xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c (revision c4a11bf4)
1 /*
2  * Copyright 2018 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 "soc15.h"
30 #include "soc15d.h"
31 #include "amdgpu_pm.h"
32 #include "amdgpu_psp.h"
33 #include "mmsch_v2_0.h"
34 #include "vcn_v2_0.h"
35 
36 #include "vcn/vcn_2_0_0_offset.h"
37 #include "vcn/vcn_2_0_0_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
39 
40 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x1fd
41 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x503
42 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x504
43 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x505
44 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x53f
45 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x54a
46 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
47 
48 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x1e1
49 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x5a6
50 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x5a7
51 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x1e2
52 
53 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev);
54 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev);
55 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev);
56 static int vcn_v2_0_set_powergating_state(void *handle,
57 				enum amd_powergating_state state);
58 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
59 				int inst_idx, struct dpg_pause_state *new_state);
60 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev);
61 /**
62  * vcn_v2_0_early_init - set function pointers
63  *
64  * @handle: amdgpu_device pointer
65  *
66  * Set ring and irq function pointers
67  */
68 static int vcn_v2_0_early_init(void *handle)
69 {
70 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
71 
72 	if (amdgpu_sriov_vf(adev))
73 		adev->vcn.num_enc_rings = 1;
74 	else
75 		adev->vcn.num_enc_rings = 2;
76 
77 	vcn_v2_0_set_dec_ring_funcs(adev);
78 	vcn_v2_0_set_enc_ring_funcs(adev);
79 	vcn_v2_0_set_irq_funcs(adev);
80 
81 	return 0;
82 }
83 
84 /**
85  * vcn_v2_0_sw_init - sw init for VCN block
86  *
87  * @handle: amdgpu_device pointer
88  *
89  * Load firmware and sw initialization
90  */
91 static int vcn_v2_0_sw_init(void *handle)
92 {
93 	struct amdgpu_ring *ring;
94 	int i, r;
95 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
96 	volatile struct amdgpu_fw_shared *fw_shared;
97 
98 	/* VCN DEC TRAP */
99 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
100 			      VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT,
101 			      &adev->vcn.inst->irq);
102 	if (r)
103 		return r;
104 
105 	/* VCN ENC TRAP */
106 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
107 		r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
108 				      i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE,
109 				      &adev->vcn.inst->irq);
110 		if (r)
111 			return r;
112 	}
113 
114 	r = amdgpu_vcn_sw_init(adev);
115 	if (r)
116 		return r;
117 
118 	amdgpu_vcn_setup_ucode(adev);
119 
120 	r = amdgpu_vcn_resume(adev);
121 	if (r)
122 		return r;
123 
124 	ring = &adev->vcn.inst->ring_dec;
125 
126 	ring->use_doorbell = true;
127 	ring->doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
128 
129 	sprintf(ring->name, "vcn_dec");
130 	r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
131 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
132 	if (r)
133 		return r;
134 
135 	adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
136 	adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
137 	adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
138 	adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
139 	adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
140 	adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
141 
142 	adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
143 	adev->vcn.inst->external.scratch9 = SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9);
144 	adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
145 	adev->vcn.inst->external.data0 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0);
146 	adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
147 	adev->vcn.inst->external.data1 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1);
148 	adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
149 	adev->vcn.inst->external.cmd = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD);
150 	adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
151 	adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP);
152 
153 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
154 		enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
155 
156 		ring = &adev->vcn.inst->ring_enc[i];
157 		ring->use_doorbell = true;
158 		if (!amdgpu_sriov_vf(adev))
159 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + i;
160 		else
161 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1 + i;
162 		sprintf(ring->name, "vcn_enc%d", i);
163 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
164 				     hw_prio, NULL);
165 		if (r)
166 			return r;
167 	}
168 
169 	adev->vcn.pause_dpg_mode = vcn_v2_0_pause_dpg_mode;
170 
171 	r = amdgpu_virt_alloc_mm_table(adev);
172 	if (r)
173 		return r;
174 
175 	fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
176 	fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
177 	return 0;
178 }
179 
180 /**
181  * vcn_v2_0_sw_fini - sw fini for VCN block
182  *
183  * @handle: amdgpu_device pointer
184  *
185  * VCN suspend and free up sw allocation
186  */
187 static int vcn_v2_0_sw_fini(void *handle)
188 {
189 	int r, idx;
190 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
191 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
192 
193 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
194 		fw_shared->present_flag_0 = 0;
195 		drm_dev_exit(idx);
196 	}
197 
198 	amdgpu_virt_free_mm_table(adev);
199 
200 	r = amdgpu_vcn_suspend(adev);
201 	if (r)
202 		return r;
203 
204 	r = amdgpu_vcn_sw_fini(adev);
205 
206 	return r;
207 }
208 
209 /**
210  * vcn_v2_0_hw_init - start and test VCN block
211  *
212  * @handle: amdgpu_device pointer
213  *
214  * Initialize the hardware, boot up the VCPU and do some testing
215  */
216 static int vcn_v2_0_hw_init(void *handle)
217 {
218 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
219 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
220 	int i, r;
221 
222 	adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
223 					     ring->doorbell_index, 0);
224 
225 	if (amdgpu_sriov_vf(adev))
226 		vcn_v2_0_start_sriov(adev);
227 
228 	r = amdgpu_ring_test_helper(ring);
229 	if (r)
230 		goto done;
231 
232 	//Disable vcn decode for sriov
233 	if (amdgpu_sriov_vf(adev))
234 		ring->sched.ready = false;
235 
236 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
237 		ring = &adev->vcn.inst->ring_enc[i];
238 		r = amdgpu_ring_test_helper(ring);
239 		if (r)
240 			goto done;
241 	}
242 
243 done:
244 	if (!r)
245 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
246 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
247 
248 	return r;
249 }
250 
251 /**
252  * vcn_v2_0_hw_fini - stop the hardware block
253  *
254  * @handle: amdgpu_device pointer
255  *
256  * Stop the VCN block, mark ring as not ready any more
257  */
258 static int vcn_v2_0_hw_fini(void *handle)
259 {
260 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
261 
262 	cancel_delayed_work_sync(&adev->vcn.idle_work);
263 
264 	if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
265 	    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
266 	      RREG32_SOC15(VCN, 0, mmUVD_STATUS)))
267 		vcn_v2_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
268 
269 	return 0;
270 }
271 
272 /**
273  * vcn_v2_0_suspend - suspend VCN block
274  *
275  * @handle: amdgpu_device pointer
276  *
277  * HW fini and suspend VCN block
278  */
279 static int vcn_v2_0_suspend(void *handle)
280 {
281 	int r;
282 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
283 
284 	r = vcn_v2_0_hw_fini(adev);
285 	if (r)
286 		return r;
287 
288 	r = amdgpu_vcn_suspend(adev);
289 
290 	return r;
291 }
292 
293 /**
294  * vcn_v2_0_resume - resume VCN block
295  *
296  * @handle: amdgpu_device pointer
297  *
298  * Resume firmware and hw init VCN block
299  */
300 static int vcn_v2_0_resume(void *handle)
301 {
302 	int r;
303 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
304 
305 	r = amdgpu_vcn_resume(adev);
306 	if (r)
307 		return r;
308 
309 	r = vcn_v2_0_hw_init(adev);
310 
311 	return r;
312 }
313 
314 /**
315  * vcn_v2_0_mc_resume - memory controller programming
316  *
317  * @adev: amdgpu_device pointer
318  *
319  * Let the VCN memory controller know it's offsets
320  */
321 static void vcn_v2_0_mc_resume(struct amdgpu_device *adev)
322 {
323 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
324 	uint32_t offset;
325 
326 	if (amdgpu_sriov_vf(adev))
327 		return;
328 
329 	/* cache window 0: fw */
330 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
331 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
332 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo));
333 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
334 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi));
335 		WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0);
336 		offset = 0;
337 	} else {
338 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
339 			lower_32_bits(adev->vcn.inst->gpu_addr));
340 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
341 			upper_32_bits(adev->vcn.inst->gpu_addr));
342 		offset = size;
343 		WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0,
344 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
345 	}
346 
347 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size);
348 
349 	/* cache window 1: stack */
350 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
351 		lower_32_bits(adev->vcn.inst->gpu_addr + offset));
352 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
353 		upper_32_bits(adev->vcn.inst->gpu_addr + offset));
354 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0);
355 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
356 
357 	/* cache window 2: context */
358 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
359 		lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
360 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
361 		upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
362 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0);
