xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c (revision 46e6acfe3501fa938af9c5bd730f0020235b08a2)
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 VCN_VID_SOC_ADDRESS_2_0					0x1fa00
41 #define VCN1_VID_SOC_ADDRESS_3_0				0x48200
42 
43 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x1fd
44 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x503
45 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x504
46 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x505
47 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x53f
48 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x54a
49 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
50 
51 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x1e1
52 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET		0x5a6
53 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x5a7
54 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x1e2
55 
56 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev);
57 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev);
58 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev);
59 static int vcn_v2_0_set_powergating_state(void *handle,
60 				enum amd_powergating_state state);
61 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
62 				int inst_idx, struct dpg_pause_state *new_state);
63 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev);
64 /**
65  * vcn_v2_0_early_init - set function pointers and load microcode
66  *
67  * @handle: amdgpu_device pointer
68  *
69  * Set ring and irq function pointers
70  * Load microcode from filesystem
71  */
72 static int vcn_v2_0_early_init(void *handle)
73 {
74 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
75 
76 	if (amdgpu_sriov_vf(adev))
77 		adev->vcn.num_enc_rings = 1;
78 	else
79 		adev->vcn.num_enc_rings = 2;
80 
81 	vcn_v2_0_set_dec_ring_funcs(adev);
82 	vcn_v2_0_set_enc_ring_funcs(adev);
83 	vcn_v2_0_set_irq_funcs(adev);
84 
85 	return amdgpu_vcn_early_init(adev);
86 }
87 
88 /**
89  * vcn_v2_0_sw_init - sw init for VCN block
90  *
91  * @handle: amdgpu_device pointer
92  *
93  * Load firmware and sw initialization
94  */
95 static int vcn_v2_0_sw_init(void *handle)
96 {
97 	struct amdgpu_ring *ring;
98 	int i, r;
99 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
100 	volatile struct amdgpu_fw_shared *fw_shared;
101 
102 	/* VCN DEC TRAP */
103 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
104 			      VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT,
105 			      &adev->vcn.inst->irq);
106 	if (r)
107 		return r;
108 
109 	/* VCN ENC TRAP */
110 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
111 		r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
112 				      i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE,
113 				      &adev->vcn.inst->irq);
114 		if (r)
115 			return r;
116 	}
117 
118 	r = amdgpu_vcn_sw_init(adev);
119 	if (r)
120 		return r;
121 
122 	amdgpu_vcn_setup_ucode(adev);
123 
124 	r = amdgpu_vcn_resume(adev);
125 	if (r)
126 		return r;
127 
128 	ring = &adev->vcn.inst->ring_dec;
129 
130 	ring->use_doorbell = true;
131 	ring->doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
132 	ring->vm_hub = AMDGPU_MMHUB0(0);
133 
134 	sprintf(ring->name, "vcn_dec");
135 	r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
136 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
137 	if (r)
138 		return r;
139 
140 	adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
141 	adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
142 	adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
143 	adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
144 	adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
145 	adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
146 
147 	adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
148 	adev->vcn.inst->external.scratch9 = SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9);
149 	adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
150 	adev->vcn.inst->external.data0 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0);
151 	adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
152 	adev->vcn.inst->external.data1 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1);
153 	adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
154 	adev->vcn.inst->external.cmd = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD);
155 	adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
156 	adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP);
157 
158 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
159 		enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
160 
161 		ring = &adev->vcn.inst->ring_enc[i];
162 		ring->use_doorbell = true;
163 		ring->vm_hub = AMDGPU_MMHUB0(0);
164 		if (!amdgpu_sriov_vf(adev))
165 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + i;
166 		else
167 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1 + i;
168 		sprintf(ring->name, "vcn_enc%d", i);
169 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
170 				     hw_prio, NULL);
171 		if (r)
172 			return r;
173 	}
174 
175 	adev->vcn.pause_dpg_mode = vcn_v2_0_pause_dpg_mode;
176 
177 	r = amdgpu_virt_alloc_mm_table(adev);
178 	if (r)
179 		return r;
180 
181 	fw_shared = adev->vcn.inst->fw_shared.cpu_addr;
182 	fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
183 
184 	if (amdgpu_vcnfw_log)
185 		amdgpu_vcn_fwlog_init(adev->vcn.inst);
186 
187 	return 0;
188 }
189 
190 /**
191  * vcn_v2_0_sw_fini - sw fini for VCN block
192  *
193  * @handle: amdgpu_device pointer
194  *
195  * VCN suspend and free up sw allocation
196  */
197 static int vcn_v2_0_sw_fini(void *handle)
198 {
199 	int r, idx;
200 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
201 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared.cpu_addr;
202 
203 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
204 		fw_shared->present_flag_0 = 0;
205 		drm_dev_exit(idx);
206 	}
207 
208 	amdgpu_virt_free_mm_table(adev);
209 
210 	r = amdgpu_vcn_suspend(adev);
211 	if (r)
212 		return r;
213 
214 	r = amdgpu_vcn_sw_fini(adev);
215 
216 	return r;
217 }
218 
219 /**
220  * vcn_v2_0_hw_init - start and test VCN block
221  *
222  * @handle: amdgpu_device pointer
223  *
224  * Initialize the hardware, boot up the VCPU and do some testing
225  */
226 static int vcn_v2_0_hw_init(void *handle)
227 {
228 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
229 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
230 	int i, r;
231 
232 	adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
233 					     ring->doorbell_index, 0);
234 
235 	if (amdgpu_sriov_vf(adev))
236 		vcn_v2_0_start_sriov(adev);
237 
238 	r = amdgpu_ring_test_helper(ring);
239 	if (r)
240 		return r;
241 
242 	//Disable vcn decode for sriov
243 	if (amdgpu_sriov_vf(adev))
244 		ring->sched.ready = false;
245 
246 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
247 		ring = &adev->vcn.inst->ring_enc[i];
248 		r = amdgpu_ring_test_helper(ring);
249 		if (r)
250 			return r;
251 	}
252 
253 	return 0;
254 }
255 
256 /**
257  * vcn_v2_0_hw_fini - stop the hardware block
258  *
259  * @handle: amdgpu_device pointer
260  *
261  * Stop the VCN block, mark ring as not ready any more
262  */
263 static int vcn_v2_0_hw_fini(void *handle)
264 {
265 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
266 
267 	cancel_delayed_work_sync(&adev->vcn.idle_work);
268 
269 	if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
270 	    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
271 	      RREG32_SOC15(VCN, 0, mmUVD_STATUS)))
272 		vcn_v2_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
273 
274 	return 0;
275 }
276 
277 /**
278  * vcn_v2_0_suspend - suspend VCN block
279  *
280  * @handle: amdgpu_device pointer
281  *
282  * HW fini and suspend VCN block
283  */
284 static int vcn_v2_0_suspend(void *handle)
285 {
286 	int r;
287 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
288 
289 	r = vcn_v2_0_hw_fini(adev);
290 	if (r)
291 		return r;
292 
293 	r = amdgpu_vcn_suspend(adev);
294 
295 	return r;
296 }
297 
298 /**
299  * vcn_v2_0_resume - resume VCN block
300  *
301  * @handle: amdgpu_device pointer
302  *
303  * Resume firmware and hw init VCN block
304  */
305 static int vcn_v2_0_resume(void *handle)
306 {
307 	int r;
308 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
309 
310 	r = amdgpu_vcn_resume(adev);
311 	if (r)
312 		return r;
313 
314 	r = vcn_v2_0_hw_init(adev);
315 
316 	return r;
317 }
318 
319 /**
320  * vcn_v2_0_mc_resume - memory controller programming
321  *
322  * @adev: amdgpu_device pointer
323  *
324  * Let the VCN memory controller know it's offsets
325  */
326 static void vcn_v2_0_mc_resume(struct amdgpu_device *adev)
327 {
