adev, int inst_idx)
104{
105 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
106
107 fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
108 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
109 fw_shared->sq.is_enabled = 1;
110
111 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
112 fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
113 AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
114
115 if (amdgpu_ip_version(adev, VCN_HWIP, 0) ==
116 IP_VERSION(4, 0, 2)) {
117 fw_shared->present_flag_0 |= AMDGPU_FW_SHARED_FLAG_0_DRM_KEY_INJECT;
118 fw_shared->drm_key_wa.method =
119 AMDGPU_DRM_KEY_INJECT_WORKAROUND_VCNFW_ASD_HANDSHAKING;
120 }
121
122 if (amdgpu_vcnfw_log)
123 amdgpu_vcn_fwlog_init(&adev->vcn.inst[inst_idx]);
124
125 return 0;
126}
127
128/**
129 * vcn_v4_0_sw_init - sw init for VCN block
130 *
131 * @handle: amdgpu_device pointer
132 *
133 * Load firmware and sw initialization
134 */
135static int vcn_v4_0_sw_init(void *handle)
136{
137 struct amdgpu_ring *ring;
138 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
139 int i, r;
140
141 r = amdgpu_vcn_sw_init(adev);
142 if (r)
143 return r;
144
145 amdgpu_vcn_setup_ucode(adev);
146
147 r = amdgpu_vcn_resume(adev);
148 if (r)
149 return r;
150
151 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
152 if (adev->vcn.harvest_config & (1 << i))
153 continue;
154
155 /* Init instance 0 sched_score to 1, so it's scheduled after other instances */
156 if (i == 0)
157 atomic_set(&adev->vcn.inst[i].sched_score, 1);
158 else
159 atomic_set(&adev->vcn.inst[i].sched_score, 0);
160
161 /* VCN UNIFIED TRAP */
162 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
163 VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
164 if (r)
165 return r;
166
167 /* VCN POISON TRAP */
168 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
169 VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].ras_poison_irq);
170 if (r)
171 return r;
172
173 ring = &adev->vcn.inst[i].ring_enc[0];
174 ring->use_doorbell = true;
175 if (amdgpu_sriov_vf(adev))
176 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + i * (adev->vcn.num_enc_rings + 1) + 1;
177 else
178 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
179 ring->vm_hub = AMDGPU_MMHUB0(0);
180 sprintf(ring->name, "vcn_unified_%d", i);
181
182 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
183 AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
184 if (r)
185 return r;
186
187 vcn_v4_0_fw_shared_init(adev, i);
188 }
189
190 if (amdgpu_sriov_vf(adev)) {
191 r = amdgpu_virt_alloc_mm_table(adev);
192 if (r)
193 return r;
194 }
195
196 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
197 adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
198
199 r = amdgpu_vcn_ras_sw_init(adev);
200 if (r)
201 return r;
202
203 return 0;
204}
205
206/**
207 * vcn_v4_0_sw_fini - sw fini for VCN block
208 *
209 * @handle: amdgpu_device pointer
210 *
211 * VCN suspend and free up sw allocation
212 */
213static int vcn_v4_0_sw_fini(void *handle)
214{
215 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
216 int i, r, idx;
217
218 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
219 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
220 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
221
222 if (adev->vcn.harvest_config & (1 << i))
223 continue;
224
225 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
226 fw_shared->present_flag_0 = 0;
227 fw_shared->sq.is_enabled = 0;
228 }
229
230 drm_dev_exit(idx);
231 }
232
233 if (amdgpu_sriov_vf(adev))
234 amdgpu_virt_free_mm_table(adev);
235
236 r = amdgpu_vcn_suspend(adev);
237 if (r)
238 return r;
239
240 r = amdgpu_vcn_sw_fini(adev);
241
242 return r;
243}
244
245/**
246 * vcn_v4_0_hw_init - start and test VCN block
247 *
248 * @handle: amdgpu_device pointer
249 *
250 * Initialize the hardware, boot up the VCPU and do some testing
251 */
252static int vcn_v4_0_hw_init(void *handle)
253{
254 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
255 struct amdgpu_ring *ring;
256 int i, r;
257
258 if (amdgpu_sriov_vf(adev)) {
259 r = vcn_v4_0_start_sriov(adev);
260 if (r)
261 goto done;
262
263 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
264 if (adev->vcn.harvest_config & (1 << i))
265 continue;
266
267 ring = &adev->vcn.inst[i].ring_enc[0];
268 ring->wptr = 0;
269 ring->wptr_old = 0;
270 vcn_v4_0_unified_ring_set_wptr(ring);
271 ring->sched.ready = true;
272
273 }
274 } else {
275 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
276 if (adev->vcn.harvest_config & (1 << i))
277 continue;
278
279 ring = &adev->vcn.inst[i].ring_enc[0];
280
281 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
282 ((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
283
284 r = amdgpu_ring_test_helper(ring);
285 if (r)
286 goto done;
287
288 }
289 }
290
291done:
292 if (!r)
293 DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
294 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
295
296 return r;
297}
298
299/**
300 * vcn_v4_0_hw_fini - stop the hardware block
301 *
302 * @handle: amdgpu_device pointer
303 *
304 * Stop the VCN block, mark ring as not ready any more
305 */
306static int vcn_v4_0_hw_fini(void *handle)
307{
308 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
309 int i;
310
311 cancel_delayed_work_sync(&adev->vcn.idle_work);
312
313 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
314 if (adev->vcn.harvest_config & (1 << i))
315 continue;
316 if (!amdgpu_sriov_vf(adev)) {
317 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
318 (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
319 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
320 vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
321 }
322 }
323 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
324 amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
325 }
326
327 return 0;
328}
329
330/**
331 * vcn_v4_0_suspend - suspend VCN block
332 *
333 * @handle: amdgpu_device pointer
334 *
335 * HW fini and suspend VCN block
336 */
337static int vcn_v4_0_suspend(void *handle)
338{
339 int r;
340 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
341
342 r = vcn_v4_0_hw_fini(adev);
343 if (r)
344 return r;
345
346 r = amdgpu_vcn_suspend(adev);
347
348 return r;
349}
350
351/**
352 * vcn_v4_0_resume - resume VCN block
353 *
354 * @handle: amdgpu_device pointer
355 *
356 * Resume firmware and hw init VCN block
357 */
358static int vcn_v4_0_resume(void *handle)
359{
360 int r;
361 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
362
363 r = amdgpu_vcn_resume(adev);
364 if (r)
365 return r;
366
367 r = vcn_v4_0_hw_init(adev);
368
369 return r;
370}
371
372/**
373 * vcn_v4_0_mc_resume - memory controller programming
374 *
375 * @adev: amdgpu_device pointer
376 * @inst: instance number
377 *
378 * Let the VCN memory controller know it's offsets
379 */
380static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
381{
382 uint32_t offset, size;
383 const struct common_firmware_header *hdr;
384
385 hdr = (const struct common_firmware_header *)adev->vcn.fw[inst]->data;
386 size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
387
388 /* cache window 0: fw */
389 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
390 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
391 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
392 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
393 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
394 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
395 offset = 0;
396 } else {
397 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
398 lower_32_bits(adev->vcn.inst[inst].gpu_addr));
399 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
400 upper_32_bits(adev->vcn.inst[inst].gpu_addr));
401 offset = size;
402 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
403 }
404 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
405
406 /* cache window 1: stack */
407 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
408 lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
409 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
410 upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
411 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
412 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
413
414 /* cache window 2: context */
415 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
416 lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
417 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
418 upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
419 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
420 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
421
422 /* non-cache window */
423 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
424 lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
425 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
426 upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
427 WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
428 WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
429 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
430}
431
432/**
433 * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
434 *
435 * @adev: amdgpu_device pointer
436 * @inst_idx: instance number index
437 * @indirect: indirectly write sram
438 *
439 * Let the VCN memory controller know it's offsets with dpg mode
440 */
441static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
442{
443 uint32_t offset, size;
444 const struct common_firmware_header *hdr;
445 hdr = (const struct common_firmware_header *)adev->vcn.fw[inst_idx]->data;
446 size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
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, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
459 } else {
460 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
461 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
462 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
463 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
464 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 VCN, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
484 else
485 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
486 VCN, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
498 } else {
499 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
500 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
501 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
502 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
503 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
505 }
506 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 VCN, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
518 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
519 VCN, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_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, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
530 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
531 VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
532 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
533
534 /* VCN global tiling registers */
535 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
536 VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
537}
538
539/**
540 * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
541 *
542 * @adev: amdgpu_device pointer
543 * @inst: instance number
544 *
545 * Disable static power gating for VCN block
546 */
547static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
548{
549 uint32_t data = 0;
550
551 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
552 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
553 | 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
554 | 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
555 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
556 | 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
557 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
558 | 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
559 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
560 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
561 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
562 | 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
563 | 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
564 | 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
565 | 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
566
567 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
568 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
569 UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
570 } else {
571 uint32_t value;
572
573 value = (inst) ? 0x2200800 : 0;
574 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
575 | 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
576 | 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
577 | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
578 | 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
579 | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
580 | 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
581 | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
582 | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
583 | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
584 | 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
585 | 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
586 | 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
587 | 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
588
589 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
590 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value, 0x3F3FFFFF);
591 }
592
593 data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
594 data &= ~0x103;
595 if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
596 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
597 UVD_POWER_STATUS__UVD_PG_EN_MASK;
598
599 WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
600
601 return;
602}
603
604/**
605 * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
606 *
607 * @adev: amdgpu_device pointer
608 * @inst: instance number
609 *
610 * Enable static power gating for VCN block
611 */
612static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
613{
614 uint32_t data;
615
616 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
617 /* Before power off, this indicator has to be turned on */
618 data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
619 data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
620 data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
621 WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
622
623 data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
624 | 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
625 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
626 | 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
627 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
628 | 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
629 | 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
630 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
631 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
632 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
633 | 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
634 | 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
635 | 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
636 | 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
637 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
638
639 data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
640 | 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
641 | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
642 | 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
643 | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
644 | 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
645 | 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
646 | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
647 | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
648 | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
649 | 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
650 | 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
651 | 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
652 | 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
653 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
654 }
655
656 return;
657}
658
659/**
660 * vcn_v4_0_disable_clock_gating - disable VCN clock gating
661 *
662 * @adev: amdgpu_device pointer
663 * @inst: instance number
664 *
665 * Disable clock gating for VCN block
666 */
667static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
668{
669 uint32_t data;
670
671 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
672 return;
673
674 /* VCN disable CGC */
675 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
676 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
677 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
678 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
679 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
680
681 data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
682 data &= ~(UVD_CGC_GATE__SYS_MASK
683 | UVD_CGC_GATE__UDEC_MASK
684 | UVD_CGC_GATE__MPEG2_MASK
685 | UVD_CGC_GATE__REGS_MASK
686 | UVD_CGC_GATE__RBC_MASK
687 | UVD_CGC_GATE__LMI_MC_MASK
688 | UVD_CGC_GATE__LMI_UMC_MASK
689 | UVD_CGC_GATE__IDCT_MASK
690 | UVD_CGC_GATE__MPRD_MASK
691 | UVD_CGC_GATE__MPC_MASK
692 | UVD_CGC_GATE__LBSI_MASK
693 | UVD_CGC_GATE__LRBBM_MASK
694 | UVD_CGC_GATE__UDEC_RE_MASK
695 | UVD_CGC_GATE__UDEC_CM_MASK
696 | UVD_CGC_GATE__UDEC_IT_MASK
697 | UVD_CGC_GATE__UDEC_DB_MASK
698 | UVD_CGC_GATE__UDEC_MP_MASK
699 | UVD_CGC_GATE__WCB_MASK
700 | UVD_CGC_GATE__VCPU_MASK
701 | UVD_CGC_GATE__MMSCH_MASK);
702
703 WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
704 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0, 0xFFFFFFFF);
705
706 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
707 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
708 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
709 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
710 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
711 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
712 | UVD_CGC_CTRL__SYS_MODE_MASK
713 | UVD_CGC_CTRL__UDEC_MODE_MASK
714 | UVD_CGC_CTRL__MPEG2_MODE_MASK
715 | UVD_CGC_CTRL__REGS_MODE_MASK