363 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
364 
365 	/* non-cache window */
366 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
367 		lower_32_bits(adev->vcn.inst->fw_shared_gpu_addr));
368 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
369 		upper_32_bits(adev->vcn.inst->fw_shared_gpu_addr));
370 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
371 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0,
372 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
373 
374 	WREG32_SOC15(UVD, 0, mmUVD_GFX10_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
375 }
376 
377 static void vcn_v2_0_mc_resume_dpg_mode(struct amdgpu_device *adev, bool indirect)
378 {
379 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
380 	uint32_t offset;
381 
382 	/* cache window 0: fw */
383 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
384 		if (!indirect) {
385 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
386 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
387 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 0, indirect);
388 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
389 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
390 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 0, indirect);
391 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
392 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
393 		} else {
394 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
395 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
396 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
397 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
398 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
399 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
400 		}
401 		offset = 0;
402 	} else {
403 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
404 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
405 			lower_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
406 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
407 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
408 			upper_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
409 		offset = size;
410 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
411 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
412 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
413 	}
414 
415 	if (!indirect)
416 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
417 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
418 	else
419 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
420 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
421 
422 	/* cache window 1: stack */
423 	if (!indirect) {
424 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
425 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
426 			lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
427 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
428 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
429 			upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
430 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
431 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
432 	} else {
433 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
434 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
435 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
436 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
437 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
438 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
439 	}
440 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
441 		UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
442 
443 	/* cache window 2: context */
444 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
445 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
446 		lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
447 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
448 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
449 		upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
450 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
451 		UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
452 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
453 		UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
454 
455 	/* non-cache window */
456 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
457 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
458 		lower_32_bits(adev->vcn.inst->fw_shared_gpu_addr), 0, indirect);
459 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
460 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
461 		upper_32_bits(adev->vcn.inst->fw_shared_gpu_addr), 0, indirect);
462 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
463 		UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
464 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
465 		UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0),
466 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
467 
468 	/* VCN global tiling registers */
469 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
470 		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
471 }
472 
473 /**
474  * vcn_v2_0_disable_clock_gating - disable VCN clock gating
475  *
476  * @adev: amdgpu_device pointer
477  *
478  * Disable clock gating for VCN block
479  */
480 static void vcn_v2_0_disable_clock_gating(struct amdgpu_device *adev)
481 {
482 	uint32_t data;
483 
484 	if (amdgpu_sriov_vf(adev))
485 		return;
486 
487 	/* UVD disable CGC */
488 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
489 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
490 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
491 	else
492 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
493 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
494 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
495 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
496 
497 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_GATE);
498 	data &= ~(UVD_CGC_GATE__SYS_MASK
499 		| UVD_CGC_GATE__UDEC_MASK
500 		| UVD_CGC_GATE__MPEG2_MASK
501 		| UVD_CGC_GATE__REGS_MASK
502 		| UVD_CGC_GATE__RBC_MASK
503 		| UVD_CGC_GATE__LMI_MC_MASK
504 		| UVD_CGC_GATE__LMI_UMC_MASK
505 		| UVD_CGC_GATE__IDCT_MASK
506 		| UVD_CGC_GATE__MPRD_MASK
507 		| UVD_CGC_GATE__MPC_MASK
508 		| UVD_CGC_GATE__LBSI_MASK
509 		| UVD_CGC_GATE__LRBBM_MASK
510 		| UVD_CGC_GATE__UDEC_RE_MASK
511 		| UVD_CGC_GATE__UDEC_CM_MASK
512 		| UVD_CGC_GATE__UDEC_IT_MASK
513 		| UVD_CGC_GATE__UDEC_DB_MASK
514 		| UVD_CGC_GATE__UDEC_MP_MASK
515 		| UVD_CGC_GATE__WCB_MASK
516 		| UVD_CGC_GATE__VCPU_MASK
517 		| UVD_CGC_GATE__SCPU_MASK);
518 	WREG32_SOC15(VCN, 0, mmUVD_CGC_GATE, data);
519 
520 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
521 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
522 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
523 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
524 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
525 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
526 		| UVD_CGC_CTRL__SYS_MODE_MASK
527 		| UVD_CGC_CTRL__UDEC_MODE_MASK
528 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
529 		| UVD_CGC_CTRL__REGS_MODE_MASK
530 		| UVD_CGC_CTRL__RBC_MODE_MASK
531 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
532 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
533 		| UVD_CGC_CTRL__IDCT_MODE_MASK
534 		| UVD_CGC_CTRL__MPRD_MODE_MASK
535 		| UVD_CGC_CTRL__MPC_MODE_MASK
536 		| UVD_CGC_CTRL__LBSI_MODE_MASK
537 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
538 		| UVD_CGC_CTRL__WCB_MODE_MASK
539 		| UVD_CGC_CTRL__VCPU_MODE_MASK
540 		| UVD_CGC_CTRL__SCPU_MODE_MASK);
541 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
542 
543 	/* turn on */
544 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE);
545 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
546 		| UVD_SUVD_CGC_GATE__SIT_MASK
547 		| UVD_SUVD_CGC_GATE__SMP_MASK
548 		| UVD_SUVD_CGC_GATE__SCM_MASK
549 		| UVD_SUVD_CGC_GATE__SDB_MASK
550 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
551 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
552 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
553 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
554 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
555 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
556 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
557 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
558 		| UVD_SUVD_CGC_GATE__SCLR_MASK
559 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
560 		| UVD_SUVD_CGC_GATE__ENT_MASK
561 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
562 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
563 		| UVD_SUVD_CGC_GATE__SITE_MASK
564 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
565 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
566 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
567 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
568 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
569 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE, data);
570 
571 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
572 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
573 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
574 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
575 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
576 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
577 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
578 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
579 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
580 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
581 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
582 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
583 }
584 
585 static void vcn_v2_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
586 		uint8_t sram_sel, uint8_t indirect)
587 {
588 	uint32_t reg_data = 0;
589 
590 	/* enable sw clock gating control */
591 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
592 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
593 	else
594 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
595 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
596 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
597 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
598 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
599 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
600 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
