1 /*
2 * Copyright 2023 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_5.h"
35
36 #include "vcn/vcn_4_0_5_offset.h"
37 #include "vcn/vcn_4_0_5_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 + 0x38000)
49
50 #define VCN_HARVEST_MMSCH 0
51
52 #define RDECODE_MSG_CREATE 0x00000000
53 #define RDECODE_MESSAGE_CREATE 0x00000001
54
55 static const struct amdgpu_hwip_reg_entry vcn_reg_list_4_0_5[] = {
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
91 static int amdgpu_ih_clientid_vcns[] = {
92 SOC15_IH_CLIENTID_VCN,
93 SOC15_IH_CLIENTID_VCN1
94 };
95
96 static void vcn_v4_0_5_set_unified_ring_funcs(struct amdgpu_device *adev);
97 static void vcn_v4_0_5_set_irq_funcs(struct amdgpu_device *adev);
98 static int vcn_v4_0_5_set_powergating_state(struct amdgpu_ip_block *ip_block,
99 enum amd_powergating_state state);
100 static int vcn_v4_0_5_pause_dpg_mode(struct amdgpu_device *adev,
101 int inst_idx, struct dpg_pause_state *new_state);
102 static void vcn_v4_0_5_unified_ring_set_wptr(struct amdgpu_ring *ring);
103
104 /**
105 * vcn_v4_0_5_early_init - set function pointers and load microcode
106 *
107 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
108 *
109 * Set ring and irq function pointers
110 * Load microcode from filesystem
111 */
vcn_v4_0_5_early_init(struct amdgpu_ip_block * ip_block)112 static int vcn_v4_0_5_early_init(struct amdgpu_ip_block *ip_block)
113 {
114 struct amdgpu_device *adev = ip_block->adev;
115
116 /* re-use enc ring as unified ring */
117 adev->vcn.num_enc_rings = 1;
118 vcn_v4_0_5_set_unified_ring_funcs(adev);
119 vcn_v4_0_5_set_irq_funcs(adev);
120
121 return amdgpu_vcn_early_init(adev);
122 }
123
124 /**
125 * vcn_v4_0_5_sw_init - sw init for VCN block
126 *
127 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
128 *
129 * Load firmware and sw initialization
130 */
vcn_v4_0_5_sw_init(struct amdgpu_ip_block * ip_block)131 static int vcn_v4_0_5_sw_init(struct amdgpu_ip_block *ip_block)
132 {
133 struct amdgpu_ring *ring;
134 struct amdgpu_device *adev = ip_block->adev;
135 int i, r;
136 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0_5);
137 uint32_t *ptr;
138
139 r = amdgpu_vcn_sw_init(adev);
140 if (r)
141 return r;
142
143 amdgpu_vcn_setup_ucode(adev);
144
145 r = amdgpu_vcn_resume(adev);
146 if (r)
147 return r;
148
149 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
150 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
151
152 if (adev->vcn.harvest_config & (1 << i))
153 continue;
154
155 atomic_set(&adev->vcn.inst[i].sched_score, 0);
156
157 /* VCN UNIFIED TRAP */
158 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
159 VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
160 if (r)
161 return r;
162
163 /* VCN POISON TRAP */
164 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
165 VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].irq);
166 if (r)
167 return r;
168
169 ring = &adev->vcn.inst[i].ring_enc[0];
170 ring->use_doorbell = true;
171 if (amdgpu_sriov_vf(adev))
172 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
173 i * (adev->vcn.num_enc_rings + 1) + 1;
174 else
175 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
176 2 + 8 * i;
177 ring->vm_hub = AMDGPU_MMHUB0(0);
178 sprintf(ring->name, "vcn_unified_%d", i);
179
180 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
181 AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
182 if (r)
183 return r;
184
185 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
186 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
187 fw_shared->sq.is_enabled = 1;
188
189 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
190 fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
191 AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
192
193 if (amdgpu_sriov_vf(adev))
194 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
195
196 if (amdgpu_vcnfw_log)
197 amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
198 }
199
200 if (amdgpu_sriov_vf(adev)) {
201 r = amdgpu_virt_alloc_mm_table(adev);
202 if (r)
203 return r;
204 }
205
206 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
207 adev->vcn.pause_dpg_mode = vcn_v4_0_5_pause_dpg_mode;
208
209 /* Allocate memory for VCN IP Dump buffer */
210 ptr = kcalloc(adev->vcn.num_vcn_inst * reg_count, sizeof(uint32_t), GFP_KERNEL);
211 if (!ptr) {
212 DRM_ERROR("Failed to allocate memory for VCN IP Dump\n");
213 adev->vcn.ip_dump = NULL;
214 } else {
215 adev->vcn.ip_dump = ptr;
216 }
217 return 0;
218 }
219
220 /**
221 * vcn_v4_0_5_sw_fini - sw fini for VCN block
222 *
223 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
224 *
225 * VCN suspend and free up sw allocation
226 */
vcn_v4_0_5_sw_fini(struct amdgpu_ip_block * ip_block)227 static int vcn_v4_0_5_sw_fini(struct amdgpu_ip_block *ip_block)
228 {
229 struct amdgpu_device *adev = ip_block->adev;
230 int i, r, idx;
231
232 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
233 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
234 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
235
236 if (adev->vcn.harvest_config & (1 << i))
237 continue;
238
239 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
240 fw_shared->present_flag_0 = 0;
241 fw_shared->sq.is_enabled = 0;
242 }
243
244 drm_dev_exit(idx);
245 }
246
247 if (amdgpu_sriov_vf(adev))
248 amdgpu_virt_free_mm_table(adev);
249
250 r = amdgpu_vcn_suspend(adev);
251 if (r)
252 return r;
253
254 r = amdgpu_vcn_sw_fini(adev);
255
256 kfree(adev->vcn.ip_dump);
257
258 return r;
259 }
260
261 /**
262 * vcn_v4_0_5_hw_init - start and test VCN block
263 *
264 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
265 *
266 * Initialize the hardware, boot up the VCPU and do some testing
267 */
vcn_v4_0_5_hw_init(struct amdgpu_ip_block * ip_block)268 static int vcn_v4_0_5_hw_init(struct amdgpu_ip_block *ip_block)
269 {
270 struct amdgpu_device *adev = ip_block->adev;
271 struct amdgpu_ring *ring;
272 int i, r;
273
274 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
275 if (adev->vcn.harvest_config & (1 << i))
276 continue;
277
278 ring = &adev->vcn.inst[i].ring_enc[0];
279
280 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
281 ((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
282
283 r = amdgpu_ring_test_helper(ring);
284 if (r)
285 return r;
286 }
287
288 return 0;
289 }
290
291 /**
292 * vcn_v4_0_5_hw_fini - stop the hardware block
293 *
294 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
295 *
296 * Stop the VCN block, mark ring as not ready any more
297 */
vcn_v4_0_5_hw_fini(struct amdgpu_ip_block * ip_block)298 static int vcn_v4_0_5_hw_fini(struct amdgpu_ip_block *ip_block)
299 {
300 struct amdgpu_device *adev = ip_block->adev;
301 int i;
302
303 cancel_delayed_work_sync(&adev->vcn.idle_work);
304
305 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
306 if (adev->vcn.harvest_config & (1 << i))
307 continue;
308 if (!amdgpu_sriov_vf(adev)) {
309 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
310 (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
311 RREG32_SOC15(VCN, i, regUVD_STATUS))) {
312 vcn_v4_0_5_set_powergating_state(ip_block, AMD_PG_STATE_GATE);
313 }
314 }
315 }
316
317 return 0;
318 }
319
320 /**
321 * vcn_v4_0_5_suspend - suspend VCN block
322 *
323 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
324 *
325 * HW fini and suspend VCN block
326 */