716 | UVD_CGC_CTRL__RBC_MODE_MASK
717 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
718 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
719 | UVD_CGC_CTRL__IDCT_MODE_MASK
720 | UVD_CGC_CTRL__MPRD_MODE_MASK
721 | UVD_CGC_CTRL__MPC_MODE_MASK
722 | UVD_CGC_CTRL__LBSI_MODE_MASK
723 | UVD_CGC_CTRL__LRBBM_MODE_MASK
724 | UVD_CGC_CTRL__WCB_MODE_MASK
725 | UVD_CGC_CTRL__VCPU_MODE_MASK
726 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
727 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
728
729 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
730 data |= (UVD_SUVD_CGC_GATE__SRE_MASK
731 | UVD_SUVD_CGC_GATE__SIT_MASK
732 | UVD_SUVD_CGC_GATE__SMP_MASK
733 | UVD_SUVD_CGC_GATE__SCM_MASK
734 | UVD_SUVD_CGC_GATE__SDB_MASK
735 | UVD_SUVD_CGC_GATE__SRE_H264_MASK
736 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
737 | UVD_SUVD_CGC_GATE__SIT_H264_MASK
738 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
739 | UVD_SUVD_CGC_GATE__SCM_H264_MASK
740 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
741 | UVD_SUVD_CGC_GATE__SDB_H264_MASK
742 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
743 | UVD_SUVD_CGC_GATE__SCLR_MASK
744 | UVD_SUVD_CGC_GATE__UVD_SC_MASK
745 | UVD_SUVD_CGC_GATE__ENT_MASK
746 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
747 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
748 | UVD_SUVD_CGC_GATE__SITE_MASK
749 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
750 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
751 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
752 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
753 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
754 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
755
756 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
757 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
758 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
759 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
760 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
761 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
762 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
763 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
764 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
765 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
766 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
767 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
768}
769
770/**
771 * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
772 *
773 * @adev: amdgpu_device pointer
774 * @sram_sel: sram select
775 * @inst_idx: instance number index
776 * @indirect: indirectly write sram
777 *
778 * Disable clock gating for VCN block with dpg mode
779 */
780static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
781 int inst_idx, uint8_t indirect)
782{
783 uint32_t reg_data = 0;
784
785 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
786 return;
787
788 /* enable sw clock gating control */
789 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
790 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
791 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
792 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
793 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
794 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
795 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
796 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
797 UVD_CGC_CTRL__SYS_MODE_MASK |
798 UVD_CGC_CTRL__UDEC_MODE_MASK |
799 UVD_CGC_CTRL__MPEG2_MODE_MASK |
800 UVD_CGC_CTRL__REGS_MODE_MASK |
801 UVD_CGC_CTRL__RBC_MODE_MASK |
802 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
803 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
804 UVD_CGC_CTRL__IDCT_MODE_MASK |
805 UVD_CGC_CTRL__MPRD_MODE_MASK |
806 UVD_CGC_CTRL__MPC_MODE_MASK |
807 UVD_CGC_CTRL__LBSI_MODE_MASK |
808 UVD_CGC_CTRL__LRBBM_MODE_MASK |
809 UVD_CGC_CTRL__WCB_MODE_MASK |
810 UVD_CGC_CTRL__VCPU_MODE_MASK);
811 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
812 VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
813
814 /* turn off clock gating */
815 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
816 VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
817
818 /* turn on SUVD clock gating */
819 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
820 VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
821
822 /* turn on sw mode in UVD_SUVD_CGC_CTRL */
823 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
824 VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
825}
826
827/**
828 * vcn_v4_0_enable_clock_gating - enable VCN clock gating
829 *
830 * @adev: amdgpu_device pointer
831 * @inst: instance number
832 *
833 * Enable clock gating for VCN block
834 */
835static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
836{
837 uint32_t data;
838
839 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
840 return;
841
842 /* enable VCN CGC */
843 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
844 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
845 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
846 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
847 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
848
849 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
850 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
851 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
852 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
853 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
854 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
855 | UVD_CGC_CTRL__SYS_MODE_MASK
856 | UVD_CGC_CTRL__UDEC_MODE_MASK
857 | UVD_CGC_CTRL__MPEG2_MODE_MASK
858 | UVD_CGC_CTRL__REGS_MODE_MASK
859 | UVD_CGC_CTRL__RBC_MODE_MASK
860 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
861 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
862 | UVD_CGC_CTRL__IDCT_MODE_MASK
863 | UVD_CGC_CTRL__MPRD_MODE_MASK
864 | UVD_CGC_CTRL__MPC_MODE_MASK
865 | UVD_CGC_CTRL__LBSI_MODE_MASK
866 | UVD_CGC_CTRL__LRBBM_MODE_MASK
867 | UVD_CGC_CTRL__WCB_MODE_MASK
868 | UVD_CGC_CTRL__VCPU_MODE_MASK
869 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
870 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
871
872 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
873 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
874 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
875 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
876 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
877 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
878 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
879 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
880 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
881 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
882 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
883 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
884}
885
886static void vcn_v4_0_enable_ras(struct amdgpu_device *adev, int inst_idx,
887 bool indirect)
888{
889 uint32_t tmp;
890
891 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
892 return;
893
894 tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
895 VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
896 VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
897 VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
898 WREG32_SOC15_DPG_MODE(inst_idx,
899 SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
900 tmp, 0, indirect);
901
902 tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
903 WREG32_SOC15_DPG_MODE(inst_idx,
904 SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
905 tmp, 0, indirect);
906}
907
908/**
909 * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
910 *
911 * @adev: amdgpu_device pointer
912 * @inst_idx: instance number index
913 * @indirect: indirectly write sram
914 *
915 * Start VCN block with dpg mode
916 */
917static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
918{
919 volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
920 struct amdgpu_ring *ring;
921 uint32_t tmp;
922
923 /* disable register anti-hang mechanism */
924 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
925 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
926 /* enable dynamic power gating mode */
927 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
928 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
929 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
930 WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
931
932 if (indirect)
933 adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
934
935 /* enable clock gating */
936 vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
937
938 /* enable VCPU clock */
939 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
940 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
941 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
942 VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
943
944 /* disable master interupt */
945 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
946 VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
947
948 /* setup regUVD_LMI_CTRL */
949 tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
950 UVD_LMI_CTRL__REQ_MODE_MASK |
951 UVD_LMI_CTRL__CRC_RESET_MASK |
952 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
953 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
954 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
955 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
956 0x00100000L);
957 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
958 VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
959
960 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
961 VCN, inst_idx, regUVD_MPC_CNTL),
962 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
963
964 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
965 VCN, inst_idx, regUVD_MPC_SET_MUXA0),
966 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
967 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
968 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
969 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
970
971 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
972 VCN, inst_idx, regUVD_MPC_SET_MUXB0),
973 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
974 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
975 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
976 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
977
978 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
979 VCN, inst_idx, regUVD_MPC_SET_MUX),
980 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
981 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
982 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
983
984 vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
985
986 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
987 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
988 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
989 VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
990
991 /* enable LMI MC and UMC channels */
992 tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
993 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
994 VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
995
996 vcn_v4_0_enable_ras(adev, inst_idx, indirect);
997
998 /* enable master interrupt */
999 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1000 VCN, inst_idx, regUVD_MASTINT_EN),
1001 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1002
1003
1004 if (indirect)
1005 amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
1006
1007 ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1008
1009 WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
1010 WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1011 WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
1012
1013 tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1014 tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1015 WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1016 fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1017 WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
1018 WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
1019
1020 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1021 WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1022 ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1023
1024 tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1025 tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1026 WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1027 fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1028
1029 WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1030 ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1031 VCN_RB1_DB_CTRL__EN_MASK);
1032
1033 return 0;
1034}
1035
1036
1037/**
1038 * vcn_v4_0_start - VCN start
1039 *
1040 * @adev: amdgpu_device pointer
1041 *
1042 * Start VCN block
1043 */
1044static int vcn_v4_0_start(struct amdgpu_device *adev)
1045{
1046 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1047 struct amdgpu_ring *ring;
1048 uint32_t tmp;
1049 int i, j, k, r;
1050
1051 if (adev->pm.dpm_enabled)
1052 amdgpu_dpm_enable_uvd(adev, true);
1053
1054 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1055 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1056
1057 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1058 r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1059 continue;
1060 }
1061
1062 /* disable VCN power gating */
1063 vcn_v4_0_disable_static_power_gating(adev, i);
1064
1065 /* set VCN status busy */
1066 tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1067 WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1068
1069 /*SW clock gating */
1070 vcn_v4_0_disable_clock_gating(adev, i);
1071
1072 /* enable VCPU clock */
1073 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1074 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1075
1076 /* disable master interrupt */
1077 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1078 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1079
1080 /* enable LMI MC and UMC channels */
1081 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1082 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1083
1084 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1085 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1086 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1087 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1088
1089 /* setup regUVD_LMI_CTRL */
1090 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1091 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1092 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1093 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1094 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1095 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1096
1097 /* setup regUVD_MPC_CNTL */
1098 tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1099 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1100 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1101 WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1102
1103 /* setup UVD_MPC_SET_MUXA0 */
1104 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1105 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1106 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1107 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1108 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1109
1110 /* setup UVD_MPC_SET_MUXB0 */
1111 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1112 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1113 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1114 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1115 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1116
1117 /* setup UVD_MPC_SET_MUX */
1118 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1119 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1120 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1121 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1122
1123 vcn_v4_0_mc_resume(adev, i);
1124
1125 /* VCN global tiling registers */
1126 WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1127 adev->gfx.config.gb_addr_config);
1128
1129 /* unblock VCPU register access */
1130 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1131 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1132
1133 /* release VCPU reset to boot */
1134 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1135 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1136
1137 for (j = 0; j < 10; ++j) {
1138 uint32_t status;
1139
1140 for (k = 0; k < 100; ++k) {
1141 status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1142 if (status & 2)
1143 break;
1144 mdelay(10);
1145 if (amdgpu_emu_mode == 1)
1146 msleep(1);
1147 }
1148
1149 if (amdgpu_emu_mode == 1) {
1150 r = -1;
1151 if (status & 2) {
1152 r = 0;
1153 break;
1154 }
1155 } else {
1156 r = 0;
1157 if (status & 2)
1158 break;
1159
1160 dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1161 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1162 UVD_VCPU_CNTL__BLK_RST_MASK,
1163 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1164 mdelay(10);
1165 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1166 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1167
1168 mdelay(10);
1169 r = -1;
1170 }
1171 }
1172
1173 if (r) {
1174 dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1175 return r;
1176 }
1177
1178 /* enable master interrupt */
1179 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1180 UVD_MASTINT_EN__VCPU_EN_MASK,
1181 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1182
1183 /* clear the busy bit of VCN_STATUS */
1184 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1185 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1186
1187 ring = &adev->vcn.inst[i].ring_enc[0];
1188 WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1189 ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1190 VCN_RB1_DB_CTRL__EN_MASK);
1191
1192 WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1193 WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1194 WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1195
1196 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1197 tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1198 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1199 fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1200 WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1201 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1202