328 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[0]->size + 4);
329 	uint32_t offset;
330 
331 	if (amdgpu_sriov_vf(adev))
332 		return;
333 
334 	/* cache window 0: fw */
335 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
336 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
337 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo));
338 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
339 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi));
340 		WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0);
341 		offset = 0;
342 	} else {
343 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
344 			lower_32_bits(adev->vcn.inst->gpu_addr));
345 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
346 			upper_32_bits(adev->vcn.inst->gpu_addr));
347 		offset = size;
348 		WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0,
349 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
350 	}
351 
352 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size);
353 
354 	/* cache window 1: stack */
355 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
356 		lower_32_bits(adev->vcn.inst->gpu_addr + offset));
357 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
358 		upper_32_bits(adev->vcn.inst->gpu_addr + offset));
359 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0);
360 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
361 
362 	/* cache window 2: context */
363 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
364 		lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
365 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
366 		upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
367 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0);
368 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
369 
370 	/* non-cache window */
371 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
372 		lower_32_bits(adev->vcn.inst->fw_shared.gpu_addr));
373 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
374 		upper_32_bits(adev->vcn.inst->fw_shared.gpu_addr));
375 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
376 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0,
377 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
378 
379 	WREG32_SOC15(UVD, 0, mmUVD_GFX10_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
380 }
381 
382 static void vcn_v2_0_mc_resume_dpg_mode(struct amdgpu_device *adev, bool indirect)
383 {
384 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[0]->size + 4);
385 	uint32_t offset;
386 
387 	/* cache window 0: fw */
388 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
389 		if (!indirect) {
390 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
391 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
392 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 0, indirect);
393 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
394 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
395 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 0, indirect);
396 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
397 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
398 		} else {
399 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
400 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
401 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
402 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
403 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
404 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
405 		}
406 		offset = 0;
407 	} else {
408 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
409 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
410 			lower_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
411 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
412 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
413 			upper_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
414 		offset = size;
415 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
416 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
417 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
418 	}
419 
420 	if (!indirect)
421 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
422 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
423 	else
424 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
425 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
426 
427 	/* cache window 1: stack */
428 	if (!indirect) {
429 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
430 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
431 			lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
432 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
433 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
434 			upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
435 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
436 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
437 	} else {
438 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
439 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
440 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
441 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
442 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
443 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
444 	}
445 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
446 		UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
447 
448 	/* cache window 2: context */
449 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
450 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
451 		lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
452 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
453 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
454 		upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
455 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
456 		UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
457 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
458 		UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
459 
460 	/* non-cache window */
461 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
462 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
463 		lower_32_bits(adev->vcn.inst->fw_shared.gpu_addr), 0, indirect);
464 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
465 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
466 		upper_32_bits(adev->vcn.inst->fw_shared.gpu_addr), 0, indirect);
467 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
468 		UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
469 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
470 		UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0),
471 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
472 
473 	/* VCN global tiling registers */
474 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
475 		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
476 }
477 
478 /**
479  * vcn_v2_0_disable_clock_gating - disable VCN clock gating
480  *
481  * @adev: amdgpu_device pointer
482  *
483  * Disable clock gating for VCN block
484  */
485 static void vcn_v2_0_disable_clock_gating(struct amdgpu_device *adev)
486 {
487 	uint32_t data;
488 
489 	if (amdgpu_sriov_vf(adev))
490 		return;
491 
492 	/* UVD disable CGC */
493 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
494 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
495 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
496 	else
497 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
498 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
499 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
500 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
501 
502 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_GATE);
503 	data &= ~(UVD_CGC_GATE__SYS_MASK
504 		| UVD_CGC_GATE__UDEC_MASK
505 		| UVD_CGC_GATE__MPEG2_MASK
506 		| UVD_CGC_GATE__REGS_MASK
507 		| UVD_CGC_GATE__RBC_MASK
508 		| UVD_CGC_GATE__LMI_MC_MASK
509 		| UVD_CGC_GATE__LMI_UMC_MASK
510 		| UVD_CGC_GATE__IDCT_MASK
511 		| UVD_CGC_GATE__MPRD_MASK
512 		| UVD_CGC_GATE__MPC_MASK
513 		| UVD_CGC_GATE__LBSI_MASK
514 		| UVD_CGC_GATE__LRBBM_MASK
515 		| UVD_CGC_GATE__UDEC_RE_MASK
516 		| UVD_CGC_GATE__UDEC_CM_MASK
517 		| UVD_CGC_GATE__UDEC_IT_MASK
518 		| UVD_CGC_GATE__UDEC_DB_MASK
519 		| UVD_CGC_GATE__UDEC_MP_MASK
520 		| UVD_CGC_GATE__WCB_MASK
521 		| UVD_CGC_GATE__VCPU_MASK
522 		| UVD_CGC_GATE__SCPU_MASK);
523 	WREG32_SOC15(VCN, 0, mmUVD_CGC_GATE, data);
524 
525 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
526 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
527 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
528 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
529 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
530 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
531 		| UVD_CGC_CTRL__SYS_MODE_MASK