601 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
602 		 UVD_CGC_CTRL__SYS_MODE_MASK |
603 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
604 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
605 		 UVD_CGC_CTRL__REGS_MODE_MASK |
606 		 UVD_CGC_CTRL__RBC_MODE_MASK |
607 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
608 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
609 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
610 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
611 		 UVD_CGC_CTRL__MPC_MODE_MASK |
612 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
613 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
614 		 UVD_CGC_CTRL__WCB_MODE_MASK |
615 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
616 		 UVD_CGC_CTRL__SCPU_MODE_MASK);
617 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
618 		UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
619 
620 	/* turn off clock gating */
621 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
622 		UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
623 
624 	/* turn on SUVD clock gating */
625 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
626 		UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
627 
628 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
629 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
630 		UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
631 }
632 
633 /**
634  * vcn_v2_0_enable_clock_gating - enable VCN clock gating
635  *
636  * @adev: amdgpu_device pointer
637  *
638  * Enable clock gating for VCN block
639  */
640 static void vcn_v2_0_enable_clock_gating(struct amdgpu_device *adev)
641 {
642 	uint32_t data = 0;
643 
644 	if (amdgpu_sriov_vf(adev))
645 		return;
646 
647 	/* enable UVD CGC */
648 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
649 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
650 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
651 	else
652 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
653 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
654 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
655 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
656 
657 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
658 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
659 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
660 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
661 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
662 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
663 		| UVD_CGC_CTRL__SYS_MODE_MASK
664 		| UVD_CGC_CTRL__UDEC_MODE_MASK
665 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
666 		| UVD_CGC_CTRL__REGS_MODE_MASK
667 		| UVD_CGC_CTRL__RBC_MODE_MASK
668 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
669 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
670 		| UVD_CGC_CTRL__IDCT_MODE_MASK
671 		| UVD_CGC_CTRL__MPRD_MODE_MASK
672 		| UVD_CGC_CTRL__MPC_MODE_MASK
673 		| UVD_CGC_CTRL__LBSI_MODE_MASK
674 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
675 		| UVD_CGC_CTRL__WCB_MODE_MASK
676 		| UVD_CGC_CTRL__VCPU_MODE_MASK
677 		| UVD_CGC_CTRL__SCPU_MODE_MASK);
678 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
679 
680 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
681 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
682 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
683 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
684 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
685 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
686 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
687 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
688 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
689 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
690 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
691 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
692 }
693 
694 static void vcn_v2_0_disable_static_power_gating(struct amdgpu_device *adev)
695 {
696 	uint32_t data = 0;
697 
698 	if (amdgpu_sriov_vf(adev))
699 		return;
700 
701 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
702 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
703 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
704 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
705 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
706 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
707 			| 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
708 			| 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
709 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
710 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
711 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
712 
713 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
714 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS,
715 			UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON_2_0, 0xFFFFF);
716 	} else {
717 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
718 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
719 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
720 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
721 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
722 			| 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
723 			| 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
724 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
725 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
726 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
727 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
728 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0,  0xFFFFF);
729 	}
730 
731 	/* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS,
732 	 * UVDU_PWR_STATUS are 0 (power on) */
733 
734 	data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
735 	data &= ~0x103;
736 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
737 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
738 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
739 
740 	WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
741 }
742 
743 static void vcn_v2_0_enable_static_power_gating(struct amdgpu_device *adev)
744 {
745 	uint32_t data = 0;
746 
747 	if (amdgpu_sriov_vf(adev))
748 		return;
749 
750 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
751 		/* Before power off, this indicator has to be turned on */
752 		data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
753 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
754 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
755 		WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
756 
757 
758 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
759 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
760 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
761 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
762 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
763 			| 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
764 			| 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
765 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
766 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
767 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
768 
769 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
770 
771 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
772 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
773 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
774 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
775 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
776 			| 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT
777 			| 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT
778 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
779 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
780 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT);
781 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFF);
782 	}
783 }
784 
785 static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
786 {
787 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
788 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
789 	uint32_t rb_bufsz, tmp;
790 
791 	vcn_v2_0_enable_static_power_gating(adev);
792 
793 	/* enable dynamic power gating mode */
794 	tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS);
795 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
796 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
797 	WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp);
798 
799 	if (indirect)
800 		adev->vcn.inst->dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst->dpg_sram_cpu_addr;
801 
802 	/* enable clock gating */
803 	vcn_v2_0_clock_gating_dpg_mode(adev, 0, indirect);
804 
805 	/* enable VCPU clock */
806 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
807 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
808 	tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK;
809 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
810 		UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
811 
812 	/* disable master interupt */
813 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
814 		UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
815 
816 	/* setup mmUVD_LMI_CTRL */
817 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
818 		UVD_LMI_CTRL__REQ_MODE_MASK |
819 		UVD_LMI_CTRL__CRC_RESET_MASK |
820 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
821 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
822 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
823 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
824 		0x00100000L);
825 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
826 		UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
827 
828 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
829 		UVD, 0, mmUVD_MPC_CNTL),
830 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
831 
832 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
833 		UVD, 0, mmUVD_MPC_SET_MUXA0),
834 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
835 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
836 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
837 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
838 
839 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
840 		UVD, 0, mmUVD_MPC_SET_MUXB0),
841 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
842 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
