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
55 static 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
91 static int amdgpu_ih_clientid_vcns[] = {
92 SOC15_IH_CLIENTID_VCN,
93 SOC15_IH_CLIENTID_VCN1
94 };
95
96 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev);
97 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev);
98 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev);
99 static int vcn_v4_0_set_powergating_state(struct amdgpu_ip_block *ip_block,
100 enum amd_powergating_state state);
101 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev,
102 int inst_idx, struct dpg_pause_state *new_state);
103 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring);
104 static 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 */
vcn_v4_0_early_init(struct amdgpu_ip_block * ip_block)114 static 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
vcn_v4_0_fw_shared_init(struct amdgpu_device * adev,int inst_idx)139 static 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 */
vcn_v4_0_sw_init(struct amdgpu_ip_block * ip_block)171 static 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 */
vcn_v4_0_sw_fini(struct amdgpu_ip_block * ip_block)268 static 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 */
vcn_v4_0_hw_init(struct amdgpu_ip_block * ip_block)310 static 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 */
vcn_v4_0_hw_fini(struct amdgpu_ip_block * ip_block)357 static 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(ip_block, 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 */
vcn_v4_0_suspend(struct amdgpu_ip_block * ip_block)388 static 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 */
vcn_v4_0_resume(struct amdgpu_ip_block * ip_block)408 static 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 */
vcn_v4_0_mc_resume(struct amdgpu_device * adev,int inst)429 static 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.inst[inst].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 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 */
vcn_v4_0_mc_resume_dpg_mode(struct amdgpu_device * adev,int inst_idx,bool indirect)490 static 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.inst[inst_idx].fw->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 */
vcn_v4_0_disable_static_power_gating(struct amdgpu_device * adev,int inst)596 static 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 */
vcn_v4_0_enable_static_power_gating(struct amdgpu_device * adev,int inst)661 static 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 */
vcn_v4_0_disable_clock_gating(struct amdgpu_device * adev,int inst)716 static 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 */
vcn_v4_0_disable_clock_gating_dpg_mode(struct amdgpu_device * adev,uint8_t sram_sel,int inst_idx,uint8_t indirect)829 static 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 */
vcn_v4_0_enable_clock_gating(struct amdgpu_device * adev,int inst)884 static 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
vcn_v4_0_enable_ras(struct amdgpu_device * adev,int inst_idx,bool indirect)935 static 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 */
vcn_v4_0_start_dpg_mode(struct amdgpu_device * adev,int inst_idx,bool indirect)966 static 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 */
vcn_v4_0_start(struct amdgpu_device * adev)1093 static 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 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1101 if (adev->pm.dpm_enabled)
1102 amdgpu_dpm_enable_vcn(adev, true, i);
1103 }
1104
1105 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1106 if (adev->vcn.harvest_config & (1 << i))
1107 continue;
1108
1109 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1110
1111 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1112 r = vcn_v4_0_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1113 continue;
1114 }
1115
1116 /* disable VCN power gating */
1117 vcn_v4_0_disable_static_power_gating(adev, i);
1118
1119 /* set VCN status busy */
1120 tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1121 WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
1122
1123 /*SW clock gating */
1124 vcn_v4_0_disable_clock_gating(adev, i);
1125
1126 /* enable VCPU clock */
1127 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1128 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1129
1130 /* disable master interrupt */
1131 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
1132 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1133
1134 /* enable LMI MC and UMC channels */
1135 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
1136 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1137
1138 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1139 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1140 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1141 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1142
1143 /* setup regUVD_LMI_CTRL */
1144 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1145 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1146 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1147 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1148 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1149 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1150
1151 /* setup regUVD_MPC_CNTL */
1152 tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1153 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1154 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1155 WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1156
1157 /* setup UVD_MPC_SET_MUXA0 */
