xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c (revision 12cecbf9)
1 /*
2  * Copyright 2019 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <drm/drm_drv.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_vcn.h"
29 #include "amdgpu_pm.h"
30 #include "soc15.h"
31 #include "soc15d.h"
32 #include "vcn_v2_0.h"
33 #include "mmsch_v1_0.h"
34 #include "vcn_v2_5.h"
35 
36 #include "vcn/vcn_2_5_offset.h"
37 #include "vcn/vcn_2_5_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
39 
40 #define VCN_VID_SOC_ADDRESS_2_0					0x1fa00
41 #define VCN1_VID_SOC_ADDRESS_3_0				0x48200
42 
43 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
44 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
45 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
46 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
47 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
48 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
49 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
50 
51 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
52 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
53 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x3b5
54 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
55 
56 #define VCN25_MAX_HW_INSTANCES_ARCTURUS			2
57 
58 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
59 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
60 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
61 static int vcn_v2_5_set_powergating_state(void *handle,
62 				enum amd_powergating_state state);
63 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
64 				int inst_idx, struct dpg_pause_state *new_state);
65 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
66 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev);
67 
68 static int amdgpu_ih_clientid_vcns[] = {
69 	SOC15_IH_CLIENTID_VCN,
70 	SOC15_IH_CLIENTID_VCN1
71 };
72 
73 /**
74  * vcn_v2_5_early_init - set function pointers
75  *
76  * @handle: amdgpu_device pointer
77  *
78  * Set ring and irq function pointers
79  */
80 static int vcn_v2_5_early_init(void *handle)
81 {
82 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
83 
84 	if (amdgpu_sriov_vf(adev)) {
85 		adev->vcn.num_vcn_inst = 2;
86 		adev->vcn.harvest_config = 0;
87 		adev->vcn.num_enc_rings = 1;
88 	} else {
89 		u32 harvest;
90 		int i;
91 
92 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
93 			harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
94 			if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
95 				adev->vcn.harvest_config |= 1 << i;
96 		}
97 		if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
98 					AMDGPU_VCN_HARVEST_VCN1))
99 			/* both instances are harvested, disable the block */
100 			return -ENOENT;
101 
102 		adev->vcn.num_enc_rings = 2;
103 	}
104 
105 	vcn_v2_5_set_dec_ring_funcs(adev);
106 	vcn_v2_5_set_enc_ring_funcs(adev);
107 	vcn_v2_5_set_irq_funcs(adev);
108 	vcn_v2_5_set_ras_funcs(adev);
109 
110 	return 0;
111 }
112 
113 /**
114  * vcn_v2_5_sw_init - sw init for VCN block
115  *
116  * @handle: amdgpu_device pointer
117  *
118  * Load firmware and sw initialization
119  */
120 static int vcn_v2_5_sw_init(void *handle)
121 {
122 	struct amdgpu_ring *ring;
123 	int i, j, r;
124 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
125 
126 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
127 		if (adev->vcn.harvest_config & (1 << j))
128 			continue;
129 		/* VCN DEC TRAP */
130 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
131 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
132 		if (r)
133 			return r;
134 
135 		/* VCN ENC TRAP */
136 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
137 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
138 				i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
139 			if (r)
140 				return r;
141 		}
142 
143 		/* VCN POISON TRAP */
144 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
145 			VCN_2_6__SRCID_UVD_POISON, &adev->vcn.inst[j].irq);
146 		if (r)
147 			return r;
148 	}
149 
150 	r = amdgpu_vcn_sw_init(adev);
151 	if (r)
152 		return r;
153 
154 	amdgpu_vcn_setup_ucode(adev);
155 
156 	r = amdgpu_vcn_resume(adev);
157 	if (r)
158 		return r;
159 
160 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
161 		volatile struct amdgpu_fw_shared *fw_shared;
162 
163 		if (adev->vcn.harvest_config & (1 << j))
164 			continue;
165 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
166 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
167 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
168 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
169 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
170 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
171 
172 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
173 		adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
174 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
175 		adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
176 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
177 		adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
178 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
179 		adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
180 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
181 		adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
182 
183 		ring = &adev->vcn.inst[j].ring_dec;
184 		ring->use_doorbell = true;
185 
186 		ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
187 				(amdgpu_sriov_vf(adev) ? 2*j : 8*j);
188 		sprintf(ring->name, "vcn_dec_%d", j);
189 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
190 				     0, AMDGPU_RING_PRIO_DEFAULT, NULL);
191 		if (r)
192 			return r;
193 
194 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
195 			enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
196 
197 			ring = &adev->vcn.inst[j].ring_enc[i];
198 			ring->use_doorbell = true;
199 
200 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
201 					(amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
202 
203 			sprintf(ring->name, "vcn_enc_%d.%d", j, i);
204 			r = amdgpu_ring_init(adev, ring, 512,
205 					     &adev->vcn.inst[j].irq, 0,
206 					     hw_prio, NULL);
207 			if (r)
208 				return r;
209 		}
210 
211 		fw_shared = adev->vcn.inst[j].fw_shared.cpu_addr;
212 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
213 
214 		if (amdgpu_vcnfw_log)
215 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
216 	}
217 
218 	if (amdgpu_sriov_vf(adev)) {
219 		r = amdgpu_virt_alloc_mm_table(adev);
220 		if (r)
221 			return r;
222 	}
223 
224 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
225 		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
226 
227 	return 0;
228 }
229 
230 /**
231  * vcn_v2_5_sw_fini - sw fini for VCN block
232  *
233  * @handle: amdgpu_device pointer
234  *
235  * VCN suspend and free up sw allocation
236  */
237 static int vcn_v2_5_sw_fini(void *handle)
238 {
239 	int i, r, idx;
240 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
241 	volatile struct amdgpu_fw_shared *fw_shared;
242 
243 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
244 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
245 			if (adev->vcn.harvest_config & (1 << i))
246 				continue;
247 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
248 			fw_shared->present_flag_0 = 0;
249 		}
250 		drm_dev_exit(idx);
251 	}
252 
253 
254 	if (amdgpu_sriov_vf(adev))
255 		amdgpu_virt_free_mm_table(adev);
256 
257 	r = amdgpu_vcn_suspend(adev);
258 	if (r)
259 		return r;
260 
261 	r = amdgpu_vcn_sw_fini(adev);
262 
263 	return r;
264 }
265 
266 /**
267  * vcn_v2_5_hw_init - start and test VCN block
268  *
269  * @handle: amdgpu_device pointer
270  *
271  * Initialize the hardware, boot up the VCPU and do some testing
272  */
273 static int vcn_v2_5_hw_init(void *handle)
274 {
275 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
276 	struct amdgpu_ring *ring;
277 	int i, j, r = 0;
278 
279 	if (amdgpu_sriov_vf(adev))
280 		r = vcn_v2_5_sriov_start(adev);
281 
282 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
283 		if (adev->vcn.harvest_config & (1 << j))
284 			continue;
285 
286 		if (amdgpu_sriov_vf(adev)) {
287 			adev->vcn.inst[j].ring_enc[0].sched.ready = true;
288 			adev->vcn.inst[j].ring_enc[1].sched.ready = false;
289 			adev->vcn.inst[j].ring_enc[2].sched.ready = false;
290 			adev->vcn.inst[j].ring_dec.sched.ready = true;
291 		} else {
292 
293 			ring = &adev->vcn.inst[j].ring_dec;
294 
295 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
296 						     ring->doorbell_index, j);
297 
298 			r = amdgpu_ring_test_helper(ring);
299 			if (r)
300 				goto done;
301 
302 			for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
303 				ring = &adev->vcn.inst[j].ring_enc[i];
304 				r = amdgpu_ring_test_helper(ring);
305 				if (r)
306 					goto done;
307 			}
308 		}
309 	}
310 
311 done:
312 	if (!r)
313 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
314 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
315 
316 	return r;