vcn_v4_0_5_suspend(struct amdgpu_ip_block * ip_block)327 static int vcn_v4_0_5_suspend(struct amdgpu_ip_block *ip_block)
328 {
329 int r;
330
331 r = vcn_v4_0_5_hw_fini(ip_block);
332 if (r)
333 return r;
334
335 r = amdgpu_vcn_suspend(ip_block->adev);
336
337 return r;
338 }
339
340 /**
341 * vcn_v4_0_5_resume - resume VCN block
342 *
343 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
344 *
345 * Resume firmware and hw init VCN block
346 */
vcn_v4_0_5_resume(struct amdgpu_ip_block * ip_block)347 static int vcn_v4_0_5_resume(struct amdgpu_ip_block *ip_block)
348 {
349 int r;
350
351 r = amdgpu_vcn_resume(ip_block->adev);
352 if (r)
353 return r;
354
355 r = vcn_v4_0_5_hw_init(ip_block);
356
357 return r;
358 }
359
360 /**
361 * vcn_v4_0_5_mc_resume - memory controller programming
362 *
363 * @adev: amdgpu_device pointer
364 * @inst: instance number
365 *
366 * Let the VCN memory controller know it's offsets
367 */
vcn_v4_0_5_mc_resume(struct amdgpu_device * adev,int inst)368 static void vcn_v4_0_5_mc_resume(struct amdgpu_device *adev, int inst)
369 {
370 uint32_t offset, size;
371 const struct common_firmware_header *hdr;
372
373 hdr = (const struct common_firmware_header *)adev->vcn.inst[inst].fw->data;
374 size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
375
376 /* cache window 0: fw */
377 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
378 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
379 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
380 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
381 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
382 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
383 offset = 0;
384 } else {
385 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
386 lower_32_bits(adev->vcn.inst[inst].gpu_addr));
387 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
388 upper_32_bits(adev->vcn.inst[inst].gpu_addr));
389 offset = size;
390 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
391 }
392 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
393
394 /* cache window 1: stack */
395 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
396 lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
397 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
398 upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
399 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
400 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
401
402 /* cache window 2: context */
403 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
404 lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
405 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
406 upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
407 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
408 WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
409
410 /* non-cache window */
411 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
412 lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
413 WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
414 upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
415 WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
416 WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
417 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
418 }
419
420 /**
421 * vcn_v4_0_5_mc_resume_dpg_mode - memory controller programming for dpg mode
422 *
423 * @adev: amdgpu_device pointer
424 * @inst_idx: instance number index
425 * @indirect: indirectly write sram
426 *
427 * Let the VCN memory controller know it's offsets with dpg mode
428 */
vcn_v4_0_5_mc_resume_dpg_mode(struct amdgpu_device * adev,int inst_idx,bool indirect)429 static void vcn_v4_0_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
430 {
431 uint32_t offset, size;
432 const struct common_firmware_header *hdr;
433
434 hdr = (const struct common_firmware_header *)adev->vcn.inst[inst_idx].fw->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 if (!indirect) {
440 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
441 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
442 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo),
443 0, indirect);
444 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
445 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
446 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi),
447 0, indirect);
448 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
449 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
450 } else {
451 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
452 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
453 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
454 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
455 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
456 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
457 }
458 offset = 0;
459 } else {
460 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
461 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
462 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
463 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
464 VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
465 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
466 offset = size;
467 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
469 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
470 }
471
472 if (!indirect)
473 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
474 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
475 else
476 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
478
479 /* cache window 1: stack */
480 if (!indirect) {
481 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
482 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
483 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
484 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
485 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
486 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
487 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
488 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
489 } else {
490 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
491 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
492 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
493 VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
494 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
495 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
496 }
497
498 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
499 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
500
501 /* cache window 2: context */
502 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
503 VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
504 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE),
505 0, indirect);
506 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
508 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE),
509 0, indirect);
510 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
512 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
513 VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
514
515 /* non-cache window */
516 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
517 VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
518 lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
519 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