1158 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1159 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1160 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1161 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1162 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1163
1164 /* setup UVD_MPC_SET_MUXB0 */
1165 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1166 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1167 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1168 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1169 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1170
1171 /* setup UVD_MPC_SET_MUX */
1172 WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1173 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1174 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1175 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1176
1177 vcn_v4_0_mc_resume(adev, i);
1178
1179 /* VCN global tiling registers */
1180 WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1181 adev->gfx.config.gb_addr_config);
1182
1183 /* unblock VCPU register access */
1184 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1185 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1186
1187 /* release VCPU reset to boot */
1188 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1189 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1190
1191 for (j = 0; j < 10; ++j) {
1192 uint32_t status;
1193
1194 for (k = 0; k < 100; ++k) {
1195 status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1196 if (status & 2)
1197 break;
1198 mdelay(10);
1199 if (amdgpu_emu_mode == 1)
1200 msleep(1);
1201 }
1202
1203 if (amdgpu_emu_mode == 1) {
1204 r = -1;
1205 if (status & 2) {
1206 r = 0;
1207 break;
1208 }
1209 } else {
1210 r = 0;
1211 if (status & 2)
1212 break;
1213
1214 dev_err(adev->dev, "VCN[%d] is not responding, trying to reset the VCPU!!!\n", i);
1215 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1216 UVD_VCPU_CNTL__BLK_RST_MASK,
1217 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1218 mdelay(10);
1219 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1220 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1221
1222 mdelay(10);
1223 r = -1;
1224 }
1225 }
1226
1227 if (r) {
1228 dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1229 return r;
1230 }
1231
1232 /* enable master interrupt */
1233 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1234 UVD_MASTINT_EN__VCPU_EN_MASK,
1235 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1236
1237 /* clear the busy bit of VCN_STATUS */
1238 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1239 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1240
1241 ring = &adev->vcn.inst[i].ring_enc[0];
1242 WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1243 ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1244 VCN_RB1_DB_CTRL__EN_MASK);
1245
1246 WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1247 WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1248 WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1249
1250 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1251 tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1252 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1253 fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1254 WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1255 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1256
1257 tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1258 WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1259 ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1260
1261 tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1262 tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1263 WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1264 fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1265 }
1266
1267 return 0;
1268 }
1269
vcn_v4_0_init_ring_metadata(struct amdgpu_device * adev,uint32_t vcn_inst,struct amdgpu_ring * ring_enc)1270 static int vcn_v4_0_init_ring_metadata(struct amdgpu_device *adev, uint32_t vcn_inst, struct amdgpu_ring *ring_enc)
1271 {
1272 struct amdgpu_vcn_rb_metadata *rb_metadata = NULL;
1273 uint8_t *rb_ptr = (uint8_t *)ring_enc->ring;
1274
1275 rb_ptr += ring_enc->ring_size;
1276 rb_metadata = (struct amdgpu_vcn_rb_metadata *)rb_ptr;
1277
1278 memset(rb_metadata, 0, sizeof(struct amdgpu_vcn_rb_metadata));
1279 rb_metadata->size = sizeof(struct amdgpu_vcn_rb_metadata);
1280 rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1281 rb_metadata->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1282 rb_metadata->version = 1;
1283 rb_metadata->ring_id = vcn_inst & 0xFF;
1284
1285 return 0;
1286 }
1287
vcn_v4_0_start_sriov(struct amdgpu_device * adev)1288 static int vcn_v4_0_start_sriov(struct amdgpu_device *adev)
1289 {
1290 int i;
1291 struct amdgpu_ring *ring_enc;
1292 uint64_t cache_addr;
1293 uint64_t rb_enc_addr;
1294 uint64_t ctx_addr;
1295 uint32_t param, resp, expected;
1296 uint32_t offset, cache_size;
1297 uint32_t tmp, timeout;
1298
1299 struct amdgpu_mm_table *table = &adev->virt.mm_table;
1300 uint32_t *table_loc;
1301 uint32_t table_size;
1302 uint32_t size, size_dw;
1303 uint32_t init_status;
1304 uint32_t enabled_vcn;
1305
1306 struct mmsch_v4_0_cmd_direct_write
1307 direct_wt = { {0} };
1308 struct mmsch_v4_0_cmd_direct_read_modify_write
1309 direct_rd_mod_wt = { {0} };
1310 struct mmsch_v4_0_cmd_end end = { {0} };
1311 struct mmsch_v4_0_init_header header;
1312
1313 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1314 volatile struct amdgpu_fw_shared_rb_setup *rb_setup;
1315
1316 direct_wt.cmd_header.command_type =
1317 MMSCH_COMMAND__DIRECT_REG_WRITE;
1318 direct_rd_mod_wt.cmd_header.command_type =