532 		| UVD_CGC_CTRL__UDEC_MODE_MASK
533 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
534 		| UVD_CGC_CTRL__REGS_MODE_MASK
535 		| UVD_CGC_CTRL__RBC_MODE_MASK
536 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
537 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
538 		| UVD_CGC_CTRL__IDCT_MODE_MASK
539 		| UVD_CGC_CTRL__MPRD_MODE_MASK
540 		| UVD_CGC_CTRL__MPC_MODE_MASK
541 		| UVD_CGC_CTRL__LBSI_MODE_MASK
542 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
543 		| UVD_CGC_CTRL__WCB_MODE_MASK
544 		| UVD_CGC_CTRL__VCPU_MODE_MASK
545 		| UVD_CGC_CTRL__SCPU_MODE_MASK);
546 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
547 
548 	/* turn on */
549 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE);
550 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
551 		| UVD_SUVD_CGC_GATE__SIT_MASK
552 		| UVD_SUVD_CGC_GATE__SMP_MASK
553 		| UVD_SUVD_CGC_GATE__SCM_MASK
554 		| UVD_SUVD_CGC_GATE__SDB_MASK
555 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
556 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
557 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
558 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
559 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
560 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
561 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
562 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
563 		| UVD_SUVD_CGC_GATE__SCLR_MASK
564 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
565 		| UVD_SUVD_CGC_GATE__ENT_MASK
566 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
567 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
568 		| UVD_SUVD_CGC_GATE__SITE_MASK
569 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
570 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
571 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
572 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
573 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
574 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE, data);
575 
576 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
577 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
578 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
579 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
580 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
581 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
582 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
583 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
584 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
585 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
586 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
587 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
588 }
589 
590 static void vcn_v2_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
591 		uint8_t sram_sel, uint8_t indirect)
592 {
593 	uint32_t reg_data = 0;
594 
595 	/* enable sw clock gating control */
596 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
597 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
598 	else
599 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
600 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
601 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
602 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
603 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
604 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
605 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
606 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
607 		 UVD_CGC_CTRL__SYS_MODE_MASK |
608 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
609 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
610 		 UVD_CGC_CTRL__REGS_MODE_MASK |
611 		 UVD_CGC_CTRL__RBC_MODE_MASK |
612 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
613 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
614 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
615 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
616 		 UVD_CGC_CTRL__MPC_MODE_MASK |
617 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
618 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
619 		 UVD_CGC_CTRL__WCB_MODE_MASK |
620 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
621 		 UVD_CGC_CTRL__SCPU_MODE_MASK);
622 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
623 		UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
624 
625 	/* turn off clock gating */
626 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
627 		UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
628 
629 	/* turn on SUVD clock gating */
630 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
631 		UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
632 
633 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
634 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
635 		UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
636 }
637 
638 /**
639  * vcn_v2_0_enable_clock_gating - enable VCN clock gating
640  *
641  * @adev: amdgpu_device pointer
642  *
643  * Enable clock gating for VCN block
644  */
645 static void vcn_v2_0_enable_clock_gating(struct amdgpu_device *adev)
646 {
647 	uint32_t data = 0;
648 
649 	if (amdgpu_sriov_vf(adev))
650 		return;
651 
652 	/* enable UVD CGC */
653 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
654 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
655 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
656 	else
657 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
658 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
659 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
660 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
661 
662 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
663 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
664 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
665 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
666 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
667 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
668 		| UVD_CGC_CTRL__SYS_MODE_MASK
669 		| UVD_CGC_CTRL__UDEC_MODE_MASK
670 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
671 		| UVD_CGC_CTRL__REGS_MODE_MASK
672 		| UVD_CGC_CTRL__RBC_MODE_MASK
673 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
674 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
675 		| UVD_CGC_CTRL__IDCT_MODE_MASK
676 		| UVD_CGC_CTRL__MPRD_MODE_MASK
677 		| UVD_CGC_CTRL__MPC_MODE_MASK
678 		| UVD_CGC_CTRL__LBSI_MODE_MASK
679 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
680 		| UVD_CGC_CTRL__WCB_MODE_MASK
681 		| UVD_CGC_CTRL__VCPU_MODE_MASK
682 		| UVD_CGC_CTRL__SCPU_MODE_MASK);
683 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
684 
685 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
686 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
687 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
688 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
689 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
690 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
691 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
692 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
693 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
694 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
695 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
696 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
697 }
698 
699 static void vcn_v2_0_disable_static_power_gating(struct amdgpu_device *adev)
700 {
701 	uint32_t data = 0;
702 
703 	if (amdgpu_sriov_vf(adev))
704 		return;
705 
706 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
707 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
708 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
709 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
710 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
711 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
712 			| 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
713 			| 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
714 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
715 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
716 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
717 
718 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
719 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS,
720 			UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON_2_0, 0xFFFFF);
721 	} else {
722 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
723 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
724 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
725 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
726 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