843 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
844 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
845 
846 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
847 		UVD, 0, mmUVD_MPC_SET_MUX),
848 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
849 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
850 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
851 
852 	vcn_v2_0_mc_resume_dpg_mode(adev, indirect);
853 
854 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
855 		UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
856 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
857 		UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
858 
859 	/* release VCPU reset to boot */
860 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
861 		UVD, 0, mmUVD_SOFT_RESET), 0, 0, indirect);
862 
863 	/* enable LMI MC and UMC channels */
864 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
865 		UVD, 0, mmUVD_LMI_CTRL2),
866 		0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 0, indirect);
867 
868 	/* enable master interrupt */
869 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
870 		UVD, 0, mmUVD_MASTINT_EN),
871 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
872 
873 	if (indirect)
874 		psp_update_vcn_sram(adev, 0, adev->vcn.inst->dpg_sram_gpu_addr,
875 				    (uint32_t)((uintptr_t)adev->vcn.inst->dpg_sram_curr_addr -
876 					       (uintptr_t)adev->vcn.inst->dpg_sram_cpu_addr));
877 
878 	/* force RBC into idle state */
879 	rb_bufsz = order_base_2(ring->ring_size);
880 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
881 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
882 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
883 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
884 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
885 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
886 
887 	/* Stall DPG before WPTR/RPTR reset */
888 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
889 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
890 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
891 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
892 
893 	/* set the write pointer delay */
894 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0);
895 
896 	/* set the wb address */
897 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR,
898 		(upper_32_bits(ring->gpu_addr) >> 2));
899 
900 	/* program the RB_BASE for ring buffer */
901 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
902 		lower_32_bits(ring->gpu_addr));
903 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
904 		upper_32_bits(ring->gpu_addr));
905 
906 	/* Initialize the ring buffer's read and write pointers */
907 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
908 
909 	WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0);
910 
911 	ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
912 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
913 		lower_32_bits(ring->wptr));
914 
915 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
916 	/* Unstall DPG */
917 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
918 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
919 	return 0;
920 }
921 
922 static int vcn_v2_0_start(struct amdgpu_device *adev)
923 {
924 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
925 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
926 	uint32_t rb_bufsz, tmp;
927 	uint32_t lmi_swap_cntl;
928 	int i, j, r;
929 
930 	if (adev->pm.dpm_enabled)
931 		amdgpu_dpm_enable_uvd(adev, true);
932 
933 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
934 		return vcn_v2_0_start_dpg_mode(adev, adev->vcn.indirect_sram);
935 
936 	vcn_v2_0_disable_static_power_gating(adev);
937 
938 	/* set uvd status busy */
939 	tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
940 	WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp);
941 
942 	/*SW clock gating */
943 	vcn_v2_0_disable_clock_gating(adev);
944 
945 	/* enable VCPU clock */
946 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL),
947 		UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
948 
949 	/* disable master interrupt */
950 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0,
951 		~UVD_MASTINT_EN__VCPU_EN_MASK);
952 
953 	/* setup mmUVD_LMI_CTRL */
954 	tmp = RREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL);
955 	WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, tmp |
956 		UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
957 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
958 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
959 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
960 
961 	/* setup mmUVD_MPC_CNTL */
962 	tmp = RREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL);
963 	tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
964 	tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
965 	WREG32_SOC15(VCN, 0, mmUVD_MPC_CNTL, tmp);
966 
967 	/* setup UVD_MPC_SET_MUXA0 */
968 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0,
969 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
970 		(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
971 		(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
972 		(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
973 
974 	/* setup UVD_MPC_SET_MUXB0 */
975 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0,
976 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
977 		(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
978 		(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
979 		(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
980 
981 	/* setup mmUVD_MPC_SET_MUX */
982 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX,
983 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
984 		(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
985 		(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
986 
987 	vcn_v2_0_mc_resume(adev);
988 
989 	/* release VCPU reset to boot */
990 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
991 		~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
992 
993 	/* enable LMI MC and UMC channels */
994 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0,
995 		~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
996 
997 	tmp = RREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET);
998 	tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
999 	tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1000 	WREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET, tmp);
1001 
1002 	/* disable byte swapping */
1003 	lmi_swap_cntl = 0;
1004 #ifdef __BIG_ENDIAN
1005 	/* swap (8 in 32) RB and IB */
1006 	lmi_swap_cntl = 0xa;
1007 #endif
1008 	WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
1009 
1010 	for (i = 0; i < 10; ++i) {
1011 		uint32_t status;
1012 
1013 		for (j = 0; j < 100; ++j) {
1014 			status = RREG32_SOC15(UVD, 0, mmUVD_STATUS);
1015 			if (status & 2)
1016 				break;
1017 			mdelay(10);
1018 		}
1019 		r = 0;
1020 		if (status & 2)
1021 			break;
1022 
1023 		DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1024 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1025 			UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
1026 			~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1027 		mdelay(10);
1028 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
1029 			~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1030 		mdelay(10);
1031 		r = -1;
1032 	}
1033 
1034 	if (r) {
1035 		DRM_ERROR("VCN decode not responding, giving up!!!\n");
1036 		return r;
1037 	}
1038 
1039 	/* enable master interrupt */
1040 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
1041 		UVD_MASTINT_EN__VCPU_EN_MASK,
1042 		~UVD_MASTINT_EN__VCPU_EN_MASK);
1043 
1044 	/* clear the busy bit of VCN_STATUS */
1045 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0,
1046 		~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1047 
1048 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_VMID, 0);
1049 
1050 	/* force RBC into idle state */
1051 	rb_bufsz = order_base_2(ring->ring_size);
1052 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1053 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1054 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1055 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1056 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1057 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
1058 
1059 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1060 	/* program the RB_BASE for ring buffer */
1061 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1062 		lower_32_bits(ring->gpu_addr));
1063 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1064 		upper_32_bits(ring->gpu_addr));
1065 
1066 	/* Initialize the ring buffer's read and write pointers */
1067 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
1068 
1069 	ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1070 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1071 			lower_32_bits(ring->wptr));
1072 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1073 
1074 	fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1075 	ring = &adev->vcn.inst->ring_enc[0];
1076 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1077 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1078 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1079 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1080 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1081 	fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1082 
1083 	fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1084 	ring = &adev->vcn.inst->ring_enc[1];
1085 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1086 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1087 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1088 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1089 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1090 	fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1091 
1092 	return 0;
1093 }
1094 
1095 static int vcn_v2_0_stop_dpg_mode(struct amdgpu_device *adev)
1096 {
1097 	uint32_t tmp;
1098 
1099 	/* Wait for power status to be 1 */
1100 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1101 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1102 
1103 	/* wait for read ptr to be equal to write ptr */
1104 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1105 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1106 
1107 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1108 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1109 