1319 MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1320 end.cmd_header.command_type =
1321 MMSCH_COMMAND__END;
1322
1323 header.version = MMSCH_VERSION;
1324 header.total_size = sizeof(struct mmsch_v4_0_init_header) >> 2;
1325 for (i = 0; i < MMSCH_V4_0_VCN_INSTANCES; i++) {
1326 header.inst[i].init_status = 0;
1327 header.inst[i].table_offset = 0;
1328 header.inst[i].table_size = 0;
1329 }
1330
1331 table_loc = (uint32_t *)table->cpu_addr;
1332 table_loc += header.total_size;
1333 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1334 if (adev->vcn.harvest_config & (1 << i))
1335 continue;
1336
1337 // Must re/init fw_shared at beginning
1338 vcn_v4_0_fw_shared_init(adev, i);
1339
1340 table_size = 0;
1341
1342 MMSCH_V4_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(VCN, i,
1343 regUVD_STATUS),
1344 ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1345
1346 cache_size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.inst[i].fw->size + 4);
1347
1348 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1349 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1350 regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1351 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1352 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1353 regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1354 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1355 offset = 0;
1356 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1357 regUVD_VCPU_CACHE_OFFSET0),
1358 0);
1359 } else {
1360 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1361 regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1362 lower_32_bits(adev->vcn.inst[i].gpu_addr));
1363 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1364 regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1365 upper_32_bits(adev->vcn.inst[i].gpu_addr));
1366 offset = cache_size;
1367 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1368 regUVD_VCPU_CACHE_OFFSET0),
1369 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1370 }
1371
1372 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1373 regUVD_VCPU_CACHE_SIZE0),
1374 cache_size);
1375
1376 cache_addr = adev->vcn.inst[i].gpu_addr + offset;
1377 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1378 regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1379 lower_32_bits(cache_addr));
1380 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1381 regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1382 upper_32_bits(cache_addr));
1383 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1384 regUVD_VCPU_CACHE_OFFSET1),
1385 0);
1386 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1387 regUVD_VCPU_CACHE_SIZE1),
1388 AMDGPU_VCN_STACK_SIZE);
1389
1390 cache_addr = adev->vcn.inst[i].gpu_addr + offset +
1391 AMDGPU_VCN_STACK_SIZE;
1392 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1393 regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1394 lower_32_bits(cache_addr));
1395 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1396 regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1397 upper_32_bits(cache_addr));
1398 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1399 regUVD_VCPU_CACHE_OFFSET2),
1400 0);
1401 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1402 regUVD_VCPU_CACHE_SIZE2),
1403 AMDGPU_VCN_CONTEXT_SIZE);
1404
1405 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1406 rb_setup = &fw_shared->rb_setup;
1407
1408 ring_enc = &adev->vcn.inst[i].ring_enc[0];
1409 ring_enc->wptr = 0;
1410 rb_enc_addr = ring_enc->gpu_addr;
1411
1412 rb_setup->is_rb_enabled_flags |= RB_ENABLED;
1413 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
1414
1415 if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
1416 vcn_v4_0_init_ring_metadata(adev, i, ring_enc);
1417
1418 memset((void *)&rb_setup->rb_info, 0, sizeof(struct amdgpu_vcn_rb_setup_info) * MAX_NUM_VCN_RB_SETUP);
1419 if (!(adev->vcn.harvest_config & (1 << 0))) {
1420 rb_setup->rb_info[0].rb_addr_lo = lower_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1421 rb_setup->rb_info[0].rb_addr_hi = upper_32_bits(adev->vcn.inst[0].ring_enc[0].gpu_addr);
1422 rb_setup->rb_info[0].rb_size = adev->vcn.inst[0].ring_enc[0].ring_size / 4;
1423 }
1424 if (!(adev->vcn.harvest_config & (1 << 1))) {
1425 rb_setup->rb_info[2].rb_addr_lo = lower_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1426 rb_setup->rb_info[2].rb_addr_hi = upper_32_bits(adev->vcn.inst[1].ring_enc[0].gpu_addr);
1427 rb_setup->rb_info[2].rb_size = adev->vcn.inst[1].ring_enc[0].ring_size / 4;
1428 }
1429 fw_shared->decouple.is_enabled = 1;
1430 fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_DECOUPLE_FLAG);
1431 } else {
1432 rb_setup->rb_addr_lo = lower_32_bits(rb_enc_addr);
1433 rb_setup->rb_addr_hi = upper_32_bits(rb_enc_addr);
1434 rb_setup->rb_size = ring_enc->ring_size / 4;
1435 }
1436
1437 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1438 regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
1439 lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1440 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1441 regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
1442 upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
1443 MMSCH_V4_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(VCN, i,
1444 regUVD_VCPU_NONCACHE_SIZE0),
1445 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
1446
1447 /* add end packet */
1448 MMSCH_V4_0_INSERT_END();
1449
1450 /* refine header */
1451 header.inst[i].init_status = 0;
1452 header.inst[i].table_offset = header.total_size;
1453 header.inst[i].table_size = table_size;
1454 header.total_size += table_size;