317 }
318 
319 /**
320  * vcn_v2_5_hw_fini - stop the hardware block
321  *
322  * @handle: amdgpu_device pointer
323  *
324  * Stop the VCN block, mark ring as not ready any more
325  */
326 static int vcn_v2_5_hw_fini(void *handle)
327 {
328 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
329 	int i;
330 
331 	cancel_delayed_work_sync(&adev->vcn.idle_work);
332 
333 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
334 		if (adev->vcn.harvest_config & (1 << i))
335 			continue;
336 
337 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
338 		    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
339 		     RREG32_SOC15(VCN, i, mmUVD_STATUS)))
340 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
341 	}
342 
343 	return 0;
344 }
345 
346 /**
347  * vcn_v2_5_suspend - suspend VCN block
348  *
349  * @handle: amdgpu_device pointer
350  *
351  * HW fini and suspend VCN block
352  */
353 static int vcn_v2_5_suspend(void *handle)
354 {
355 	int r;
356 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
357 
358 	r = vcn_v2_5_hw_fini(adev);
359 	if (r)
360 		return r;
361 
362 	r = amdgpu_vcn_suspend(adev);
363 
364 	return r;
365 }
366 
367 /**
368  * vcn_v2_5_resume - resume VCN block
369  *
370  * @handle: amdgpu_device pointer
371  *
372  * Resume firmware and hw init VCN block
373  */
374 static int vcn_v2_5_resume(void *handle)
375 {
376 	int r;
377 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
378 
379 	r = amdgpu_vcn_resume(adev);
380 	if (r)
381 		return r;
382 
383 	r = vcn_v2_5_hw_init(adev);
384 
385 	return r;
386 }
387 
388 /**
389  * vcn_v2_5_mc_resume - memory controller programming
390  *
391  * @adev: amdgpu_device pointer
392  *
393  * Let the VCN memory controller know it's offsets
394  */
395 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
396 {
397 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
398 	uint32_t offset;
399 	int i;
400 
401 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
402 		if (adev->vcn.harvest_config & (1 << i))
403 			continue;
404 		/* cache window 0: fw */
405 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
406 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
407 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
408 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
409 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
410 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
411 			offset = 0;
412 		} else {
413 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
414 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
415 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
416 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
417 			offset = size;
418 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
419 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
420 		}
421 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
422 
423 		/* cache window 1: stack */
424 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
425 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
426 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
427 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
428 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
429 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
430 
431 		/* cache window 2: context */
432 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
433 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
434 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
435 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
436 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
437 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
438 
439 		/* non-cache window */
440 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
441 			lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
442 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
443 			upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
444 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
445 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
446 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
447 	}
448 }
449 
450 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
451 {
452 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
453 	uint32_t offset;
454 
455 	/* cache window 0: fw */
456 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
457 		if (!indirect) {
458 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
459 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
460 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
461 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
462 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
463 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
464 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
466 		} else {
467 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
468 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
469 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
470 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
471 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
472 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
473 		}
474 		offset = 0;
475 	} else {
476 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
478 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
479 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
480 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
481 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
482 		offset = size;
483 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
484 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
485 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
486 	}
487 
488 	if (!indirect)
489 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
490 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
491 	else
492 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
493 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
494 
495 	/* cache window 1: stack */
496 	if (!indirect) {
497 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
498 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
499 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
500 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
501 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
502 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
503 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
504 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
505 	} else {
506 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
507 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
508 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
509 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
510 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
511 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
512 	}
513 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
514 		VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
515 
516 	/* cache window 2: context */
517 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
518 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
519 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
520 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
521 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
522 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
523 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
524 		VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
525 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
526 		VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
527 
528 	/* non-cache window */
529 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
530 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
531 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
532 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
533 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
534 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
535 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
536 		VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
537 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
538 		VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
539 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
540 
541 	/* VCN global tiling registers */
542 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
543 		VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
544 }
545 
546 /**
547  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
548  *
549  * @adev: amdgpu_device pointer
550  *
551  * Disable clock gating for VCN block
552  */
553 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
554 {
555 	uint32_t data;
556 	int i;
557 
558 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
559 		if (adev->vcn.harvest_config & (1 << i))
560 			continue;
561 		/* UVD disable CGC */
562 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
563 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
564 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
565 		else
566 			data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