1203 tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1204 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1205 ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1206
1207 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1208 tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1209 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1210 fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1211 }
1212
1213 return 0;
1214}
1215
1216static int vcn_v4_0_init_ring_metadata(struct amdgpu_device *adev, uint32_t vcn_inst, struct amdgpu_ring *ring_enc)
1217{
1218 struct amdgpu_vcn_rb_metadata *rb_metadata = NULL;
1219 uint8_t *rb_ptr = (uint8_t *)ring_enc->ring;
1220
1221 rb_ptr += ring_enc->ring_size;
1222 rb_metadata = (struct amdgpu_vcn_rb_metadata *)rb_ptr;
1223
1224 memset(rb_metadata, 0, sizeof(struct amdgpu_vcn_rb_metadata));
1225 rb_metadata->size = sizeof(struct amdgpu_vcn_rb_metadata);
1226 rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1227 rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1228 rb_metadata->version = 1;
1229 rb_metadata->ring_id = vcn_inst & 0xFF;
1230
1231 return 0;
1232}
1233
1234static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1235{
1236 int i;
1237 struct amdgpu_ring *ring_enc;
1238 uint64_t cache_addr;
1239 uint64_t rb_enc_addr;
1240 uint64_t ctx_addr;
1241 uint32_t param, resp, expected;
1242 uint32_t offset, cache_size;
1243 uint32_t tmp, timeout;
1244
1245 struct amdgpu_mm_table *table = &adev->virt.mm_table;
1246 uint32_t *table_loc;
1247 uint32_t table_size;
1248 uint32_t size, size_dw;
1249 uint32_t init_status;
1250 uint32_t enabled_vcn;
1251
1252 struct mmsch_v4_0_cmd_direct_write
1253 direct_wt = { {0} };
1254 struct mmsch_v4_0_cmd_direct_read_modify_write
1255 direct_rd_mod_wt = { {0} };
1256 struct mmsch_v4_0_cmd_end end = { {0} };
1257 struct mmsch_v4_0_init_header header;
1258
1259 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1260 volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1261
1262 direct_wt.cmd_header.command_type =
1263 MMSCH_COMMAND__DIRECT_REG_WRITE;
1264 direct_rd_mod_wt.cmd_header.command_type =
1265 MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1266 end.cmd_header.command_type =
1267 MMSCH_COMMAND__END;
1268
1269 header.version = MMSCH_VERSION;
1270 header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1271 for (i = 0; i < MMSCH_V4_0_VCN_INSTANCES; i++) {
1272 header.inst[i].init_status = 0;
1273 header.inst[i].table_offset = 0;
1274 header.inst[i].table_size = 0;
1275 }
1276
1277 table_loc = (uint32_t *)table->cpu_addr;
1278 table_loc += header.total_size;
1279 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1280 if (adev->vcn.harvest_config & (1 << i))
1281 continue;
1282
1283 // Must re/init fw_shared at beginning
1284 vcn_v4_0_fw_shared_init(adev, i);
1285
1286 table_size = 0;
1287
1288 MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1289 regUVD_STATUS),
1290 ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1291
1292 cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4);
1293
1294 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1295 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1296 regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1297 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1298 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1299 regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1300 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1301 offset = 0;
1302 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1303 regUVD_VCPU_CACHE_OFFSET0),
1304 0);
1305 } else {
1306 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1307 regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1308 lower_32_bits(adev->vcn.inst[i].gpu_addr));
1309 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1310 regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1311 upper_32_bits(adev->vcn.inst[i].gpu_addr));
1312 offset = cache_size;
1313 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1314 regUVD_VCPU_CACHE_OFFSET0),
1315 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1316 }
1317
1318 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1319 regUVD_VCPU_CACHE_SIZE0),
1320 cache_size);
1321
1322 cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1323 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1324 regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1325 lower_32_bits(cache_addr));
1326 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1327 regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1328 upper_32_bits(cache_addr));
1329 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1330 regUVD_VCPU_CACHE_OFFSET1),
1331 0);
1332 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1333 regUVD_VCPU_CACHE_SIZE1),
1334 AMDGPU_VCN_STACK_SIZE);
1335
1336 cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1337 AMDGPU_VCN_STACK_SIZE;
1338 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1339 regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1340 lower_32_bits(cache_addr));
1341 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1342 regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1343 upper_32_bits(cache_addr));
1344 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1345 regUVD_VCPU_CACHE_OFFSET2),
1346 0);
1347 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1348 regUVD_VCPU_CACHE_SIZE2),
1349 AMDGPU_VCN_CONTEXT_SIZE);
1350
1351 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1352 rb_setup = &fw_shared->rb_setup;
1353
1354 ring_enc = &adev->vcn.inst[i].ring_enc[0];
1355 ring_enc->wptr = 0;
1356 rb_enc_addr = ring_enc->gpu_addr;
1357
1358 rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1359 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1360
1361 if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
1362 vcn_v4_0_init_ring_metadata(adev, i, ring_enc);
1363
1364 memset((void *)&rb_setup->rb_info, 0, sizeof(struct amdgpu_vcn_rb_setup_info) * MAX_NUM_VCN_RB_SETUP);
1365 if (!(adev->vcn.harvest_config & (1 << 0))) {
1366 rb_setup->rb_info[0].rb_addr_lo = lower_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1367 rb_setup->rb_info[0].rb_addr_hi = upper_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1368 rb_setup->rb_info[0].rb_size = adev->vcn.inst[0].ring_enc[0].ring_size / 4;
1369 }
1370 if (!(adev->vcn.harvest_config & (1 << 1))) {
1371 rb_setup->rb_info[2].rb_addr_lo = lower_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1372 rb_setup->rb_info[2].rb_addr_hi = upper_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1373 rb_setup->rb_info[2].rb_size = adev->vcn.inst[1].ring_enc[0].ring_size / 4;
1374 }
1375 fw_shared->decouple.is_enabled = 1;
1376 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1377 } else {
1378 rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1379 rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1380 rb_setup->rb_size = ring_enc->ring_size / 4;
1381 }
1382
1383 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1384 regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1385 lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1386 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1387 regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1388 upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1389 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1390 regUVD_VCPU_NONCACHE_SIZE0),
1391 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1392
1393 /* add end packet */
1394 MMSCH_V4_0_INSERT_END();
1395
1396 /* refine header */
1397 header.inst[i].init_status = 0;
1398 header.inst[i].table_offset = header.total_size;
1399 header.inst[i].table_size = table_size;
1400 header.total_size += table_size;
1401 }
1402
1403 /* Update init table header in memory */
1404 size = sizeof(struct mmsch_v4_0_init_header);
1405 table_loc = (uint32_t *)table->cpu_addr;
1406 memcpy((void *)table_loc, &header, size);
1407
1408 /* message MMSCH (in VCN[0]) to initialize this client
1409 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1410 * of memory descriptor location
1411 */
1412 ctx_addr = table->gpu_addr;
1413 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1414 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1415
1416 /* 2, update vmid of descriptor */
1417 tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1418 tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1419 /* use domain0 for MM scheduler */
1420 tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1421 WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1422
1423 /* 3, notify mmsch about the size of this descriptor */
1424 size = header.total_size;
1425 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1426
1427 /* 4, set resp to zero */
1428 WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1429
1430 /* 5, kick off the initialization and wait until
1431 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1432 */
1433 param = 0x00000001;
1434 WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1435 tmp = 0;
1436 timeout = 1000;
1437 resp = 0;
1438 expected = MMSCH_VF_MAILBOX_RESP__OK;
1439 while (resp != expected) {
1440 resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1441 if (resp != 0)
1442 break;
1443
1444 udelay(10);
1445 tmp = tmp + 10;
1446 if (tmp >= timeout) {
1447 DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1448 " waiting for regMMSCH_VF_MAILBOX_RESP "\
1449 "(expected=0x%08x, readback=0x%08x)\n",
1450 tmp, expected, resp);
1451 return -EBUSY;
1452 }
1453 }
1454 enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1455 init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1456 if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1457 && init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1458 DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1459 "status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1460
1461 return 0;
1462}
1463
1464/**
1465 * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1466 *
1467 * @adev: amdgpu_device pointer
1468 * @inst_idx: instance number index
1469 *
1470 * Stop VCN block with dpg mode
1471 */
1472static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1473{
1474 struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1475 uint32_t tmp;
1476
1477 vcn_v4_0_pause_dpg_mode(adev, inst_idx, &state);
1478 /* Wait for power status to be 1 */
1479 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1480 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1481
1482 /* wait for read ptr to be equal to write ptr */
1483 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1484 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1485
1486 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1487 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1488
1489 /* disable dynamic power gating mode */
1490 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1491 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1492}
1493
1494/**
1495 * vcn_v4_0_stop - VCN stop
1496 *
1497 * @adev: amdgpu_device pointer
1498 *
1499 * Stop VCN block
1500 */
1501static int vcn_v4_0_stop(struct amdgpu_device *adev)
1502{
1503 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1504 uint32_t tmp;
1505 int i, r = 0;
1506
1507 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1508 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1509 fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1510
1511 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1512 vcn_v4_0_stop_dpg_mode(adev, i);
1513 continue;
1514 }
1515
1516 /* wait for vcn idle */
1517 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1518 if (r)
1519 return r;
1520
1521 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1522 UVD_LMI_STATUS__READ_CLEAN_MASK |
1523 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1524 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1525 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1526 if (r)
1527 return r;
1528
1529 /* disable LMI UMC channel */
1530 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1531 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1532 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1533 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1534 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1535 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1536 if (r)
1537 return r;
1538
1539 /* block VCPU register access */
1540 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1541 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1542 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1543
1544 /* reset VCPU */
1545 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1546 UVD_VCPU_CNTL__BLK_RST_MASK,
1547 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1548
1549 /* disable VCPU clock */
1550 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1551 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1552
1553 /* apply soft reset */
1554 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1555 tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1556 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1557 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1558 tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1559 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1560
1561 /* clear status */
1562 WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1563
1564 /* apply HW clock gating */
1565 vcn_v4_0_enable_clock_gating(adev, i);
1566
1567 /* enable VCN power gating */
1568 vcn_v4_0_enable_static_power_gating(adev, i);
1569 }
1570
1571 if (adev->pm.dpm_enabled)
1572 amdgpu_dpm_enable_uvd(adev, false);
1573
1574 return 0;
1575}
1576
1577/**
1578 * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1579 *
1580 * @adev: amdgpu_device pointer
1581 * @inst_idx: instance number index
1582 * @new_state: pause state
1583 *
1584 * Pause dpg mode for VCN block
1585 */
1586static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1587 struct dpg_pause_state *new_state)
1588{
1589 uint32_t reg_data = 0;
1590 int ret_code;
1591
1592 /* pause/unpause if state is changed */
1593 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1594 DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1595 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based);
1596 reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1597 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1598
1599 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1600 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1601 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1602
1603 if (!ret_code) {
1604 /* pause DPG */
1605 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1606 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1607
1608 /* wait for ACK */
1609 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1610 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1611 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1612
1613 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1614 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1615 }
1616 } else {
1617 /* unpause dpg, no need to wait */
1618 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1619 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1620 }
1621 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1622 }
1623
1624 return 0;
1625}
1626
1627/**
1628 * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1629 *
1630 * @ring: amdgpu_ring pointer
1631 *
1632 * Returns the current hardware unified read pointer
1633 */
1634static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1635{
1636 struct amdgpu_device *adev = ring->adev;
1637
1638 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1639 DRM_ERROR("wrong ring id is identified in %s", __func__);
1640
1641 return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1642}
1643
1644/**
1645 * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1646 *
1647 * @ring: amdgpu_ring pointer
1648 *
1649 * Returns the current hardware unified write pointer
1650 */
1651static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1652{
1653 struct amdgpu_device *adev = ring->adev;
1654
1655 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1656 DRM_ERROR("wrong ring id is identified in %s", __func__);
1657
1658 if (ring->use_doorbell)
1659 return *ring->wptr_cpu_addr;
1660 else
1661 return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1662}
1663
1664/**
1665 * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1666 *
1667 * @ring: amdgpu_ring pointer
1668 *
1669 * Commits the enc write pointer to the hardware
1670 */
1671static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1672{
1673 struct amdgpu_device *adev = ring->adev;
1674
1675 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1676 DRM_ERROR("wrong ring id is identified in %s", __func__);
1677
1678 if (ring->use_doorbell) {
1679 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1680 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1681 } else {
1682 WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1683 }
1684}
1685
1686static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1687 struct amdgpu_job *job)
1688{
1689 struct drm_gpu_scheduler **scheds;
1690
1691 /* The create msg must be in the first IB submitted */
1692 if (atomic_read(&job->base.entity->fence_seq))
1693 return -EINVAL;
1694
1695 /* if VCN0 is harvested, we can't support AV1 */
1696 if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1697 return -EINVAL;
1698
1699 scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1700 [AMDGPU_RING_PRIO_0].sched;
1701 drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1702 return 0;
1703}
1704
1705static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1706 uint64_t addr)
1707{
1708 struct ttm_operation_ctx ctx = { false, false };
1709 struct amdgpu_bo_va_mapping *map;
1710 uint32_t *msg, num_buffers;
1711 struct amdgpu_bo *bo;
1712 uint64_t start, end;
1713 unsigned int i;
1714 void *ptr;
1715 int r;
1716
1717 addr &= AMDGPU_GMC_HOLE_MASK;
1718 r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1719 if (r) {
1720 DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1721 return r;
1722 }
1723
1724 start = map->start * AMDGPU_GPU_PAGE_SIZE;
1725 end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1726 if (addr & 0x7) {