520 VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
521 upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
522 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
523 VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
524 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
525 VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
526 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
527
528 /* VCN global tiling registers */
529 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
530 VCN, inst_idx, regUVD_GFX10_ADDR_CONFIG),
531 adev->gfx.config.gb_addr_config, 0, indirect);
532 }
533
534 /**
535 * vcn_v4_0_5_disable_static_power_gating - disable VCN static power gating
536 *
537 * @adev: amdgpu_device pointer
538 * @inst: instance number
539 *
540 * Disable static power gating for VCN block
541 */
vcn_v4_0_5_disable_static_power_gating(struct amdgpu_device * adev,int inst)542 static void vcn_v4_0_5_disable_static_power_gating(struct amdgpu_device *adev, int inst)
543 {
544 uint32_t data = 0;
545
546 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
547 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
548 1 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT);
549 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, 0,
550 UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK);
551 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
552 2 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT);
553 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
554 1 << UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS__SHIFT,
555 UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK);
556 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
557 2 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT);
558 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
559 1 << UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS__SHIFT,
560 UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK);
561 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
562 2 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT);
563 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
564 1 << UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS__SHIFT,
565 UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK);
566 } else {
567 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
568 1 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT);
569 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
570 0, UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK);
571 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
572 1 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT);
573 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
574 0, UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK);
575 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
576 1 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT);
577 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
578 0, UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK);
579 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
580 1 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT);
581 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
582 0, UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK);
583 }
584
585 data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
586 data &= ~0x103;
587 if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
588 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
589 UVD_POWER_STATUS__UVD_PG_EN_MASK;
590 WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
591 }
592
593 /**
594 * vcn_v4_0_5_enable_static_power_gating - enable VCN static power gating
595 *
596 * @adev: amdgpu_device pointer
597 * @inst: instance number
598 *
599 * Enable static power gating for VCN block
600 */
vcn_v4_0_5_enable_static_power_gating(struct amdgpu_device * adev,int inst)601 static void vcn_v4_0_5_enable_static_power_gating(struct amdgpu_device *adev, int inst)
602 {
603 uint32_t data;
604
605 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
606 /* Before power off, this indicator has to be turned on */
607 data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
608 data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
609 data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
610 WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
611
612 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
613 2 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT);
614 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
615 1 << UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS__SHIFT,
616 UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK);
617 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
618 2 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT);
619 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
620 1 << UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS__SHIFT,
621 UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK);
622 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
623 2 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT);
624 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
625 1 << UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS__SHIFT,
626 UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK);
627 WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
628 2 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT);
629 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
630 1 << UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS__SHIFT,
631 UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK);
632 }
633 }
634
635 /**
636 * vcn_v4_0_5_disable_clock_gating - disable VCN clock gating
637 *
638 * @adev: amdgpu_device pointer
639 * @inst: instance number
640 *
641 * Disable clock gating for VCN block
642 */
vcn_v4_0_5_disable_clock_gating(struct amdgpu_device * adev,int inst)643 static void vcn_v4_0_5_disable_clock_gating(struct amdgpu_device *adev, int inst)
644 {
645 uint32_t data;
646
647 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
648 return;
649
650 /* VCN disable CGC */
651 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
652 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
653 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
654 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
655 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
656
657 data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
658 data &= ~(UVD_CGC_GATE__SYS_MASK
659 | UVD_CGC_GATE__UDEC_MASK
660 | UVD_CGC_GATE__MPEG2_MASK
661 | UVD_CGC_GATE__REGS_MASK
662 | UVD_CGC_GATE__RBC_MASK
663 | UVD_CGC_GATE__LMI_MC_MASK
664 | UVD_CGC_GATE__LMI_UMC_MASK
665 | UVD_CGC_GATE__IDCT_MASK
666 | UVD_CGC_GATE__MPRD_MASK
667 | UVD_CGC_GATE__MPC_MASK
668 | UVD_CGC_GATE__LBSI_MASK
669 | UVD_CGC_GATE__LRBBM_MASK
670 | UVD_CGC_GATE__UDEC_RE_MASK
671 | UVD_CGC_GATE__UDEC_CM_MASK
672 | UVD_CGC_GATE__UDEC_IT_MASK
673 | UVD_CGC_GATE__UDEC_DB_MASK
674 | UVD_CGC_GATE__UDEC_MP_MASK
675 | UVD_CGC_GATE__WCB_MASK
676 | UVD_CGC_GATE__VCPU_MASK
677 | UVD_CGC_GATE__MMSCH_MASK);
678
679 WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
680 SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0, 0xFFFFFFFF);
681
682 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
683 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
684 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