567 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
568 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
569 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
570 
571 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
572 		data &= ~(UVD_CGC_GATE__SYS_MASK
573 			| UVD_CGC_GATE__UDEC_MASK
574 			| UVD_CGC_GATE__MPEG2_MASK
575 			| UVD_CGC_GATE__REGS_MASK
576 			| UVD_CGC_GATE__RBC_MASK
577 			| UVD_CGC_GATE__LMI_MC_MASK
578 			| UVD_CGC_GATE__LMI_UMC_MASK
579 			| UVD_CGC_GATE__IDCT_MASK
580 			| UVD_CGC_GATE__MPRD_MASK
581 			| UVD_CGC_GATE__MPC_MASK
582 			| UVD_CGC_GATE__LBSI_MASK
583 			| UVD_CGC_GATE__LRBBM_MASK
584 			| UVD_CGC_GATE__UDEC_RE_MASK
585 			| UVD_CGC_GATE__UDEC_CM_MASK
586 			| UVD_CGC_GATE__UDEC_IT_MASK
587 			| UVD_CGC_GATE__UDEC_DB_MASK
588 			| UVD_CGC_GATE__UDEC_MP_MASK
589 			| UVD_CGC_GATE__WCB_MASK
590 			| UVD_CGC_GATE__VCPU_MASK
591 			| UVD_CGC_GATE__MMSCH_MASK);
592 
593 		WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
594 
595 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0,  0xFFFFFFFF);
596 
597 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
598 		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
599 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
600 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
601 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
602 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
603 			| UVD_CGC_CTRL__SYS_MODE_MASK
604 			| UVD_CGC_CTRL__UDEC_MODE_MASK
605 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
606 			| UVD_CGC_CTRL__REGS_MODE_MASK
607 			| UVD_CGC_CTRL__RBC_MODE_MASK
608 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
609 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
610 			| UVD_CGC_CTRL__IDCT_MODE_MASK
611 			| UVD_CGC_CTRL__MPRD_MODE_MASK
612 			| UVD_CGC_CTRL__MPC_MODE_MASK
613 			| UVD_CGC_CTRL__LBSI_MODE_MASK
614 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
615 			| UVD_CGC_CTRL__WCB_MODE_MASK
616 			| UVD_CGC_CTRL__VCPU_MODE_MASK
617 			| UVD_CGC_CTRL__MMSCH_MODE_MASK);
618 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
619 
620 		/* turn on */
621 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
622 		data |= (UVD_SUVD_CGC_GATE__SRE_MASK
623 			| UVD_SUVD_CGC_GATE__SIT_MASK
624 			| UVD_SUVD_CGC_GATE__SMP_MASK
625 			| UVD_SUVD_CGC_GATE__SCM_MASK
626 			| UVD_SUVD_CGC_GATE__SDB_MASK
627 			| UVD_SUVD_CGC_GATE__SRE_H264_MASK
628 			| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
629 			| UVD_SUVD_CGC_GATE__SIT_H264_MASK
630 			| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
631 			| UVD_SUVD_CGC_GATE__SCM_H264_MASK
632 			| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
633 			| UVD_SUVD_CGC_GATE__SDB_H264_MASK
634 			| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
635 			| UVD_SUVD_CGC_GATE__SCLR_MASK
636 			| UVD_SUVD_CGC_GATE__UVD_SC_MASK
637 			| UVD_SUVD_CGC_GATE__ENT_MASK
638 			| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
639 			| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
640 			| UVD_SUVD_CGC_GATE__SITE_MASK
641 			| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
642 			| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
643 			| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
644 			| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
645 			| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
646 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
647 
648 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
649 		data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
650 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
651 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
652 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
653 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
654 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
655 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
656 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
657 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
658 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
659 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
660 	}
661 }
662 
663 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
664 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
665 {
666 	uint32_t reg_data = 0;
667 
668 	/* enable sw clock gating control */
669 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
670 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
671 	else
672 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
673 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
674 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
675 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
676 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
677 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
678 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
679 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
680 		 UVD_CGC_CTRL__SYS_MODE_MASK |
681 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
682 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
683 		 UVD_CGC_CTRL__REGS_MODE_MASK |
684 		 UVD_CGC_CTRL__RBC_MODE_MASK |
685 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
686 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
687 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
688 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
689 		 UVD_CGC_CTRL__MPC_MODE_MASK |
690 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
691 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
692 		 UVD_CGC_CTRL__WCB_MODE_MASK |
693 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
694 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
695 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
696 		VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
697 
698 	/* turn off clock gating */
699 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
700 		VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
701 
702 	/* turn on SUVD clock gating */
703 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
704 		VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
705 
706 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
707 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
708 		VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
709 }
710 
711 /**
712  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
713  *
714  * @adev: amdgpu_device pointer
715  *
716  * Enable clock gating for VCN block
717  */
718 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
719 {
720 	uint32_t data = 0;
721 	int i;
722 
723 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
724 		if (adev->vcn.harvest_config & (1 << i))
725 			continue;
726 		/* enable UVD CGC */
727 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
728 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
729 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
730 		else
731 			data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
732 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
733 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
734 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
735 
736 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
737 		data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
738 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
739 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
740 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
741 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
742 			| UVD_CGC_CTRL__SYS_MODE_MASK
743 			| UVD_CGC_CTRL__UDEC_MODE_MASK
744 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
745 			| UVD_CGC_CTRL__REGS_MODE_MASK
746 			| UVD_CGC_CTRL__RBC_MODE_MASK
747 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
748 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
749 			| UVD_CGC_CTRL__IDCT_MODE_MASK
750 			| UVD_CGC_CTRL__MPRD_MODE_MASK
751 			| UVD_CGC_CTRL__MPC_MODE_MASK
752 			| UVD_CGC_CTRL__LBSI_MODE_MASK
753 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
754 			| UVD_CGC_CTRL__WCB_MODE_MASK
755 			| UVD_CGC_CTRL__VCPU_MODE_MASK);
756 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
757 
758 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
759 		data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
760 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
761 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
762 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
763 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
764 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
765 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
766 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
767 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
768 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
769 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
770 	}
771 }
772 
773 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
774 {
775 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