1727 DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1728 return -EINVAL;
1729 }
1730
1731 bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1732 amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1733 r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1734 if (r) {
1735 DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1736 return r;
1737 }
1738
1739 r = amdgpu_bo_kmap(bo, &ptr);
1740 if (r) {
1741 DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1742 return r;
1743 }
1744
1745 msg = ptr + addr - start;
1746
1747 /* Check length */
1748 if (msg[1] > end - addr) {
1749 r = -EINVAL;
1750 goto out;
1751 }
1752
1753 if (msg[3] != RDECODE_MSG_CREATE)
1754 goto out;
1755
1756 num_buffers = msg[2];
1757 for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1758 uint32_t offset, size, *create;
1759
1760 if (msg[0] != RDECODE_MESSAGE_CREATE)
1761 continue;
1762
1763 offset = msg[1];
1764 size = msg[2];
1765
1766 if (offset + size > end) {
1767 r = -EINVAL;
1768 goto out;
1769 }
1770
1771 create = ptr + addr + offset - start;
1772
1773 /* H264, HEVC and VP9 can run on any instance */
1774 if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1775 continue;
1776
1777 r = vcn_v4_0_limit_sched(p, job);
1778 if (r)
1779 goto out;
1780 }
1781
1782out:
1783 amdgpu_bo_kunmap(bo);
1784 return r;
1785}
1786
1787#define RADEON_VCN_ENGINE_TYPE_ENCODE (0x00000002)
1788#define RADEON_VCN_ENGINE_TYPE_DECODE (0x00000003)
1789
1790#define RADEON_VCN_ENGINE_INFO (0x30000001)
1791#define RADEON_VCN_ENGINE_INFO_MAX_OFFSET 16
1792
1793#define RENCODE_ENCODE_STANDARD_AV1 2
1794#define RENCODE_IB_PARAM_SESSION_INIT 0x00000003
1795#define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET 64
1796
1797/* return the offset in ib if id is found, -1 otherwise
1798 * to speed up the searching we only search upto max_offset
1799 */
1800static int vcn_v4_0_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1801{
1802 int i;
1803
1804 for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1805 if (ib->ptr[i + 1] == id)
1806 return i;
1807 }
1808 return -1;
1809}
1810
1811static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1812 struct amdgpu_job *job,
1813 struct amdgpu_ib *ib)
1814{
1815 struct amdgpu_ring *ring = amdgpu_job_ring(job);
1816 struct amdgpu_vcn_decode_buffer *decode_buffer;
1817 uint64_t addr;
1818 uint32_t val;
1819 int idx;
1820
1821 /* The first instance can decode anything */
1822 if (!ring->me)
1823 return 0;
1824
1825 /* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1826 idx = vcn_v4_0_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1827 RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1828 if (idx < 0) /* engine info is missing */
1829 return 0;
1830
1831 val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1832 if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1833 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1834
1835 if (!(decode_buffer->valid_buf_flag & 0x1))
1836 return 0;
1837
1838 addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1839 decode_buffer->msg_buffer_address_lo;
1840 return vcn_v4_0_dec_msg(p, job, addr);
1841 } else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1842 idx = vcn_v4_0_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1843 RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1844 if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1845 return vcn_v4_0_limit_sched(p, job);
1846 }
1847 return 0;
1848}
1849
1850static struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1851 .type = AMDGPU_RING_TYPE_VCN_ENC,
1852 .align_mask = 0x3f,
1853 .nop = VCN_ENC_CMD_NO_OP,
1854 .extra_dw = sizeof(struct amdgpu_vcn_rb_metadata),
1855 .get_rptr = vcn_v4_0_unified_ring_get_rptr,
1856 .get_wptr = vcn_v4_0_unified_ring_get_wptr,
1857 .set_wptr = vcn_v4_0_unified_ring_set_wptr,
1858 .patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1859 .emit_frame_size =
1860 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1861 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1862 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1863 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1864 1, /* vcn_v2_0_enc_ring_insert_end */
1865 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1866 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1867 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1868 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1869 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1870 .test_ib = amdgpu_vcn_unified_ring_test_ib,
1871 .insert_nop = amdgpu_ring_insert_nop,
1872 .insert_end = vcn_v2_0_enc_ring_insert_end,
1873 .pad_ib = amdgpu_ring_generic_pad_ib,
1874 .begin_use = amdgpu_vcn_ring_begin_use,
1875 .end_use = amdgpu_vcn_ring_end_use,
1876 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1877 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1878 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1879};
1880
1881/**
1882 * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1883 *
1884 * @adev: amdgpu_device pointer
1885 *
1886 * Set unified ring functions
1887 */
1888static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1889{
1890 int i;
1891
1892 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1893 if (adev->vcn.harvest_config & (1 << i))
1894 continue;
1895
1896 if (amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 2))
1897 vcn_v4_0_unified_ring_vm_funcs.secure_submission_supported = true;
1898
1899 adev->vcn.inst[i].ring_enc[0].funcs =
1900 (const struct amdgpu_ring_funcs *)&vcn_v4_0_unified_ring_vm_funcs;
1901 adev->vcn.inst[i].ring_enc[0].me = i;
1902
1903 DRM_INFO("VCN(%d) encode/decode are enabled in VM mode\n", i);
1904 }
1905}
1906
1907/**
1908 * vcn_v4_0_is_idle - check VCN block is idle
1909 *
1910 * @handle: amdgpu_device pointer
1911 *
1912 * Check whether VCN block is idle
1913 */
1914static bool vcn_v4_0_is_idle(void *handle)
1915{
1916 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1917 int i, ret = 1;
1918
1919 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1920 if (adev->vcn.harvest_config & (1 << i))
1921 continue;
1922
1923 ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1924 }
1925
1926 return ret;
1927}
1928
1929/**
1930 * vcn_v4_0_wait_for_idle - wait for VCN block idle
1931 *
1932 * @handle: amdgpu_device pointer
1933 *
1934 * Wait for VCN block idle
1935 */
1936static int vcn_v4_0_wait_for_idle(void *handle)
1937{
1938 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1939 int i, ret = 0;
1940
1941 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1942 if (adev->vcn.harvest_config & (1 << i))
1943 continue;
1944
1945 ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1946 UVD_STATUS__IDLE);
1947 if (ret)
1948 return ret;
1949 }
1950
1951 return ret;
1952}
1953
1954/**
1955 * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
1956 *
1957 * @handle: amdgpu_device pointer
1958 * @state: clock gating state
1959 *
1960 * Set VCN block clockgating state
1961 */
1962static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1963{
1964 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1965 bool enable = state == AMD_CG_STATE_GATE;
1966 int i;
1967
1968 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1969 if (adev->vcn.harvest_config & (1 << i))
1970 continue;
1971
1972 if (enable) {
1973 if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1974 return -EBUSY;
1975 vcn_v4_0_enable_clock_gating(adev, i);
1976 } else {
1977 vcn_v4_0_disable_clock_gating(adev, i);
1978 }
1979 }
1980
1981 return 0;
1982}
1983
1984/**
1985 * vcn_v4_0_set_powergating_state - set VCN block powergating state
1986 *
1987 * @handle: amdgpu_device pointer
1988 * @state: power gating state
1989 *
1990 * Set VCN block powergating state
1991 */
1992static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
1993{
1994 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1995 int ret;
1996
1997 /* for SRIOV, guest should not control VCN Power-gating
1998 * MMSCH FW should control Power-gating and clock-gating
1999 * guest should avoid touching CGC and PG
2000 */
2001 if (amdgpu_sriov_vf(adev)) {
2002 adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2003 return 0;
2004 }
2005
2006 if (state == adev->vcn.cur_state)
2007 return 0;
2008
2009 if (state == AMD_PG_STATE_GATE)
2010 ret = vcn_v4_0_stop(adev);
2011 else
2012 ret = vcn_v4_0_start(adev);
2013
2014 if (!ret)
2015 adev->vcn.cur_state = state;
2016
2017 return ret;
2018}
2019
2020/**
2021 * vcn_v4_0_set_ras_interrupt_state - set VCN block RAS interrupt state
2022 *
2023 * @adev: amdgpu_device pointer
2024 * @source: interrupt sources
2025 * @type: interrupt types
2026 * @state: interrupt states
2027 *
2028 * Set VCN block RAS interrupt state
2029 */
2030static int vcn_v4_0_set_ras_interrupt_state(struct amdgpu_device *adev,
2031 struct amdgpu_irq_src *source,
2032 unsigned int type,
2033 enum amdgpu_interrupt_state state)
2034{
2035 return 0;
2036}
2037
2038/**
2039 * vcn_v4_0_process_interrupt - process VCN block interrupt
2040 *
2041 * @adev: amdgpu_device pointer
2042 * @source: interrupt sources
2043 * @entry: interrupt entry from clients and sources
2044 *
2045 * Process VCN block interrupt
2046 */
2047static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
2048 struct amdgpu_iv_entry *entry)
2049{
2050 uint32_t ip_instance;
2051
2052 if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
2053 ip_instance = entry->ring_id;
2054 } else {
2055 switch (entry->client_id) {
2056 case SOC15_IH_CLIENTID_VCN:
2057 ip_instance = 0;
2058 break;
2059 case SOC15_IH_CLIENTID_VCN1:
2060 ip_instance = 1;
2061 break;
2062 default:
2063 DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2064 return 0;
2065 }
2066 }
2067
2068 DRM_DEBUG("IH: VCN TRAP\n");
2069
2070 switch (entry->src_id) {
2071 case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2072 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2073 break;
2074 default:
2075 DRM_ERROR("Unhandled interrupt: %d %d\n",
2076 entry->src_id, entry->src_data[0]);
2077 break;
2078 }
2079
2080 return 0;
2081}
2082
2083static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
2084 .process = vcn_v4_0_process_interrupt,
2085};
2086
2087static const struct amdgpu_irq_src_funcs vcn_v4_0_ras_irq_funcs = {
2088 .set = vcn_v4_0_set_ras_interrupt_state,
2089 .process = amdgpu_vcn_process_poison_irq,
2090};
2091
2092/**
2093 * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
2094 *
2095 * @adev: amdgpu_device pointer
2096 *
2097 * Set VCN block interrupt irq functions
2098 */
2099static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2100{
2101 int i;
2102
2103 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2104 if (adev->vcn.harvest_config & (1 << i))
2105 continue;
2106
2107 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2108 adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2109
2110 adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
2111 adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v4_0_ras_irq_funcs;
2112 }
2113}
2114
2115static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2116 .name = "vcn_v4_0",
2117 .early_init = vcn_v4_0_early_init,
2118 .late_init = NULL,
2119 .sw_init = vcn_v4_0_sw_init,
2120 .sw_fini = vcn_v4_0_sw_fini,
2121 .hw_init = vcn_v4_0_hw_init,
2122 .hw_fini = vcn_v4_0_hw_fini,
2123 .suspend = vcn_v4_0_suspend,
2124 .resume = vcn_v4_0_resume,
2125 .is_idle = vcn_v4_0_is_idle,
2126 .wait_for_idle = vcn_v4_0_wait_for_idle,
2127 .check_soft_reset = NULL,
2128 .pre_soft_reset = NULL,
2129 .soft_reset = NULL,
2130 .post_soft_reset = NULL,
2131 .set_clockgating_state = vcn_v4_0_set_clockgating_state,
2132 .set_powergating_state = vcn_v4_0_set_powergating_state,
2133};
2134
2135const struct amdgpu_ip_block_version vcn_v4_0_ip_block = {
2136 .type = AMD_IP_BLOCK_TYPE_VCN,
2137 .major = 4,
2138 .minor = 0,
2139 .rev = 0,
2140 .funcs = &vcn_v4_0_ip_funcs,
2141};
2142
2143static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2144 uint32_t instance, uint32_t sub_block)
2145{
2146 uint32_t poison_stat = 0, reg_value = 0;
2147
2148 switch (sub_block) {
2149 case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2150 reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2151 poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2152 break;
2153 default:
2154 break;
2155 }
2156
2157 if (poison_stat)
2158 dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2159 instance, sub_block);
2160
2161 return poison_stat;
2162}
2163
2164static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2165{
2166 uint32_t inst, sub;
2167 uint32_t poison_stat = 0;
2168
2169 for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2170 for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2171 poison_stat +=
2172 vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2173
2174 return !!poison_stat;
2175}
2176
2177const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2178 .query_poison_status = vcn_v4_0_query_ras_poison_status,
2179};
2180
2181static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2182 .ras_block = {
2183 .hw_ops = &vcn_v4_0_ras_hw_ops,
2184 .ras_late_init = amdgpu_vcn_ras_late_init,
2185 },
2186};
2187
2188static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2189{
2190 switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
2191 case IP_VERSION(4, 0, 0):
2192 adev->vcn.ras = &vcn_v4_0_ras;
2193 break;
2194 default:
2195 break;
2196 }
2197}
1/*
2 * Copyright 2021 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 "amdgpu.h"
26#include "amdgpu_vcn.h"
27#include "amdgpu_pm.h"
28#include "amdgpu_cs.h"
29#include "soc15.h"
30#include "soc15d.h"
31#include "soc15_hw_ip.h"
32#include "vcn_v2_0.h"
33#include "mmsch_v4_0.h"
34#include "vcn_v4_0.h"
35
36#include "vcn/vcn_4_0_0_offset.h"
37#include "vcn/vcn_4_0_0_sh_mask.h"
38#include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
39
40#include <drm/drm_drv.h>
41
42#define mmUVD_DPG_LMA_CTL regUVD_DPG_LMA_CTL
43#define mmUVD_DPG_LMA_CTL_BASE_IDX regUVD_DPG_LMA_CTL_BASE_IDX
44#define mmUVD_DPG_LMA_DATA regUVD_DPG_LMA_DATA
45#define mmUVD_DPG_LMA_DATA_BASE_IDX regUVD_DPG_LMA_DATA_BASE_IDX
46
47#define VCN_VID_SOC_ADDRESS_2_0 0x1fb00
48#define VCN1_VID_SOC_ADDRESS_3_0 0x48300
49
50#define VCN_HARVEST_MMSCH 0
51
52#define RDECODE_MSG_CREATE 0x00000000
53#define RDECODE_MESSAGE_CREATE 0x00000001
54
55static const struct amdgpu_hwip_reg_entry vcn_reg_list_4_0[] = {
56 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_POWER_STATUS),
57 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_STATUS),
58 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_CONTEXT_ID),
59 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_CONTEXT_ID2),
60 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_DATA0),
61 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_DATA1),
62 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_GPCOM_VCPU_CMD),
63 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI),
64 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO),
65 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI2),
66 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO2),
67 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI3),
68 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO3),
69 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_HI4),
70 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_BASE_LO4),
71 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR),
72 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR),
73 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR2),
74 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR2),
75 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR3),
76 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR3),
77 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_RPTR4),
78 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_WPTR4),
79 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE),
80 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE2),
81 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE3),
82 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_RB_SIZE4),
83 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_PGFSM_CONFIG),
84 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_PGFSM_STATUS),
85 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_CTL),
86 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_DATA),
87 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_LMA_MASK),
88 SOC15_REG_ENTRY_STR(VCN, 0, regUVD_DPG_PAUSE)
89};
90
91static int amdgpu_ih_clientid_vcns[] = {
92 SOC15_IH_CLIENTID_VCN,
93 SOC15_IH_CLIENTID_VCN1
94};
95
96static int vcn_v4_0_start_sriov(struct amdgpu_device *adev);
97static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev);
98static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
99static int vcn_v4_0_set_powergating_state(void *handle,
100 enum amd_powergating_state state);
101static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
102 int inst_idx, struct dpg_pause_state *new_state);
103static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring);
104static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev);
105
106/**
107 * vcn_v4_0_early_init - set function pointers and load microcode
108 *
109 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
110 *
111 * Set ring and irq function pointers
112 * Load microcode from filesystem
113 */
114static int vcn_v4_0_early_init(struct amdgpu_ip_block *ip_block)
115{
116 struct amdgpu_device *adev = ip_block->adev;
117 int i;
118
119 if (amdgpu_sriov_vf(adev)) {
120 adev->vcn.harvest_config = VCN_HARVEST_MMSCH;
121 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
122 if (amdgpu_vcn_is_disabled_vcn(adev, VCN_ENCODE_RING, i)) {
123 adev->vcn.harvest_config |= 1 << i;
124 dev_info(adev->dev, "VCN%d is disabled by hypervisor\n", i);
125 }
126 }
127 }
128
129 /* re-use enc ring as unified ring */
130 adev->vcn.num_enc_rings = 1;
131
132 vcn_v4_0_set_unified_ring_funcs(adev);
133 vcn_v4_0_set_irq_funcs(adev);
134 vcn_v4_0_set_ras_funcs(adev);
135
136 return amdgpu_vcn_early_init(adev);
137}
138
139static int vcn_v4_0_fw_shared_init(struct amdgpu_device *adev, int inst_idx)
140{
141 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
142
143 fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
144 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
145 fw_shared->sq.is_enabled = 1;
146
147 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