685 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
686 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
687 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
688 | UVD_CGC_CTRL__SYS_MODE_MASK
689 | UVD_CGC_CTRL__UDEC_MODE_MASK
690 | UVD_CGC_CTRL__MPEG2_MODE_MASK
691 | UVD_CGC_CTRL__REGS_MODE_MASK
692 | UVD_CGC_CTRL__RBC_MODE_MASK
693 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
694 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
695 | UVD_CGC_CTRL__IDCT_MODE_MASK
696 | UVD_CGC_CTRL__MPRD_MODE_MASK
697 | UVD_CGC_CTRL__MPC_MODE_MASK
698 | UVD_CGC_CTRL__LBSI_MODE_MASK
699 | UVD_CGC_CTRL__LRBBM_MODE_MASK
700 | UVD_CGC_CTRL__WCB_MODE_MASK
701 | UVD_CGC_CTRL__VCPU_MODE_MASK
702 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
703 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
704
705 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
706 data |= (UVD_SUVD_CGC_GATE__SRE_MASK
707 | UVD_SUVD_CGC_GATE__SIT_MASK
708 | UVD_SUVD_CGC_GATE__SMP_MASK
709 | UVD_SUVD_CGC_GATE__SCM_MASK
710 | UVD_SUVD_CGC_GATE__SDB_MASK
711 | UVD_SUVD_CGC_GATE__SRE_H264_MASK
712 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
713 | UVD_SUVD_CGC_GATE__SIT_H264_MASK
714 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
715 | UVD_SUVD_CGC_GATE__SCM_H264_MASK
716 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
717 | UVD_SUVD_CGC_GATE__SDB_H264_MASK
718 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
719 | UVD_SUVD_CGC_GATE__SCLR_MASK
720 | UVD_SUVD_CGC_GATE__UVD_SC_MASK
721 | UVD_SUVD_CGC_GATE__ENT_MASK
722 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
723 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
724 | UVD_SUVD_CGC_GATE__SITE_MASK
725 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
726 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
727 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
728 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
729 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
730 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
731
732 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
733 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
734 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
735 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
736 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
737 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
738 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
739 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
740 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
741 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
742 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
743 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
744 }
745
746 /**
747 * vcn_v4_0_5_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
748 *
749 * @adev: amdgpu_device pointer
750 * @sram_sel: sram select
751 * @inst_idx: instance number index
752 * @indirect: indirectly write sram
753 *
754 * Disable clock gating for VCN block with dpg mode
755 */
vcn_v4_0_5_disable_clock_gating_dpg_mode(struct amdgpu_device * adev,uint8_t sram_sel,int inst_idx,uint8_t indirect)756 static void vcn_v4_0_5_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
757 int inst_idx, uint8_t indirect)
758 {
759 uint32_t reg_data = 0;
760
761 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
762 return;
763
764 /* enable sw clock gating control */
765 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
766 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
767 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
768 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
769 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
770 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
771 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
772 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
773 UVD_CGC_CTRL__SYS_MODE_MASK |
774 UVD_CGC_CTRL__UDEC_MODE_MASK |
775 UVD_CGC_CTRL__MPEG2_MODE_MASK |
776 UVD_CGC_CTRL__REGS_MODE_MASK |
777 UVD_CGC_CTRL__RBC_MODE_MASK |
778 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
779 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
780 UVD_CGC_CTRL__IDCT_MODE_MASK |
781 UVD_CGC_CTRL__MPRD_MODE_MASK |
782 UVD_CGC_CTRL__MPC_MODE_MASK |
783 UVD_CGC_CTRL__LBSI_MODE_MASK |
784 UVD_CGC_CTRL__LRBBM_MODE_MASK |
785 UVD_CGC_CTRL__WCB_MODE_MASK |
786 UVD_CGC_CTRL__VCPU_MODE_MASK);
787 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
788 VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
789
790 /* turn off clock gating */
791 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
792 VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
793
794 /* turn on SUVD clock gating */
795 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
796 VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
797
798 /* turn on sw mode in UVD_SUVD_CGC_CTRL */
799 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
800 VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
801 }
802
803 /**
804 * vcn_v4_0_5_enable_clock_gating - enable VCN clock gating
805 *
806 * @adev: amdgpu_device pointer
807 * @inst: instance number
808 *
809 * Enable clock gating for VCN block
810 */
vcn_v4_0_5_enable_clock_gating(struct amdgpu_device * adev,int inst)811 static void vcn_v4_0_5_enable_clock_gating(struct amdgpu_device *adev, int inst)
812 {
813 uint32_t data;
814
815 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
816 return;
817
818 /* enable VCN CGC */
819 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
820 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
821 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
822 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
823 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
824
825 data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
826 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
827 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
828 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
829 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
830 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
831 | UVD_CGC_CTRL__SYS_MODE_MASK
832 | UVD_CGC_CTRL__UDEC_MODE_MASK
833 | UVD_CGC_CTRL__MPEG2_MODE_MASK
834 | UVD_CGC_CTRL__REGS_MODE_MASK
835 | UVD_CGC_CTRL__RBC_MODE_MASK
836 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
837 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
838 | UVD_CGC_CTRL__IDCT_MODE_MASK
839 | UVD_CGC_CTRL__MPRD_MODE_MASK
840 | UVD_CGC_CTRL__MPC_MODE_MASK
841 | UVD_CGC_CTRL__LBSI_MODE_MASK
842 | UVD_CGC_CTRL__LRBBM_MODE_MASK
843 | UVD_CGC_CTRL__WCB_MODE_MASK
844 | UVD_CGC_CTRL__VCPU_MODE_MASK
845 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
846 WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
847
848 data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
849 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
850 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
851 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
852 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
853 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
854 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
855 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
856 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