727 			| 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
728 			| 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
729 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
730 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
731 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
732 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
733 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0,  0xFFFFF);
734 	}
735 
736 	/* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS,
737 	 * UVDU_PWR_STATUS are 0 (power on) */
738 
739 	data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
740 	data &= ~0x103;
741 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
742 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
743 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
744 
745 	WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
746 }
747 
748 static void vcn_v2_0_enable_static_power_gating(struct amdgpu_device *adev)
749 {
750 	uint32_t data = 0;
751 
752 	if (amdgpu_sriov_vf(adev))
753 		return;
754 
755 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
756 		/* Before power off, this indicator has to be turned on */
757 		data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
758 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
759 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
760 		WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
761 
762 
763 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
764 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
765 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
766 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
767 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
768 			| 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
769 			| 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
770 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
771 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
772 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
773 
774 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
775 
776 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
777 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
778 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
779 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
780 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
781 			| 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT
782 			| 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT
783 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
784 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
785 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT);
786 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFF);
787 	}
788 }
789 
790 static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
791 {
792 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared.cpu_addr;
793 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
794 	uint32_t rb_bufsz, tmp;
795 
796 	vcn_v2_0_enable_static_power_gating(adev);
797 
798 	/* enable dynamic power gating mode */
799 	tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS);
800 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
801 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
802 	WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp);
803 
804 	if (indirect)
805 		adev->vcn.inst->dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst->dpg_sram_cpu_addr;
806 
807 	/* enable clock gating */
808 	vcn_v2_0_clock_gating_dpg_mode(adev, 0, indirect);
809 
810 	/* enable VCPU clock */
811 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
812 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
813 	tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK;
814 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
815 		UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
816 
817 	/* disable master interupt */
818 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
819 		UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
820 
821 	/* setup mmUVD_LMI_CTRL */
822 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
823 		UVD_LMI_CTRL__REQ_MODE_MASK |
824 		UVD_LMI_CTRL__CRC_RESET_MASK |
825 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
826 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
827 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
828 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
829 		0x00100000L);
830 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
831 		UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
832 
833 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
834 		UVD, 0, mmUVD_MPC_CNTL),
835 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
836 
837 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
838 		UVD, 0, mmUVD_MPC_SET_MUXA0),
839 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
840 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
841 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
842 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
843 
844 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
845 		UVD, 0, mmUVD_MPC_SET_MUXB0),
846 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
847 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
848 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
849 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
850 
851 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
852 		UVD, 0, mmUVD_MPC_SET_MUX),
853 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
854 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
855 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
856 
857 	vcn_v2_0_mc_resume_dpg_mode(adev, indirect);
858 
859 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
860 		UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
861 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
862 		UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
863 
864 	/* release VCPU reset to boot */
865 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
866 		UVD, 0, mmUVD_SOFT_RESET), 0, 0, indirect);
867 
868 	/* enable LMI MC and UMC channels */
869 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
870 		UVD, 0, mmUVD_LMI_CTRL2),
871 		0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 0, indirect);
872 
873 	/* enable master interrupt */
874 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
875 		UVD, 0, mmUVD_MASTINT_EN),
876 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
877 
878 	if (indirect)
879 		amdgpu_vcn_psp_update_sram(adev, 0, 0);
880 
881 	/* force RBC into idle state */
882 	rb_bufsz = order_base_2(ring->ring_size);
883 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
884 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
885 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
886 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
887 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
888 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
889 
890 	/* Stall DPG before WPTR/RPTR reset */
891 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
892 		UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
893 		~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
894 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
895 
896 	/* set the write pointer delay */
897 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0);
898 
899 	/* set the wb address */
900 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR,
901 		(upper_32_bits(ring->gpu_addr) >> 2));
902 
903 	/* program the RB_BASE for ring buffer */
904 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
905 		lower_32_bits(ring->gpu_addr));
906 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
907 		upper_32_bits(ring->gpu_addr));
908 
909 	/* Initialize the ring buffer's read and write pointers */
910 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
911 
912 	WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0);
913 
914 	ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
915 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
916 		lower_32_bits(ring->wptr));
917 
918 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
919 	/* Unstall DPG */
920 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
921 		0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
922 	return 0;
923 }
924 
925 static int vcn_v2_0_start(struct amdgpu_device *adev)
926 {
927 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared.cpu_addr;
928 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
929 	uint32_t rb_bufsz, tmp;
930 	uint32_t lmi_swap_cntl;
931 	int i, j, r;
932 
933 	if (adev->pm.dpm_enabled)
934 		amdgpu_dpm_enable_uvd(adev, true);
935 
936 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
937 		return vcn_v2_0_start_dpg_mode(adev, adev->vcn.indirect_sram);