776 	struct amdgpu_ring *ring;
777 	uint32_t rb_bufsz, tmp;
778 
779 	/* disable register anti-hang mechanism */
780 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
781 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
782 	/* enable dynamic power gating mode */
783 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
784 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
785 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
786 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
787 
788 	if (indirect)
789 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
790 
791 	/* enable clock gating */
792 	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
793 
794 	/* enable VCPU clock */
795 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
796 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
797 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
798 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
799 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
800 
801 	/* disable master interupt */
802 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
803 		VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
804 
805 	/* setup mmUVD_LMI_CTRL */
806 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
807 		UVD_LMI_CTRL__REQ_MODE_MASK |
808 		UVD_LMI_CTRL__CRC_RESET_MASK |
809 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
810 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
811 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
812 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
813 		0x00100000L);
814 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
815 		VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
816 
817 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
818 		VCN, 0, mmUVD_MPC_CNTL),
819 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
820 
821 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
822 		VCN, 0, mmUVD_MPC_SET_MUXA0),
823 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
824 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
825 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
826 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
827 
828 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
829 		VCN, 0, mmUVD_MPC_SET_MUXB0),
830 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
831 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
832 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
833 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
834 
835 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
836 		VCN, 0, mmUVD_MPC_SET_MUX),
837 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
838 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
839 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
840 
841 	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
842 
843 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
844 		VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
845 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
846 		VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
847 
848 	/* enable LMI MC and UMC channels */
849 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
850 		VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
851 
852 	/* unblock VCPU register access */
853 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
854 		VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
855 
856 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
857 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
858 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
859 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
860 
861 	/* enable master interrupt */
862 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
863 		VCN, 0, mmUVD_MASTINT_EN),
864 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
865 
866 	if (indirect)
867 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
868 				    (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
869 					       (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
870 
871 	ring = &adev->vcn.inst[inst_idx].ring_dec;
872 	/* force RBC into idle state */
873 	rb_bufsz = order_base_2(ring->ring_size);
874 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
875 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
876 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
877 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
878 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
879 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
880 
881 	/* Stall DPG before WPTR/RPTR reset */
882 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
883 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
884 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
885 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
886 
887 	/* set the write pointer delay */
888 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
889 
890 	/* set the wb address */
891 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
892 		(upper_32_bits(ring->gpu_addr) >> 2));
893 
894 	/* program the RB_BASE for ring buffer */
895 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
896 		lower_32_bits(ring->gpu_addr));
897 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
898 		upper_32_bits(ring->gpu_addr));
899 
900 	/* Initialize the ring buffer's read and write pointers */
901 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
902 
903 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
904 
905 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
906 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
907 		lower_32_bits(ring->wptr));
908 
909 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
910 	/* Unstall DPG */
911 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
912 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
913 
914 	return 0;
915 }
916 
917 static int vcn_v2_5_start(struct amdgpu_device *adev)
918 {
919 	struct amdgpu_ring *ring;
920 	uint32_t rb_bufsz, tmp;
921 	int i, j, k, r;
922 
923 	if (adev->pm.dpm_enabled)
924 		amdgpu_dpm_enable_uvd(adev, true);
925 
926 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
927 		if (adev->vcn.harvest_config & (1 << i))
928 			continue;
929 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
930 			r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
931 			continue;
932 		}
933 
934 		/* disable register anti-hang mechanism */
935 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
936 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
937 
938 		/* set uvd status busy */
939 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
940 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
941 	}
942 
943 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
944 		return 0;
945 
946 	/*SW clock gating */
947 	vcn_v2_5_disable_clock_gating(adev);
948 
949 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
950 		if (adev->vcn.harvest_config & (1 << i))
951 			continue;
952 		/* enable VCPU clock */
953 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
954 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
955 
956 		/* disable master interrupt */
957 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
958 			~UVD_MASTINT_EN__VCPU_EN_MASK);
959 
960 		/* setup mmUVD_LMI_CTRL */
961 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
962 		tmp &= ~0xff;
963 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
964 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
965 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
966 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
967 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
968 
969 		/* setup mmUVD_MPC_CNTL */
970 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
971 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
972 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
973 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
974 
975 		/* setup UVD_MPC_SET_MUXA0 */
976 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
977 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
978 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
979 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
980 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
981 
982 		/* setup UVD_MPC_SET_MUXB0 */
983 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
984 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
985 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
986 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
987 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
988 
989 		/* setup mmUVD_MPC_SET_MUX */
990 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
991 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
992 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
993 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
994 	}
995 
996 	vcn_v2_5_mc_resume(adev);
997 
998 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
999 		volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1000 		if (adev->vcn.harvest_config & (1 << i))