857 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
858 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
859 WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
860 }
861
862 /**
863 * vcn_v4_0_5_start_dpg_mode - VCN start with dpg mode
864 *
865 * @adev: amdgpu_device pointer
866 * @inst_idx: instance number index
867 * @indirect: indirectly write sram
868 *
869 * Start VCN block with dpg mode
870 */
vcn_v4_0_5_start_dpg_mode(struct amdgpu_device * adev,int inst_idx,bool indirect)871 static int vcn_v4_0_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
872 {
873 volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
874 struct amdgpu_ring *ring;
875 uint32_t tmp;
876
877 /* disable register anti-hang mechanism */
878 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
879 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
880 /* enable dynamic power gating mode */
881 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
882 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
883 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
884 WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
885
886 if (indirect)
887 adev->vcn.inst[inst_idx].dpg_sram_curr_addr =
888 (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
889
890 /* enable clock gating */
891 vcn_v4_0_5_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
892
893 /* enable VCPU clock */
894 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
895 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
896 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
897 VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
898
899 /* disable master interrupt */
900 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
901 VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
902
903 /* setup regUVD_LMI_CTRL */
904 tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
905 UVD_LMI_CTRL__REQ_MODE_MASK |
906 UVD_LMI_CTRL__CRC_RESET_MASK |
907 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
908 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
909 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
910 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
911 0x00100000L);
912 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
913 VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
914
915 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
916 VCN, inst_idx, regUVD_MPC_CNTL),
917 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
918
919 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
920 VCN, inst_idx, regUVD_MPC_SET_MUXA0),
921 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
922 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
923 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
924 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
925
926 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
927 VCN, inst_idx, regUVD_MPC_SET_MUXB0),
928 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
929 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
930 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
931 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
932
933 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
934 VCN, inst_idx, regUVD_MPC_SET_MUX),
935 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
936 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
937 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
938
939 vcn_v4_0_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
940
941 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
942 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
943 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
944 VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
945
946 /* enable LMI MC and UMC channels */
947 tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
948 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
949 VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
950
951 /* enable master interrupt */
952 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
953 VCN, inst_idx, regUVD_MASTINT_EN),
954 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
955
956 if (indirect)
957 amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
958
959 ring = &adev->vcn.inst[inst_idx].ring_enc[0];
960
961 WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
962 WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
963 WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
964
965 tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
966 tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
967 WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
968 fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
969 WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
970 WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
971
972 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
973 WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
974 ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
975
976 tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
977 tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
978 WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
979 fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
980
981 WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
982 ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
983 VCN_RB1_DB_CTRL__EN_MASK);
984
985 return 0;
986 }
987
988
989 /**
990 * vcn_v4_0_5_start - VCN start
991 *
992 * @adev: amdgpu_device pointer
993 *
994 * Start VCN block
995 */
vcn_v4_0_5_start(struct amdgpu_device * adev)996 static int vcn_v4_0_5_start(struct amdgpu_device *adev)
997 {
998 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
999 struct amdgpu_ring *ring;
1000 uint32_t tmp;
1001 int i, j, k, r;
1002
1003 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1004 if (adev->pm.dpm_enabled)
1005 amdgpu_dpm_enable_vcn(adev, true, i);
1006 }
1007
1008 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1009 if (adev->vcn.harvest_config & (1 << i))
1010 continue;
1011
1012 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1013
1014 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1015 r = vcn_v4_0_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1016 continue;
1017 }
1018
1019 /* disable VCN power gating */
1020 vcn_v4_0_5_disable_static_power_gating(adev, i);
1021
1022 /* set VCN status busy */
1023 tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1024 WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1025
1026 /*SW clock gating */
1027 vcn_v4_0_5_disable_clock_gating(adev, i);
1028
1029 /* enable VCPU clock */
1030 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1031 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1032
1033 /* disable master interrupt */
1034 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1035 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1036
1037 /* enable LMI MC and UMC channels */
1038 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1039 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1040
1041 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1042 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1043 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1044 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1045
1046 /* setup regUVD_LMI_CTRL */