938 
939 	vcn_v2_0_disable_static_power_gating(adev);
940 
941 	/* set uvd status busy */
942 	tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
943 	WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp);
944 
945 	/*SW clock gating */
946 	vcn_v2_0_disable_clock_gating(adev);
947 
948 	/* enable VCPU clock */
949 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL),
950 		UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
951 
952 	/* disable master interrupt */
953 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0,
954 		~UVD_MASTINT_EN__VCPU_EN_MASK);
955 
956 	/* setup mmUVD_LMI_CTRL */
957 	tmp = RREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL);
958 	WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, tmp |
959 		UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK	|
960 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
961 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
962 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
963 
964 	/* setup mmUVD_MPC_CNTL */
965 	tmp = RREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL);
966 	tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
967 	tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
968 	WREG32_SOC15(VCN, 0, mmUVD_MPC_CNTL, tmp);
969 
970 	/* setup UVD_MPC_SET_MUXA0 */
971 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0,
972 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
973 		(0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
974 		(0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
975 		(0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
976 
977 	/* setup UVD_MPC_SET_MUXB0 */
978 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0,
979 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
980 		(0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
981 		(0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
982 		(0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
983 
984 	/* setup mmUVD_MPC_SET_MUX */
985 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX,
986 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
987 		(0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
988 		(0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
989 
990 	vcn_v2_0_mc_resume(adev);
991 
992 	/* release VCPU reset to boot */
993 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
994 		~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
995 
996 	/* enable LMI MC and UMC channels */
997 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0,
998 		~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
999 
1000 	tmp = RREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET);
1001 	tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1002 	tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1003 	WREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET, tmp);
1004 
1005 	/* disable byte swapping */
1006 	lmi_swap_cntl = 0;
1007 #ifdef __BIG_ENDIAN
1008 	/* swap (8 in 32) RB and IB */
1009 	lmi_swap_cntl = 0xa;
1010 #endif
1011 	WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
1012 
1013 	for (i = 0; i < 10; ++i) {
1014 		uint32_t status;
1015 
1016 		for (j = 0; j < 100; ++j) {
1017 			status = RREG32_SOC15(UVD, 0, mmUVD_STATUS);
1018 			if (status & 2)
1019 				break;
1020 			mdelay(10);
1021 		}
1022 		r = 0;
1023 		if (status & 2)
1024 			break;
1025 
1026 		DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1027 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1028 			UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
1029 			~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1030 		mdelay(10);
1031 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
1032 			~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1033 		mdelay(10);
1034 		r = -1;
1035 	}
1036 
1037 	if (r) {
1038 		DRM_ERROR("VCN decode not responding, giving up!!!\n");
1039 		return r;
1040 	}
1041 
1042 	/* enable master interrupt */
1043 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
1044 		UVD_MASTINT_EN__VCPU_EN_MASK,
1045 		~UVD_MASTINT_EN__VCPU_EN_MASK);
1046 
1047 	/* clear the busy bit of VCN_STATUS */
1048 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0,
1049 		~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1050 
1051 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_VMID, 0);
1052 
1053 	/* force RBC into idle state */
1054 	rb_bufsz = order_base_2(ring->ring_size);
1055 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1056 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1057 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1058 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1059 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1060 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
1061 
1062 	fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1063 	/* program the RB_BASE for ring buffer */
1064 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1065 		lower_32_bits(ring->gpu_addr));
1066 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1067 		upper_32_bits(ring->gpu_addr));
1068 
1069 	/* Initialize the ring buffer's read and write pointers */
1070 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
1071 
1072 	ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1073 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1074 			lower_32_bits(ring->wptr));
1075 	fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1076 
1077 	fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1078 	ring = &adev->vcn.inst->ring_enc[0];
1079 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1080 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1081 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1082 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1083 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1084 	fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1085 
1086 	fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1087 	ring = &adev->vcn.inst->ring_enc[1];
1088 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1089 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1090 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1091 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1092 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1093 	fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1094 
1095 	return 0;
1096 }
1097 
1098 static int vcn_v2_0_stop_dpg_mode(struct amdgpu_device *adev)
1099 {
1100 	struct dpg_pause_state state = {.fw_based = VCN_DPG_STATE__UNPAUSE};
1101 	uint32_t tmp;
1102 
1103 	vcn_v2_0_pause_dpg_mode(adev, 0, &state);
1104 	/* Wait for power status to be 1 */
1105 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1106 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1107 
1108 	/* wait for read ptr to be equal to write ptr */
1109 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1110 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1111 
1112 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1113 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1114 
1115 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1116 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1117 
1118 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1119 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1120 
1121 	/* disable dynamic power gating mode */
1122 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0,
1123 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1124 
1125 	return 0;
1126 }
1127 
1128 static int vcn_v2_0_stop(struct amdgpu_device *adev)
1129 {
1130 	uint32_t tmp;
1131 	int r;
1132 
1133 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1134 		r = vcn_v2_0_stop_dpg_mode(adev);
1135 		if (r)
1136 			return r;
1137 		goto power_off;
1138 	}
1139 
1140 	/* wait for uvd idle */
1141 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1142 	if (r)
1143 		return r;
1144 
1145 	tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1146 		UVD_LMI_STATUS__READ_CLEAN_MASK |
1147 		UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1148 		UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1149 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp);
1150 	if (r)
1151 		return r;
1152 
1153 	/* stall UMC channel */
1154 	tmp = RREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2);
1155 	tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1156 	WREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2, tmp);
1157 
1158 	tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1159 		UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1160 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp);
1161 	if (r)
1162 		return r;
1163 
1164 	/* disable VCPU clock */
1165 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0,