1001 			continue;
1002 		/* VCN global tiling registers */
1003 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1004 			adev->gfx.config.gb_addr_config);
1005 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1006 			adev->gfx.config.gb_addr_config);
1007 
1008 		/* enable LMI MC and UMC channels */
1009 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1010 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1011 
1012 		/* unblock VCPU register access */
1013 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1014 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1015 
1016 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1017 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1018 
1019 		for (k = 0; k < 10; ++k) {
1020 			uint32_t status;
1021 
1022 			for (j = 0; j < 100; ++j) {
1023 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1024 				if (status & 2)
1025 					break;
1026 				if (amdgpu_emu_mode == 1)
1027 					msleep(500);
1028 				else
1029 					mdelay(10);
1030 			}
1031 			r = 0;
1032 			if (status & 2)
1033 				break;
1034 
1035 			DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1036 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1037 				UVD_VCPU_CNTL__BLK_RST_MASK,
1038 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1039 			mdelay(10);
1040 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1041 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1042 
1043 			mdelay(10);
1044 			r = -1;
1045 		}
1046 
1047 		if (r) {
1048 			DRM_ERROR("VCN decode not responding, giving up!!!\n");
1049 			return r;
1050 		}
1051 
1052 		/* enable master interrupt */
1053 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1054 			UVD_MASTINT_EN__VCPU_EN_MASK,
1055 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1056 
1057 		/* clear the busy bit of VCN_STATUS */
1058 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1059 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1060 
1061 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1062 
1063 		ring = &adev->vcn.inst[i].ring_dec;
1064 		/* force RBC into idle state */
1065 		rb_bufsz = order_base_2(ring->ring_size);
1066 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1067 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1068 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1069 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1070 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1071 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1072 
1073 		fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1074 		/* program the RB_BASE for ring buffer */
1075 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1076 			lower_32_bits(ring->gpu_addr));
1077 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1078 			upper_32_bits(ring->gpu_addr));
1079 
1080 		/* Initialize the ring buffer's read and write pointers */
1081 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1082 
1083 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1084 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1085 				lower_32_bits(ring->wptr));
1086 		fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1087 
1088 		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1089 		ring = &adev->vcn.inst[i].ring_enc[0];
1090 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1091 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1092 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1093 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1094 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1095 		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1096 
1097 		fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1098 		ring = &adev->vcn.inst[i].ring_enc[1];
1099 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1100 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1101 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1102 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1103 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1104 		fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1105 	}
1106 
1107 	return 0;
1108 }
1109 
1110 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1111 				struct amdgpu_mm_table *table)
1112 {
1113 	uint32_t data = 0, loop = 0, size = 0;
1114 	uint64_t addr = table->gpu_addr;
1115 	struct mmsch_v1_1_init_header *header = NULL;
1116 
1117 	header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1118 	size = header->total_size;
1119 
1120 	/*
1121 	 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1122 	 *  memory descriptor location
1123 	 */
1124 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1125 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1126 
1127 	/* 2, update vmid of descriptor */
1128 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1129 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1130 	/* use domain0 for MM scheduler */
1131 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1132 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1133 
1134 	/* 3, notify mmsch about the size of this descriptor */
1135 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1136 
1137 	/* 4, set resp to zero */
1138 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1139 
1140 	/*
1141 	 * 5, kick off the initialization and wait until
1142 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1143 	 */
1144 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1145 
1146 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1147 	loop = 10;
1148 	while ((data & 0x10000002) != 0x10000002) {
1149 		udelay(100);
1150 		data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1151 		loop--;
1152 		if (!loop)
1153 			break;
1154 	}
1155 
1156 	if (!loop) {
1157 		dev_err(adev->dev,
1158 			"failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1159 			data);
1160 		return -EBUSY;
1161 	}
1162 
1163 	return 0;
1164 }
1165 
1166 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1167 {
1168 	struct amdgpu_ring *ring;
1169 	uint32_t offset, size, tmp, i, rb_bufsz;
1170 	uint32_t table_size = 0;
1171 	struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1172 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1173 	struct mmsch_v1_0_cmd_end end = { { 0 } };
1174 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1175 	struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1176 
1177 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1178 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1179 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1180 
1181 	header->version = MMSCH_VERSION;
1182 	header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1183 	init_table += header->total_size;
1184 
1185 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1186 		header->eng[i].table_offset = header->total_size;
1187 		header->eng[i].init_status = 0;
1188 		header->eng[i].table_size = 0;
1189 
1190 		table_size = 0;
1191 
1192 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1193 			SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1194 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1195 
1196 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1197 		/* mc resume*/
1198 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1199 			MMSCH_V1_0_INSERT_DIRECT_WT(
1200 				SOC15_REG_OFFSET(VCN, i,
1201 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1202 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1203 			MMSCH_V1_0_INSERT_DIRECT_WT(
1204 				SOC15_REG_OFFSET(VCN, i,
1205 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1206 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1207 			offset = 0;
1208 			MMSCH_V1_0_INSERT_DIRECT_WT(
1209 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1210 		} else {
1211 			MMSCH_V1_0_INSERT_DIRECT_WT(
1212 				SOC15_REG_OFFSET(VCN, i,
1213 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1214 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1215 			MMSCH_V1_0_INSERT_DIRECT_WT(
1216 				SOC15_REG_OFFSET(VCN, i,
1217 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1218 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1219 			offset = size;
1220 			MMSCH_V1_0_INSERT_DIRECT_WT(
1221 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1222 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1223 		}
1224 
1225 		MMSCH_V1_0_INSERT_DIRECT_WT(
1226 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1227 			size);
1228 		MMSCH_V1_0_INSERT_DIRECT_WT(
1229 			SOC15_REG_OFFSET(VCN, i,
1230 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1231 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1232 		MMSCH_V1_0_INSERT_DIRECT_WT(
1233 			SOC15_REG_OFFSET(VCN, i,
1234 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1235 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1236 		MMSCH_V1_0_INSERT_DIRECT_WT(
1237 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1238 			0);
1239 		MMSCH_V1_0_INSERT_DIRECT_WT(
1240 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1241 			AMDGPU_VCN_STACK_SIZE);
1242 		MMSCH_V1_0_INSERT_DIRECT_WT(
