xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c (revision 82df5b73)
1 /*
2  * Copyright 2019 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 
26 #include "amdgpu.h"
27 #include "amdgpu_vcn.h"
28 #include "amdgpu_pm.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "vcn_v2_0.h"
32 #include "mmsch_v1_0.h"
33 
34 #include "vcn/vcn_2_5_offset.h"
35 #include "vcn/vcn_2_5_sh_mask.h"
36 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
37 
38 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x27
39 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x0f
40 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x10
41 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x11
42 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x29
43 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x66
44 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
45 
46 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x431
47 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x3b4
48 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 	0x3b5
49 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x25c
50 
51 #define VCN25_MAX_HW_INSTANCES_ARCTURUS			2
52 
53 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
54 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
55 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
56 static int vcn_v2_5_set_powergating_state(void *handle,
57 				enum amd_powergating_state state);
58 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
59 				int inst_idx, struct dpg_pause_state *new_state);
60 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
61 
62 static int amdgpu_ih_clientid_vcns[] = {
63 	SOC15_IH_CLIENTID_VCN,
64 	SOC15_IH_CLIENTID_VCN1
65 };
66 
67 /**
68  * vcn_v2_5_early_init - set function pointers
69  *
70  * @handle: amdgpu_device pointer
71  *
72  * Set ring and irq function pointers
73  */
74 static int vcn_v2_5_early_init(void *handle)
75 {
76 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
77 
78 	if (amdgpu_sriov_vf(adev)) {
79 		adev->vcn.num_vcn_inst = 2;
80 		adev->vcn.harvest_config = 0;
81 		adev->vcn.num_enc_rings = 1;
82 	} else {
83 		if (adev->asic_type == CHIP_ARCTURUS) {
84 			u32 harvest;
85 			int i;
86 
87 			adev->vcn.num_vcn_inst = VCN25_MAX_HW_INSTANCES_ARCTURUS;
88 			for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
89 				harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
90 				if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
91 					adev->vcn.harvest_config |= 1 << i;
92 			}
93 
94 			if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
95 						AMDGPU_VCN_HARVEST_VCN1))
96 				/* both instances are harvested, disable the block */
97 				return -ENOENT;
98 		} else
99 			adev->vcn.num_vcn_inst = 1;
100 
101 		adev->vcn.num_enc_rings = 2;
102 	}
103 
104 	vcn_v2_5_set_dec_ring_funcs(adev);
105 	vcn_v2_5_set_enc_ring_funcs(adev);
106 	vcn_v2_5_set_irq_funcs(adev);
107 
108 	return 0;
109 }
110 
111 /**
112  * vcn_v2_5_sw_init - sw init for VCN block
113  *
114  * @handle: amdgpu_device pointer
115  *
116  * Load firmware and sw initialization
117  */
118 static int vcn_v2_5_sw_init(void *handle)
119 {
120 	struct amdgpu_ring *ring;
121 	int i, j, r;
122 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
123 
124 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
125 		if (adev->vcn.harvest_config & (1 << j))
126 			continue;
127 		/* VCN DEC TRAP */
128 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
129 				VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
130 		if (r)
131 			return r;
132 
133 		/* VCN ENC TRAP */
134 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
135 			r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
136 				i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
137 			if (r)
138 				return r;
139 		}
140 	}
141 
142 	r = amdgpu_vcn_sw_init(adev);
143 	if (r)
144 		return r;
145 
146 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
147 		const struct common_firmware_header *hdr;
148 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
149 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
150 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
151 		adev->firmware.fw_size +=
152 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
153 
154 		if (adev->vcn.num_vcn_inst == VCN25_MAX_HW_INSTANCES_ARCTURUS) {
155 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].ucode_id = AMDGPU_UCODE_ID_VCN1;
156 			adev->firmware.ucode[AMDGPU_UCODE_ID_VCN1].fw = adev->vcn.fw;
157 			adev->firmware.fw_size +=
158 				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
159 		}
160 		DRM_INFO("PSP loading VCN firmware\n");
161 	}
162 
163 	r = amdgpu_vcn_resume(adev);
164 	if (r)
165 		return r;
166 
167 	for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
168 		volatile struct amdgpu_fw_shared *fw_shared;
169 
170 		if (adev->vcn.harvest_config & (1 << j))
171 			continue;
172 		adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
173 		adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
174 		adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
175 		adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
176 		adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
177 		adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
178 
179 		adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
180 		adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
181 		adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
182 		adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
183 		adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
184 		adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
185 		adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
186 		adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
187 		adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
188 		adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
189 
190 		ring = &adev->vcn.inst[j].ring_dec;
191 		ring->use_doorbell = true;
192 
193 		ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
194 				(amdgpu_sriov_vf(adev) ? 2*j : 8*j);
195 		sprintf(ring->name, "vcn_dec_%d", j);
196 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
197 				     0, AMDGPU_RING_PRIO_DEFAULT);
198 		if (r)
199 			return r;
200 
201 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
202 			ring = &adev->vcn.inst[j].ring_enc[i];
203 			ring->use_doorbell = true;
204 
205 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
206 					(amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
207 
208 			sprintf(ring->name, "vcn_enc_%d.%d", j, i);
209 			r = amdgpu_ring_init(adev, ring, 512,
210 					     &adev->vcn.inst[j].irq, 0,
211 					     AMDGPU_RING_PRIO_DEFAULT);
212 			if (r)
213 				return r;
214 		}
215 
216 		fw_shared = adev->vcn.inst[j].fw_shared_cpu_addr;
217 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
218 	}
219 
220 	if (amdgpu_sriov_vf(adev)) {
221 		r = amdgpu_virt_alloc_mm_table(adev);
222 		if (r)
223 			return r;
224 	}
225 
226 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
227 		adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
228 
229 	return 0;
230 }
231 
232 /**
233  * vcn_v2_5_sw_fini - sw fini for VCN block
234  *
235  * @handle: amdgpu_device pointer
236  *
237  * VCN suspend and free up sw allocation
238  */
239 static int vcn_v2_5_sw_fini(void *handle)
240 {
241 	int i, r;
242 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
243 	volatile struct amdgpu_fw_shared *fw_shared;
244 
245 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
246 		if (adev->vcn.harvest_config & (1 << i))
247 			continue;
248 		fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
249 		fw_shared->present_flag_0 = 0;
250 	}
251 
252 	if (amdgpu_sriov_vf(adev))
253 		amdgpu_virt_free_mm_table(adev);
254 
255 	r = amdgpu_vcn_suspend(adev);
256 	if (r)
257 		return r;
258 
259 	r = amdgpu_vcn_sw_fini(adev);
260 
261 	return r;
262 }
263 
264 /**
265  * vcn_v2_5_hw_init - start and test VCN block
266  *
267  * @handle: amdgpu_device pointer
268  *
269  * Initialize the hardware, boot up the VCPU and do some testing
270  */
271 static int vcn_v2_5_hw_init(void *handle)
272 {
273 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
274 	struct amdgpu_ring *ring;
275 	int i, j, r = 0;
276 
277 	if (amdgpu_sriov_vf(adev))