1455 }
1456
1457 /* Update init table header in memory */
1458 size = sizeof(struct mmsch_v4_0_init_header);
1459 table_loc = (uint32_t *)table->cpu_addr;
1460 memcpy((void *)table_loc, &header, size);
1461
1462 /* message MMSCH (in VCN[0]) to initialize this client
1463 * 1, write to mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1464 * of memory descriptor location
1465 */
1466 ctx_addr = table->gpu_addr;
1467 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_LO, lower_32_bits(ctx_addr));
1468 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_ADDR_HI, upper_32_bits(ctx_addr));
1469
1470 /* 2, update vmid of descriptor */
1471 tmp = RREG32_SOC15(VCN, 0, regMMSCH_VF_VMID);
1472 tmp &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1473 /* use domain0 for MM scheduler */
1474 tmp |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1475 WREG32_SOC15(VCN, 0, regMMSCH_VF_VMID, tmp);
1476
1477 /* 3, notify mmsch about the size of this descriptor */
1478 size = header.total_size;
1479 WREG32_SOC15(VCN, 0, regMMSCH_VF_CTX_SIZE, size);
1480
1481 /* 4, set resp to zero */
1482 WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP, 0);
1483
1484 /* 5, kick off the initialization and wait until
1485 * MMSCH_VF_MAILBOX_RESP becomes non-zero
1486 */
1487 param = 0x00000001;
1488 WREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_HOST, param);
1489 tmp = 0;
1490 timeout = 1000;
1491 resp = 0;
1492 expected = MMSCH_VF_MAILBOX_RESP__OK;
1493 while (resp != expected) {
1494 resp = RREG32_SOC15(VCN, 0, regMMSCH_VF_MAILBOX_RESP);
1495 if (resp != 0)
1496 break;
1497
1498 udelay(10);
1499 tmp = tmp + 10;
1500 if (tmp >= timeout) {
1501 DRM_ERROR("failed to init MMSCH. TIME-OUT after %d usec"\
1502 " waiting for regMMSCH_VF_MAILBOX_RESP "\
1503 "(expected=0x%08x, readback=0x%08x)\n",
1504 tmp, expected, resp);
1505 return -EBUSY;
1506 }
1507 }
1508 enabled_vcn = amdgpu_vcn_is_disabled_vcn(adev, VCN_DECODE_RING, 0) ? 1 : 0;
1509 init_status = ((struct mmsch_v4_0_init_header *)(table_loc))->inst[enabled_vcn].init_status;
1510 if (resp != expected && resp != MMSCH_VF_MAILBOX_RESP__INCOMPLETE
1511 && init_status != MMSCH_VF_ENGINE_STATUS__PASS)
1512 DRM_ERROR("MMSCH init status is incorrect! readback=0x%08x, header init "\
1513 "status for VCN%x: 0x%x\n", resp, enabled_vcn, init_status);
1514
1515 return 0;
1516 }
1517
1518 /**
1519 * vcn_v4_0_stop_dpg_mode - VCN stop with dpg mode
1520 *
1521 * @adev: amdgpu_device pointer
1522 * @inst_idx: instance number index
1523 *
1524 * Stop VCN block with dpg mode
1525 */
vcn_v4_0_stop_dpg_mode(struct amdgpu_device * adev,int inst_idx)1526 static void vcn_v4_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1527 {
1528 struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1529 uint32_t tmp;
1530
1531 vcn_v4_0_pause_dpg_mode(adev, inst_idx, &state);
1532 /* Wait for power status to be 1 */
1533 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1534 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1535
1536 /* wait for read ptr to be equal to write ptr */
1537 tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1538 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1539
1540 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1541 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1542
1543 /* disable dynamic power gating mode */
1544 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1545 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1546 }
1547
1548 /**
1549 * vcn_v4_0_stop - VCN stop
1550 *
1551 * @adev: amdgpu_device pointer
1552 *
1553 * Stop VCN block
1554 */
vcn_v4_0_stop(struct amdgpu_device * adev)1555 static int vcn_v4_0_stop(struct amdgpu_device *adev)
1556 {
1557 volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1558 uint32_t tmp;
1559 int i, r = 0;
1560
1561 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1562 if (adev->vcn.harvest_config & (1 << i))
1563 continue;
1564
1565 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1566 fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1567
1568 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1569 vcn_v4_0_stop_dpg_mode(adev, i);
1570 continue;
1571 }
1572
1573 /* wait for vcn idle */
1574 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1575 if (r)
1576 return r;
1577
1578 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1579 UVD_LMI_STATUS__READ_CLEAN_MASK |
1580 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1581 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1582 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1583 if (r)
1584 return r;
1585
1586 /* disable LMI UMC channel */
1587 tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1588 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1589 WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1590 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1591 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1592 r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1593 if (r)
1594 return r;
1595
1596 /* block VCPU register access */
1597 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1598 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1599 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1600
1601 /* reset VCPU */
1602 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1603 UVD_VCPU_CNTL__BLK_RST_MASK,
1604 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1605
1606 /* disable VCPU clock */
1607 WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1608 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1609
1610 /* apply soft reset */