1243 			SOC15_REG_OFFSET(VCN, i,
1244 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1245 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1246 				AMDGPU_VCN_STACK_SIZE));
1247 		MMSCH_V1_0_INSERT_DIRECT_WT(
1248 			SOC15_REG_OFFSET(VCN, i,
1249 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1250 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1251 				AMDGPU_VCN_STACK_SIZE));
1252 		MMSCH_V1_0_INSERT_DIRECT_WT(
1253 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1254 			0);
1255 		MMSCH_V1_0_INSERT_DIRECT_WT(
1256 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1257 			AMDGPU_VCN_CONTEXT_SIZE);
1258 
1259 		ring = &adev->vcn.inst[i].ring_enc[0];
1260 		ring->wptr = 0;
1261 
1262 		MMSCH_V1_0_INSERT_DIRECT_WT(
1263 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1264 			lower_32_bits(ring->gpu_addr));
1265 		MMSCH_V1_0_INSERT_DIRECT_WT(
1266 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1267 			upper_32_bits(ring->gpu_addr));
1268 		MMSCH_V1_0_INSERT_DIRECT_WT(
1269 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1270 			ring->ring_size / 4);
1271 
1272 		ring = &adev->vcn.inst[i].ring_dec;
1273 		ring->wptr = 0;
1274 		MMSCH_V1_0_INSERT_DIRECT_WT(
1275 			SOC15_REG_OFFSET(VCN, i,
1276 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1277 			lower_32_bits(ring->gpu_addr));
1278 		MMSCH_V1_0_INSERT_DIRECT_WT(
1279 			SOC15_REG_OFFSET(VCN, i,
1280 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1281 			upper_32_bits(ring->gpu_addr));
1282 
1283 		/* force RBC into idle state */
1284 		rb_bufsz = order_base_2(ring->ring_size);
1285 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1286 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1287 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1288 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1289 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1290 		MMSCH_V1_0_INSERT_DIRECT_WT(
1291 			SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1292 
1293 		/* add end packet */
1294 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1295 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1296 		init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1297 
1298 		/* refine header */
1299 		header->eng[i].table_size = table_size;
1300 		header->total_size += table_size;
1301 	}
1302 
1303 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1304 }
1305 
1306 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1307 {
1308 	uint32_t tmp;
1309 
1310 	/* Wait for power status to be 1 */
1311 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1312 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1313 
1314 	/* wait for read ptr to be equal to write ptr */
1315 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1316 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1317 
1318 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1319 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1320 
1321 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1322 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1323 
1324 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1325 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1326 
1327 	/* disable dynamic power gating mode */
1328 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1329 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1330 
1331 	return 0;
1332 }
1333 
1334 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1335 {
1336 	uint32_t tmp;
1337 	int i, r = 0;
1338 
1339 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1340 		if (adev->vcn.harvest_config & (1 << i))
1341 			continue;
1342 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1343 			r = vcn_v2_5_stop_dpg_mode(adev, i);
1344 			continue;
1345 		}
1346 
1347 		/* wait for vcn idle */
1348 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1349 		if (r)
1350 			return r;
1351 
1352 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1353 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1354 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1355 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1356 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1357 		if (r)
1358 			return r;
1359 
1360 		/* block LMI UMC channel */
1361 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1362 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1363 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1364 
1365 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1366 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1367 		r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1368 		if (r)
1369 			return r;
1370 
1371 		/* block VCPU register access */
1372 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1373 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1374 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1375 
1376 		/* reset VCPU */
1377 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1378 			UVD_VCPU_CNTL__BLK_RST_MASK,
1379 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1380 
1381 		/* disable VCPU clock */
1382 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1383 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1384 
1385 		/* clear status */
1386 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1387 
1388 		vcn_v2_5_enable_clock_gating(adev);
1389 
1390 		/* enable register anti-hang mechanism */
1391 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1392 			UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1393 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1394 	}
1395 
1396 	if (adev->pm.dpm_enabled)
1397 		amdgpu_dpm_enable_uvd(adev, false);
1398 
1399 	return 0;
1400 }
1401 
1402 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1403 				int inst_idx, struct dpg_pause_state *new_state)
1404 {
1405 	struct amdgpu_ring *ring;
1406 	uint32_t reg_data = 0;
1407 	int ret_code = 0;
1408 
1409 	/* pause/unpause if state is changed */
1410 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1411 		DRM_DEBUG("dpg pause state changed %d -> %d",
1412 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1413 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1414 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1415 
1416 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1417 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1418 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1419 
1420 			if (!ret_code) {
1421 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1422 
1423 				/* pause DPG */
1424 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1425 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1426 
1427 				/* wait for ACK */
1428 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1429 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1430 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1431 
1432 				/* Stall DPG before WPTR/RPTR reset */
1433 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1434 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1435 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1436 
1437 				/* Restore */
1438 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1439 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1440 				ring->wptr = 0;
1441 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1442 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1443 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1444 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1445 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1446 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1447 
1448 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1449 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1450 				ring->wptr = 0;
1451 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1452 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1453 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1454 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1455 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1456 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1457 
1458 				/* Unstall DPG */
1459 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1460 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1461 
1462 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1463 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1464 			}
1465 		} else {
1466 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1467 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1468 			SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1469 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1470 		}
1471 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1472 	}
1473 
1474 	return 0;
1475 }
1476 
1477 /**
1478  * vcn_v2_5_dec_ring_get_rptr - get read pointer
1479  *
1480  * @ring: amdgpu_ring pointer
1481  *
1482  * Returns the current hardware read pointer
1483  */
1484 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1485 {
1486 	struct amdgpu_device *adev = ring->adev;
1487 
1488 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1489 }