1047 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1048 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1049 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1050 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1051 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1052 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1053
1054 /* setup regUVD_MPC_CNTL */
1055 tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1056 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1057 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1058 WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1059
1060 /* setup UVD_MPC_SET_MUXA0 */
1061 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1062 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1063 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1064 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1065 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1066
1067 /* setup UVD_MPC_SET_MUXB0 */
1068 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1069 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1070 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1071 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1072 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1073
1074 /* setup UVD_MPC_SET_MUX */
1075 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1076 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1077 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1078 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1079
1080 vcn_v4_0_5_mc_resume(adev, i);
1081
1082 /* VCN global tiling registers */
1083 WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1084 adev->gfx.config.gb_addr_config);
1085
1086 /* unblock VCPU register access */
1087 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1088 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1089
1090 /* release VCPU reset to boot */
1091 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1092 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1093
1094 for (j = 0; j < 10; ++j) {
1095 uint32_t status;
1096
1097 for (k = 0; k < 100; ++k) {
1098 status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1099 if (status & 2)
1100 break;
1101 mdelay(10);
1102 if (amdgpu_emu_mode == 1)
1103 msleep(1);
1104 }
1105
1106 if (amdgpu_emu_mode == 1) {
1107 r = -1;
1108 if (status & 2) {
1109 r = 0;
1110 break;
1111 }
1112 } else {
1113 r = 0;
1114 if (status & 2)
1115 break;
1116
1117 dev_err(adev->dev,
1118 "VCN[%d] is not responding, trying to reset VCPU!!!\n", i);
1119 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1120 UVD_VCPU_CNTL__BLK_RST_MASK,
1121 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1122 mdelay(10);
1123 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1124 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1125
1126 mdelay(10);
1127 r = -1;
1128 }
1129 }
1130
1131 if (r) {
1132 dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1133 return r;
1134 }
1135
1136 /* enable master interrupt */
1137 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1138 UVD_MASTINT_EN__VCPU_EN_MASK,
1139 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1140
1141 /* clear the busy bit of VCN_STATUS */
1142 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1143 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1144
1145 ring = &adev->vcn.inst[i].ring_enc[0];
1146 WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1147 ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1148 VCN_RB1_DB_CTRL__EN_MASK);
1149
1150 WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1151 WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1152 WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1153
1154 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1155 tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1156 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1157 fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1158 WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1159 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1160
1161 tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1162 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1163 ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1164
1165 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1166 tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1167 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1168 fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1169 }
1170
1171 return 0;
1172 }
1173
1174 /**
1175 * vcn_v4_0_5_stop_dpg_mode - VCN stop with dpg mode
1176 *
1177 * @adev: amdgpu_device pointer
1178 * @inst_idx: instance number index
1179 *
1180 * Stop VCN block with dpg mode
1181 */
vcn_v4_0_5_stop_dpg_mode(struct amdgpu_device * adev,int inst_idx)1182 static void vcn_v4_0_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1183 {
1184 uint32_t tmp;
1185
1186 /* Wait for power status to be 1 */
1187 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1188 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1189
1190 /* wait for read ptr to be equal to write ptr */
1191 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1192 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1193
1194 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1195 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1196
1197 /* disable dynamic power gating mode */
1198 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1199 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1200 }
1201
1202 /**
1203 * vcn_v4_0_5_stop - VCN stop
1204 *
1205 * @adev: amdgpu_device pointer
1206 *
1207 * Stop VCN block
1208 */
vcn_v4_0_5_stop(struct amdgpu_device * adev)1209 static int vcn_v4_0_5_stop(struct amdgpu_device *adev)
1210 {
1211 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1212 uint32_t tmp;
1213 int i, r = 0;
1214
1215 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1216 if (adev->vcn.harvest_config & (1 << i))
1217 continue;
1218
1219 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1220 fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1221
1222 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1223 vcn_v4_0_5_stop_dpg_mode(adev, i);
1224 continue;
1225 }
1226
1227 /* wait for vcn idle */
1228 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1229 if (r)
1230 return r;
1231
1232 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1233 UVD_LMI_STATUS__READ_CLEAN_MASK |
1234 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1235 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1236 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1237 if (r)
1238 return r;
1239
1240 /* disable LMI UMC channel */
1241 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1242 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1243 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1244 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1245 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1246 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1247 if (r)
1248 return r;
1249
1250 /* block VCPU register access */
1251 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1252 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1253 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1254
1255 /* reset VCPU */
1256 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1257 UVD_VCPU_CNTL__BLK_RST_MASK,