1166 		~(UVD_VCPU_CNTL__CLK_EN_MASK));
1167 
1168 	/* reset LMI UMC */
1169 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1170 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK,
1171 		~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
1172 
1173 	/* reset LMI */
1174 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1175 		UVD_SOFT_RESET__LMI_SOFT_RESET_MASK,
1176 		~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
1177 
1178 	/* reset VCPU */
1179 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1180 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
1181 		~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1182 
1183 	/* clear status */
1184 	WREG32_SOC15(VCN, 0, mmUVD_STATUS, 0);
1185 
1186 	vcn_v2_0_enable_clock_gating(adev);
1187 	vcn_v2_0_enable_static_power_gating(adev);
1188 
1189 power_off:
1190 	if (adev->pm.dpm_enabled)
1191 		amdgpu_dpm_enable_uvd(adev, false);
1192 
1193 	return 0;
1194 }
1195 
1196 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
1197 				int inst_idx, struct dpg_pause_state *new_state)
1198 {
1199 	struct amdgpu_ring *ring;
1200 	uint32_t reg_data = 0;
1201 	int ret_code;
1202 
1203 	/* pause/unpause if state is changed */
1204 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1205 		DRM_DEBUG("dpg pause state changed %d -> %d",
1206 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1207 		reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) &
1208 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1209 
1210 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1211 			ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 0x1,
1212 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1213 
1214 			if (!ret_code) {
1215 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared.cpu_addr;
1216 				/* pause DPG */
1217 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1218 				WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1219 
1220 				/* wait for ACK */
1221 				SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE,
1222 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1223 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1224 
1225 				/* Stall DPG before WPTR/RPTR reset */
1226 				WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
1227 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1228 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1229 				/* Restore */
1230 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1231 				ring = &adev->vcn.inst->ring_enc[0];
1232 				ring->wptr = 0;
1233 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1234 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1235 				WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1236 				WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1237 				WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1238 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1239 
1240 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1241 				ring = &adev->vcn.inst->ring_enc[1];
1242 				ring->wptr = 0;
1243 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1244 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1245 				WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1246 				WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1247 				WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1248 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1249 
1250 				fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1251 				WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1252 					   RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF);
1253 				fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1254 				/* Unstall DPG */
1255 				WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
1256 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1257 
1258 				SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS,
1259 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON,
1260 					   UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1261 			}
1262 		} else {
1263 			/* unpause dpg, no need to wait */
1264 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1265 			WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1266 		}
1267 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1268 	}
1269 
1270 	return 0;
1271 }
1272 
1273 static bool vcn_v2_0_is_idle(void *handle)
1274 {
1275 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1276 
1277 	return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE);
1278 }
1279 
1280 static int vcn_v2_0_wait_for_idle(void *handle)
1281 {
1282 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1283 	int ret;
1284 
1285 	ret = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE,
1286 		UVD_STATUS__IDLE);
1287 
1288 	return ret;
1289 }
1290 
1291 static int vcn_v2_0_set_clockgating_state(void *handle,
1292 					  enum amd_clockgating_state state)
1293 {
1294 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1295 	bool enable = (state == AMD_CG_STATE_GATE);
1296 
1297 	if (amdgpu_sriov_vf(adev))
1298 		return 0;
1299 
1300 	if (enable) {
1301 		/* wait for STATUS to clear */
1302 		if (!vcn_v2_0_is_idle(handle))
1303 			return -EBUSY;
1304 		vcn_v2_0_enable_clock_gating(adev);
1305 	} else {
1306 		/* disable HW gating and enable Sw gating */
1307 		vcn_v2_0_disable_clock_gating(adev);
1308 	}
1309 	return 0;
1310 }
1311 
1312 /**
1313  * vcn_v2_0_dec_ring_get_rptr - get read pointer
1314  *
1315  * @ring: amdgpu_ring pointer
1316  *
1317  * Returns the current hardware read pointer
1318  */
1319 static uint64_t vcn_v2_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1320 {
1321 	struct amdgpu_device *adev = ring->adev;
1322 
1323 	return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1324 }
1325 
1326 /**
1327  * vcn_v2_0_dec_ring_get_wptr - get write pointer
1328  *
1329  * @ring: amdgpu_ring pointer
1330  *
1331  * Returns the current hardware write pointer
1332  */
1333 static uint64_t vcn_v2_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1334 {
1335 	struct amdgpu_device *adev = ring->adev;
1336 
1337 	if (ring->use_doorbell)
1338 		return *ring->wptr_cpu_addr;
1339 	else
1340 		return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR);
1341 }
1342 
1343 /**
1344  * vcn_v2_0_dec_ring_set_wptr - set write pointer
1345  *
1346  * @ring: amdgpu_ring pointer
1347  *
1348  * Commits the write pointer to the hardware
1349  */
1350 static void vcn_v2_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1351 {
1352 	struct amdgpu_device *adev = ring->adev;
1353 
1354 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
1355 		WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2,
1356 			lower_32_bits(ring->wptr) | 0x80000000);
1357 
1358 	if (ring->use_doorbell) {
1359 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1360 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1361 	} else {
1362 		WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1363 	}
1364 }
1365 
1366 /**
1367  * vcn_v2_0_dec_ring_insert_start - insert a start command
1368  *
1369  * @ring: amdgpu_ring pointer
1370  *
1371  * Write a start command to the ring.
1372  */
1373 void vcn_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring)
1374 {
1375 	struct amdgpu_device *adev = ring->adev;
1376 
1377 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1378 	amdgpu_ring_write(ring, 0);
1379 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1380 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1381 }
1382 
1383 /**
1384  * vcn_v2_0_dec_ring_insert_end - insert a end command
1385  *
1386  * @ring: amdgpu_ring pointer
1387  *
1388  * Write a end command to the ring.
1389  */
1390 void vcn_v2_0_dec_ring_insert_end(struct amdgpu_ring *ring)
1391 {
1392 	struct amdgpu_device *adev = ring->adev;
1393 
1394 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1395 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_END << 1));
1396 }
1397 
1398 /**
1399  * vcn_v2_0_dec_ring_insert_nop - insert a nop command
1400  *
1401  * @ring: amdgpu_ring pointer
1402  * @count: the number of NOP packets to insert
1403  *
1404  * Write a nop command to the ring.
1405  */
1406 void vcn_v2_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
1407 {
1408 	struct amdgpu_device *adev = ring->adev;
1409 	int i;
1410 
1411 	WARN_ON(ring->wptr % 2 || count % 2);
1412 
1413 	for (i = 0; i < count / 2; i++) {
1414 		amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.nop, 0));
1415 		amdgpu_ring_write(ring, 0);
1416 	}
1417 }
1418 
1419 /**
1420  * vcn_v2_0_dec_ring_emit_fence - emit an fence & trap command
1421  *
1422  * @ring: amdgpu_ring pointer
1423  * @addr: address
1424  * @seq: sequence number
1425  * @flags: fence related flags
1426  *
1427  * Write a fence and a trap command to the ring.