1611 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1612 tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1613 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1614 tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1615 tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1616 WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1617
1618 /* clear status */
1619 WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1620
1621 /* apply HW clock gating */
1622 vcn_v4_0_enable_clock_gating(adev, i);
1623
1624 /* enable VCN power gating */
1625 vcn_v4_0_enable_static_power_gating(adev, i);
1626 }
1627
1628 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1629 if (adev->pm.dpm_enabled)
1630 amdgpu_dpm_enable_vcn(adev, false, i);
1631 }
1632
1633 return 0;
1634 }
1635
1636 /**
1637 * vcn_v4_0_pause_dpg_mode - VCN pause with dpg mode
1638 *
1639 * @adev: amdgpu_device pointer
1640 * @inst_idx: instance number index
1641 * @new_state: pause state
1642 *
1643 * Pause dpg mode for VCN block
1644 */
vcn_v4_0_pause_dpg_mode(struct amdgpu_device * adev,int inst_idx,struct dpg_pause_state * new_state)1645 static int vcn_v4_0_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1646 struct dpg_pause_state *new_state)
1647 {
1648 uint32_t reg_data = 0;
1649 int ret_code;
1650
1651 /* pause/unpause if state is changed */
1652 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1653 DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1654 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based);
1655 reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1656 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1657
1658 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1659 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1660 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1661
1662 if (!ret_code) {
1663 /* pause DPG */
1664 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1665 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1666
1667 /* wait for ACK */
1668 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1669 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1670 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1671
1672 SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1673 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1674 }
1675 } else {
1676 /* unpause dpg, no need to wait */
1677 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1678 WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1679 }
1680 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1681 }
1682
1683 return 0;
1684 }
1685
1686 /**
1687 * vcn_v4_0_unified_ring_get_rptr - get unified read pointer
1688 *
1689 * @ring: amdgpu_ring pointer
1690 *
1691 * Returns the current hardware unified read pointer
1692 */
vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring * ring)1693 static uint64_t vcn_v4_0_unified_ring_get_rptr(struct amdgpu_ring *ring)
1694 {
1695 struct amdgpu_device *adev = ring->adev;
1696
1697 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1698 DRM_ERROR("wrong ring id is identified in %s", __func__);
1699
1700 return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1701 }
1702
1703 /**
1704 * vcn_v4_0_unified_ring_get_wptr - get unified write pointer
1705 *
1706 * @ring: amdgpu_ring pointer
1707 *
1708 * Returns the current hardware unified write pointer
1709 */
vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring * ring)1710 static uint64_t vcn_v4_0_unified_ring_get_wptr(struct amdgpu_ring *ring)
1711 {
1712 struct amdgpu_device *adev = ring->adev;
1713
1714 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1715 DRM_ERROR("wrong ring id is identified in %s", __func__);
1716
1717 if (ring->use_doorbell)
1718 return *ring->wptr_cpu_addr;
1719 else
1720 return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1721 }
1722
1723 /**
1724 * vcn_v4_0_unified_ring_set_wptr - set enc write pointer
1725 *
1726 * @ring: amdgpu_ring pointer
1727 *
1728 * Commits the enc write pointer to the hardware
1729 */
vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring * ring)1730 static void vcn_v4_0_unified_ring_set_wptr(struct amdgpu_ring *ring)
1731 {
1732 struct amdgpu_device *adev = ring->adev;
1733
1734 if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1735 DRM_ERROR("wrong ring id is identified in %s", __func__);
1736
1737 if (ring->use_doorbell) {
1738 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1739 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1740 } else {
1741 WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1742 }
1743 }
1744
vcn_v4_0_limit_sched(struct amdgpu_cs_parser * p,struct amdgpu_job * job)1745 static int vcn_v4_0_limit_sched(struct amdgpu_cs_parser *p,
1746 struct amdgpu_job *job)
1747 {
1748 struct drm_gpu_scheduler **scheds;
1749
1750 /* The create msg must be in the first IB submitted */
1751 if (atomic_read(&job->base.entity->fence_seq))
1752 return -EINVAL;
1753
1754 /* if VCN0 is harvested, we can't support AV1 */
1755 if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1756 return -EINVAL;
1757
1758 scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1759 [AMDGPU_RING_PRIO_0].sched;
1760 drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1761 return 0;
1762 }
1763
vcn_v4_0_dec_msg(struct amdgpu_cs_parser * p,struct amdgpu_job * job,uint64_t addr)1764 static int vcn_v4_0_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1765 uint64_t addr)
1766 {
1767 struct ttm_operation_ctx ctx = { false, false };
1768 struct amdgpu_bo_va_mapping *map;
1769 uint32_t *msg, num_buffers;
1770 struct amdgpu_bo *bo;
1771 uint64_t start, end;
1772 unsigned int i;
1773 void *ptr;
1774 int r;
1775
1776 addr &= AMDGPU_GMC_HOLE_MASK;