278 		r = vcn_v2_5_sriov_start(adev);
279 
280 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
281 		if (adev->vcn.harvest_config & (1 << j))
282 			continue;
283 
284 		if (amdgpu_sriov_vf(adev)) {
285 			adev->vcn.inst[j].ring_enc[0].sched.ready = true;
286 			adev->vcn.inst[j].ring_enc[1].sched.ready = false;
287 			adev->vcn.inst[j].ring_enc[2].sched.ready = false;
288 			adev->vcn.inst[j].ring_dec.sched.ready = true;
289 		} else {
290 
291 			ring = &adev->vcn.inst[j].ring_dec;
292 
293 			adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
294 						     ring->doorbell_index, j);
295 
296 			r = amdgpu_ring_test_helper(ring);
297 			if (r)
298 				goto done;
299 
300 			for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
301 				ring = &adev->vcn.inst[j].ring_enc[i];
302 				r = amdgpu_ring_test_helper(ring);
303 				if (r)
304 					goto done;
305 			}
306 		}
307 	}
308 
309 done:
310 	if (!r)
311 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
312 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
313 
314 	return r;
315 }
316 
317 /**
318  * vcn_v2_5_hw_fini - stop the hardware block
319  *
320  * @handle: amdgpu_device pointer
321  *
322  * Stop the VCN block, mark ring as not ready any more
323  */
324 static int vcn_v2_5_hw_fini(void *handle)
325 {
326 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
327 	int i;
328 
329 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
330 		if (adev->vcn.harvest_config & (1 << i))
331 			continue;
332 
333 		if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
334 		    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
335 		     RREG32_SOC15(VCN, i, mmUVD_STATUS)))
336 			vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
337 	}
338 
339 	return 0;
340 }
341 
342 /**
343  * vcn_v2_5_suspend - suspend VCN block
344  *
345  * @handle: amdgpu_device pointer
346  *
347  * HW fini and suspend VCN block
348  */
349 static int vcn_v2_5_suspend(void *handle)
350 {
351 	int r;
352 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
353 
354 	r = vcn_v2_5_hw_fini(adev);
355 	if (r)
356 		return r;
357 
358 	r = amdgpu_vcn_suspend(adev);
359 
360 	return r;
361 }
362 
363 /**
364  * vcn_v2_5_resume - resume VCN block
365  *
366  * @handle: amdgpu_device pointer
367  *
368  * Resume firmware and hw init VCN block
369  */
370 static int vcn_v2_5_resume(void *handle)
371 {
372 	int r;
373 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
374 
375 	r = amdgpu_vcn_resume(adev);
376 	if (r)
377 		return r;
378 
379 	r = vcn_v2_5_hw_init(adev);
380 
381 	return r;
382 }
383 
384 /**
385  * vcn_v2_5_mc_resume - memory controller programming
386  *
387  * @adev: amdgpu_device pointer
388  *
389  * Let the VCN memory controller know it's offsets
390  */
391 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
392 {
393 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
394 	uint32_t offset;
395 	int i;
396 
397 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
398 		if (adev->vcn.harvest_config & (1 << i))
399 			continue;
400 		/* cache window 0: fw */
401 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
402 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
403 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
404 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
405 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
406 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
407 			offset = 0;
408 		} else {
409 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
410 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
411 			WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
412 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
413 			offset = size;
414 			WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
415 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
416 		}
417 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
418 
419 		/* cache window 1: stack */
420 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
421 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
422 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
423 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
424 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
425 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
426 
427 		/* cache window 2: context */
428 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
429 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
430 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
431 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
432 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
433 		WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
434 
435 		/* non-cache window */
436 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
437 			lower_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
438 		WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
439 			upper_32_bits(adev->vcn.inst[i].fw_shared_gpu_addr));
440 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
441 		WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
442 			AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
443 	}
444 }
445 
446 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
447 {
448 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
449 	uint32_t offset;
450 
451 	/* cache window 0: fw */
452 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
453 		if (!indirect) {
454 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
455 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
456 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
457 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
458 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
459 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
460 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
461 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
462 		} else {
463 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
464 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
465 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
466 				VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
467 			WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
468 				VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
469 		}
470 		offset = 0;
471 	} else {
472 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
473 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
474 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
475 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
476 			VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
477 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
478 		offset = size;
479 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
480 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
481 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
482 	}
483 
484 	if (!indirect)
485 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
486 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
487 	else
488 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
489 			VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
490 
491 	/* cache window 1: stack */
492 	if (!indirect) {
493 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
494 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
495 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
496 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
497 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
498 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
499 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
500 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
501 	} else {
502 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
503 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
504 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
505 			VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
506 		WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
507 			VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
508 	}
509 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
510 		VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
511 
512 	/* cache window 2: context */
513 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
514 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
515 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