148 fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
149 AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
150
151 if (amdgpu_ip_version(adev, VCN_HWIP, 0) ==
152 IP_VERSION(4, 0, 2)) {
153 fw_shared->present_flag_0 |= AMDGPU_FW_SHARED_FLAG_0_DRM_KEY_INJECT;
154 fw_shared->drm_key_wa.method =
155 AMDGPU_DRM_KEY_INJECT_WORKAROUND_VCNFW_ASD_HANDSHAKING;
156 }
157
158 if (amdgpu_vcnfw_log)
159 amdgpu_vcn_fwlog_init(&adev->vcn.inst[inst_idx]);
160
161 return 0;
162}
163
164/**
165 * vcn_v4_0_sw_init - sw init for VCN block
166 *
167 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
168 *
169 * Load firmware and sw initialization
170 */
171static int vcn_v4_0_sw_init(struct amdgpu_ip_block *ip_block)
172{
173 struct amdgpu_ring *ring;
174 struct amdgpu_device *adev = ip_block->adev;
175 int i, r;
176 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0);
177 uint32_t *ptr;
178
179 r = amdgpu_vcn_sw_init(adev);
180 if (r)
181 return r;
182
183 amdgpu_vcn_setup_ucode(adev);
184
185 r = amdgpu_vcn_resume(adev);
186 if (r)
187 return r;
188
189 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
190 if (adev->vcn.harvest_config & (1 << i))
191 continue;
192
193 /* Init instance 0 sched_score to 1, so it's scheduled after other instances */
194 if (i == 0)
195 atomic_set(&adev->vcn.inst[i].sched_score, 1);
196 else
197 atomic_set(&adev->vcn.inst[i].sched_score, 0);
198
199 /* VCN UNIFIED TRAP */
200 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
201 VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
202 if (r)
203 return r;
204
205 /* VCN POISON TRAP */
206 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
207 VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].ras_poison_irq);
208 if (r)
209 return r;
210
211 ring = &adev->vcn.inst[i].ring_enc[0];
212 ring->use_doorbell = true;
213 if (amdgpu_sriov_vf(adev))
214 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + i * (adev->vcn.num_enc_rings + 1) + 1;
215 else
216 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + 8 * i;
217 ring->vm_hub = AMDGPU_MMHUB0(0);
218 sprintf(ring->name, "vcn_unified_%d", i);
219
220 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
221 AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
222 if (r)
223 return r;
224
225 vcn_v4_0_fw_shared_init(adev, i);
226 }
227
228 /* TODO: Add queue reset mask when FW fully supports it */
229 adev->vcn.supported_reset =
230 amdgpu_get_soft_full_reset_mask(&adev->vcn.inst[0].ring_enc[0]);
231
232 if (amdgpu_sriov_vf(adev)) {
233 r = amdgpu_virt_alloc_mm_table(adev);
234 if (r)
235 return r;
236 }
237
238 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
239 adev->vcn.pause_dpg_mode = vcn_v4_0_pause_dpg_mode;
240
241 r = amdgpu_vcn_ras_sw_init(adev);
242 if (r)
243 return r;
244
245 /* Allocate memory for VCN IP Dump buffer */
246 ptr = kcalloc(adev->vcn.num_vcn_inst * reg_count, sizeof(uint32_t), GFP_KERNEL);
247 if (!ptr) {
248 DRM_ERROR("Failed to allocate memory for VCN IP Dump\n");
249 adev->vcn.ip_dump = NULL;
250 } else {
251 adev->vcn.ip_dump = ptr;
252 }
253
254 r = amdgpu_vcn_sysfs_reset_mask_init(adev);
255 if (r)
256 return r;
257
258 return 0;
259}
260
261/**
262 * vcn_v4_0_sw_fini - sw fini for VCN block
263 *
264 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
265 *
266 * VCN suspend and free up sw allocation
267 */
268static int vcn_v4_0_sw_fini(struct amdgpu_ip_block *ip_block)
269{
270 struct amdgpu_device *adev = ip_block->adev;
271 int i, r, idx;
272
273 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
274 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
275 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
276
277 if (adev->vcn.harvest_config & (1 << i))
278 continue;
279
280 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
281 fw_shared->present_flag_0 = 0;
282 fw_shared->sq.is_enabled = 0;
283 }
284
285 drm_dev_exit(idx);
286 }
287
288 if (amdgpu_sriov_vf(adev))
289 amdgpu_virt_free_mm_table(adev);
290
291 r = amdgpu_vcn_suspend(adev);
292 if (r)
293 return r;
294
295 amdgpu_vcn_sysfs_reset_mask_fini(adev);
296 r = amdgpu_vcn_sw_fini(adev);
297
298 kfree(adev->vcn.ip_dump);
299
300 return r;
301}
302
303/**
304 * vcn_v4_0_hw_init - start and test VCN block
305 *
306 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
307 *
308 * Initialize the hardware, boot up the VCPU and do some testing
309 */
310static int vcn_v4_0_hw_init(struct amdgpu_ip_block *ip_block)
311{
312 struct amdgpu_device *adev = ip_block->adev;
313 struct amdgpu_ring *ring;
314 int i, r;
315
316 if (amdgpu_sriov_vf(adev)) {
317 r = vcn_v4_0_start_sriov(adev);
318 if (r)
319 return r;
320
321 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
322 if (adev->vcn.harvest_config & (1 << i))
323 continue;
324
325 ring = &adev->vcn.inst[i].ring_enc[0];
326 ring->wptr = 0;
327 ring->wptr_old = 0;
328 vcn_v4_0_unified_ring_set_wptr(ring);
329 ring->sched.ready = true;
330 }
331 } else {
332 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
333 if (adev->vcn.harvest_config & (1 << i))
334 continue;
335
336 ring = &adev->vcn.inst[i].ring_enc[0];
337
338 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
339 ((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
340
341 r = amdgpu_ring_test_helper(ring);
342 if (r)
343 return r;
344 }
345 }
346
347 return 0;
348}
349
350/**
351 * vcn_v4_0_hw_fini - stop the hardware block
352 *
353 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
354 *
355 * Stop the VCN block, mark ring as not ready any more
356 */
357static int vcn_v4_0_hw_fini(struct amdgpu_ip_block *ip_block)
358{
359 struct amdgpu_device *adev = ip_block->adev;
360 int i;
361
362 cancel_delayed_work_sync(&adev->vcn.idle_work);
363
364 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
365 if (adev->vcn.harvest_config & (1 << i))
366 continue;
367 if (!amdgpu_sriov_vf(adev)) {
368 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
369 (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
370 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
371 vcn_v4_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
372 }
373 }
374 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
375 amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
376 }
377
378 return 0;
379}
380
381/**
382 * vcn_v4_0_suspend - suspend VCN block
383 *
384 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
385 *
386 * HW fini and suspend VCN block
387 */
388static int vcn_v4_0_suspend(struct amdgpu_ip_block *ip_block)
389{
390 int r;
391
392 r = vcn_v4_0_hw_fini(ip_block);
393 if (r)
394 return r;
395
396 r = amdgpu_vcn_suspend(ip_block->adev);
397
398 return r;
399}
400
401/**
402 * vcn_v4_0_resume - resume VCN block
403 *
404 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
405 *
406 * Resume firmware and hw init VCN block
407 */
408static int vcn_v4_0_resume(struct amdgpu_ip_block *ip_block)
409{
410 int r;
411
412 r = amdgpu_vcn_resume(ip_block->adev);
413 if (r)
414 return r;
415
416 r = vcn_v4_0_hw_init(ip_block);
417
418 return r;
419}
420
421/**
422 * vcn_v4_0_mc_resume - memory controller programming
423 *
424 * @adev: amdgpu_device pointer
425 * @inst: instance number
426 *
427 * Let the VCN memory controller know it's offsets
428 */
429static void vcn_v4_0_mc_resume(struct amdgpu_device *adev, int inst)
430{
431 uint32_t offset, size;
432 const struct common_firmware_header *hdr;
433
434 hdr = (const struct common_firmware_header *)adev->vcn.fw[inst]->data;
435 size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
436
437 /* cache window 0: fw */
438 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
439 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
440 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
441 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
442 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
443 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
444 offset = 0;
445 } else {
446 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
447 lower_32_bits(adev->vcn.inst[inst].gpu_addr));
448 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
449 upper_32_bits(adev->vcn.inst[inst].gpu_addr));
450 offset = size;
451 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
452 }
453 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
454
455 /* cache window 1: stack */
456 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
457 lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
458 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
459 upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
460 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
461 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
462
463 /* cache window 2: context */
464 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
465 lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
466 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
467 upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
468 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
469 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
470
471 /* non-cache window */
472 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
473 lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
474 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
475 upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
476 WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
477 WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
478 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
479}
480
481/**
482 * vcn_v4_0_mc_resume_dpg_mode - memory controller programming for dpg mode
483 *
484 * @adev: amdgpu_device pointer
485 * @inst_idx: instance number index
486 * @indirect: indirectly write sram
487 *
488 * Let the VCN memory controller know it's offsets with dpg mode
489 */
490static void vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
491{
492 uint32_t offset, size;
493 const struct common_firmware_header *hdr;
494 hdr = (const struct common_firmware_header *)adev->vcn.fw[inst_idx]->data;
495 size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
496
497 /* cache window 0: fw */
498 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
499 if (!indirect) {
500 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
501 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
502 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
503 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
505 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
506 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
508 } else {
509 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
510 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
511 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
512 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
513 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
515 }
516 offset = 0;
517 } else {
518 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
519 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
520 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
521 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
522 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
523 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
524 offset = size;
525 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
527 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
528 }
529
530 if (!indirect)
531 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
532 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
533 else
534 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
535 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
536
537 /* cache window 1: stack */
538 if (!indirect) {
539 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
540 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
541 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
542 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
543 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
544 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
545 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
546 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
547 } else {
548 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
549 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
550 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
551 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
552 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
553 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
554 }
555 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
556 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
557
558 /* cache window 2: context */
559 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
560 VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
561 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
562 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
563 VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
564 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
565 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
566 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
567 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
568 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
569
570 /* non-cache window */
571 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
572 VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
573 lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
574 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
575 VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
576 upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
577 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
578 VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
579 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
580 VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
581 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
582
583 /* VCN global tiling registers */
584 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
585 VCN, 0, regUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
586}
587
588/**
589 * vcn_v4_0_disable_static_power_gating - disable VCN static power gating
590 *
591 * @adev: amdgpu_device pointer
592 * @inst: instance number
593 *
594 * Disable static power gating for VCN block
595 */
596static void vcn_v4_0_disable_static_power_gating(struct amdgpu_device *adev, int inst)
597{
598 uint32_t data = 0;
599
600 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
601 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
602 | 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
603 | 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
604 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
605 | 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
606 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
607 | 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
608 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
609 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
610 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
611 | 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
612 | 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
613 | 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
614 | 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
615
616 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
617 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS,
618 UVD_PGFSM_STATUS__UVDM_UVDU_UVDLM_PWR_ON_3_0, 0x3F3FFFFF);
619 } else {
620 uint32_t value;
621
622 value = (inst) ? 0x2200800 : 0;
623 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
624 | 1 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
625 | 1 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
626 | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
627 | 1 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
628 | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
629 | 1 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
630 | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
631 | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
632 | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
633 | 1 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
634 | 1 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
635 | 1 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
636 | 1 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
637
638 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
639 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, value, 0x3F3FFFFF);
640 }
641
642 data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
643 data &= ~0x103;
644 if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
645 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
646 UVD_POWER_STATUS__UVD_PG_EN_MASK;
647
648 WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
649
650 return;
651}
652
653/**
654 * vcn_v4_0_enable_static_power_gating - enable VCN static power gating
655 *
656 * @adev: amdgpu_device pointer
657 * @inst: instance number
658 *
659 * Enable static power gating for VCN block
660 */
661static void vcn_v4_0_enable_static_power_gating(struct amdgpu_device *adev, int inst)
662{
663 uint32_t data;
664
665 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
666 /* Before power off, this indicator has to be turned on */
667 data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
668 data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
669 data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
670 WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
671
672 data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
673 | 2 << UVD_PGFSM_CONFIG__UVDS_PWR_CONFIG__SHIFT
674 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
675 | 2 << UVD_PGFSM_CONFIG__UVDTC_PWR_CONFIG__SHIFT
676 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
677 | 2 << UVD_PGFSM_CONFIG__UVDTA_PWR_CONFIG__SHIFT
678 | 2 << UVD_PGFSM_CONFIG__UVDLM_PWR_CONFIG__SHIFT
679 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
680 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
681 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT
682 | 2 << UVD_PGFSM_CONFIG__UVDAB_PWR_CONFIG__SHIFT
683 | 2 << UVD_PGFSM_CONFIG__UVDTB_PWR_CONFIG__SHIFT
684 | 2 << UVD_PGFSM_CONFIG__UVDNA_PWR_CONFIG__SHIFT
685 | 2 << UVD_PGFSM_CONFIG__UVDNB_PWR_CONFIG__SHIFT);
686 WREG32_SOC15(VCN, inst, regUVD_PGFSM_CONFIG, data);
687
688 data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