1777 r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1778 if (r) {
1779 DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1780 return r;
1781 }
1782
1783 start = map->start * AMDGPU_GPU_PAGE_SIZE;
1784 end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1785 if (addr & 0x7) {
1786 DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1787 return -EINVAL;
1788 }
1789
1790 bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1791 amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1792 r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1793 if (r) {
1794 DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1795 return r;
1796 }
1797
1798 r = amdgpu_bo_kmap(bo, &ptr);
1799 if (r) {
1800 DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1801 return r;
1802 }
1803
1804 msg = ptr + addr - start;
1805
1806 /* Check length */
1807 if (msg[1] > end - addr) {
1808 r = -EINVAL;
1809 goto out;
1810 }
1811
1812 if (msg[3] != RDECODE_MSG_CREATE)
1813 goto out;
1814
1815 num_buffers = msg[2];
1816 for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1817 uint32_t offset, size, *create;
1818
1819 if (msg[0] != RDECODE_MESSAGE_CREATE)
1820 continue;
1821
1822 offset = msg[1];
1823 size = msg[2];
1824
1825 if (offset + size > end) {
1826 r = -EINVAL;
1827 goto out;
1828 }
1829
1830 create = ptr + addr + offset - start;
1831
1832 /* H264, HEVC and VP9 can run on any instance */
1833 if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1834 continue;
1835
1836 r = vcn_v4_0_limit_sched(p, job);
1837 if (r)
1838 goto out;
1839 }
1840
1841 out:
1842 amdgpu_bo_kunmap(bo);
1843 return r;
1844 }
1845
1846 #define RADEON_VCN_ENGINE_TYPE_ENCODE (0x00000002)
1847 #define RADEON_VCN_ENGINE_TYPE_DECODE (0x00000003)
1848
1849 #define RADEON_VCN_ENGINE_INFO (0x30000001)
1850 #define RADEON_VCN_ENGINE_INFO_MAX_OFFSET 16
1851
1852 #define RENCODE_ENCODE_STANDARD_AV1 2
1853 #define RENCODE_IB_PARAM_SESSION_INIT 0x00000003
1854 #define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET 64
1855
1856 /* return the offset in ib if id is found, -1 otherwise
1857 * to speed up the searching we only search upto max_offset
1858 */
vcn_v4_0_enc_find_ib_param(struct amdgpu_ib * ib,uint32_t id,int max_offset)1859 static int vcn_v4_0_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1860 {
1861 int i;
1862
1863 for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1864 if (ib->ptr[i + 1] == id)
1865 return i;
1866 }
1867 return -1;
1868 }
1869
vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser * p,struct amdgpu_job * job,struct amdgpu_ib * ib)1870 static int vcn_v4_0_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1871 struct amdgpu_job *job,
1872 struct amdgpu_ib *ib)
1873 {
1874 struct amdgpu_ring *ring = amdgpu_job_ring(job);
1875 struct amdgpu_vcn_decode_buffer *decode_buffer;
1876 uint64_t addr;
1877 uint32_t val;
1878 int idx;
1879
1880 /* The first instance can decode anything */
1881 if (!ring->me)
1882 return 0;
1883
1884 /* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1885 idx = vcn_v4_0_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1886 RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1887 if (idx < 0) /* engine info is missing */
1888 return 0;
1889
1890 val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1891 if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1892 decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1893
1894 if (!(decode_buffer->valid_buf_flag & 0x1))
1895 return 0;
1896
1897 addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1898 decode_buffer->msg_buffer_address_lo;
1899 return vcn_v4_0_dec_msg(p, job, addr);
1900 } else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1901 idx = vcn_v4_0_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1902 RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1903 if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1904 return vcn_v4_0_limit_sched(p, job);
1905 }
1906 return 0;
1907 }
1908
1909 static struct amdgpu_ring_funcs vcn_v4_0_unified_ring_vm_funcs = {
1910 .type = AMDGPU_RING_TYPE_VCN_ENC,
1911 .align_mask = 0x3f,
1912 .nop = VCN_ENC_CMD_NO_OP,
1913 .extra_dw = sizeof(struct amdgpu_vcn_rb_metadata),
1914 .get_rptr = vcn_v4_0_unified_ring_get_rptr,
1915 .get_wptr = vcn_v4_0_unified_ring_get_wptr,
1916 .set_wptr = vcn_v4_0_unified_ring_set_wptr,
1917 .patch_cs_in_place = vcn_v4_0_ring_patch_cs_in_place,
1918 .emit_frame_size =
1919 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1920 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1921 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1922 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1923 1, /* vcn_v2_0_enc_ring_insert_end */
1924 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1925 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1926 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1927 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1928 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1929 .test_ib = amdgpu_vcn_unified_ring_test_ib,
1930 .insert_nop = amdgpu_ring_insert_nop,
1931 .insert_end = vcn_v2_0_enc_ring_insert_end,
1932 .pad_ib = amdgpu_ring_generic_pad_ib,
1933 .begin_use = amdgpu_vcn_ring_begin_use,
1934 .end_use = amdgpu_vcn_ring_end_use,
1935 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1936 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1937 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1938 };
1939
1940 /**
1941 * vcn_v4_0_set_unified_ring_funcs - set unified ring functions
1942 *
1943 * @adev: amdgpu_device pointer
1944 *
1945 * Set unified ring functions
1946 */
vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device * adev)1947 static void vcn_v4_0_set_unified_ring_funcs(struct amdgpu_device *adev)
1948 {
1949 int i;
1950
1951 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1952 if (adev->vcn.harvest_config & (1 << i))
1953 continue;
1954
1955 if (amdgpu_ip_version(adev, VCN_HWIP, 0) == IP_VERSION(4, 0, 2))
1956 vcn_v4_0_unified_ring_vm_funcs.secure_submission_supported = true;
1957
1958 adev->vcn.inst[i].ring_enc[0].funcs =
1959 (const struct amdgpu_ring_funcs *)&vcn_v4_0_unified_ring_vm_funcs;
1960 adev->vcn.inst[i].ring_enc[0].me = i;
1961 }
1962 }
1963
1964 /**
1965 * vcn_v4_0_is_idle - check VCN block is idle
1966 *
1967 * @handle: amdgpu_device pointer
1968 *
1969 * Check whether VCN block is idle
1970 */
vcn_v4_0_is_idle(void * handle)1971 static bool vcn_v4_0_is_idle(void *handle)
1972 {
1973 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1974 int i, ret = 1;
1975
1976 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1977 if (adev->vcn.harvest_config & (1 << i))
1978 continue;
1979
1980 ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1981 }
1982
1983 return ret;
1984 }
1985
1986 /**
1987 * vcn_v4_0_wait_for_idle - wait for VCN block idle
1988 *
1989 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
1990 *
1991 * Wait for VCN block idle
1992 */
vcn_v4_0_wait_for_idle(struct amdgpu_ip_block * ip_block)1993 static int vcn_v4_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
1994 {
1995 struct amdgpu_device *adev = ip_block->adev;
1996 int i, ret = 0;
1997
1998 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1999 if (adev->vcn.harvest_config & (1 << i))
2000 continue;
2001
2002 ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
2003 UVD_STATUS__IDLE);
2004 if (ret)
2005 return ret;
2006 }
2007
2008 return ret;
2009 }
2010
2011 /**
2012 * vcn_v4_0_set_clockgating_state - set VCN block clockgating state
2013 *
2014 * @ip_block: amdgpu_ip_block pointer
2015 * @state: clock gating state
2016 *
2017 * Set VCN block clockgating state
2018 */
vcn_v4_0_set_clockgating_state(struct amdgpu_ip_block * ip_block,enum amd_clockgating_state state)2019 static int vcn_v4_0_set_clockgating_state(struct amdgpu_ip_block *ip_block,
2020 enum amd_clockgating_state state)
2021 {
2022 struct amdgpu_device *adev = ip_block->adev;
2023 bool enable = state == AMD_CG_STATE_GATE;
2024 int i;
2025
2026 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2027 if (adev->vcn.harvest_config & (1 << i))
2028 continue;
2029
2030 if (enable) {
2031 if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
2032 return -EBUSY;
2033 vcn_v4_0_enable_clock_gating(adev, i);
2034 } else {
2035 vcn_v4_0_disable_clock_gating(adev, i);
2036 }
2037 }
2038
2039 return 0;
2040 }
2041
2042 /**
2043 * vcn_v4_0_set_powergating_state - set VCN block powergating state
2044 *
2045 * @ip_block: amdgpu_ip_block pointer
2046 * @state: power gating state
2047 *
2048 * Set VCN block powergating state
2049 */
vcn_v4_0_set_powergating_state(struct amdgpu_ip_block * ip_block,enum amd_powergating_state state)2050 static int vcn_v4_0_set_powergating_state(struct amdgpu_ip_block *ip_block,
2051 enum amd_powergating_state state)
2052 {
2053 struct amdgpu_device *adev = ip_block->adev;
2054 int ret;
2055
2056 /* for SRIOV, guest should not control VCN Power-gating
2057 * MMSCH FW should control Power-gating and clock-gating
2058 * guest should avoid touching CGC and PG
2059 */
2060 if (amdgpu_sriov_vf(adev)) {
2061 adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
2062 return 0;
2063 }
2064
2065 if (state == adev->vcn.cur_state)
2066 return 0;
2067
2068 if (state == AMD_PG_STATE_GATE)
2069 ret = vcn_v4_0_stop(adev);
2070 else
2071 ret = vcn_v4_0_start(adev);
2072
2073 if (!ret)
2074 adev->vcn.cur_state = state;
2075
2076 return ret;
2077 }
2078
2079 /**
2080 * vcn_v4_0_set_ras_interrupt_state - set VCN block RAS interrupt state
2081 *
2082 * @adev: amdgpu_device pointer
2083 * @source: interrupt sources
2084 * @type: interrupt types
2085 * @state: interrupt states
2086 *
2087 * Set VCN block RAS interrupt state
2088 */
vcn_v4_0_set_ras_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned int type,enum amdgpu_interrupt_state state)2089 static int vcn_v4_0_set_ras_interrupt_state(struct amdgpu_device *adev,
2090 struct amdgpu_irq_src *source,
2091 unsigned int type,
2092 enum amdgpu_interrupt_state state)
2093 {
2094 return 0;
2095 }
2096
2097 /**
2098 * vcn_v4_0_process_interrupt - process VCN block interrupt
2099 *
2100 * @adev: amdgpu_device pointer
2101 * @source: interrupt sources
2102 * @entry: interrupt entry from clients and sources
2103 *
2104 * Process VCN block interrupt
2105 */
vcn_v4_0_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2106 static int vcn_v4_0_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
2107 struct amdgpu_iv_entry *entry)
2108 {
2109 uint32_t ip_instance;
2110
2111 if (amdgpu_sriov_is_vcn_rb_decouple(adev)) {
2112 ip_instance = entry->ring_id;
2113 } else {
2114 switch (entry->client_id) {
2115 case SOC15_IH_CLIENTID_VCN:
2116 ip_instance = 0;
2117 break;
2118 case SOC15_IH_CLIENTID_VCN1:
2119 ip_instance = 1;
2120 break;
2121 default:
2122 DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
2123 return 0;
2124 }
2125 }
2126
2127 DRM_DEBUG("IH: VCN TRAP\n");
2128
2129 switch (entry->src_id) {
2130 case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
2131 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
2132 break;
2133 default:
2134 DRM_ERROR("Unhandled interrupt: %d %d\n",
2135 entry->src_id, entry->src_data[0]);
2136 break;
2137 }
2138
2139 return 0;
2140 }
2141
2142 static const struct amdgpu_irq_src_funcs vcn_v4_0_irq_funcs = {