516 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
517 		VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
518 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
519 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
520 		VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
521 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
522 		VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
523 
524 	/* non-cache window */
525 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
526 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
527 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
528 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
529 		VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
530 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared_gpu_addr), 0, indirect);
531 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
532 		VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
533 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
534 		VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
535 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
536 
537 	/* VCN global tiling registers */
538 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
539 		VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
540 }
541 
542 /**
543  * vcn_v2_5_disable_clock_gating - disable VCN clock gating
544  *
545  * @adev: amdgpu_device pointer
546  *
547  * Disable clock gating for VCN block
548  */
549 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
550 {
551 	uint32_t data;
552 	int ret = 0;
553 	int i;
554 
555 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
556 		if (adev->vcn.harvest_config & (1 << i))
557 			continue;
558 		/* UVD disable CGC */
559 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
560 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
561 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
562 		else
563 			data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
564 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
565 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
566 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
567 
568 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
569 		data &= ~(UVD_CGC_GATE__SYS_MASK
570 			| UVD_CGC_GATE__UDEC_MASK
571 			| UVD_CGC_GATE__MPEG2_MASK
572 			| UVD_CGC_GATE__REGS_MASK
573 			| UVD_CGC_GATE__RBC_MASK
574 			| UVD_CGC_GATE__LMI_MC_MASK
575 			| UVD_CGC_GATE__LMI_UMC_MASK
576 			| UVD_CGC_GATE__IDCT_MASK
577 			| UVD_CGC_GATE__MPRD_MASK
578 			| UVD_CGC_GATE__MPC_MASK
579 			| UVD_CGC_GATE__LBSI_MASK
580 			| UVD_CGC_GATE__LRBBM_MASK
581 			| UVD_CGC_GATE__UDEC_RE_MASK
582 			| UVD_CGC_GATE__UDEC_CM_MASK
583 			| UVD_CGC_GATE__UDEC_IT_MASK
584 			| UVD_CGC_GATE__UDEC_DB_MASK
585 			| UVD_CGC_GATE__UDEC_MP_MASK
586 			| UVD_CGC_GATE__WCB_MASK
587 			| UVD_CGC_GATE__VCPU_MASK
588 			| UVD_CGC_GATE__MMSCH_MASK);
589 
590 		WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
591 
592 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0,  0xFFFFFFFF, ret);
593 
594 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
595 		data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
596 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
597 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
598 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
599 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
600 			| UVD_CGC_CTRL__SYS_MODE_MASK
601 			| UVD_CGC_CTRL__UDEC_MODE_MASK
602 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
603 			| UVD_CGC_CTRL__REGS_MODE_MASK
604 			| UVD_CGC_CTRL__RBC_MODE_MASK
605 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
606 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
607 			| UVD_CGC_CTRL__IDCT_MODE_MASK
608 			| UVD_CGC_CTRL__MPRD_MODE_MASK
609 			| UVD_CGC_CTRL__MPC_MODE_MASK
610 			| UVD_CGC_CTRL__LBSI_MODE_MASK
611 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
612 			| UVD_CGC_CTRL__WCB_MODE_MASK
613 			| UVD_CGC_CTRL__VCPU_MODE_MASK
614 			| UVD_CGC_CTRL__MMSCH_MODE_MASK);
615 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
616 
617 		/* turn on */
618 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
619 		data |= (UVD_SUVD_CGC_GATE__SRE_MASK
620 			| UVD_SUVD_CGC_GATE__SIT_MASK
621 			| UVD_SUVD_CGC_GATE__SMP_MASK
622 			| UVD_SUVD_CGC_GATE__SCM_MASK
623 			| UVD_SUVD_CGC_GATE__SDB_MASK
624 			| UVD_SUVD_CGC_GATE__SRE_H264_MASK
625 			| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
626 			| UVD_SUVD_CGC_GATE__SIT_H264_MASK
627 			| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
628 			| UVD_SUVD_CGC_GATE__SCM_H264_MASK
629 			| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
630 			| UVD_SUVD_CGC_GATE__SDB_H264_MASK
631 			| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
632 			| UVD_SUVD_CGC_GATE__SCLR_MASK
633 			| UVD_SUVD_CGC_GATE__UVD_SC_MASK
634 			| UVD_SUVD_CGC_GATE__ENT_MASK
635 			| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
636 			| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
637 			| UVD_SUVD_CGC_GATE__SITE_MASK
638 			| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
639 			| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
640 			| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
641 			| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
642 			| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
643 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
644 
645 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
646 		data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
647 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
648 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
649 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
650 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
651 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
652 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
653 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
654 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
655 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
656 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
657 	}
658 }
659 
660 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
661 		uint8_t sram_sel, int inst_idx, uint8_t indirect)
662 {
663 	uint32_t reg_data = 0;
664 
665 	/* enable sw clock gating control */
666 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
667 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
668 	else
669 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
670 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
671 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
672 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
673 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
674 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
675 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
676 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
677 		 UVD_CGC_CTRL__SYS_MODE_MASK |
678 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
679 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
680 		 UVD_CGC_CTRL__REGS_MODE_MASK |
681 		 UVD_CGC_CTRL__RBC_MODE_MASK |
682 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
683 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
684 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
685 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
686 		 UVD_CGC_CTRL__MPC_MODE_MASK |
687 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
688 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
689 		 UVD_CGC_CTRL__WCB_MODE_MASK |
690 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
691 		 UVD_CGC_CTRL__MMSCH_MODE_MASK);
692 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
693 		VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
694 
695 	/* turn off clock gating */
696 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
697 		VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
698 
699 	/* turn on SUVD clock gating */
700 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
701 		VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
702 
703 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
704 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
705 		VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