1258 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1259
1260 /* disable VCPU clock */
1261 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1262 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1263
1264 /* apply soft reset */
1265 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1266 tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1267 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1268 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1269 tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1270 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1271
1272 /* clear status */
1273 WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1274
1275 /* apply HW clock gating */
1276 vcn_v4_0_5_enable_clock_gating(adev, i);
1277
1278 /* enable VCN power gating */
1279 vcn_v4_0_5_enable_static_power_gating(adev, i);
1280 }
1281
1282 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1283 if (adev->pm.dpm_enabled)
1284 amdgpu_dpm_enable_vcn(adev, false, i);
1285 }
1286
1287 return 0;
1288 }
1289
1290 /**
1291 * vcn_v4_0_5_pause_dpg_mode - VCN pause with dpg mode
1292 *
1293 * @adev: amdgpu_device pointer
1294 * @inst_idx: instance number index
1295 * @new_state: pause state
1296 *
1297 * Pause dpg mode for VCN block
1298 */
vcn_v4_0_5_pause_dpg_mode(struct amdgpu_device * adev,int inst_idx,struct dpg_pause_state * new_state)1299 static int vcn_v4_0_5_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1300 struct dpg_pause_state *new_state)
1301 {
1302 uint32_t reg_data = 0;
1303 int ret_code;
1304
1305 /* pause/unpause if state is changed */
1306 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1307 DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1308 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based);
1309 reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1310 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1311
1312 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1313 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1314 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1315
1316 if (!ret_code) {
1317 /* pause DPG */
1318 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1319 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1320
1321 /* wait for ACK */
1322 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1323 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1324 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1325
1326 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1327 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON,
1328 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1329 }
1330 } else {
1331 /* unpause dpg, no need to wait */
1332 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1333 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1334 }
1335 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1336 }
1337
1338 return 0;
1339 }
1340
1341 /**
1342 * vcn_v4_0_5_unified_ring_get_rptr - get unified read pointer
1343 *
1344 * @ring: amdgpu_ring pointer
1345 *
1346 * Returns the current hardware unified read pointer
1347 */
vcn_v4_0_5_unified_ring_get_rptr(struct amdgpu_ring * ring)1348 static uint64_t vcn_v4_0_5_unified_ring_get_rptr(struct amdgpu_ring *ring)
1349 {
1350 struct amdgpu_device *adev = ring->adev;
1351
1352 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1353 DRM_ERROR("wrong ring id is identified in %s", __func__);
1354
1355 return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1356 }
1357
1358 /**
1359 * vcn_v4_0_5_unified_ring_get_wptr - get unified write pointer
1360 *
1361 * @ring: amdgpu_ring pointer
1362 *
1363 * Returns the current hardware unified write pointer
1364 */
vcn_v4_0_5_unified_ring_get_wptr(struct amdgpu_ring * ring)1365 static uint64_t vcn_v4_0_5_unified_ring_get_wptr(struct amdgpu_ring *ring)
1366 {
1367 struct amdgpu_device *adev = ring->adev;
1368
1369 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1370 DRM_ERROR("wrong ring id is identified in %s", __func__);
1371
1372 if (ring->use_doorbell)
1373 return *ring->wptr_cpu_addr;
1374 else
1375 return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1376 }
1377
1378 /**
1379 * vcn_v4_0_5_unified_ring_set_wptr - set enc write pointer
1380 *
1381 * @ring: amdgpu_ring pointer
1382 *
1383 * Commits the enc write pointer to the hardware
1384 */
vcn_v4_0_5_unified_ring_set_wptr(struct amdgpu_ring * ring)1385 static void vcn_v4_0_5_unified_ring_set_wptr(struct amdgpu_ring *ring)
1386 {
1387 struct amdgpu_device *adev = ring->adev;
1388
1389 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1390 DRM_ERROR("wrong ring id is identified in %s", __func__);
1391
1392 if (ring->use_doorbell) {
1393 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1394 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1395 } else {
1396 WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1397 }
1398 }
1399
1400 static const struct amdgpu_ring_funcs vcn_v4_0_5_unified_ring_vm_funcs = {
1401 .type = AMDGPU_RING_TYPE_VCN_ENC,
1402 .align_mask = 0x3f,
1403 .nop = VCN_ENC_CMD_NO_OP,
1404 .get_rptr = vcn_v4_0_5_unified_ring_get_rptr,
1405 .get_wptr = vcn_v4_0_5_unified_ring_get_wptr,
1406 .set_wptr = vcn_v4_0_5_unified_ring_set_wptr,
1407 .emit_frame_size =
1408 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1409 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1410 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1411 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1412 1, /* vcn_v2_0_enc_ring_insert_end */
1413 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1414 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1415 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1416 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1417 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1418 .test_ib = amdgpu_vcn_unified_ring_test_ib,
1419 .insert_nop = amdgpu_ring_insert_nop,
1420 .insert_end = vcn_v2_0_enc_ring_insert_end,
1421 .pad_ib = amdgpu_ring_generic_pad_ib,
1422 .begin_use = amdgpu_vcn_ring_begin_use,
1423 .end_use = amdgpu_vcn_ring_end_use,
1424 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1425 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1426 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1427 };
1428
1429 /**
1430 * vcn_v4_0_5_set_unified_ring_funcs - set unified ring functions
1431 *
1432 * @adev: amdgpu_device pointer
1433 *
1434 * Set unified ring functions
1435 */
vcn_v4_0_5_set_unified_ring_funcs(struct amdgpu_device * adev)1436 static void vcn_v4_0_5_set_unified_ring_funcs(struct amdgpu_device *adev)
1437 {
1438 int i;
1439
1440 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1441 if (adev->vcn.harvest_config & (1 << i))
1442 continue;
1443
1444 adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v4_0_5_unified_ring_vm_funcs;
1445 adev->vcn.inst[i].ring_enc[0].me = i;
1446 }
1447 }
1448
1449 /**
1450 * vcn_v4_0_5_is_idle - check VCN block is idle
1451 *
1452 * @handle: amdgpu_device pointer
1453 *
1454 * Check whether VCN block is idle
1455 */
vcn_v4_0_5_is_idle(void * handle)1456 static bool vcn_v4_0_5_is_idle(void *handle)
1457 {
1458 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1459 int i, ret = 1;
1460
1461 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1462 if (adev->vcn.harvest_config & (1 << i))
1463 continue;
1464
1465 ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1466 }