689 | 2 << UVD_PGFSM_STATUS__UVDS_PWR_STATUS__SHIFT
690 | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
691 | 2 << UVD_PGFSM_STATUS__UVDTC_PWR_STATUS__SHIFT
692 | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
693 | 2 << UVD_PGFSM_STATUS__UVDTA_PWR_STATUS__SHIFT
694 | 2 << UVD_PGFSM_STATUS__UVDLM_PWR_STATUS__SHIFT
695 | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
696 | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
697 | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT
698 | 2 << UVD_PGFSM_STATUS__UVDAB_PWR_STATUS__SHIFT
699 | 2 << UVD_PGFSM_STATUS__UVDTB_PWR_STATUS__SHIFT
700 | 2 << UVD_PGFSM_STATUS__UVDNA_PWR_STATUS__SHIFT
701 | 2 << UVD_PGFSM_STATUS__UVDNB_PWR_STATUS__SHIFT);
702 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_PGFSM_STATUS, data, 0x3F3FFFFF);
703 }
704
705 return;
706}
707
708/**
709 * vcn_v4_0_disable_clock_gating - disable VCN clock gating
710 *
711 * @adev: amdgpu_device pointer
712 * @inst: instance number
713 *
714 * Disable clock gating for VCN block
715 */
716static void vcn_v4_0_disable_clock_gating(struct amdgpu_device *adev, int inst)
717{
718 uint32_t data;
719
720 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
721 return;
722
723 /* VCN disable CGC */
724 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
725 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
726 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
727 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
728 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
729
730 data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
731 data &= ~(UVD_CGC_GATE__SYS_MASK
732 | UVD_CGC_GATE__UDEC_MASK
733 | UVD_CGC_GATE__MPEG2_MASK
734 | UVD_CGC_GATE__REGS_MASK
735 | UVD_CGC_GATE__RBC_MASK
736 | UVD_CGC_GATE__LMI_MC_MASK
737 | UVD_CGC_GATE__LMI_UMC_MASK
738 | UVD_CGC_GATE__IDCT_MASK
739 | UVD_CGC_GATE__MPRD_MASK
740 | UVD_CGC_GATE__MPC_MASK
741 | UVD_CGC_GATE__LBSI_MASK
742 | UVD_CGC_GATE__LRBBM_MASK
743 | UVD_CGC_GATE__UDEC_RE_MASK
744 | UVD_CGC_GATE__UDEC_CM_MASK
745 | UVD_CGC_GATE__UDEC_IT_MASK
746 | UVD_CGC_GATE__UDEC_DB_MASK
747 | UVD_CGC_GATE__UDEC_MP_MASK
748 | UVD_CGC_GATE__WCB_MASK
749 | UVD_CGC_GATE__VCPU_MASK
750 | UVD_CGC_GATE__MMSCH_MASK);
751
752 WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
753 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0, 0xFFFFFFFF);
754
755 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
756 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
757 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
758 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
759 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
760 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
761 | UVD_CGC_CTRL__SYS_MODE_MASK
762 | UVD_CGC_CTRL__UDEC_MODE_MASK
763 | UVD_CGC_CTRL__MPEG2_MODE_MASK
764 | UVD_CGC_CTRL__REGS_MODE_MASK
765 | UVD_CGC_CTRL__RBC_MODE_MASK
766 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
767 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
768 | UVD_CGC_CTRL__IDCT_MODE_MASK
769 | UVD_CGC_CTRL__MPRD_MODE_MASK
770 | UVD_CGC_CTRL__MPC_MODE_MASK
771 | UVD_CGC_CTRL__LBSI_MODE_MASK
772 | UVD_CGC_CTRL__LRBBM_MODE_MASK
773 | UVD_CGC_CTRL__WCB_MODE_MASK
774 | UVD_CGC_CTRL__VCPU_MODE_MASK
775 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
776 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
777
778 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
779 data |= (UVD_SUVD_CGC_GATE__SRE_MASK
780 | UVD_SUVD_CGC_GATE__SIT_MASK
781 | UVD_SUVD_CGC_GATE__SMP_MASK
782 | UVD_SUVD_CGC_GATE__SCM_MASK
783 | UVD_SUVD_CGC_GATE__SDB_MASK
784 | UVD_SUVD_CGC_GATE__SRE_H264_MASK
785 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
786 | UVD_SUVD_CGC_GATE__SIT_H264_MASK
787 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
788 | UVD_SUVD_CGC_GATE__SCM_H264_MASK
789 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
790 | UVD_SUVD_CGC_GATE__SDB_H264_MASK
791 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
792 | UVD_SUVD_CGC_GATE__SCLR_MASK
793 | UVD_SUVD_CGC_GATE__UVD_SC_MASK
794 | UVD_SUVD_CGC_GATE__ENT_MASK
795 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
796 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
797 | UVD_SUVD_CGC_GATE__SITE_MASK
798 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
799 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
800 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
801 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
802 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
803 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
804
805 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
806 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
807 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
808 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
809 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
810 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
811 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
812 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
813 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
814 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
815 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
816 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
817}
818
819/**
820 * vcn_v4_0_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
821 *
822 * @adev: amdgpu_device pointer
823 * @sram_sel: sram select
824 * @inst_idx: instance number index
825 * @indirect: indirectly write sram
826 *
827 * Disable clock gating for VCN block with dpg mode
828 */
829static void vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
830 int inst_idx, uint8_t indirect)
831{
832 uint32_t reg_data = 0;
833
834 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
835 return;
836
837 /* enable sw clock gating control */
838 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
839 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
840 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
841 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
842 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
843 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
844 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
845 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
846 UVD_CGC_CTRL__SYS_MODE_MASK |
847 UVD_CGC_CTRL__UDEC_MODE_MASK |
848 UVD_CGC_CTRL__MPEG2_MODE_MASK |
849 UVD_CGC_CTRL__REGS_MODE_MASK |
850 UVD_CGC_CTRL__RBC_MODE_MASK |
851 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
852 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
853 UVD_CGC_CTRL__IDCT_MODE_MASK |
854 UVD_CGC_CTRL__MPRD_MODE_MASK |
855 UVD_CGC_CTRL__MPC_MODE_MASK |
856 UVD_CGC_CTRL__LBSI_MODE_MASK |
857 UVD_CGC_CTRL__LRBBM_MODE_MASK |
858 UVD_CGC_CTRL__WCB_MODE_MASK |
859 UVD_CGC_CTRL__VCPU_MODE_MASK);
860 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
861 VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
862
863 /* turn off clock gating */
864 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
865 VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
866
867 /* turn on SUVD clock gating */
868 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
869 VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
870
871 /* turn on sw mode in UVD_SUVD_CGC_CTRL */
872 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
873 VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
874}
875
876/**
877 * vcn_v4_0_enable_clock_gating - enable VCN clock gating
878 *
879 * @adev: amdgpu_device pointer
880 * @inst: instance number
881 *
882 * Enable clock gating for VCN block
883 */
884static void vcn_v4_0_enable_clock_gating(struct amdgpu_device *adev, int inst)
885{
886 uint32_t data;
887
888 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
889 return;
890
891 /* enable VCN CGC */
892 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
893 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
894 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
895 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
896 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
897
898 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
899 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
900 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
901 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
902 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
903 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
904 | UVD_CGC_CTRL__SYS_MODE_MASK
905 | UVD_CGC_CTRL__UDEC_MODE_MASK
906 | UVD_CGC_CTRL__MPEG2_MODE_MASK
907 | UVD_CGC_CTRL__REGS_MODE_MASK
908 | UVD_CGC_CTRL__RBC_MODE_MASK
909 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
910 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
911 | UVD_CGC_CTRL__IDCT_MODE_MASK
912 | UVD_CGC_CTRL__MPRD_MODE_MASK
913 | UVD_CGC_CTRL__MPC_MODE_MASK
914 | UVD_CGC_CTRL__LBSI_MODE_MASK
915 | UVD_CGC_CTRL__LRBBM_MODE_MASK
916 | UVD_CGC_CTRL__WCB_MODE_MASK
917 | UVD_CGC_CTRL__VCPU_MODE_MASK
918 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
919 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
920
921 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
922 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
923 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
924 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
925 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
926 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
927 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
928 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
929 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
930 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
931 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
932 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
933}
934
935static void vcn_v4_0_enable_ras(struct amdgpu_device *adev, int inst_idx,
936 bool indirect)
937{
938 uint32_t tmp;
939
940 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
941 return;
942
943 tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
944 VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
945 VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
946 VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
947 WREG32_SOC15_DPG_MODE(inst_idx,
948 SOC15_DPG_MODE_OFFSET(VCN, 0, regVCN_RAS_CNTL),
949 tmp, 0, indirect);
950
951 tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
952 WREG32_SOC15_DPG_MODE(inst_idx,
953 SOC15_DPG_MODE_OFFSET(VCN, 0, regUVD_SYS_INT_EN),
954 tmp, 0, indirect);
955}
956
957/**
958 * vcn_v4_0_start_dpg_mode - VCN start with dpg mode
959 *
960 * @adev: amdgpu_device pointer
961 * @inst_idx: instance number index
962 * @indirect: indirectly write sram
963 *
964 * Start VCN block with dpg mode
965 */
966static int vcn_v4_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
967{
968 volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
969 struct amdgpu_ring *ring;
970 uint32_t tmp;
971
972 /* disable register anti-hang mechanism */
973 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
974 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
975 /* enable dynamic power gating mode */
976 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
977 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
978 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
979 WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
980
981 if (indirect)
982 adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
983
984 /* enable clock gating */
985 vcn_v4_0_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
986
987 /* enable VCPU clock */
988 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
989 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
990 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
991 VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
992
993 /* disable master interupt */
994 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
995 VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
996
997 /* setup regUVD_LMI_CTRL */
998 tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
999 UVD_LMI_CTRL__REQ_MODE_MASK |
1000 UVD_LMI_CTRL__CRC_RESET_MASK |
1001 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1002 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1003 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
1004 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
1005 0x00100000L);
1006 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1007 VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
1008
1009 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1010 VCN, inst_idx, regUVD_MPC_CNTL),
1011 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
1012
1013 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1014 VCN, inst_idx, regUVD_MPC_SET_MUXA0),
1015 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1016 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1017 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1018 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
1019
1020 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1021 VCN, inst_idx, regUVD_MPC_SET_MUXB0),
1022 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1023 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1024 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1025 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
1026
1027 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1028 VCN, inst_idx, regUVD_MPC_SET_MUX),
1029 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1030 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1031 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
1032
1033 vcn_v4_0_mc_resume_dpg_mode(adev, inst_idx, indirect);
1034
1035 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
1036 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
1037 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1038 VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
1039
1040 /* enable LMI MC and UMC channels */
1041 tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
1042 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1043 VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
1044
1045 vcn_v4_0_enable_ras(adev, inst_idx, indirect);
1046
1047 /* enable master interrupt */
1048 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
1049 VCN, inst_idx, regUVD_MASTINT_EN),
1050 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
1051
1052
1053 if (indirect)
1054 amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
1055
1056 ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1057
1058 WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
1059 WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1060 WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
1061
1062 tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1063 tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1064 WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1065 fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1066 WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
1067 WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
1068
1069 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
1070 WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
1071 ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1072
1073 tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
1074 tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1075 WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
1076 fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1077
1078 WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
1079 ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1080 VCN_RB1_DB_CTRL__EN_MASK);
1081
1082 return 0;
1083}
1084
1085
1086/**
1087 * vcn_v4_0_start - VCN start
1088 *
1089 * @adev: amdgpu_device pointer
1090 *
1091 * Start VCN block
1092 */
1093static int vcn_v4_0_start(struct amdgpu_device *adev)
1094{
1095 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1096 struct amdgpu_ring *ring;
1097 uint32_t tmp;
1098 int i, j, k, r;
1099
1100 if (adev->pm.dpm_enabled)
1101 amdgpu_dpm_enable_uvd(adev, true);
1102
1103 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1104 if (adev->vcn.harvest_config & (1 << i))
1105 continue;
1106
1107 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1108
1109 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1110 r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1111 continue;
1112 }
1113
1114 /* disable VCN power gating */
1115 vcn_v4_0_disable_static_power_gating(adev, i);
1116
1117 /* set VCN status busy */
1118 tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1119 WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1120
1121 /*SW clock gating */
1122 vcn_v4_0_disable_clock_gating(adev, i);
1123
1124 /* enable VCPU clock */
1125 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1126 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1127
1128 /* disable master interrupt */
1129 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1130 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1131
1132 /* enable LMI MC and UMC channels */
1133 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1134 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1135
1136 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1137 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1138 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1139 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1140
1141 /* setup regUVD_LMI_CTRL */
1142 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1143 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1144 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1145 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1146 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1147 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1148
1149 /* setup regUVD_MPC_CNTL */
1150 tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1151 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1152 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1153 WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1154
1155 /* setup UVD_MPC_SET_MUXA0 */
1156 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1157 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1158 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1159 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1160 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1161