706 }
707 
708 /**
709  * vcn_v2_5_enable_clock_gating - enable VCN clock gating
710  *
711  * @adev: amdgpu_device pointer
712  *
713  * Enable clock gating for VCN block
714  */
715 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
716 {
717 	uint32_t data = 0;
718 	int i;
719 
720 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
721 		if (adev->vcn.harvest_config & (1 << i))
722 			continue;
723 		/* enable UVD CGC */
724 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
725 		if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
726 			data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
727 		else
728 			data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
729 		data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
730 		data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
731 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
732 
733 		data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
734 		data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
735 			| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
736 			| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
737 			| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
738 			| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
739 			| UVD_CGC_CTRL__SYS_MODE_MASK
740 			| UVD_CGC_CTRL__UDEC_MODE_MASK
741 			| UVD_CGC_CTRL__MPEG2_MODE_MASK
742 			| UVD_CGC_CTRL__REGS_MODE_MASK
743 			| UVD_CGC_CTRL__RBC_MODE_MASK
744 			| UVD_CGC_CTRL__LMI_MC_MODE_MASK
745 			| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
746 			| UVD_CGC_CTRL__IDCT_MODE_MASK
747 			| UVD_CGC_CTRL__MPRD_MODE_MASK
748 			| UVD_CGC_CTRL__MPC_MODE_MASK
749 			| UVD_CGC_CTRL__LBSI_MODE_MASK
750 			| UVD_CGC_CTRL__LRBBM_MODE_MASK
751 			| UVD_CGC_CTRL__WCB_MODE_MASK
752 			| UVD_CGC_CTRL__VCPU_MODE_MASK);
753 		WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
754 
755 		data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
756 		data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
757 			| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
758 			| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
759 			| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
760 			| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
761 			| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
762 			| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
763 			| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
764 			| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
765 			| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
766 		WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
767 	}
768 }
769 
770 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
771 {
772 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
773 	struct amdgpu_ring *ring;
774 	uint32_t rb_bufsz, tmp;
775 
776 	/* disable register anti-hang mechanism */
777 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
778 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
779 	/* enable dynamic power gating mode */
780 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
781 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
782 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
783 	WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
784 
785 	if (indirect)
786 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t*)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
787 
788 	/* enable clock gating */
789 	vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
790 
791 	/* enable VCPU clock */
792 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
793 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
794 	tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
795 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
796 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
797 
798 	/* disable master interupt */
799 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
800 		VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
801 
802 	/* setup mmUVD_LMI_CTRL */
803 	tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
804 		UVD_LMI_CTRL__REQ_MODE_MASK |
805 		UVD_LMI_CTRL__CRC_RESET_MASK |
806 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
807 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
808 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
809 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
810 		0x00100000L);
811 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
812 		VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
813 
814 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
815 		VCN, 0, mmUVD_MPC_CNTL),
816 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
817 
818 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
819 		VCN, 0, mmUVD_MPC_SET_MUXA0),
820 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
821 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
822 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
823 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
824 
825 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
826 		VCN, 0, mmUVD_MPC_SET_MUXB0),
827 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
828 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
829 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
830 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
831 
832 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
833 		VCN, 0, mmUVD_MPC_SET_MUX),
834 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
835 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
836 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
837 
838 	vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
839 
840 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
841 		VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
842 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
843 		VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
844 
845 	/* enable LMI MC and UMC channels */
846 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
847 		VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
848 
849 	/* unblock VCPU register access */
850 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
851 		VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
852 
853 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
854 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
855 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
856 		VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
857 
858 	/* enable master interrupt */
859 	WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
860 		VCN, 0, mmUVD_MASTINT_EN),
861 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
862 
863 	if (indirect)
864 		psp_update_vcn_sram(adev, inst_idx, adev->vcn.inst[inst_idx].dpg_sram_gpu_addr,
865 				    (uint32_t)((uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_curr_addr -
866 					       (uintptr_t)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr));
867 
868 	ring = &adev->vcn.inst[inst_idx].ring_dec;
869 	/* force RBC into idle state */
870 	rb_bufsz = order_base_2(ring->ring_size);
871 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
872 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
873 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
874 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
875 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
876 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
877 
878 	/* Stall DPG before WPTR/RPTR reset */
879 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
880 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
881 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
882 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
883 
884 	/* set the write pointer delay */
885 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
886 
887 	/* set the wb address */
888 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
889 		(upper_32_bits(ring->gpu_addr) >> 2));
890 
891 	/* programm the RB_BASE for ring buffer */
892 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
893 		lower_32_bits(ring->gpu_addr));
894 	WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
895 		upper_32_bits(ring->gpu_addr));
896 
897 	/* Initialize the ring buffer's read and write pointers */
898 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