1110 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1111 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1112 
1113 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1114 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1115 
1116 	/* disable dynamic power gating mode */
1117 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0,
1118 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1119 
1120 	return 0;
1121 }
1122 
1123 static int vcn_v2_0_stop(struct amdgpu_device *adev)
1124 {
1125 	uint32_t tmp;
1126 	int r;
1127 
1128 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1129 		r = vcn_v2_0_stop_dpg_mode(adev);
1130 		if (r)
1131 			return r;
1132 		goto power_off;
1133 	}
1134 
1135 	/* wait for uvd idle */
1136 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1137 	if (r)
1138 		return r;
1139 
1140 	tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1141 		UVD_LMI_STATUS__READ_CLEAN_MASK |
1142 		UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1143 		UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1144 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp);
1145 	if (r)
1146 		return r;
1147 
1148 	/* stall UMC channel */
1149 	tmp = RREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2);
1150 	tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1151 	WREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2, tmp);
1152 
1153 	tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1154 		UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1155 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp);
1156 	if (r)
1157 		return r;
1158 
1159 	/* disable VCPU clock */
1160 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0,
1161 		~(UVD_VCPU_CNTL__CLK_EN_MASK));
1162 
1163 	/* reset LMI UMC */
1164 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1165 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK,
1166 		~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
1167 
1168 	/* reset LMI */
1169 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1170 		UVD_SOFT_RESET__LMI_SOFT_RESET_MASK,
1171 		~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
1172 
1173 	/* reset VCPU */
1174 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1175 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
1176 		~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1177 
1178 	/* clear status */
1179 	WREG32_SOC15(VCN, 0, mmUVD_STATUS, 0);
1180 
1181 	vcn_v2_0_enable_clock_gating(adev);
1182 	vcn_v2_0_enable_static_power_gating(adev);
1183 
1184 power_off:
1185 	if (adev->pm.dpm_enabled)
1186 		amdgpu_dpm_enable_uvd(adev, false);
1187 
1188 	return 0;
1189 }
1190 
1191 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
1192 				int inst_idx, struct dpg_pause_state *new_state)
1193 {
1194 	struct amdgpu_ring *ring;
1195 	uint32_t reg_data = 0;
1196 	int ret_code;
1197 
1198 	/* pause/unpause if state is changed */
1199 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1200 		DRM_DEBUG("dpg pause state changed %d -> %d",
1201 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1202 		reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) &
1203 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1204 
1205 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1206 			ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 0x1,
1207 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1208 
1209 			if (!ret_code) {
1210 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
1211 				/* pause DPG */
1212 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1213 				WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1214 
1215 				/* wait for ACK */
1216 				SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE,
1217 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1218 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1219 
1220 				/* Stall DPG before WPTR/RPTR reset */
1221 				WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
1222 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1223 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1224 				/* Restore */
1225 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1226 				ring = &adev->vcn.inst->ring_enc[0];
1227 				ring->wptr = 0;
1228 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1229 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1230 				WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1231 				WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1232 				WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1233 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1234 
1235 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1236 				ring = &adev->vcn.inst->ring_enc[1];
1237 				ring->wptr = 0;
1238 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1239 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1240 				WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1241 				WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1242 				WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1243 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1244 
1245 				fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1246 				WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1247 					   RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF);
1248 				fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1249 				/* Unstall DPG */
1250 				WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
1251 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1252 
1253 				SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS,
1254 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON,
1255 					   UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1256 			}
1257 		} else {
1258 			/* unpause dpg, no need to wait */
1259 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1260 			WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1261 		}
1262 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1263 	}
1264 
1265 	return 0;
1266 }
1267 
1268 static bool vcn_v2_0_is_idle(void *handle)
1269 {
1270 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1271 
1272 	return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE);
1273 }
1274 
1275 static int vcn_v2_0_wait_for_idle(void *handle)
1276 {
1277 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1278 	int ret;
1279 
1280 	ret = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE,
1281 		UVD_STATUS__IDLE);
1282 
1283 	return ret;
1284 }
1285 
1286 static int vcn_v2_0_set_clockgating_state(void *handle,
1287 					  enum amd_clockgating_state state)
1288 {
1289 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1290 	bool enable = (state == AMD_CG_STATE_GATE);
1291 
1292 	if (amdgpu_sriov_vf(adev))
1293 		return 0;
1294 
1295 	if (enable) {
1296 		/* wait for STATUS to clear */
1297 		if (!vcn_v2_0_is_idle(handle))
1298 			return -EBUSY;
1299 		vcn_v2_0_enable_clock_gating(adev);
1300 	} else {
1301 		/* disable HW gating and enable Sw gating */
1302 		vcn_v2_0_disable_clock_gating(adev);
1303 	}
1304 	return 0;
1305 }
1306 
1307 /**
1308  * vcn_v2_0_dec_ring_get_rptr - get read pointer
1309  *
1310  * @ring: amdgpu_ring pointer
1311  *
1312  * Returns the current hardware read pointer
1313  */
1314 static uint64_t vcn_v2_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1315 {
1316 	struct amdgpu_device *adev = ring->adev;
1317 
1318 	return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1319 }
1320 
1321 /**
1322  * vcn_v2_0_dec_ring_get_wptr - get write pointer
1323  *
1324  * @ring: amdgpu_ring pointer
1325  *
1326  * Returns the current hardware write pointer
1327  */
1328 static uint64_t vcn_v2_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1329 {
1330 	struct amdgpu_device *adev = ring->adev;
1331 
1332 	if (ring->use_doorbell)
1333 		return adev->wb.wb[ring->wptr_offs];
1334 	else
1335 		return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR);
1336 }
1337 
1338 /**
1339  * vcn_v2_0_dec_ring_set_wptr - set write pointer
1340  *
1341  * @ring: amdgpu_ring pointer
1342  *
1343  * Commits the write pointer to the hardware
1344  */
1345 static void vcn_v2_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1346 {
1347 	struct amdgpu_device *adev = ring->adev;
1348 
1349 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
1350 		WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2,
1351 			lower_32_bits(ring->wptr) | 0x80000000);
1352 
1353 	if (ring->use_doorbell) {
1354 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1355 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1356 	} else {
1357 		WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1358 	}
1359 }
1360 
1361 /**
1362  * vcn_v2_0_dec_ring_insert_start - insert a start command
1363  *
1364  * @ring: amdgpu_ring pointer
1365  *
1366  * Write a start command to the ring.
1367  */
1368 void vcn_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring)
1369 {
1370 	struct amdgpu_device *adev = ring->adev;
1371 
1372 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1373 	amdgpu_ring_write(ring, 0);
1374 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1375 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1376 }
1377 
1378 /**
1379  * vcn_v2_0_dec_ring_insert_end - insert a end command
1380  *
1381  * @ring: amdgpu_ring pointer
1382  *
1383  * Write a end command to the ring.
1384  */
1385 void vcn_v2_0_dec_ring_insert_end(struct amdgpu_ring *ring)
1386 {
1387 	struct amdgpu_device *adev = ring->adev;
1388 
1389 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1390 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_END << 1));
1391 }
1392 
1393 /**
1394  * vcn_v2_0_dec_ring_insert_nop - insert a nop command
1395  *
1396  * @ring: amdgpu_ring pointer
1397  * @count: the number of NOP packets to insert
1398  *
1399  * Write a nop command to the ring.
1400  */
1401 void vcn_v2_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
1402 {
1403 	struct amdgpu_device *adev = ring->adev;
1404 	int i;
1405 
1406 	WARN_ON(ring->wptr % 2 || count % 2);
1407 
1408 	for (i = 0; i < count / 2; i++) {
1409 		amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.nop, 0));
1410 		amdgpu_ring_write(ring, 0);
1411 	}
1412 }
1413 
1414 /**
1415  * vcn_v2_0_dec_ring_emit_fence - emit an fence & trap command
1416  *
1417  * @ring: amdgpu_ring pointer
1418  * @addr: address
1419  * @seq: sequence number
1420  * @flags: fence related flags
1421  *
1422  * Write a fence and a trap command to the ring.