1162 /* setup UVD_MPC_SET_MUXB0 */
1163 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1164 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1165 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1166 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1167 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1168
1169 /* setup UVD_MPC_SET_MUX */
1170 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1171 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1172 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1173 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1174
1175 vcn_v4_0_mc_resume(adev, i);
1176
1177 /* VCN global tiling registers */
1178 WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1179 adev->gfx.config.gb_addr_config);
1180
1181 /* unblock VCPU register access */
1182 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1183 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1184
1185 /* release VCPU reset to boot */
1186 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1187 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1188
1189 for (j = 0; j < 10; ++j) {
1190 uint32_t status;
1191
1192 for (k = 0; k < 100; ++k) {
1193 status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1194 if (status & 2)
1195 break;
1196 mdelay(10);
1197 if (amdgpu_emu_mode == 1)
1198 msleep(1);
1199 }
1200
1201 if (amdgpu_emu_mode == 1) {
1202 r = -1;
1203 if (status & 2) {
1204 r = 0;
1205 break;
1206 }
1207 } else {
1208 r = 0;
1209 if (status & 2)
1210 break;
1211
1212 dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1213 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1214 UVD_VCPU_CNTL__BLK_RST_MASK,
1215 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1216 mdelay(10);
1217 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1218 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1219
1220 mdelay(10);
1221 r = -1;
1222 }
1223 }
1224
1225 if (r) {
1226 dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1227 return r;
1228 }
1229
1230 /* enable master interrupt */
1231 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1232 UVD_MASTINT_EN__VCPU_EN_MASK,
1233 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1234
1235 /* clear the busy bit of VCN_STATUS */
1236 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1237 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1238
1239 ring = &adev->vcn.inst[i].ring_enc[0];
1240 WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1241 ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1242 VCN_RB1_DB_CTRL__EN_MASK);
1243
1244 WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1245 WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1246 WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1247
1248 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1249 tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1250 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1251 fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1252 WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1253 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1254
1255 tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1256 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1257 ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1258
1259 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1260 tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1261 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1262 fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1263 }
1264
1265 return 0;
1266}
1267
1268static int vcn_v4_0_init_ring_metadata(struct amdgpu_device *adev, uint32_t vcn_inst, struct amdgpu_ring *ring_enc)
1269{
1270 struct amdgpu_vcn_rb_metadata *rb_metadata = NULL;
1271 uint8_t *rb_ptr = (uint8_t *)ring_enc->ring;
1272
1273 rb_ptr += ring_enc->ring_size;
1274 rb_metadata = (struct amdgpu_vcn_rb_metadata *)rb_ptr;
1275
1276 memset(rb_metadata, 0, sizeof(struct amdgpu_vcn_rb_metadata));
1277 rb_metadata->size = sizeof(struct amdgpu_vcn_rb_metadata);
1278 rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1279 rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1280 rb_metadata->version = 1;
1281 rb_metadata->ring_id = vcn_inst & 0xFF;
1282
1283 return 0;
1284}
1285
1286static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1287{
1288 int i;
1289 struct amdgpu_ring *ring_enc;
1290 uint64_t cache_addr;
1291 uint64_t rb_enc_addr;
1292 uint64_t ctx_addr;
1293 uint32_t param, resp, expected;
1294 uint32_t offset, cache_size;
1295 uint32_t tmp, timeout;
1296
1297 struct amdgpu_mm_table *table = &adev->virt.mm_table;
1298 uint32_t *table_loc;
1299 uint32_t table_size;
1300 uint32_t size, size_dw;
1301 uint32_t init_status;
1302 uint32_t enabled_vcn;
1303
1304 struct mmsch_v4_0_cmd_direct_write
1305 direct_wt = { {0} };
1306 struct mmsch_v4_0_cmd_direct_read_modify_write
1307 direct_rd_mod_wt = { {0} };
1308 struct mmsch_v4_0_cmd_end end = { {0} };
1309 struct mmsch_v4_0_init_header header;
1310
1311 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1312 volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1313
1314 direct_wt.cmd_header.command_type =
1315 MMSCH_COMMAND__DIRECT_REG_WRITE;
1316 direct_rd_mod_wt.cmd_header.command_type =
1317 MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1318 end.cmd_header.command_type =
1319 MMSCH_COMMAND__END;
1320
1321 header.version = MMSCH_VERSION;
1322 header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1323 for (i = 0; i < MMSCH_V4_0_VCN_INSTANCES; i++) {
1324 header.inst[i].init_status = 0;
1325 header.inst[i].table_offset = 0;
1326 header.inst[i].table_size = 0;
1327 }
1328
1329 table_loc = (uint32_t *)table->cpu_addr;
1330 table_loc += header.total_size;
1331 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1332 if (adev->vcn.harvest_config & (1 << i))
1333 continue;
1334
1335 // Must re/init fw_shared at beginning
1336 vcn_v4_0_fw_shared_init(adev, i);
1337
1338 table_size = 0;
1339
1340 MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1341 regUVD_STATUS),
1342 ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1343
1344 cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4);
1345
1346 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1347 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1348 regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1349 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1350 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1351 regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1352 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1353 offset = 0;
1354 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1355 regUVD_VCPU_CACHE_OFFSET0),
1356 0);
1357 } else {
1358 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1359 regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1360 lower_32_bits(adev->vcn.inst[i].gpu_addr));
1361 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1362 regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1363 upper_32_bits(adev->vcn.inst[i].gpu_addr));
1364 offset = cache_size;
1365 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1366 regUVD_VCPU_CACHE_OFFSET0),
1367 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1368 }
1369
1370 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1371 regUVD_VCPU_CACHE_SIZE0),
1372 cache_size);
1373
1374 cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1375 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1376 regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1377 lower_32_bits(cache_addr));
1378 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1379 regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1380 upper_32_bits(cache_addr));
1381 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1382 regUVD_VCPU_CACHE_OFFSET1),
1383 0);
1384 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1385 regUVD_VCPU_CACHE_SIZE1),
1386 AMDGPU_VCN_STACK_SIZE);
1387
1388 cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1389 AMDGPU_VCN_STACK_SIZE;
1390 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1391 regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1392 lower_32_bits(cache_addr));
1393 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1394 regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1395 upper_32_bits(cache_addr));
1396 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1397 regUVD_VCPU_CACHE_OFFSET2),
1398 0);
1399 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1400 regUVD_VCPU_CACHE_SIZE2),
1401 AMDGPU_VCN_CONTEXT_SIZE);
1402
1403 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1404 rb_setup = &fw_shared->rb_setup;
1405
1406 ring_enc = &adev->vcn.inst[i].ring_enc[0];
1407 ring_enc->wptr = 0;
1408 rb_enc_addr = ring_enc->gpu_addr;
1409
1410 rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1411 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1412
1413 if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
1414 vcn_v4_0_init_ring_metadata(adev, i, ring_enc);
1415
1416 memset((void *)&rb_setup->rb_info, 0, sizeof(struct amdgpu_vcn_rb_setup_info) * MAX_NUM_VCN_RB_SETUP);
1417 if (!(adev->vcn.harvest_config & (1 << 0))) {
1418 rb_setup->rb_info[0].rb_addr_lo = lower_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1419 rb_setup->rb_info[0].rb_addr_hi = upper_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1420 rb_setup->rb_info[0].rb_size = adev->vcn.inst[0].ring_enc[0].ring_size / 4;
1421 }
1422 if (!(adev->vcn.harvest_config & (1 << 1))) {
1423 rb_setup->rb_info[2].rb_addr_lo = lower_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1424 rb_setup->rb_info[2].rb_addr_hi = upper_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1425 rb_setup->rb_info[2].rb_size = adev->vcn.inst[1].ring_enc[0].ring_size / 4;
1426 }
1427 fw_shared->decouple.is_enabled = 1;
1428 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1429 } else {
1430 rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1431 rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1432 rb_setup->rb_size = ring_enc->ring_size / 4;
1433 }
1434
1435 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1436 regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1437 lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1438 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1439 regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1440 upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1441 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1442 regUVD_VCPU_NONCACHE_SIZE0),
1443 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1444
1445 /* add end packet */
1446 MMSCH_V4_0_INSERT_END();
1447
1448 /* refine header */
1449 header.inst[i].init_status = 0;
1450 header.inst[i].table_offset = header.total_size;
1451 header.inst[i].table_size = table_size;
1452 header.total_size += table_size;
1453 }
1454
1455 /* Update init table header in memory */
1456 size = sizeof(struct mmsch_v4_0_init_header);
1457 table_loc = (uint32_t *)table->cpu_addr;
1458 memcpy((void *)table_loc, &header, size);
1459
1460 /* message MMSCH (in VCN[0]) to initialize this client
1461 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1462 * of memory descriptor location
1463 */
1464 ctx_addr = table->gpu_addr;
1465 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1466 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1467
1468 /* 2, update vmid of descriptor */
1469 tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1470 tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1471 /* use domain0 for MM scheduler */
1472 tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1473 WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1474
1475 /* 3, notify mmsch about the size of this descriptor */
1476 size = header.total_size;
1477 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1478
1479 /* 4, set resp to zero */
1480 WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1481
1482 /* 5, kick off the initialization and wait until
1483 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1484 */
1485 param = 0x00000001;
1486 WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1487 tmp = 0;
1488 timeout = 1000;
1489 resp = 0;
1490 expected = MMSCH_VF_MAILBOX_RESP__OK;
1491 while (resp != expected) {
1492 resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1493 if (resp != 0)
1494 break;
1495
1496 udelay(10);
1497 tmp = tmp + 10;
1498 if (tmp >= timeout) {
1499 DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1500 " waiting for regMMSCH_VF_MAILBOX_RESP "\
1501 "(expected=0x%08x, readback=0x%08x)\n",
1502 tmp, expected, resp);
1503 return -EBUSY;
1504 }
1505 }
1506 enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1507 init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1508 if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1509 && init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1510 DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1511 "status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1512
1513 return 0;
1514}
1515
1516/**
1517 * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1518 *
1519 * @adev: amdgpu_device pointer
1520 * @inst_idx: instance number index
1521 *
1522 * Stop VCN block with dpg mode
1523 */
1524static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1525{
1526 struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1527 uint32_t tmp;
1528
1529 vcn_v4_0_pause_dpg_mode(adev, inst_idx, &state);
1530 /* Wait for power status to be 1 */
1531 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1532 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1533
1534 /* wait for read ptr to be equal to write ptr */
1535 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1536 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1537
1538 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1539 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1540
1541 /* disable dynamic power gating mode */
1542 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1543 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1544}
1545
1546/**
1547 * vcn_v4_0_stop - VCN stop
1548 *
1549 * @adev: amdgpu_device pointer
1550 *
1551 * Stop VCN block
1552 */
1553static int vcn_v4_0_stop(struct amdgpu_device *adev)
1554{
1555 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1556 uint32_t tmp;
1557 int i, r = 0;
1558
1559 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1560 if (adev->vcn.harvest_config & (1 << i))
1561 continue;
1562
1563 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1564 fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1565
1566 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1567 vcn_v4_0_stop_dpg_mode(adev, i);
1568 continue;
1569 }
1570
1571 /* wait for vcn idle */
1572 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1573 if (r)
1574 return r;
1575
1576 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1577 UVD_LMI_STATUS__READ_CLEAN_MASK |
1578 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1579 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1580 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1581 if (r)
1582 return r;
1583
1584 /* disable LMI UMC channel */
1585 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1586 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1587 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1588 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1589 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1590 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1591 if (r)
1592 return r;
1593
1594 /* block VCPU register access */
1595 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1596 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1597 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1598
1599 /* reset VCPU */
1600 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1601 UVD_VCPU_CNTL__BLK_RST_MASK,
1602 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1603
1604 /* disable VCPU clock */
1605 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1606 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1607
1608 /* apply soft reset */
1609 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1610 tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1611 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1612 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1613 tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1614 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1615
1616 /* clear status */
1617 WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1618
1619 /* apply HW clock gating */
1620 vcn_v4_0_enable_clock_gating(adev, i);
1621
1622 /* enable VCN power gating */
1623 vcn_v4_0_enable_static_power_gating(adev, i);
1624 }
1625
1626 if (adev->pm.dpm_enabled)
1627 amdgpu_dpm_enable_uvd(adev, false);
1628
1629 return 0;
1630}
1631
1632/**
1633 * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1634 *
1635 * @adev: amdgpu_device pointer
1636 * @inst_idx: instance number index
1637 * @new_state: pause state
1638 *
1639 * Pause dpg mode for VCN block
1640 */
1641static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1642 struct dpg_pause_state *new_state)
1643{
1644 uint32_t reg_data = 0;
1645 int ret_code;
1646
1647 /* pause/unpause if state is changed */
1648 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1649 DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1650 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based);
1651 reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1652 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1653
1654 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1655 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1656 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1657
1658 if (!ret_code) {
1659 /* pause DPG */
1660 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1661 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1662
1663 /* wait for ACK */
1664 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1665 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1666 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1667
1668 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1669 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1670 }
1671 } else {
1672 /* unpause dpg, no need to wait */
1673 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1674 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1675 }