899 
900 	WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
901 
902 	ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
903 	WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
904 		lower_32_bits(ring->wptr));
905 
906 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
907 	/* Unstall DPG */
908 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
909 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
910 
911 	return 0;
912 }
913 
914 static int vcn_v2_5_start(struct amdgpu_device *adev)
915 {
916 	struct amdgpu_ring *ring;
917 	uint32_t rb_bufsz, tmp;
918 	int i, j, k, r;
919 
920 	if (adev->pm.dpm_enabled)
921 		amdgpu_dpm_enable_uvd(adev, true);
922 
923 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
924 		if (adev->vcn.harvest_config & (1 << i))
925 			continue;
926 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
927 			r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
928 			continue;
929 		}
930 
931 		/* disable register anti-hang mechanism */
932 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
933 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
934 
935 		/* set uvd status busy */
936 		tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
937 		WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
938 	}
939 
940 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
941 		return 0;
942 
943 	/*SW clock gating */
944 	vcn_v2_5_disable_clock_gating(adev);
945 
946 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
947 		if (adev->vcn.harvest_config & (1 << i))
948 			continue;
949 		/* enable VCPU clock */
950 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
951 			UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
952 
953 		/* disable master interrupt */
954 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
955 			~UVD_MASTINT_EN__VCPU_EN_MASK);
956 
957 		/* setup mmUVD_LMI_CTRL */
958 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
959 		tmp &= ~0xff;
960 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
961 			UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
962 			UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
963 			UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
964 			UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
965 
966 		/* setup mmUVD_MPC_CNTL */
967 		tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
968 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
969 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
970 		WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
971 
972 		/* setup UVD_MPC_SET_MUXA0 */
973 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
974 			((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
975 			(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
976 			(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
977 			(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
978 
979 		/* setup UVD_MPC_SET_MUXB0 */
980 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
981 			((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
982 			(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
983 			(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
984 			(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
985 
986 		/* setup mmUVD_MPC_SET_MUX */
987 		WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
988 			((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
989 			(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
990 			(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
991 	}
992 
993 	vcn_v2_5_mc_resume(adev);
994 
995 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
996 		volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;
997 		if (adev->vcn.harvest_config & (1 << i))
998 			continue;
999 		/* VCN global tiling registers */
1000 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1001 			adev->gfx.config.gb_addr_config);
1002 		WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1003 			adev->gfx.config.gb_addr_config);
1004 
1005 		/* enable LMI MC and UMC channels */
1006 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1007 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1008 
1009 		/* unblock VCPU register access */
1010 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1011 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1012 
1013 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1014 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1015 
1016 		for (k = 0; k < 10; ++k) {
1017 			uint32_t status;
1018 
1019 			for (j = 0; j < 100; ++j) {
1020 				status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1021 				if (status & 2)
1022 					break;
1023 				if (amdgpu_emu_mode == 1)
1024 					msleep(500);
1025 				else
1026 					mdelay(10);
1027 			}
1028 			r = 0;
1029 			if (status & 2)
1030 				break;
1031 
1032 			DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1033 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1034 				UVD_VCPU_CNTL__BLK_RST_MASK,
1035 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1036 			mdelay(10);
1037 			WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1038 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1039 
1040 			mdelay(10);
1041 			r = -1;
1042 		}
1043 
1044 		if (r) {
1045 			DRM_ERROR("VCN decode not responding, giving up!!!\n");
1046 			return r;
1047 		}
1048 
1049 		/* enable master interrupt */
1050 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1051 			UVD_MASTINT_EN__VCPU_EN_MASK,
1052 			~UVD_MASTINT_EN__VCPU_EN_MASK);
1053 
1054 		/* clear the busy bit of VCN_STATUS */
1055 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1056 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1057 
1058 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1059 
1060 		ring = &adev->vcn.inst[i].ring_dec;
1061 		/* force RBC into idle state */
1062 		rb_bufsz = order_base_2(ring->ring_size);
1063 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1064 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1065 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1066 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1067 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1068 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1069 
1070 		fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1071 		/* programm the RB_BASE for ring buffer */
1072 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1073 			lower_32_bits(ring->gpu_addr));
1074 		WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1075 			upper_32_bits(ring->gpu_addr));
1076 
1077 		/* Initialize the ring buffer's read and write pointers */
1078 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1079 
1080 		ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1081 		WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1082 				lower_32_bits(ring->wptr));
1083 		fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1084 
1085 		fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1086 		ring = &adev->vcn.inst[i].ring_enc[0];
1087 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1088 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1089 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1090 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1091 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1092 		fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1093 
1094 		fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1095 		ring = &adev->vcn.inst[i].ring_enc[1];
1096 		WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1097 		WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1098 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1099 		WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1100 		WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1101 		fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1102 	}
1103 
1104 	return 0;
1105 }
1106 
1107 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1108 				struct amdgpu_mm_table *table)
1109 {
1110 	uint32_t data = 0, loop = 0, size = 0;
1111 	uint64_t addr = table->gpu_addr;
1112 	struct mmsch_v1_1_init_header *header = NULL;;
1113 
1114 	header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1115 	size = header->total_size;
1116 
1117 	/*
1118 	 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1119 	 *  memory descriptor location