1490 
1491 /**
1492  * vcn_v2_5_dec_ring_get_wptr - get write pointer
1493  *
1494  * @ring: amdgpu_ring pointer
1495  *
1496  * Returns the current hardware write pointer
1497  */
1498 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1499 {
1500 	struct amdgpu_device *adev = ring->adev;
1501 
1502 	if (ring->use_doorbell)
1503 		return *ring->wptr_cpu_addr;
1504 	else
1505 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1506 }
1507 
1508 /**
1509  * vcn_v2_5_dec_ring_set_wptr - set write pointer
1510  *
1511  * @ring: amdgpu_ring pointer
1512  *
1513  * Commits the write pointer to the hardware
1514  */
1515 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1516 {
1517 	struct amdgpu_device *adev = ring->adev;
1518 
1519 	if (ring->use_doorbell) {
1520 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1521 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1522 	} else {
1523 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1524 	}
1525 }
1526 
1527 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1528 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1529 	.align_mask = 0xf,
1530 	.secure_submission_supported = true,
1531 	.vmhub = AMDGPU_MMHUB_1,
1532 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1533 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1534 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1535 	.emit_frame_size =
1536 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1537 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1538 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1539 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1540 		6,
1541 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1542 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1543 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1544 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1545 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1546 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1547 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1548 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1549 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1550 	.pad_ib = amdgpu_ring_generic_pad_ib,
1551 	.begin_use = amdgpu_vcn_ring_begin_use,
1552 	.end_use = amdgpu_vcn_ring_end_use,
1553 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1554 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1555 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1556 };
1557 
1558 static const struct amdgpu_ring_funcs vcn_v2_6_dec_ring_vm_funcs = {
1559 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1560 	.align_mask = 0xf,
1561 	.secure_submission_supported = true,
1562 	.vmhub = AMDGPU_MMHUB_0,
1563 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1564 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1565 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1566 	.emit_frame_size =
1567 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1568 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1569 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1570 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1571 		6,
1572 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1573 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1574 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1575 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1576 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1577 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1578 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1579 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1580 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1581 	.pad_ib = amdgpu_ring_generic_pad_ib,
1582 	.begin_use = amdgpu_vcn_ring_begin_use,
1583 	.end_use = amdgpu_vcn_ring_end_use,
1584 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1585 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1586 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1587 };
1588 
1589 /**
1590  * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1591  *
1592  * @ring: amdgpu_ring pointer
1593  *
1594  * Returns the current hardware enc read pointer
1595  */
1596 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1597 {
1598 	struct amdgpu_device *adev = ring->adev;
1599 
1600 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1601 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1602 	else
1603 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1604 }
1605 
1606 /**
1607  * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1608  *
1609  * @ring: amdgpu_ring pointer
1610  *
1611  * Returns the current hardware enc write pointer
1612  */
1613 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1614 {
1615 	struct amdgpu_device *adev = ring->adev;
1616 
1617 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1618 		if (ring->use_doorbell)
1619 			return *ring->wptr_cpu_addr;
1620 		else
1621 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1622 	} else {
1623 		if (ring->use_doorbell)
1624 			return *ring->wptr_cpu_addr;
1625 		else
1626 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1627 	}
1628 }
1629 
1630 /**
1631  * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1632  *
1633  * @ring: amdgpu_ring pointer
1634  *
1635  * Commits the enc write pointer to the hardware
1636  */
1637 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1638 {
1639 	struct amdgpu_device *adev = ring->adev;
1640 
1641 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1642 		if (ring->use_doorbell) {
1643 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1644 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1645 		} else {
1646 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1647 		}
1648 	} else {
1649 		if (ring->use_doorbell) {
1650 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1651 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1652 		} else {
1653 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1654 		}
1655 	}
1656 }
1657 
1658 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1659 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1660 	.align_mask = 0x3f,
1661 	.nop = VCN_ENC_CMD_NO_OP,
1662 	.vmhub = AMDGPU_MMHUB_1,
1663 	.get_rptr = vcn_v2_5_enc_ring_get_rptr,
1664 	.get_wptr = vcn_v2_5_enc_ring_get_wptr,
1665 	.set_wptr = vcn_v2_5_enc_ring_set_wptr,
1666 	.emit_frame_size =
1667 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1668 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1669 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1670 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1671 		1, /* vcn_v2_0_enc_ring_insert_end */
1672 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1673 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1674 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1675 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1676 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1677 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1678 	.insert_nop = amdgpu_ring_insert_nop,
1679 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1680 	.pad_ib = amdgpu_ring_generic_pad_ib,
1681 	.begin_use = amdgpu_vcn_ring_begin_use,
1682 	.end_use = amdgpu_vcn_ring_end_use,
1683 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1684 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1685 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1686 };
1687 
1688 static const struct amdgpu_ring_funcs vcn_v2_6_enc_ring_vm_funcs = {
1689         .type = AMDGPU_RING_TYPE_VCN_ENC,
1690         .align_mask = 0x3f,
1691         .nop = VCN_ENC_CMD_NO_OP,
1692         .vmhub = AMDGPU_MMHUB_0,
1693         .get_rptr = vcn_v2_5_enc_ring_get_rptr,
1694         .get_wptr = vcn_v2_5_enc_ring_get_wptr,
1695         .set_wptr = vcn_v2_5_enc_ring_set_wptr,
1696         .emit_frame_size =
1697                 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1698                 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1699                 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1700                 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1701                 1, /* vcn_v2_0_enc_ring_insert_end */
1702         .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1703         .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1704         .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1705         .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1706         .test_ring = amdgpu_vcn_enc_ring_test_ring,
1707         .test_ib = amdgpu_vcn_enc_ring_test_ib,
1708         .insert_nop = amdgpu_ring_insert_nop,
1709         .insert_end = vcn_v2_0_enc_ring_insert_end,
1710         .pad_ib = amdgpu_ring_generic_pad_ib,
1711         .begin_use = amdgpu_vcn_ring_begin_use,
1712         .end_use = amdgpu_vcn_ring_end_use,
1713         .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1714         .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1715         .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1716 };
1717 