1423  */
1424 void vcn_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
1425 				unsigned flags)
1426 {
1427 	struct amdgpu_device *adev = ring->adev;
1428 
1429 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1430 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.context_id, 0));
1431 	amdgpu_ring_write(ring, seq);
1432 
1433 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1434 	amdgpu_ring_write(ring, addr & 0xffffffff);
1435 
1436 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1437 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
1438 
1439 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1440 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_FENCE << 1));
1441 
1442 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1443 	amdgpu_ring_write(ring, 0);
1444 
1445 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1446 	amdgpu_ring_write(ring, 0);
1447 
1448 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1449 
1450 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_TRAP << 1));
1451 }
1452 
1453 /**
1454  * vcn_v2_0_dec_ring_emit_ib - execute indirect buffer
1455  *
1456  * @ring: amdgpu_ring pointer
1457  * @job: job to retrieve vmid from
1458  * @ib: indirect buffer to execute
1459  * @flags: unused
1460  *
1461  * Write ring commands to execute the indirect buffer
1462  */
1463 void vcn_v2_0_dec_ring_emit_ib(struct amdgpu_ring *ring,
1464 			       struct amdgpu_job *job,
1465 			       struct amdgpu_ib *ib,
1466 			       uint32_t flags)
1467 {
1468 	struct amdgpu_device *adev = ring->adev;
1469 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1470 
1471 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_vmid, 0));
1472 	amdgpu_ring_write(ring, vmid);
1473 
1474 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_bar_low, 0));
1475 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1476 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_bar_high, 0));
1477 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1478 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_size, 0));
1479 	amdgpu_ring_write(ring, ib->length_dw);
1480 }
1481 
1482 void vcn_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1483 				uint32_t val, uint32_t mask)
1484 {
1485 	struct amdgpu_device *adev = ring->adev;
1486 
1487 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1488 	amdgpu_ring_write(ring, reg << 2);
1489 
1490 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1491 	amdgpu_ring_write(ring, val);
1492 
1493 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.gp_scratch8, 0));
1494 	amdgpu_ring_write(ring, mask);
1495 
1496 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1497 
1498 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_REG_READ_COND_WAIT << 1));
1499 }
1500 
1501 void vcn_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
1502 				unsigned vmid, uint64_t pd_addr)
1503 {
1504 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1505 	uint32_t data0, data1, mask;
1506 
1507 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1508 
1509 	/* wait for register write */
1510 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1511 	data1 = lower_32_bits(pd_addr);
1512 	mask = 0xffffffff;
1513 	vcn_v2_0_dec_ring_emit_reg_wait(ring, data0, data1, mask);
1514 }
1515 
1516 void vcn_v2_0_dec_ring_emit_wreg(struct amdgpu_ring *ring,
1517 				uint32_t reg, uint32_t val)
1518 {
1519 	struct amdgpu_device *adev = ring->adev;
1520 
1521 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1522 	amdgpu_ring_write(ring, reg << 2);
1523 
1524 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1525 	amdgpu_ring_write(ring, val);
1526 
1527 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1528 
1529 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_WRITE_REG << 1));
1530 }
1531 
1532 /**
1533  * vcn_v2_0_enc_ring_get_rptr - get enc read pointer
1534  *
1535  * @ring: amdgpu_ring pointer
1536  *
1537  * Returns the current hardware enc read pointer
1538  */
1539 static uint64_t vcn_v2_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1540 {
1541 	struct amdgpu_device *adev = ring->adev;
1542 
1543 	if (ring == &adev->vcn.inst->ring_enc[0])
1544 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR);
1545 	else
1546 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2);
1547 }
1548 
1549  /**
1550  * vcn_v2_0_enc_ring_get_wptr - get enc write pointer
1551  *
1552  * @ring: amdgpu_ring pointer
1553  *
1554  * Returns the current hardware enc write pointer
1555  */
1556 static uint64_t vcn_v2_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1557 {
1558 	struct amdgpu_device *adev = ring->adev;
1559 
1560 	if (ring == &adev->vcn.inst->ring_enc[0]) {
1561 		if (ring->use_doorbell)
1562 			return adev->wb.wb[ring->wptr_offs];
1563 		else
1564 			return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1565 	} else {
1566 		if (ring->use_doorbell)
1567 			return adev->wb.wb[ring->wptr_offs];
1568 		else
1569 			return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1570 	}
1571 }
1572 
1573  /**
1574  * vcn_v2_0_enc_ring_set_wptr - set enc write pointer
1575  *
1576  * @ring: amdgpu_ring pointer
1577  *
1578  * Commits the enc write pointer to the hardware
1579  */
1580 static void vcn_v2_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1581 {
1582 	struct amdgpu_device *adev = ring->adev;
1583 
1584 	if (ring == &adev->vcn.inst->ring_enc[0]) {
1585 		if (ring->use_doorbell) {
1586 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1587 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1588 		} else {
1589 			WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1590 		}
1591 	} else {
1592 		if (ring->use_doorbell) {
1593 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1594 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1595 		} else {
1596 			WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1597 		}
1598 	}
1599 }
1600 
1601 /**
1602  * vcn_v2_0_enc_ring_emit_fence - emit an enc fence & trap command
1603  *
1604  * @ring: amdgpu_ring pointer
1605  * @addr: address
1606  * @seq: sequence number
1607  * @flags: fence related flags
1608  *
1609  * Write enc a fence and a trap command to the ring.
1610  */
1611 void vcn_v2_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1612 				u64 seq, unsigned flags)
1613 {
1614 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1615 
1616 	amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE);
1617 	amdgpu_ring_write(ring, addr);
1618 	amdgpu_ring_write(ring, upper_32_bits(addr));
1619 	amdgpu_ring_write(ring, seq);
1620 	amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP);
1621 }
1622 
1623 void vcn_v2_0_enc_ring_insert_end(struct amdgpu_ring *ring)
1624 {
1625 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
1626 }
1627 
1628 /**
1629  * vcn_v2_0_enc_ring_emit_ib - enc execute indirect buffer
1630  *
1631  * @ring: amdgpu_ring pointer
1632  * @job: job to retrive vmid from
1633  * @ib: indirect buffer to execute
1634  * @flags: unused
1635  *
1636  * Write enc ring commands to execute the indirect buffer
1637  */
1638 void vcn_v2_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
1639 			       struct amdgpu_job *job,
1640 			       struct amdgpu_ib *ib,
1641 			       uint32_t flags)
1642 {
1643 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1644 
1645 	amdgpu_ring_write(ring, VCN_ENC_CMD_IB);
1646 	amdgpu_ring_write(ring, vmid);
1647 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1648 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1649 	amdgpu_ring_write(ring, ib->length_dw);
1650 }
1651 
1652 void vcn_v2_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1653 				uint32_t val, uint32_t mask)
1654 {
1655 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
1656 	amdgpu_ring_write(ring, reg << 2);
1657 	amdgpu_ring_write(ring, mask);
1658 	amdgpu_ring_write(ring, val);
1659 }
1660 
1661 void vcn_v2_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
1662 				unsigned int vmid, uint64_t pd_addr)
1663 {
1664 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1665 
1666 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1667 
1668 	/* wait for reg writes */
1669 	vcn_v2_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 +
1670 					vmid * hub->ctx_addr_distance,
1671 					lower_32_bits(pd_addr), 0xffffffff);
1672 }
1673 
1674 void vcn_v2_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1675 {
1676 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
1677 	amdgpu_ring_write(ring,	reg << 2);
1678 	amdgpu_ring_write(ring, val);
1679 }
1680 
1681 static int vcn_v2_0_set_interrupt_state(struct amdgpu_device *adev,
1682 					struct amdgpu_irq_src *source,
1683 					unsigned type,
1684 					enum amdgpu_interrupt_state state)
1685 {
1686 	return 0;
1687 }
1688 
1689 static int vcn_v2_0_process_interrupt(struct amdgpu_device *adev,
1690 				      struct amdgpu_irq_src *source,
1691 				      struct amdgpu_iv_entry *entry)
1692 {
1693 	DRM_DEBUG("IH: VCN TRAP\n");
1694 
1695 	switch (entry->src_id) {
1696 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1697 		amdgpu_fence_process(&adev->vcn.inst->ring_dec);
1698 		break;
1699 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1700 		amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]);
1701 		break;
1702 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1703 		amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]);
1704 		break;
1705 	default:
1706 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1707 			  entry->src_id, entry->src_data[0]);
1708 		break;
1709 	}
1710 
1711 	return 0;
1712 }
1713 
1714 int vcn_v2_0_dec_ring_test_ring(struct amdgpu_ring *ring)
1715 {
1716 	struct amdgpu_device *adev = ring->adev;
1717 	uint32_t tmp = 0;
1718 	unsigned i;
1719 	int r;
1720 
1721 	if (amdgpu_sriov_vf(adev))
1722 		return 0;
1723 
1724 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
1725 	r = amdgpu_ring_alloc(ring, 4);
1726 	if (r)
1727 		return r;
1728 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1729 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1730 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
1731 	amdgpu_ring_write(ring, 0xDEADBEEF);
1732 	amdgpu_ring_commit(ring);
1733 	for (i = 0; i < adev->usec_timeout; i++) {
1734 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
1735 		if (tmp == 0xDEADBEEF)
1736 			break;
1737 		udelay(1);
1738 	}
1739 
1740 	if (i >= adev->usec_timeout)
1741 		r = -ETIMEDOUT;
1742 
1743 	return r;
1744 }
1745 
1746 
1747 static int vcn_v2_0_set_powergating_state(void *handle,
1748 					  enum amd_powergating_state state)
1749 {
1750 	/* This doesn't actually powergate the VCN block.