1120 	 */
1121 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1122 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1123 
1124 	/* 2, update vmid of descriptor */
1125 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1126 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1127 	/* use domain0 for MM scheduler */
1128 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1129 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1130 
1131 	/* 3, notify mmsch about the size of this descriptor */
1132 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1133 
1134 	/* 4, set resp to zero */
1135 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1136 
1137 	/*
1138 	 * 5, kick off the initialization and wait until
1139 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1140 	 */
1141 	WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1142 
1143 	data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1144 	loop = 10;
1145 	while ((data & 0x10000002) != 0x10000002) {
1146 		udelay(100);
1147 		data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1148 		loop--;
1149 		if (!loop)
1150 			break;
1151 	}
1152 
1153 	if (!loop) {
1154 		dev_err(adev->dev,
1155 			"failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1156 			data);
1157 		return -EBUSY;
1158 	}
1159 
1160 	return 0;
1161 }
1162 
1163 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1164 {
1165 	struct amdgpu_ring *ring;
1166 	uint32_t offset, size, tmp, i, rb_bufsz;
1167 	uint32_t table_size = 0;
1168 	struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1169 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1170 	struct mmsch_v1_0_cmd_end end = { { 0 } };
1171 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1172 	struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1173 
1174 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1175 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1176 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1177 
1178 	header->version = MMSCH_VERSION;
1179 	header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1180 	init_table += header->total_size;
1181 
1182 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1183 		header->eng[i].table_offset = header->total_size;
1184 		header->eng[i].init_status = 0;
1185 		header->eng[i].table_size = 0;
1186 
1187 		table_size = 0;
1188 
1189 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1190 			SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1191 			~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1192 
1193 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1194 		/* mc resume*/
1195 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1196 			MMSCH_V1_0_INSERT_DIRECT_WT(
1197 				SOC15_REG_OFFSET(VCN, i,
1198 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1199 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1200 			MMSCH_V1_0_INSERT_DIRECT_WT(
1201 				SOC15_REG_OFFSET(VCN, i,
1202 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1203 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1204 			offset = 0;
1205 			MMSCH_V1_0_INSERT_DIRECT_WT(
1206 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1207 		} else {
1208 			MMSCH_V1_0_INSERT_DIRECT_WT(
1209 				SOC15_REG_OFFSET(VCN, i,
1210 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1211 				lower_32_bits(adev->vcn.inst[i].gpu_addr));
1212 			MMSCH_V1_0_INSERT_DIRECT_WT(
1213 				SOC15_REG_OFFSET(VCN, i,
1214 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1215 				upper_32_bits(adev->vcn.inst[i].gpu_addr));
1216 			offset = size;
1217 			MMSCH_V1_0_INSERT_DIRECT_WT(
1218 				SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1219 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1220 		}
1221 
1222 		MMSCH_V1_0_INSERT_DIRECT_WT(
1223 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1224 			size);
1225 		MMSCH_V1_0_INSERT_DIRECT_WT(
1226 			SOC15_REG_OFFSET(VCN, i,
1227 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1228 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1229 		MMSCH_V1_0_INSERT_DIRECT_WT(
1230 			SOC15_REG_OFFSET(VCN, i,
1231 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1232 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1233 		MMSCH_V1_0_INSERT_DIRECT_WT(
1234 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1235 			0);
1236 		MMSCH_V1_0_INSERT_DIRECT_WT(
1237 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1238 			AMDGPU_VCN_STACK_SIZE);
1239 		MMSCH_V1_0_INSERT_DIRECT_WT(
1240 			SOC15_REG_OFFSET(VCN, i,
1241 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1242 			lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1243 				AMDGPU_VCN_STACK_SIZE));
1244 		MMSCH_V1_0_INSERT_DIRECT_WT(
1245 			SOC15_REG_OFFSET(VCN, i,
1246 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1247 			upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1248 				AMDGPU_VCN_STACK_SIZE));
1249 		MMSCH_V1_0_INSERT_DIRECT_WT(
1250 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1251 			0);
1252 		MMSCH_V1_0_INSERT_DIRECT_WT(
1253 			SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1254 			AMDGPU_VCN_CONTEXT_SIZE);
1255 
1256 		ring = &adev->vcn.inst[i].ring_enc[0];
1257 		ring->wptr = 0;
1258 
1259 		MMSCH_V1_0_INSERT_DIRECT_WT(
1260 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1261 			lower_32_bits(ring->gpu_addr));
1262 		MMSCH_V1_0_INSERT_DIRECT_WT(
1263 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1264 			upper_32_bits(ring->gpu_addr));
1265 		MMSCH_V1_0_INSERT_DIRECT_WT(
1266 			SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1267 			ring->ring_size / 4);
1268 
1269 		ring = &adev->vcn.inst[i].ring_dec;
1270 		ring->wptr = 0;
1271 		MMSCH_V1_0_INSERT_DIRECT_WT(
1272 			SOC15_REG_OFFSET(VCN, i,
1273 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1274 			lower_32_bits(ring->gpu_addr));
1275 		MMSCH_V1_0_INSERT_DIRECT_WT(
1276 			SOC15_REG_OFFSET(VCN, i,
1277 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1278 			upper_32_bits(ring->gpu_addr));
1279 
1280 		/* force RBC into idle state */
1281 		rb_bufsz = order_base_2(ring->ring_size);
1282 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1283 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1284 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1285 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1286 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1287 		MMSCH_V1_0_INSERT_DIRECT_WT(
1288 			SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1289 
1290 		/* add end packet */
1291 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1292 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1293 		init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1294 
1295 		/* refine header */
1296 		header->eng[i].table_size = table_size;
1297 		header->total_size += table_size;
1298 	}
1299 
1300 	return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1301 }
1302 
1303 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1304 {
1305 	int ret_code = 0;
1306 	uint32_t tmp;
1307 
1308 	/* Wait for power status to be 1 */
1309 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1310 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1311 
1312 	/* wait for read ptr to be equal to write ptr */
1313 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1314 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
1315 
1316 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1317 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code);
1318 
1319 	tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1320 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
1321 
1322 	SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1323 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1324 
1325 	/* disable dynamic power gating mode */
1326 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1327 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1328 
1329 	return 0;
1330 }
1331 
1332 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1333 {
1334 	uint32_t tmp;
1335 	int i, r = 0;
1336 
1337 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1338 		if (adev->vcn.harvest_config & (1 << i))
1339 			continue;
1340 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1341 			r = vcn_v2_5_stop_dpg_mode(adev, i);
1342 			continue;
1343 		}