1467
1468 return ret;
1469 }
1470
1471 /**
1472 * vcn_v4_0_5_wait_for_idle - wait for VCN block idle
1473 *
1474 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
1475 *
1476 * Wait for VCN block idle
1477 */
vcn_v4_0_5_wait_for_idle(struct amdgpu_ip_block * ip_block)1478 static int vcn_v4_0_5_wait_for_idle(struct amdgpu_ip_block *ip_block)
1479 {
1480 struct amdgpu_device *adev = ip_block->adev;
1481 int i, ret = 0;
1482
1483 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1484 if (adev->vcn.harvest_config & (1 << i))
1485 continue;
1486
1487 ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1488 UVD_STATUS__IDLE);
1489 if (ret)
1490 return ret;
1491 }
1492
1493 return ret;
1494 }
1495
1496 /**
1497 * vcn_v4_0_5_set_clockgating_state - set VCN block clockgating state
1498 *
1499 * @ip_block: amdgpu_ip_block pointer
1500 * @state: clock gating state
1501 *
1502 * Set VCN block clockgating state
1503 */
vcn_v4_0_5_set_clockgating_state(struct amdgpu_ip_block * ip_block,enum amd_clockgating_state state)1504 static int vcn_v4_0_5_set_clockgating_state(struct amdgpu_ip_block *ip_block,
1505 enum amd_clockgating_state state)
1506 {
1507 struct amdgpu_device *adev = ip_block->adev;
1508 bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1509 int i;
1510
1511 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1512 if (adev->vcn.harvest_config & (1 << i))
1513 continue;
1514
1515 if (enable) {
1516 if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1517 return -EBUSY;
1518 vcn_v4_0_5_enable_clock_gating(adev, i);
1519 } else {
1520 vcn_v4_0_5_disable_clock_gating(adev, i);
1521 }
1522 }
1523
1524 return 0;
1525 }
1526
1527 /**
1528 * vcn_v4_0_5_set_powergating_state - set VCN block powergating state
1529 *
1530 * @ip_block: amdgpu_ip_block pointer
1531 * @state: power gating state
1532 *
1533 * Set VCN block powergating state
1534 */
vcn_v4_0_5_set_powergating_state(struct amdgpu_ip_block * ip_block,enum amd_powergating_state state)1535 static int vcn_v4_0_5_set_powergating_state(struct amdgpu_ip_block *ip_block,
1536 enum amd_powergating_state state)
1537 {
1538 struct amdgpu_device *adev = ip_block->adev;
1539 int ret;
1540
1541 if (state == adev->vcn.cur_state)
1542 return 0;
1543
1544 if (state == AMD_PG_STATE_GATE)
1545 ret = vcn_v4_0_5_stop(adev);
1546 else
1547 ret = vcn_v4_0_5_start(adev);
1548
1549 if (!ret)
1550 adev->vcn.cur_state = state;
1551
1552 return ret;
1553 }
1554
1555 /**
1556 * vcn_v4_0_5_process_interrupt - process VCN block interrupt
1557 *
1558 * @adev: amdgpu_device pointer
1559 * @source: interrupt sources
1560 * @entry: interrupt entry from clients and sources
1561 *
1562 * Process VCN block interrupt
1563 */
vcn_v4_0_5_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)1564 static int vcn_v4_0_5_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1565 struct amdgpu_iv_entry *entry)
1566 {
1567 uint32_t ip_instance;
1568
1569 switch (entry->client_id) {
1570 case SOC15_IH_CLIENTID_VCN:
1571 ip_instance = 0;
1572 break;
1573 case SOC15_IH_CLIENTID_VCN1:
1574 ip_instance = 1;
1575 break;
1576 default:
1577 DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1578 return 0;
1579 }
1580
1581 DRM_DEBUG("IH: VCN TRAP\n");
1582
1583 switch (entry->src_id) {
1584 case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1585 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1586 break;
1587 case VCN_4_0__SRCID_UVD_POISON:
1588 amdgpu_vcn_process_poison_irq(adev, source, entry);
1589 break;
1590 default:
1591 DRM_ERROR("Unhandled interrupt: %d %d\n",
1592 entry->src_id, entry->src_data[0]);
1593 break;
1594 }
1595
1596 return 0;
1597 }
1598
1599 static const struct amdgpu_irq_src_funcs vcn_v4_0_5_irq_funcs = {
1600 .process = vcn_v4_0_5_process_interrupt,
1601 };
1602
1603 /**
1604 * vcn_v4_0_5_set_irq_funcs - set VCN block interrupt irq functions
1605 *
1606 * @adev: amdgpu_device pointer
1607 *
1608 * Set VCN block interrupt irq functions
1609 */
vcn_v4_0_5_set_irq_funcs(struct amdgpu_device * adev)1610 static void vcn_v4_0_5_set_irq_funcs(struct amdgpu_device *adev)
1611 {
1612 int i;
1613
1614 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1615 if (adev->vcn.harvest_config & (1 << i))
1616 continue;
1617
1618 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1619 adev->vcn.inst[i].irq.funcs = &vcn_v4_0_5_irq_funcs;
1620 }
1621 }
1622
vcn_v4_0_5_print_ip_state(struct amdgpu_ip_block * ip_block,struct drm_printer * p)1623 static void vcn_v4_0_5_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p)
1624 {
1625 struct amdgpu_device *adev = ip_block->adev;
1626 int i, j;
1627 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0_5);
1628 uint32_t inst_off, is_powered;
1629
1630 if (!adev->vcn.ip_dump)
1631 return;
1632
1633 drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst);
1634 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1635 if (adev->vcn.harvest_config & (1 << i)) {
1636 drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i);
1637 continue;
1638 }
1639
1640 inst_off = i * reg_count;
1641 is_powered = (adev->vcn.ip_dump[inst_off] &
1642 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
1643
1644 if (is_powered) {
1645 drm_printf(p, "\nActive Instance:VCN%d\n", i);
1646 for (j = 0; j < reg_count; j++)
1647 drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_4_0_5[j].reg_name,
1648 adev->vcn.ip_dump[inst_off + j]);
1649 } else {
1650 drm_printf(p, "\nInactive Instance:VCN%d\n", i);
1651 }
1652 }
1653 }
1654
vcn_v4_0_5_dump_ip_state(struct amdgpu_ip_block * ip_block)1655 static void vcn_v4_0_5_dump_ip_state(struct amdgpu_ip_block *ip_block)
1656 {
1657 struct amdgpu_device *adev = ip_block->adev;
1658 int i, j;
1659 bool is_powered;
1660 uint32_t inst_off;
1661 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0_5);
1662
1663 if (!adev->vcn.ip_dump)
1664 return;
1665
1666 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1667 if (adev->vcn.harvest_config & (1 << i))
1668 continue;
1669
1670 inst_off = i * reg_count;
1671 /* mmUVD_POWER_STATUS is always readable and is first element of the array */
1672 adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, i, regUVD_POWER_STATUS);
1673 is_powered = (adev->vcn.ip_dump[inst_off] &
1674 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
1675
1676 if (is_powered)
1677 for (j = 1; j < reg_count; j++)
1678 adev->vcn.ip_dump[inst_off + j] =
1679 RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_4_0_5[j],
1680 i));
1681 }
1682 }
1683
1684 static const struct amd_ip_funcs vcn_v4_0_5_ip_funcs = {
1685 .name = "vcn_v4_0_5",
1686 .early_init = vcn_v4_0_5_early_init,
1687 .sw_init = vcn_v4_0_5_sw_init,
1688 .sw_fini = vcn_v4_0_5_sw_fini,
1689 .hw_init = vcn_v4_0_5_hw_init,
1690 .hw_fini = vcn_v4_0_5_hw_fini,
1691 .suspend = vcn_v4_0_5_suspend,
1692 .resume = vcn_v4_0_5_resume,
1693 .is_idle = vcn_v4_0_5_is_idle,
1694 .wait_for_idle = vcn_v4_0_5_wait_for_idle,
1695 .set_clockgating_state = vcn_v4_0_5_set_clockgating_state,
1696 .set_powergating_state = vcn_v4_0_5_set_powergating_state,
1697 .dump_ip_state = vcn_v4_0_5_dump_ip_state,
1698 .print_ip_state = vcn_v4_0_5_print_ip_state,
1699 };
1700
1701 const struct amdgpu_ip_block_version vcn_v4_0_5_ip_block = {
1702 .type = AMD_IP_BLOCK_TYPE_VCN,
1703 .major = 4,
1704 .minor = 0,
1705 .rev = 5,
1706 .funcs = &vcn_v4_0_5_ip_funcs,
1707 };
1708