2143 .process = vcn_v4_0_process_interrupt,
2144 };
2145
2146 static const struct amdgpu_irq_src_funcs vcn_v4_0_ras_irq_funcs = {
2147 .set = vcn_v4_0_set_ras_interrupt_state,
2148 .process = amdgpu_vcn_process_poison_irq,
2149 };
2150
2151 /**
2152 * vcn_v4_0_set_irq_funcs - set VCN block interrupt irq functions
2153 *
2154 * @adev: amdgpu_device pointer
2155 *
2156 * Set VCN block interrupt irq functions
2157 */
vcn_v4_0_set_irq_funcs(struct amdgpu_device * adev)2158 static void vcn_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2159 {
2160 int i;
2161
2162 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
2163 if (adev->vcn.harvest_config & (1 << i))
2164 continue;
2165
2166 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
2167 adev->vcn.inst[i].irq.funcs = &vcn_v4_0_irq_funcs;
2168
2169 adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
2170 adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v4_0_ras_irq_funcs;
2171 }
2172 }
2173
vcn_v4_0_print_ip_state(struct amdgpu_ip_block * ip_block,struct drm_printer * p)2174 static void vcn_v4_0_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p)
2175 {
2176 struct amdgpu_device *adev = ip_block->adev;
2177 int i, j;
2178 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0);
2179 uint32_t inst_off, is_powered;
2180
2181 if (!adev->vcn.ip_dump)
2182 return;
2183
2184 drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst);
2185 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
2186 if (adev->vcn.harvest_config & (1 << i)) {
2187 drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i);
2188 continue;
2189 }
2190
2191 inst_off = i * reg_count;
2192 is_powered = (adev->vcn.ip_dump[inst_off] &
2193 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
2194
2195 if (is_powered) {
2196 drm_printf(p, "\nActive Instance:VCN%d\n", i);
2197 for (j = 0; j < reg_count; j++)
2198 drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_4_0[j].reg_name,
2199 adev->vcn.ip_dump[inst_off + j]);
2200 } else {
2201 drm_printf(p, "\nInactive Instance:VCN%d\n", i);
2202 }
2203 }
2204 }
2205
vcn_v4_0_dump_ip_state(struct amdgpu_ip_block * ip_block)2206 static void vcn_v4_0_dump_ip_state(struct amdgpu_ip_block *ip_block)
2207 {
2208 struct amdgpu_device *adev = ip_block->adev;
2209 int i, j;
2210 bool is_powered;
2211 uint32_t inst_off;
2212 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_4_0);
2213
2214 if (!adev->vcn.ip_dump)
2215 return;
2216
2217 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
2218 if (adev->vcn.harvest_config & (1 << i))
2219 continue;
2220
2221 inst_off = i * reg_count;
2222 /* mmUVD_POWER_STATUS is always readable and is first element of the array */
2223 adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, i, regUVD_POWER_STATUS);
2224 is_powered = (adev->vcn.ip_dump[inst_off] &
2225 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
2226
2227 if (is_powered)
2228 for (j = 1; j < reg_count; j++)
2229 adev->vcn.ip_dump[inst_off + j] =
2230 RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_4_0[j],
2231 i));
2232 }
2233 }
2234
2235 static const struct amd_ip_funcs vcn_v4_0_ip_funcs = {
2236 .name = "vcn_v4_0",
2237 .early_init = vcn_v4_0_early_init,
2238 .sw_init = vcn_v4_0_sw_init,
2239 .sw_fini = vcn_v4_0_sw_fini,
2240 .hw_init = vcn_v4_0_hw_init,
2241 .hw_fini = vcn_v4_0_hw_fini,
2242 .suspend = vcn_v4_0_suspend,
2243 .resume = vcn_v4_0_resume,
2244 .is_idle = vcn_v4_0_is_idle,
2245 .wait_for_idle = vcn_v4_0_wait_for_idle,
2246 .set_clockgating_state = vcn_v4_0_set_clockgating_state,
2247 .set_powergating_state = vcn_v4_0_set_powergating_state,
2248 .dump_ip_state = vcn_v4_0_dump_ip_state,
2249 .print_ip_state = vcn_v4_0_print_ip_state,
2250 };
2251
2252 const struct amdgpu_ip_block_version vcn_v4_0_ip_block = {
2253 .type = AMD_IP_BLOCK_TYPE_VCN,
2254 .major = 4,
2255 .minor = 0,
2256 .rev = 0,
2257 .funcs = &vcn_v4_0_ip_funcs,
2258 };
2259
vcn_v4_0_query_poison_by_instance(struct amdgpu_device * adev,uint32_t instance,uint32_t sub_block)2260 static uint32_t vcn_v4_0_query_poison_by_instance(struct amdgpu_device *adev,
2261 uint32_t instance, uint32_t sub_block)
2262 {
2263 uint32_t poison_stat = 0, reg_value = 0;
2264
2265 switch (sub_block) {
2266 case AMDGPU_VCN_V4_0_VCPU_VCODEC:
2267 reg_value = RREG32_SOC15(VCN, instance, regUVD_RAS_VCPU_VCODEC_STATUS);
2268 poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2269 break;
2270 default:
2271 break;
2272 }
2273
2274 if (poison_stat)
2275 dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2276 instance, sub_block);
2277
2278 return poison_stat;
2279 }
2280
vcn_v4_0_query_ras_poison_status(struct amdgpu_device * adev)2281 static bool vcn_v4_0_query_ras_poison_status(struct amdgpu_device *adev)
2282 {
2283 uint32_t inst, sub;
2284 uint32_t poison_stat = 0;
2285
2286 for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2287 for (sub = 0; sub < AMDGPU_VCN_V4_0_MAX_SUB_BLOCK; sub++)
2288 poison_stat +=
2289 vcn_v4_0_query_poison_by_instance(adev, inst, sub);
2290
2291 return !!poison_stat;
2292 }
2293
2294 const struct amdgpu_ras_block_hw_ops vcn_v4_0_ras_hw_ops = {
2295 .query_poison_status = vcn_v4_0_query_ras_poison_status,
2296 };
2297
2298 static struct amdgpu_vcn_ras vcn_v4_0_ras = {
2299 .ras_block = {
2300 .hw_ops = &vcn_v4_0_ras_hw_ops,
2301 .ras_late_init = amdgpu_vcn_ras_late_init,
2302 },
2303 };
2304
vcn_v4_0_set_ras_funcs(struct amdgpu_device * adev)2305 static void vcn_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2306 {
2307 switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
2308 case IP_VERSION(4, 0, 0):
2309 adev->vcn.ras = &vcn_v4_0_ras;
2310 break;
2311 default:
2312 break;
2313 }
2314 }
2315