1344 
1345 		/* wait for vcn idle */
1346 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7, r);
1347 		if (r)
1348 			return r;
1349 
1350 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1351 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1352 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1353 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1354 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp, r);
1355 		if (r)
1356 			return r;
1357 
1358 		/* block LMI UMC channel */
1359 		tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1360 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1361 		WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1362 
1363 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1364 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1365 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp, r);
1366 		if (r)
1367 			return r;
1368 
1369 		/* block VCPU register access */
1370 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1371 			UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1372 			~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1373 
1374 		/* reset VCPU */
1375 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1376 			UVD_VCPU_CNTL__BLK_RST_MASK,
1377 			~UVD_VCPU_CNTL__BLK_RST_MASK);
1378 
1379 		/* disable VCPU clock */
1380 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1381 			~(UVD_VCPU_CNTL__CLK_EN_MASK));
1382 
1383 		/* clear status */
1384 		WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1385 
1386 		vcn_v2_5_enable_clock_gating(adev);
1387 
1388 		/* enable register anti-hang mechanism */
1389 		WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1390 			UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1391 			~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1392 	}
1393 
1394 	if (adev->pm.dpm_enabled)
1395 		amdgpu_dpm_enable_uvd(adev, false);
1396 
1397 	return 0;
1398 }
1399 
1400 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1401 				int inst_idx, struct dpg_pause_state *new_state)
1402 {
1403 	struct amdgpu_ring *ring;
1404 	uint32_t reg_data = 0;
1405 	int ret_code = 0;
1406 
1407 	/* pause/unpause if state is changed */
1408 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1409 		DRM_DEBUG("dpg pause state changed %d -> %d",
1410 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1411 		reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1412 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1413 
1414 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1415 			SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1416 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1417 
1418 			if (!ret_code) {
1419 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;
1420 
1421 				/* pause DPG */
1422 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1423 				WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1424 
1425 				/* wait for ACK */
1426 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1427 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1428 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, ret_code);
1429 
1430 				/* Stall DPG before WPTR/RPTR reset */
1431 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1432 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1433 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1434 
1435 				/* Restore */
1436 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1437 				ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1438 				ring->wptr = 0;
1439 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1440 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1441 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1442 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1443 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1444 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1445 
1446 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1447 				ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1448 				ring->wptr = 0;
1449 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1450 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1451 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1452 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1453 				WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1454 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1455 
1456 				/* Unstall DPG */
1457 				WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1458 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1459 
1460 				SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1461 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1462 			}
1463 		} else {
1464 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1465 			WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1466 			SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1467 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1468 		}
1469 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1470 	}
1471 
1472 	return 0;
1473 }
1474 
1475 /**
1476  * vcn_v2_5_dec_ring_get_rptr - get read pointer
1477  *
1478  * @ring: amdgpu_ring pointer
1479  *
1480  * Returns the current hardware read pointer
1481  */
1482 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1483 {
1484 	struct amdgpu_device *adev = ring->adev;
1485 
1486 	return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1487 }
1488 
1489 /**
1490  * vcn_v2_5_dec_ring_get_wptr - get write pointer
1491  *
1492  * @ring: amdgpu_ring pointer
1493  *
1494  * Returns the current hardware write pointer
1495  */
1496 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1497 {
1498 	struct amdgpu_device *adev = ring->adev;
1499 
1500 	if (ring->use_doorbell)
1501 		return adev->wb.wb[ring->wptr_offs];
1502 	else
1503 		return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1504 }
1505 
1506 /**
1507  * vcn_v2_5_dec_ring_set_wptr - set write pointer
1508  *
1509  * @ring: amdgpu_ring pointer
1510  *
1511  * Commits the write pointer to the hardware
1512  */
1513 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1514 {
1515 	struct amdgpu_device *adev = ring->adev;
1516 
1517 	if (ring->use_doorbell) {
1518 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1519 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1520 	} else {
1521 		WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1522 	}
1523 }
1524 
1525 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1526 	.type = AMDGPU_RING_TYPE_VCN_DEC,
1527 	.align_mask = 0xf,
1528 	.vmhub = AMDGPU_MMHUB_1,
1529 	.get_rptr = vcn_v2_5_dec_ring_get_rptr,
1530 	.get_wptr = vcn_v2_5_dec_ring_get_wptr,
1531 	.set_wptr = vcn_v2_5_dec_ring_set_wptr,
1532 	.emit_frame_size =
1533 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1534 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1535 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1536 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1537 		6,
1538 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1539 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
1540 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
1541 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1542 	.test_ring = vcn_v2_0_dec_ring_test_ring,
1543 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
1544 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
1545 	.insert_start = vcn_v2_0_dec_ring_insert_start,
1546 	.insert_end = vcn_v2_0_dec_ring_insert_end,
1547 	.pad_ib = amdgpu_ring_generic_pad_ib,
1548 	.begin_use = amdgpu_vcn_ring_begin_use,
1549 	.end_use = amdgpu_vcn_ring_end_use,
1550 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1551 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1552 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1553 };
1554 
1555 /**
1556  * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1557  *
1558  * @ring: amdgpu_ring pointer
1559  *
1560  * Returns the current hardware enc read pointer
1561  */
1562 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1563 {
1564 	struct amdgpu_device *adev = ring->adev;
1565 
1566 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1567 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1568 	else
1569 		return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1570 }
1571 
1572 /**
1573  * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1574  *