1428  */
1429 void vcn_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
1430 				unsigned flags)
1431 {
1432 	struct amdgpu_device *adev = ring->adev;
1433 
1434 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1435 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.context_id, 0));
1436 	amdgpu_ring_write(ring, seq);
1437 
1438 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1439 	amdgpu_ring_write(ring, addr & 0xffffffff);
1440 
1441 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1442 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
1443 
1444 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1445 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_FENCE << 1));
1446 
1447 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1448 	amdgpu_ring_write(ring, 0);
1449 
1450 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1451 	amdgpu_ring_write(ring, 0);
1452 
1453 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1454 
1455 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_TRAP << 1));
1456 }
1457 
1458 /**
1459  * vcn_v2_0_dec_ring_emit_ib - execute indirect buffer
1460  *
1461  * @ring: amdgpu_ring pointer
1462  * @job: job to retrieve vmid from
1463  * @ib: indirect buffer to execute
1464  * @flags: unused
1465  *
1466  * Write ring commands to execute the indirect buffer
1467  */
1468 void vcn_v2_0_dec_ring_emit_ib(struct amdgpu_ring *ring,
1469 			       struct amdgpu_job *job,
1470 			       struct amdgpu_ib *ib,
1471 			       uint32_t flags)
1472 {
1473 	struct amdgpu_device *adev = ring->adev;
1474 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1475 
1476 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_vmid, 0));
1477 	amdgpu_ring_write(ring, vmid);
1478 
1479 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_bar_low, 0));
1480 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1481 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_bar_high, 0));
1482 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1483 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_size, 0));
1484 	amdgpu_ring_write(ring, ib->length_dw);
1485 }
1486 
1487 void vcn_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1488 				uint32_t val, uint32_t mask)
1489 {
1490 	struct amdgpu_device *adev = ring->adev;
1491 
1492 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1493 	amdgpu_ring_write(ring, reg << 2);
1494 
1495 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1496 	amdgpu_ring_write(ring, val);
1497 
1498 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.gp_scratch8, 0));
1499 	amdgpu_ring_write(ring, mask);
1500 
1501 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1502 
1503 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_REG_READ_COND_WAIT << 1));
1504 }
1505 
1506 void vcn_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
1507 				unsigned vmid, uint64_t pd_addr)
1508 {
1509 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub];
1510 	uint32_t data0, data1, mask;
1511 
1512 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1513 
1514 	/* wait for register write */
1515 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1516 	data1 = lower_32_bits(pd_addr);
1517 	mask = 0xffffffff;
1518 	vcn_v2_0_dec_ring_emit_reg_wait(ring, data0, data1, mask);
1519 }
1520 
1521 void vcn_v2_0_dec_ring_emit_wreg(struct amdgpu_ring *ring,
1522 				uint32_t reg, uint32_t val)
1523 {
1524 	struct amdgpu_device *adev = ring->adev;
1525 
1526 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1527 	amdgpu_ring_write(ring, reg << 2);
1528 
1529 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1530 	amdgpu_ring_write(ring, val);
1531 
1532 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1533 
1534 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_WRITE_REG << 1));
1535 }
1536 
1537 /**
1538  * vcn_v2_0_enc_ring_get_rptr - get enc read pointer
1539  *
1540  * @ring: amdgpu_ring pointer
1541  *
1542  * Returns the current hardware enc read pointer
1543  */
1544 static uint64_t vcn_v2_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1545 {
1546 	struct amdgpu_device *adev = ring->adev;
1547 
1548 	if (ring == &adev->vcn.inst->ring_enc[0])
1549 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR);
1550 	else
1551 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2);
1552 }
1553 
1554  /**
1555  * vcn_v2_0_enc_ring_get_wptr - get enc write pointer
1556  *
1557  * @ring: amdgpu_ring pointer
1558  *
1559  * Returns the current hardware enc write pointer
1560  */
1561 static uint64_t vcn_v2_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1562 {
1563 	struct amdgpu_device *adev = ring->adev;
1564 
1565 	if (ring == &adev->vcn.inst->ring_enc[0]) {
1566 		if (ring->use_doorbell)
1567 			return *ring->wptr_cpu_addr;
1568 		else
1569 			return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1570 	} else {
1571 		if (ring->use_doorbell)
1572 			return *ring->wptr_cpu_addr;
1573 		else
1574 			return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1575 	}
1576 }
1577 
1578  /**
1579  * vcn_v2_0_enc_ring_set_wptr - set enc write pointer
1580  *
1581  * @ring: amdgpu_ring pointer
1582  *
1583  * Commits the enc write pointer to the hardware
1584  */
1585 static void vcn_v2_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1586 {
1587 	struct amdgpu_device *adev = ring->adev;
1588 
1589 	if (ring == &adev->vcn.inst->ring_enc[0]) {
1590 		if (ring->use_doorbell) {
1591 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1592 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1593 		} else {
1594 			WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1595 		}
1596 	} else {
1597 		if (ring->use_doorbell) {
1598 			*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1599 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1600 		} else {
1601 			WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1602 		}
1603 	}
1604 }
1605 
1606 /**
1607  * vcn_v2_0_enc_ring_emit_fence - emit an enc fence & trap command
1608  *
1609  * @ring: amdgpu_ring pointer
1610  * @addr: address
1611  * @seq: sequence number
1612  * @flags: fence related flags
1613  *
1614  * Write enc a fence and a trap command to the ring.