1751 	 * That's done in the dpm code via the SMC.  This
1752 	 * just re-inits the block as necessary.  The actual
1753 	 * gating still happens in the dpm code.  We should
1754 	 * revisit this when there is a cleaner line between
1755 	 * the smc and the hw blocks
1756 	 */
1757 	int ret;
1758 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1759 
1760 	if (amdgpu_sriov_vf(adev)) {
1761 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1762 		return 0;
1763 	}
1764 
1765 	if (state == adev->vcn.cur_state)
1766 		return 0;
1767 
1768 	if (state == AMD_PG_STATE_GATE)
1769 		ret = vcn_v2_0_stop(adev);
1770 	else
1771 		ret = vcn_v2_0_start(adev);
1772 
1773 	if (!ret)
1774 		adev->vcn.cur_state = state;
1775 	return ret;
1776 }
1777 
1778 static int vcn_v2_0_start_mmsch(struct amdgpu_device *adev,
1779 				struct amdgpu_mm_table *table)
1780 {
1781 	uint32_t data = 0, loop;
1782 	uint64_t addr = table->gpu_addr;
1783 	struct mmsch_v2_0_init_header *header;
1784 	uint32_t size;
1785 	int i;
1786 
1787 	header = (struct mmsch_v2_0_init_header *)table->cpu_addr;
1788 	size = header->header_size + header->vcn_table_size;
1789 
1790 	/* 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1791 	 * of memory descriptor location
1792 	 */
1793 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1794 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1795 
1796 	/* 2, update vmid of descriptor */
1797 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID);
1798 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1799 	/* use domain0 for MM scheduler */
1800 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1801 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID, data);
1802 
1803 	/* 3, notify mmsch about the size of this descriptor */
1804 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_SIZE, size);
1805 
1806 	/* 4, set resp to zero */
1807 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1808 
1809 	adev->vcn.inst->ring_dec.wptr = 0;
1810 	adev->vcn.inst->ring_dec.wptr_old = 0;
1811 	vcn_v2_0_dec_ring_set_wptr(&adev->vcn.inst->ring_dec);
1812 
1813 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1814 		adev->vcn.inst->ring_enc[i].wptr = 0;
1815 		adev->vcn.inst->ring_enc[i].wptr_old = 0;
1816 		vcn_v2_0_enc_ring_set_wptr(&adev->vcn.inst->ring_enc[i]);
1817 	}
1818 
1819 	/* 5, kick off the initialization and wait until
1820 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1821 	 */
1822 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1823 
1824 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1825 	loop = 1000;
1826 	while ((data & 0x10000002) != 0x10000002) {
1827 		udelay(10);
1828 		data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1829 		loop--;
1830 		if (!loop)
1831 			break;
1832 	}
1833 
1834 	if (!loop) {
1835 		DRM_ERROR("failed to init MMSCH, " \
1836 			"mmMMSCH_VF_MAILBOX_RESP = 0x%08x\n", data);
1837 		return -EBUSY;
1838 	}
1839 
1840 	return 0;
1841 }
1842 
1843 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev)
1844 {
1845 	int r;
1846 	uint32_t tmp;
1847 	struct amdgpu_ring *ring;
1848 	uint32_t offset, size;
1849 	uint32_t table_size = 0;
1850 	struct mmsch_v2_0_cmd_direct_write direct_wt = { {0} };
1851 	struct mmsch_v2_0_cmd_direct_read_modify_write direct_rd_mod_wt = { {0} };
1852 	struct mmsch_v2_0_cmd_end end = { {0} };
1853 	struct mmsch_v2_0_init_header *header;
1854 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1855 	uint8_t i = 0;
1856 
1857 	header = (struct mmsch_v2_0_init_header *)init_table;
1858 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1859 	direct_rd_mod_wt.cmd_header.command_type =
1860 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1861 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1862 
1863 	if (header->vcn_table_offset == 0 && header->vcn_table_size == 0) {
1864 		header->version = MMSCH_VERSION;
1865 		header->header_size = sizeof(struct mmsch_v2_0_init_header) >> 2;
1866 
1867 		header->vcn_table_offset = header->header_size;
1868 
1869 		init_table += header->vcn_table_offset;
1870 
1871 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1872 
1873 		MMSCH_V2_0_INSERT_DIRECT_RD_MOD_WT(
1874 			SOC15_REG_OFFSET(UVD, i, mmUVD_STATUS),
1875 			0xFFFFFFFF, 0x00000004);
1876 
1877 		/* mc resume*/
1878 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1879 			MMSCH_V2_0_INSERT_DIRECT_WT(
1880 				SOC15_REG_OFFSET(UVD, i,
1881 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1882 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo);
1883 			MMSCH_V2_0_INSERT_DIRECT_WT(
1884 				SOC15_REG_OFFSET(UVD, i,
1885 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1886 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi);
1887 			offset = 0;
1888 		} else {
1889 			MMSCH_V2_0_INSERT_DIRECT_WT(
1890 				SOC15_REG_OFFSET(UVD, i,
1891 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1892 				lower_32_bits(adev->vcn.inst->gpu_addr));
1893 			MMSCH_V2_0_INSERT_DIRECT_WT(
1894 				SOC15_REG_OFFSET(UVD, i,
1895 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1896 				upper_32_bits(adev->vcn.inst->gpu_addr));
1897 			offset = size;
1898 		}
1899 
1900 		MMSCH_V2_0_INSERT_DIRECT_WT(
1901 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0),
1902 			0);
1903 		MMSCH_V2_0_INSERT_DIRECT_WT(
1904 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE0),
1905 			size);
1906 
1907 		MMSCH_V2_0_INSERT_DIRECT_WT(
1908 			SOC15_REG_OFFSET(UVD, i,
1909 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1910 			lower_32_bits(adev->vcn.inst->gpu_addr + offset));
1911 		MMSCH_V2_0_INSERT_DIRECT_WT(
1912 			SOC15_REG_OFFSET(UVD, i,
1913 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1914 			upper_32_bits(adev->vcn.inst->gpu_addr + offset));
1915 		MMSCH_V2_0_INSERT_DIRECT_WT(
1916 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET1),
1917 			0);
1918 		MMSCH_V2_0_INSERT_DIRECT_WT(
1919 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE1),
1920 			AMDGPU_VCN_STACK_SIZE);
1921 
1922 		MMSCH_V2_0_INSERT_DIRECT_WT(
1923 			SOC15_REG_OFFSET(UVD, i,
1924 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1925 			lower_32_bits(adev->vcn.inst->gpu_addr + offset +
1926 				AMDGPU_VCN_STACK_SIZE));
1927 		MMSCH_V2_0_INSERT_DIRECT_WT(
1928 			SOC15_REG_OFFSET(UVD, i,
1929 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1930 			upper_32_bits(adev->vcn.inst->gpu_addr + offset +
1931 				AMDGPU_VCN_STACK_SIZE));
1932 		MMSCH_V2_0_INSERT_DIRECT_WT(