1575  * @ring: amdgpu_ring pointer
1576  *
1577  * Returns the current hardware enc write pointer
1578  */
1579 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1580 {
1581 	struct amdgpu_device *adev = ring->adev;
1582 
1583 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1584 		if (ring->use_doorbell)
1585 			return adev->wb.wb[ring->wptr_offs];
1586 		else
1587 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1588 	} else {
1589 		if (ring->use_doorbell)
1590 			return adev->wb.wb[ring->wptr_offs];
1591 		else
1592 			return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1593 	}
1594 }
1595 
1596 /**
1597  * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1598  *
1599  * @ring: amdgpu_ring pointer
1600  *
1601  * Commits the enc write pointer to the hardware
1602  */
1603 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1604 {
1605 	struct amdgpu_device *adev = ring->adev;
1606 
1607 	if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1608 		if (ring->use_doorbell) {
1609 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1610 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1611 		} else {
1612 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1613 		}
1614 	} else {
1615 		if (ring->use_doorbell) {
1616 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1617 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1618 		} else {
1619 			WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1620 		}
1621 	}
1622 }
1623 
1624 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1625 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1626 	.align_mask = 0x3f,
1627 	.nop = VCN_ENC_CMD_NO_OP,
1628 	.vmhub = AMDGPU_MMHUB_1,
1629 	.get_rptr = vcn_v2_5_enc_ring_get_rptr,
1630 	.get_wptr = vcn_v2_5_enc_ring_get_wptr,
1631 	.set_wptr = vcn_v2_5_enc_ring_set_wptr,
1632 	.emit_frame_size =
1633 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1634 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1635 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1636 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1637 		1, /* vcn_v2_0_enc_ring_insert_end */
1638 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1639 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1640 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1641 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1642 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1643 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
1644 	.insert_nop = amdgpu_ring_insert_nop,
1645 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1646 	.pad_ib = amdgpu_ring_generic_pad_ib,
1647 	.begin_use = amdgpu_vcn_ring_begin_use,
1648 	.end_use = amdgpu_vcn_ring_end_use,
1649 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1650 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1651 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1652 };
1653 
1654 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1655 {
1656 	int i;
1657 
1658 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1659 		if (adev->vcn.harvest_config & (1 << i))
1660 			continue;
1661 		adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1662 		adev->vcn.inst[i].ring_dec.me = i;
1663 		DRM_INFO("VCN(%d) decode is enabled in VM mode\n", i);
1664 	}
1665 }
1666 
1667 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1668 {
1669 	int i, j;
1670 
1671 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1672 		if (adev->vcn.harvest_config & (1 << j))
1673 			continue;
1674 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1675 			adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1676 			adev->vcn.inst[j].ring_enc[i].me = j;
1677 		}
1678 		DRM_INFO("VCN(%d) encode is enabled in VM mode\n", j);
1679 	}
1680 }
1681 
1682 static bool vcn_v2_5_is_idle(void *handle)
1683 {
1684 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1685 	int i, ret = 1;
1686 
1687 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1688 		if (adev->vcn.harvest_config & (1 << i))
1689 			continue;
1690 		ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1691 	}
1692 
1693 	return ret;
1694 }
1695 
1696 static int vcn_v2_5_wait_for_idle(void *handle)
1697 {
1698 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1699 	int i, ret = 0;
1700 
1701 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1702 		if (adev->vcn.harvest_config & (1 << i))
1703 			continue;
1704 		SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1705 			UVD_STATUS__IDLE, ret);
1706 		if (ret)
1707 			return ret;
1708 	}
1709 
1710 	return ret;
1711 }
1712 
1713 static int vcn_v2_5_set_clockgating_state(void *handle,
1714 					  enum amd_clockgating_state state)
1715 {
1716 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1717 	bool enable = (state == AMD_CG_STATE_GATE);
1718 
1719 	if (amdgpu_sriov_vf(adev))
1720 		return 0;
1721 
1722 	if (enable) {
1723 		if (!vcn_v2_5_is_idle(handle))
1724 			return -EBUSY;
1725 		vcn_v2_5_enable_clock_gating(adev);
1726 	} else {
1727 		vcn_v2_5_disable_clock_gating(adev);
1728 	}
1729 
1730 	return 0;
1731 }
1732 
1733 static int vcn_v2_5_set_powergating_state(void *handle,
1734 					  enum amd_powergating_state state)
1735 {
1736 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1737 	int ret;
1738 
1739 	if (amdgpu_sriov_vf(adev))
1740 		return 0;
1741 
1742 	if(state == adev->vcn.cur_state)
1743 		return 0;
1744 
1745 	if (state == AMD_PG_STATE_GATE)
1746 		ret = vcn_v2_5_stop(adev);
1747 	else
1748 		ret = vcn_v2_5_start(adev);
1749 
1750 	if(!ret)
1751 		adev->vcn.cur_state = state;
1752 
1753 	return ret;
1754 }
1755 
1756 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1757 					struct amdgpu_irq_src *source,
1758 					unsigned type,
1759 					enum amdgpu_interrupt_state state)
1760 {
1761 	return 0;
1762 }
1763 
1764 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1765 				      struct amdgpu_irq_src *source,
1766 				      struct amdgpu_iv_entry *entry)
1767 {
1768 	uint32_t ip_instance;
1769 
1770 	switch (entry->client_id) {
1771 	case SOC15_IH_CLIENTID_VCN:
1772 		ip_instance = 0;
1773 		break;
1774 	case SOC15_IH_CLIENTID_VCN1:
1775 		ip_instance = 1;
1776 		break;
1777 	default:
1778 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1779 		return 0;
1780 	}
1781 
1782 	DRM_DEBUG("IH: VCN TRAP\n");
1783 
1784 	switch (entry->src_id) {
1785 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1786 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1787 		break;
1788 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1789 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1790 		break;
1791 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1792 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1793 		break;
1794 	default:
1795 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1796 			  entry->src_id, entry->src_data[0]);
1797 		break;
1798 	}
1799 
1800 	return 0;
1801 }
1802 
1803 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1804 	.set = vcn_v2_5_set_interrupt_state,
1805 	.process = vcn_v2_5_process_interrupt,
1806 };
1807 
1808 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1809 {
1810 	int i;
1811 
1812 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1813 		if (adev->vcn.harvest_config & (1 << i))
1814 			continue;
1815 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1816 		adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1817 	}
1818 }
1819 
1820 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1821 	.name = "vcn_v2_5",
1822 	.early_init = vcn_v2_5_early_init,
1823 	.late_init = NULL,
1824 	.sw_init = vcn_v2_5_sw_init,
1825 	.sw_fini = vcn_v2_5_sw_fini,
1826 	.hw_init = vcn_v2_5_hw_init,
1827 	.hw_fini = vcn_v2_5_hw_fini,
1828 	.suspend = vcn_v2_5_suspend,
1829 	.resume = vcn_v2_5_resume,
1830 	.is_idle = vcn_v2_5_is_idle,
1831 	.wait_for_idle = vcn_v2_5_wait_for_idle,
1832 	.check_soft_reset = NULL,
1833 	.pre_soft_reset = NULL,
1834 	.soft_reset = NULL,
1835 	.post_soft_reset = NULL,
1836 	.set_clockgating_state = vcn_v2_5_set_clockgating_state,
1837 	.set_powergating_state = vcn_v2_5_set_powergating_state,
1838 };
1839 
1840 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
1841 {
1842 		.type = AMD_IP_BLOCK_TYPE_VCN,
1843 		.major = 2,
1844 		.minor = 5,
1845 		.rev = 0,
1846 		.funcs = &vcn_v2_5_ip_funcs,
1847 };
1848