1676 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1677 }
1678
1679 return 0;
1680}
1681
1682/**
1683 * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1684 *
1685 * @ring: amdgpu_ring pointer
1686 *
1687 * Returns the current hardware unified read pointer
1688 */
1689static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1690{
1691 struct amdgpu_device *adev = ring->adev;
1692
1693 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1694 DRM_ERROR("wrong ring id is identified in %s", __func__);
1695
1696 return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1697}
1698
1699/**
1700 * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1701 *
1702 * @ring: amdgpu_ring pointer
1703 *
1704 * Returns the current hardware unified write pointer
1705 */
1706static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1707{
1708 struct amdgpu_device *adev = ring->adev;
1709
1710 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1711 DRM_ERROR("wrong ring id is identified in %s", __func__);
1712
1713 if (ring->use_doorbell)
1714 return *ring->wptr_cpu_addr;
1715 else
1716 return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1717}
1718
1719/**
1720 * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1721 *
1722 * @ring: amdgpu_ring pointer
1723 *
1724 * Commits the enc write pointer to the hardware
1725 */
1726static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1727{
1728 struct amdgpu_device *adev = ring->adev;
1729
1730 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1731 DRM_ERROR("wrong ring id is identified in %s", __func__);
1732
1733 if (ring->use_doorbell) {
1734 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1735 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1736 } else {
1737 WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1738 }
1739}
1740
1741static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1742 struct amdgpu_job *job)
1743{
1744 struct drm_gpu_scheduler **scheds;
1745
1746 /* The create msg must be in the first IB submitted */
1747 if (atomic_read(&job->base.entity->fence_seq))
1748 return -EINVAL;
1749
1750 /* if VCN0 is harvested, we can't support AV1 */
1751 if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1752 return -EINVAL;
1753
1754 scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1755 [AMDGPU_RING_PRIO_0].sched;
1756 drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1757 return 0;
1758}
1759
1760static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1761 uint64_t addr)
1762{
1763 struct ttm_operation_ctx ctx = { false, false };
1764 struct amdgpu_bo_va_mapping *map;
1765 uint32_t *msg, num_buffers;
1766 struct amdgpu_bo *bo;
1767 uint64_t start, end;
1768 unsigned int i;
1769 void *ptr;
1770 int r;
1771
1772 addr &= AMDGPU_GMC_HOLE_MASK;
1773 r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1774 if (r) {
1775 DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1776 return r;
1777 }
1778
1779 start = map->start * AMDGPU_GPU_PAGE_SIZE;
1780 end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1781 if (addr & 0x7) {
1782 DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1783 return -EINVAL;
1784 }
1785
1786 bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1787 amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1788 r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1789 if (r) {
1790 DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1791 return r;
1792 }
1793
1794 r = amdgpu_bo_kmap(bo, &ptr);
1795 if (r) {
1796 DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1797 return r;
1798 }
1799
1800 msg = ptr + addr - start;
1801
1802 /* Check length */
1803 if (msg[1] > end - addr) {
1804 r = -EINVAL;
1805 goto out;
1806 }
1807
1808 if (msg[3] != RDECODE_MSG_CREATE)
1809 goto out;
1810
1811 num_buffers = msg[2];
1812 for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1813 uint32_t offset, size, *create;
1814
1815 if (msg[0] != RDECODE_MESSAGE_CREATE)
1816 continue;
1817
1818 offset = msg[1];
1819 size = msg[2];
1820
1821 if (offset + size > end) {
1822 r = -EINVAL;
1823 goto out;
1824 }
1825
1826 create = ptr + addr + offset - start;
1827
1828 /* H264, HEVC and VP9 can run on any instance */
1829 if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1830 continue;
1831
1832 r = vcn_v4_0_limit_sched(p, job);
1833 if (r)
1834 goto out;
1835 }
1836
1837out:
1838 amdgpu_bo_kunmap(bo);
1839 return r;
1840}
1841
1842#define RADEON_VCN_ENGINE_TYPE_ENCODE (0x00000002)
1843#define RADEON_VCN_ENGINE_TYPE_DECODE (0x00000003)
1844
1845#define RADEON_VCN_ENGINE_INFO (0x30000001)
1846#define RADEON_VCN_ENGINE_INFO_MAX_OFFSET 16
1847
1848#define RENCODE_ENCODE_STANDARD_AV1 2
1849#define RENCODE_IB_PARAM_SESSION_INIT 0x00000003
1850#define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET 64
1851
1852/* return the offset in ib if id is found, -1 otherwise
1853 * to speed up the searching we only search upto max_offset
1854 */
1855static int vcn_v4_0_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1856{
1857 int i;
1858
1859 for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1860 if (ib->ptr[i + 1] == id)
1861 return i;
1862 }
1863 return -1;
1864}
1865
1866static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1867 struct amdgpu_job *job,
1868 struct amdgpu_ib *ib)
1869{
1870 struct amdgpu_ring *ring = amdgpu_job_ring(job);
1871 struct amdgpu_vcn_decode_buffer *decode_buffer;
1872 uint64_t addr;
1873 uint32_t val;
1874 int idx;
1875
1876 /* The first instance can decode anything */
1877 if (!ring->me)
1878 return 0;
1879
1880 /* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1881 idx = vcn_v4_0_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1882 RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1883 if (idx < 0) /* engine info is missing */
1884 return 0;
1885
1886 val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1887 if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1888 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1889
1890 if (!(decode_buffer->valid_buf_flag & 0x1))
1891 return 0;
1892
1893 addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1894 decode_buffer->msg_buffer_address_lo;
1895 return vcn_v4_0_dec_msg(p, job, addr);
1896 } else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1897 idx = vcn_v4_0_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1898 RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1899 if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1900 return vcn_v4_0_limit_sched(p, job);
1901 }
1902 return 0;
1903}
1904
1905static struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1906 .type = AMDGPU_RING_TYPE_VCN_ENC,
1907 .align_mask = 0x3f,
1908 .nop = VCN_ENC_CMD_NO_OP,
1909 .extra_dw = sizeof(struct amdgpu_vcn_rb_metadata),
1910 .get_rptr = vcn_v4_0_unified_ring_get_rptr,
1911 .get_wptr = vcn_v4_0_unified_ring_get_wptr,
1912 .set_wptr = vcn_v4_0_unified_ring_set_wptr,
1913 .patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1914 .emit_frame_size =
1915 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1916 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1917 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1918 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1919 1, /* vcn_v2_0_enc_ring_insert_end */
1920 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1921 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1922 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1923 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1924 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1925 .test_ib = amdgpu_vcn_unified_ring_test_ib,
1926 .insert_nop = amdgpu_ring_insert_nop,
1927 .insert_end = vcn_v2_0_enc_ring_insert_end,
1928 .pad_ib = amdgpu_ring_generic_pad_ib,
1929 .begin_use = amdgpu_vcn_ring_begin_use,
1930 .end_use = amdgpu_vcn_ring_end_use,
1931 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1932 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1933 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1934};
1935
1936/**
1937 * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1938 *
1939 * @adev: amdgpu_device pointer
1940 *
1941 * Set unified ring functions
1942 */
1943static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1944{
1945 int i;
1946
1947 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1948 if (adev->vcn.harvest_config & (1 << i))
1949 continue;
1950
1951 if (amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 2))
1952 vcn_v4_0_unified_ring_vm_funcs.secure_submission_supported = true;
1953
1954 adev->vcn.inst[i].ring_enc[0].funcs =
1955 (const struct amdgpu_ring_funcs *)&vcn_v4_0_unified_ring_vm_funcs;
1956 adev->vcn.inst[i].ring_enc[0].me = i;
1957 }
1958}
1959
1960/**
1961 * vcn_v4_0_is_idle - check VCN block is idle
1962 *
1963 * @handle: amdgpu_device pointer
1964 *
1965 * Check whether VCN block is idle
1966 */
1967static bool vcn_v4_0_is_idle(void *handle)
1968{
1969 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1970 int i, ret = 1;
1971
1972 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1973 if (adev->vcn.harvest_config & (1 << i))
1974 continue;
1975
1976 ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1977 }
1978
1979 return ret;
1980}
1981
1982/**
1983 * vcn_v4_0_wait_for_idle - wait for VCN block idle
1984 *
1985 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
1986 *
1987 * Wait for VCN block idle
1988 */
1989static int vcn_v4_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
1990{
1991 struct amdgpu_device *adev = ip_block->adev;
1992 int i, ret = 0;
1993
1994 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1995 if (adev->vcn.harvest_config & (1 << i))
1996 continue;
1997
1998 ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1999 UVD_STATUS__IDLE);
2000 if (ret)
2001 return ret;
2002 }
2003
2004 return ret;
2005}
2006
2007/**
2008 * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
2009 *
2010 * @handle: amdgpu_device pointer
2011 * @state: clock gating state
2012 *
2013 * Set VCN block clockgating state
2014 */
2015static int vcn_v4_0_set_clockgating_state(void *handle, enum amd_clockgating_state state)
2016{
2017 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2018 bool enable = state == AMD_CG_STATE_GATE;
2019 int i;
2020
2021 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2022 if (adev->vcn.harvest_config & (1 << i))
2023 continue;
2024
2025 if (enable) {
2026 if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
2027 return -EBUSY;
2028 vcn_v4_0_enable_clock_gating(adev, i);
2029 } else {
2030 vcn_v4_0_disable_clock_gating(adev, i);
2031 }
2032 }
2033
2034 return 0;
2035}
2036
2037/**
2038 * vcn_v4_0_set_powergating_state - set VCN block powergating state
2039 *
2040 * @handle: amdgpu_device pointer
2041 * @state: power gating state
2042 *
2043 * Set VCN block powergating state
2044 */
2045static int vcn_v4_0_set_powergating_state(void *handle, enum amd_powergating_state state)
2046{
2047 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2048 int ret;
2049
2050 /* for SRIOV, guest should not control VCN Power-gating
2051 * MMSCH FW should control Power-gating and clock-gating
2052 * guest should avoid touching CGC and PG
2053 */
2054 if (amdgpu_sriov_vf(adev)) {
2055 adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2056 return 0;
2057 }
2058
2059 if (state == adev->vcn.cur_state)
2060 return 0;
2061
2062 if (state == AMD_PG_STATE_GATE)
2063 ret = vcn_v4_0_stop(adev);
2064 else
2065 ret = vcn_v4_0_start(adev);
2066
2067 if (!ret)
2068 adev->vcn.cur_state = state;
2069
2070 return ret;
2071}
2072
2073/**
2074 * vcn_v4_0_set_ras_interrupt_state - set VCN block RAS interrupt state
2075 *
2076 * @adev: amdgpu_device pointer
2077 * @source: interrupt sources
2078 * @type: interrupt types
2079 * @state: interrupt states
2080 *
2081 * Set VCN block RAS interrupt state
2082 */
2083static int vcn_v4_0_set_ras_interrupt_state(struct amdgpu_device *adev,
2084 struct amdgpu_irq_src *source,
2085 unsigned int type,
2086 enum amdgpu_interrupt_state state)
2087{
2088 return 0;
2089}
2090
2091/**
2092 * vcn_v4_0_process_interrupt - process VCN block interrupt
2093 *
2094 * @adev: amdgpu_device pointer
2095 * @source: interrupt sources
2096 * @entry: interrupt entry from clients and sources
2097 *
2098 * Process VCN block interrupt
2099 */
2100static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
2101 struct amdgpu_iv_entry *entry)
2102{
2103 uint32_t ip_instance;
2104
2105 if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
2106 ip_instance = entry->ring_id;
2107 } else {
2108 switch (entry->client_id) {
2109 case SOC15_IH_CLIENTID_VCN:
2110 ip_instance = 0;
2111 break;
2112 case SOC15_IH_CLIENTID_VCN1:
2113 ip_instance = 1;
2114 break;
2115 default:
2116 DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2117 return 0;
2118 }
2119 }
2120
2121 DRM_DEBUG("IH: VCN TRAP\n");
2122
2123 switch (entry->src_id) {
2124 case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2125 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2126 break;
2127 default:
2128 DRM_ERROR("Unhandled interrupt: %d %d\n",
2129 entry->src_id, entry->src_data[0]);
2130 break;
2131 }
2132
2133 return 0;
2134}
2135
2136static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
2137 .process = vcn_v4_0_process_interrupt,
2138};
2139
2140static const struct amdgpu_irq_src_funcs vcn_v4_0_ras_irq_funcs = {
2141 .set = vcn_v4_0_set_ras_interrupt_state,
2142 .process = amdgpu_vcn_process_poison_irq,
2143};
2144
2145/**
2146 * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
2147 *
2148 * @adev: amdgpu_device pointer
2149 *
2150 * Set VCN block interrupt irq functions
2151 */
2152static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2153{
2154 int i;
2155
2156 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2157 if (adev->vcn.harvest_config & (1 << i))
2158 continue;
2159
2160 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2161 adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2162
2163 adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
2164 adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v4_0_ras_irq_funcs;
2165 }
2166}
2167
2168static void vcn_v4_0_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p)
2169{
2170 struct amdgpu_device *adev = ip_block->adev;
2171 int i, j;
2172 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0);
2173 uint32_t inst_off, is_powered;
2174
2175 if (!adev->vcn.ip_dump)
2176 return;
2177
2178 drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst);
2179 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
2180 if (adev->vcn.harvest_config & (1 << i)) {
2181 drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i);
2182 continue;
2183 }
2184
2185 inst_off = i * reg_count;
2186 is_powered = (adev->vcn.ip_dump[inst_off] &
2187 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
2188
2189 if (is_powered) {
2190 drm_printf(p, "\nActive Instance:VCN%d\n", i);
2191 for (j = 0; j < reg_count; j++)
2192 drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_4_0[j].reg_name,
2193 adev->vcn.ip_dump[inst_off + j]);
2194 } else {
2195 drm_printf(p, "\nInactive Instance:VCN%d\n", i);
2196 }
2197 }
2198}
2199
2200static void vcn_v4_0_dump_ip_state(struct amdgpu_ip_block *ip_block)
2201{
2202 struct amdgpu_device *adev = ip_block->adev;
2203 int i, j;
2204 bool is_powered;
2205 uint32_t inst_off;
2206 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0);
2207
2208 if (!adev->vcn.ip_dump)
2209 return;
2210
2211 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
2212 if (adev->vcn.harvest_config & (1 << i))
2213 continue;
2214
2215 inst_off = i * reg_count;
2216 /* mmUVD_POWER_STATUS is always readable and is first element of the array */
2217 adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, i, regUVD_POWER_STATUS);
2218 is_powered = (adev->vcn.ip_dump[inst_off] &
2219 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
2220
2221 if (is_powered)
2222 for (j = 1; j < reg_count; j++)
2223 adev->vcn.ip_dump[inst_off + j] =
2224 RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_4_0[j],
2225 i));
2226 }
2227}
2228
2229static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2230 .name = "vcn_v4_0",
2231 .early_init = vcn_v4_0_early_init,
2232 .sw_init = vcn_v4_0_sw_init,
2233 .sw_fini = vcn_v4_0_sw_fini,
2234 .hw_init = vcn_v4_0_hw_init,
2235 .hw_fini = vcn_v4_0_hw_fini,
2236 .suspend = vcn_v4_0_suspend,
2237 .resume = vcn_v4_0_resume,
2238 .is_idle = vcn_v4_0_is_idle,
2239 .wait_for_idle = vcn_v4_0_wait_for_idle,
2240 .set_clockgating_state = vcn_v4_0_set_clockgating_state,
2241 .set_powergating_state = vcn_v4_0_set_powergating_state,
2242 .dump_ip_state = vcn_v4_0_dump_ip_state,
2243 .print_ip_state = vcn_v4_0_print_ip_state,
2244};
2245
2246const struct amdgpu_ip_block_version vcn_v4_0_ip_block = {
2247 .type = AMD_IP_BLOCK_TYPE_VCN,
2248 .major = 4,
2249 .minor = 0,
2250 .rev = 0,
2251 .funcs = &vcn_v4_0_ip_funcs,
2252};
2253
2254static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2255 uint32_t instance, uint32_t sub_block)
2256{
2257 uint32_t poison_stat = 0, reg_value = 0;
2258
2259 switch (sub_block) {
2260 case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2261 reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2262 poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2263 break;
2264 default:
2265 break;
2266 }
2267
2268 if (poison_stat)
2269 dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2270 instance, sub_block);
2271
2272 return poison_stat;
2273}
2274
2275static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2276{
2277 uint32_t inst, sub;
2278 uint32_t poison_stat = 0;
2279
2280 for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2281 for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2282 poison_stat +=
2283 vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2284
2285 return !!poison_stat;
2286}
2287
2288const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2289 .query_poison_status = vcn_v4_0_query_ras_poison_status,
2290};
2291
2292static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2293 .ras_block = {
2294 .hw_ops = &vcn_v4_0_ras_hw_ops,
2295 .ras_late_init = amdgpu_vcn_ras_late_init,
2296 },
2297};
2298
2299static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2300{
2301 switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
2302 case IP_VERSION(4, 0, 0):
2303 adev->vcn.ras = &vcn_v4_0_ras;
2304 break;
2305 default:
2306 break;
2307 }
2308}