1933 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET2),
1934 			0);
1935 		MMSCH_V2_0_INSERT_DIRECT_WT(
1936 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE2),
1937 			AMDGPU_VCN_CONTEXT_SIZE);
1938 
1939 		for (r = 0; r < adev->vcn.num_enc_rings; ++r) {
1940 			ring = &adev->vcn.inst->ring_enc[r];
1941 			ring->wptr = 0;
1942 			MMSCH_V2_0_INSERT_DIRECT_WT(
1943 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_LO),
1944 				lower_32_bits(ring->gpu_addr));
1945 			MMSCH_V2_0_INSERT_DIRECT_WT(
1946 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_HI),
1947 				upper_32_bits(ring->gpu_addr));
1948 			MMSCH_V2_0_INSERT_DIRECT_WT(
1949 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_SIZE),
1950 				ring->ring_size / 4);
1951 		}
1952 
1953 		ring = &adev->vcn.inst->ring_dec;
1954 		ring->wptr = 0;
1955 		MMSCH_V2_0_INSERT_DIRECT_WT(
1956 			SOC15_REG_OFFSET(UVD, i,
1957 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1958 			lower_32_bits(ring->gpu_addr));
1959 		MMSCH_V2_0_INSERT_DIRECT_WT(
1960 			SOC15_REG_OFFSET(UVD, i,
1961 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1962 			upper_32_bits(ring->gpu_addr));
1963 		/* force RBC into idle state */
1964 		tmp = order_base_2(ring->ring_size);
1965 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1966 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1967 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1968 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1969 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1970 		MMSCH_V2_0_INSERT_DIRECT_WT(
1971 			SOC15_REG_OFFSET(UVD, i, mmUVD_RBC_RB_CNTL), tmp);
1972 
1973 		/* add end packet */
1974 		tmp = sizeof(struct mmsch_v2_0_cmd_end);
1975 		memcpy((void *)init_table, &end, tmp);
1976 		table_size += (tmp / 4);
1977 		header->vcn_table_size = table_size;
1978 
1979 	}
1980 	return vcn_v2_0_start_mmsch(adev, &adev->virt.mm_table);
1981 }
1982 
1983 static const struct amd_ip_funcs vcn_v2_0_ip_funcs = {
1984 	.name = "vcn_v2_0",
1985 	.early_init = vcn_v2_0_early_init,
1986 	.late_init = NULL,
1987 	.sw_init = vcn_v2_0_sw_init,
1988 	.sw_fini = vcn_v2_0_sw_fini,
1989 	.hw_init = vcn_v2_0_hw_init,
1990 	.hw_fini = vcn_v2_0_hw_fini,
1991 	.suspend = vcn_v2_0_suspend,
1992 	.resume = vcn_v2_0_resume,
1993 	.is_idle = vcn_v2_0_is_idle,
1994 	.wait_for_idle = vcn_v2_0_wait_for_idle,
1995 	.check_soft_reset = NULL,
1996 	.pre_soft_reset = NULL,
1997 	.soft_reset = NULL,
1998 	.post_soft_reset = NULL,
1999 	.set_clockgating_state = vcn_v2_0_set_clockgating_state,
2000 	.set_powergating_state = vcn_v2_0_set_powergating_state,
2001 };
2002 
2003 static const struct amdgpu_ring_funcs vcn_v2_0_dec_ring_vm_funcs = {
2004 	.type = AMDGPU_RING_TYPE_VCN_DEC,
2005 	.align_mask = 0xf,
2006 	.vmhub = AMDGPU_MMHUB_0,
2007 	.get_rptr = vcn_v2_0_dec_ring_get_rptr,
2008 	.get_wptr = vcn_v2_0_dec_ring_get_wptr,
2009 	.set_wptr = vcn_v2_0_dec_ring_set_wptr,
2010 	.emit_frame_size =
2011 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
2012 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
2013 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
2014 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
2015 		6,
2016 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
2017 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
2018 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
2019 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
2020 	.test_ring = vcn_v2_0_dec_ring_test_ring,
2021 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
2022 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
2023 	.insert_start = vcn_v2_0_dec_ring_insert_start,
2024 	.insert_end = vcn_v2_0_dec_ring_insert_end,
2025 	.pad_ib = amdgpu_ring_generic_pad_ib,
2026 	.begin_use = amdgpu_vcn_ring_begin_use,
2027 	.end_use = amdgpu_vcn_ring_end_use,
2028 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
2029 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
2030 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2031 };
2032 
2033 static const struct amdgpu_ring_funcs vcn_v2_0_enc_ring_vm_funcs = {
2034 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2035 	.align_mask = 0x3f,
2036 	.nop = VCN_ENC_CMD_NO_OP,
2037 	.vmhub = AMDGPU_MMHUB_0,
2038 	.get_rptr = vcn_v2_0_enc_ring_get_rptr,
2039 	.get_wptr = vcn_v2_0_enc_ring_get_wptr,
2040 	.set_wptr = vcn_v2_0_enc_ring_set_wptr,
2041 	.emit_frame_size =
2042 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2043 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2044 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2045 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2046 		1, /* vcn_v2_0_enc_ring_insert_end */
2047 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2048 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2049 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2050 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2051 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2052 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2053 	.insert_nop = amdgpu_ring_insert_nop,
2054 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2055 	.pad_ib = amdgpu_ring_generic_pad_ib,
2056 	.begin_use = amdgpu_vcn_ring_begin_use,
2057 	.end_use = amdgpu_vcn_ring_end_use,
2058 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2059 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2060 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2061 };
2062 
2063 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2064 {
2065 	adev->vcn.inst->ring_dec.funcs = &vcn_v2_0_dec_ring_vm_funcs;
2066 	DRM_INFO("VCN decode is enabled in VM mode\n");
2067 }
2068 
2069 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2070 {
2071 	int i;
2072 
2073 	for (i = 0; i < adev->vcn.num_enc_rings; ++i)
2074 		adev->vcn.inst->ring_enc[i].funcs = &vcn_v2_0_enc_ring_vm_funcs;
2075 
2076 	DRM_INFO("VCN encode is enabled in VM mode\n");
2077 }
2078 
2079 static const struct amdgpu_irq_src_funcs vcn_v2_0_irq_funcs = {
2080 	.set = vcn_v2_0_set_interrupt_state,
2081 	.process = vcn_v2_0_process_interrupt,
2082 };
2083 
2084 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev)
2085 {
2086 	adev->vcn.inst->irq.num_types = adev->vcn.num_enc_rings + 1;
2087 	adev->vcn.inst->irq.funcs = &vcn_v2_0_irq_funcs;
2088 }
2089 
2090 const struct amdgpu_ip_block_version vcn_v2_0_ip_block =
2091 {
2092 		.type = AMD_IP_BLOCK_TYPE_VCN,
2093 		.major = 2,
2094 		.minor = 0,
2095 		.rev = 0,
2096 		.funcs = &vcn_v2_0_ip_funcs,
2097 };
2098