1718 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1719 {
1720 	int i;
1721 
1722 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1723 		if (adev->vcn.harvest_config & (1 << i))
1724 			continue;
1725 		if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(2, 5, 0))
1726 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1727 		else /* CHIP_ALDEBARAN */
1728 			adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_6_dec_ring_vm_funcs;
1729 		adev->vcn.inst[i].ring_dec.me = i;
1730 		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1731 	}
1732 }
1733 
1734 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1735 {
1736 	int i, j;
1737 
1738 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1739 		if (adev->vcn.harvest_config & (1 << j))
1740 			continue;
1741 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1742 			if (adev->ip_versions[UVD_HWIP][0] == IP_VERSION(2, 5, 0))
1743 				adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1744 			else /* CHIP_ALDEBARAN */
1745 				adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_6_enc_ring_vm_funcs;
1746 			adev->vcn.inst[j].ring_enc[i].me = j;
1747 		}
1748 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1749 	}
1750 }
1751 
1752 static bool vcn_v2_5_is_idle(void *handle)
1753 {
1754 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1755 	int i, ret = 1;
1756 
1757 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1758 		if (adev->vcn.harvest_config & (1 << i))
1759 			continue;
1760 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1761 	}
1762 
1763 	return ret;
1764 }
1765 
1766 static int vcn_v2_5_wait_for_idle(void *handle)
1767 {
1768 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1769 	int i, ret = 0;
1770 
1771 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1772 		if (adev->vcn.harvest_config & (1 << i))
1773 			continue;
1774 		ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1775 			UVD_STATUS__IDLE);
1776 		if (ret)
1777 			return ret;
1778 	}
1779 
1780 	return ret;
1781 }
1782 
1783 static int vcn_v2_5_set_clockgating_state(void *handle,
1784 					  enum amd_clockgating_state state)
1785 {
1786 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1787 	bool enable = (state == AMD_CG_STATE_GATE);
1788 
1789 	if (amdgpu_sriov_vf(adev))
1790 		return 0;
1791 
1792 	if (enable) {
1793 		if (!vcn_v2_5_is_idle(handle))
1794 			return -EBUSY;
1795 		vcn_v2_5_enable_clock_gating(adev);
1796 	} else {
1797 		vcn_v2_5_disable_clock_gating(adev);
1798 	}
1799 
1800 	return 0;
1801 }
1802 
1803 static int vcn_v2_5_set_powergating_state(void *handle,
1804 					  enum amd_powergating_state state)
1805 {
1806 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1807 	int ret;
1808 
1809 	if (amdgpu_sriov_vf(adev))
1810 		return 0;
1811 
1812 	if(state == adev->vcn.cur_state)
1813 		return 0;
1814 
1815 	if (state == AMD_PG_STATE_GATE)
1816 		ret = vcn_v2_5_stop(adev);
1817 	else
1818 		ret = vcn_v2_5_start(adev);
1819 
1820 	if(!ret)
1821 		adev->vcn.cur_state = state;
1822 
1823 	return ret;
1824 }
1825 
1826 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1827 					struct amdgpu_irq_src *source,
1828 					unsigned type,
1829 					enum amdgpu_interrupt_state state)
1830 {
1831 	return 0;
1832 }
1833 
1834 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1835 				      struct amdgpu_irq_src *source,
1836 				      struct amdgpu_iv_entry *entry)
1837 {
1838 	uint32_t ip_instance;
1839 
1840 	switch (entry->client_id) {
1841 	case SOC15_IH_CLIENTID_VCN:
1842 		ip_instance = 0;
1843 		break;
1844 	case SOC15_IH_CLIENTID_VCN1:
1845 		ip_instance = 1;
1846 		break;
1847 	default:
1848 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1849 		return 0;
1850 	}
1851 
1852 	DRM_DEBUG("IH: VCN TRAP\n");
1853 
1854 	switch (entry->src_id) {
1855 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1856 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1857 		break;
1858 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1859 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1860 		break;
1861 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1862 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1863 		break;
1864 	case VCN_2_6__SRCID_UVD_POISON:
1865 		amdgpu_vcn_process_poison_irq(adev, source, entry);
1866 		break;
1867 	default:
1868 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1869 			  entry->src_id, entry->src_data[0]);
1870 		break;
1871 	}
1872 
1873 	return 0;
1874 }
1875 
1876 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1877 	.set = vcn_v2_5_set_interrupt_state,
1878 	.process = vcn_v2_5_process_interrupt,
1879 };
1880 
1881 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1882 {
1883 	int i;
1884 
1885 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1886 		if (adev->vcn.harvest_config & (1 << i))
1887 			continue;
1888 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1889 		adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1890 	}
1891 }
1892 
1893 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1894 	.name = "vcn_v2_5",
1895 	.early_init = vcn_v2_5_early_init,
1896 	.late_init = NULL,
1897 	.sw_init = vcn_v2_5_sw_init,
1898 	.sw_fini = vcn_v2_5_sw_fini,
1899 	.hw_init = vcn_v2_5_hw_init,
1900 	.hw_fini = vcn_v2_5_hw_fini,
1901 	.suspend = vcn_v2_5_suspend,
1902 	.resume = vcn_v2_5_resume,
1903 	.is_idle = vcn_v2_5_is_idle,
1904 	.wait_for_idle = vcn_v2_5_wait_for_idle,
1905 	.check_soft_reset = NULL,
1906 	.pre_soft_reset = NULL,
1907 	.soft_reset = NULL,
1908 	.post_soft_reset = NULL,
1909 	.set_clockgating_state = vcn_v2_5_set_clockgating_state,
1910 	.set_powergating_state = vcn_v2_5_set_powergating_state,
1911 };
1912 
1913 static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
1914         .name = "vcn_v2_6",
1915         .early_init = vcn_v2_5_early_init,
1916         .late_init = NULL,
1917         .sw_init = vcn_v2_5_sw_init,
1918         .sw_fini = vcn_v2_5_sw_fini,
1919         .hw_init = vcn_v2_5_hw_init,
1920         .hw_fini = vcn_v2_5_hw_fini,
1921         .suspend = vcn_v2_5_suspend,
1922         .resume = vcn_v2_5_resume,
1923         .is_idle = vcn_v2_5_is_idle,
1924         .wait_for_idle = vcn_v2_5_wait_for_idle,
1925         .check_soft_reset = NULL,
1926         .pre_soft_reset = NULL,
1927         .soft_reset = NULL,
1928         .post_soft_reset = NULL,
1929         .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1930         .set_powergating_state = vcn_v2_5_set_powergating_state,
1931 };
1932 
1933 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1934 {
1935 		.type = AMD_IP_BLOCK_TYPE_VCN,
1936 		.major = 2,
1937 		.minor = 5,
1938 		.rev = 0,
1939 		.funcs = &vcn_v2_5_ip_funcs,
1940 };
1941 
1942 const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
1943 {
1944 		.type = AMD_IP_BLOCK_TYPE_VCN,
1945 		.major = 2,
1946 		.minor = 6,
1947 		.rev = 0,
1948 		.funcs = &vcn_v2_6_ip_funcs,
1949 };
1950 
1951 static uint32_t vcn_v2_6_query_poison_by_instance(struct amdgpu_device *adev,
1952 			uint32_t instance, uint32_t sub_block)
1953 {
1954 	uint32_t poison_stat = 0, reg_value = 0;
1955 
1956 	switch (sub_block) {
1957 	case AMDGPU_VCN_V2_6_VCPU_VCODEC:
1958 		reg_value = RREG32_SOC15(VCN, instance, mmUVD_RAS_VCPU_VCODEC_STATUS);
1959 		poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
1960 		break;
1961 	default:
1962 		break;
1963 	}
1964 
1965 	if (poison_stat)
1966 		dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
1967 			instance, sub_block);
1968 
1969 	return poison_stat;
1970 }
1971 
1972 static bool vcn_v2_6_query_poison_status(struct amdgpu_device *adev)
1973 {
1974 	uint32_t inst, sub;
1975 	uint32_t poison_stat = 0;
1976 
1977 	for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
1978 		for (sub = 0; sub < AMDGPU_VCN_V2_6_MAX_SUB_BLOCK; sub++)
1979 			poison_stat +=
1980 			vcn_v2_6_query_poison_by_instance(adev, inst, sub);
1981 
1982 	return !!poison_stat;
1983 }
1984 
1985 const struct amdgpu_ras_block_hw_ops vcn_v2_6_ras_hw_ops = {
1986 	.query_poison_status = vcn_v2_6_query_poison_status,
1987 };
1988 
1989 static struct amdgpu_vcn_ras vcn_v2_6_ras = {
1990 	.ras_block = {
1991 		.hw_ops = &vcn_v2_6_ras_hw_ops,
1992 	},
1993 };
1994 
1995 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev)
1996 {
1997 	switch (adev->ip_versions[VCN_HWIP][0]) {
1998 	case IP_VERSION(2, 6, 0):
1999 		adev->vcn.ras = &vcn_v2_6_ras;
2000 		break;
2001 	default:
2002 		break;
2003 	}
2004 
2005 	if (adev->vcn.ras) {
2006 		amdgpu_ras_register_ras_block(adev, &adev->vcn.ras->ras_block);
2007 
2008 		strcpy(adev->vcn.ras->ras_block.ras_comm.name, "vcn");
2009 		adev->vcn.ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__VCN;
2010 		adev->vcn.ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__POISON;
2011 		adev->vcn.ras_if = &adev->vcn.ras->ras_block.ras_comm;
2012 
2013 		/* If don't define special ras_late_init function, use default ras_late_init */
2014 		if (!adev->vcn.ras->ras_block.ras_late_init)
2015 			adev->vcn.ras->ras_block.ras_late_init = amdgpu_ras_block_late_init;
2016 	}
2017 }
2018