1615  */
1616 void vcn_v2_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1617 				u64 seq, unsigned flags)
1618 {
1619 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1620 
1621 	amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE);
1622 	amdgpu_ring_write(ring, addr);
1623 	amdgpu_ring_write(ring, upper_32_bits(addr));
1624 	amdgpu_ring_write(ring, seq);
1625 	amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP);
1626 }
1627 
1628 void vcn_v2_0_enc_ring_insert_end(struct amdgpu_ring *ring)
1629 {
1630 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
1631 }
1632 
1633 /**
1634  * vcn_v2_0_enc_ring_emit_ib - enc execute indirect buffer
1635  *
1636  * @ring: amdgpu_ring pointer
1637  * @job: job to retrive vmid from
1638  * @ib: indirect buffer to execute
1639  * @flags: unused
1640  *
1641  * Write enc ring commands to execute the indirect buffer
1642  */
1643 void vcn_v2_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
1644 			       struct amdgpu_job *job,
1645 			       struct amdgpu_ib *ib,
1646 			       uint32_t flags)
1647 {
1648 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1649 
1650 	amdgpu_ring_write(ring, VCN_ENC_CMD_IB);
1651 	amdgpu_ring_write(ring, vmid);
1652 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1653 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1654 	amdgpu_ring_write(ring, ib->length_dw);
1655 }
1656 
1657 void vcn_v2_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1658 				uint32_t val, uint32_t mask)
1659 {
1660 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
1661 	amdgpu_ring_write(ring, reg << 2);
1662 	amdgpu_ring_write(ring, mask);
1663 	amdgpu_ring_write(ring, val);
1664 }
1665 
1666 void vcn_v2_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
1667 				unsigned int vmid, uint64_t pd_addr)
1668 {
1669 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub];
1670 
1671 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1672 
1673 	/* wait for reg writes */
1674 	vcn_v2_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 +
1675 					vmid * hub->ctx_addr_distance,
1676 					lower_32_bits(pd_addr), 0xffffffff);
1677 }
1678 
1679 void vcn_v2_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1680 {
1681 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
1682 	amdgpu_ring_write(ring,	reg << 2);
1683 	amdgpu_ring_write(ring, val);
1684 }
1685 
1686 static int vcn_v2_0_set_interrupt_state(struct amdgpu_device *adev,
1687 					struct amdgpu_irq_src *source,
1688 					unsigned type,
1689 					enum amdgpu_interrupt_state state)
1690 {
1691 	return 0;
1692 }
1693 
1694 static int vcn_v2_0_process_interrupt(struct amdgpu_device *adev,
1695 				      struct amdgpu_irq_src *source,
1696 				      struct amdgpu_iv_entry *entry)
1697 {
1698 	DRM_DEBUG("IH: VCN TRAP\n");
1699 
1700 	switch (entry->src_id) {
1701 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1702 		amdgpu_fence_process(&adev->vcn.inst->ring_dec);
1703 		break;
1704 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1705 		amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]);
1706 		break;
1707 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1708 		amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]);
1709 		break;
1710 	default:
1711 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1712 			  entry->src_id, entry->src_data[0]);
1713 		break;
1714 	}
1715 
1716 	return 0;
1717 }
1718 
1719 int vcn_v2_0_dec_ring_test_ring(struct amdgpu_ring *ring)
1720 {
1721 	struct amdgpu_device *adev = ring->adev;
1722 	uint32_t tmp = 0;
1723 	unsigned i;
1724 	int r;
1725 
1726 	if (amdgpu_sriov_vf(adev))
1727 		return 0;
1728 
1729 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
1730 	r = amdgpu_ring_alloc(ring, 4);
1731 	if (r)
1732 		return r;
1733 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1734 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1735 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
1736 	amdgpu_ring_write(ring, 0xDEADBEEF);
1737 	amdgpu_ring_commit(ring);
1738 	for (i = 0; i < adev->usec_timeout; i++) {
1739 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
1740 		if (tmp == 0xDEADBEEF)
1741 			break;
1742 		udelay(1);
1743 	}
1744 
1745 	if (i >= adev->usec_timeout)
1746 		r = -ETIMEDOUT;
1747 
1748 	return r;
1749 }
1750 
1751 
1752 static int vcn_v2_0_set_powergating_state(void *handle,
1753 					  enum amd_powergating_state state)
1754 {
1755 	/* This doesn't actually powergate the VCN block.
1756 	 * That's done in the dpm code via the SMC.  This
1757 	 * just re-inits the block as necessary.  The actual
1758 	 * gating still happens in the dpm code.  We should
1759 	 * revisit this when there is a cleaner line between
1760 	 * the smc and the hw blocks
1761 	 */
1762 	int ret;
1763 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1764 
1765 	if (amdgpu_sriov_vf(adev)) {
1766 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1767 		return 0;
1768 	}
1769 
1770 	if (state == adev->vcn.cur_state)
1771 		return 0;
1772 
1773 	if (state == AMD_PG_STATE_GATE)
1774 		ret = vcn_v2_0_stop(adev);
1775 	else
1776 		ret = vcn_v2_0_start(adev);
1777 
1778 	if (!ret)
1779 		adev->vcn.cur_state = state;
1780 	return ret;
1781 }
1782 
1783 static int vcn_v2_0_start_mmsch(struct amdgpu_device *adev,
1784 				struct amdgpu_mm_table *table)
1785 {
1786 	uint32_t data = 0, loop;
1787 	uint64_t addr = table->gpu_addr;
1788 	struct mmsch_v2_0_init_header *header;
1789 	uint32_t size;
1790 	int i;
1791 
1792 	header = (struct mmsch_v2_0_init_header *)table->cpu_addr;
1793 	size = header->header_size + header->vcn_table_size;
1794 
1795 	/* 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1796 	 * of memory descriptor location
1797 	 */
1798 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1799 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1800 
1801 	/* 2, update vmid of descriptor */
1802 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID);
1803 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1804 	/* use domain0 for MM scheduler */
1805 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1806 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID, data);
1807 
1808 	/* 3, notify mmsch about the size of this descriptor */
1809 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_SIZE, size);
1810 
1811 	/* 4, set resp to zero */
1812 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1813 
1814 	adev->vcn.inst->ring_dec.wptr = 0;
1815 	adev->vcn.inst->ring_dec.wptr_old = 0;
1816 	vcn_v2_0_dec_ring_set_wptr(&adev->vcn.inst->ring_dec);
1817 
1818 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1819 		adev->vcn.inst->ring_enc[i].wptr = 0;
1820 		adev->vcn.inst->ring_enc[i].wptr_old = 0;
1821 		vcn_v2_0_enc_ring_set_wptr(&adev->vcn.inst->ring_enc[i]);
1822 	}
1823 
1824 	/* 5, kick off the initialization and wait until
1825 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1826 	 */
1827 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1828 
1829 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1830 	loop = 1000;
1831 	while ((data & 0x10000002) != 0x10000002) {
1832 		udelay(10);
1833 		data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1834 		loop--;
1835 		if (!loop)
1836 			break;
1837 	}
1838 
1839 	if (!loop) {
1840 		DRM_ERROR("failed to init MMSCH, " \
1841 			"mmMMSCH_VF_MAILBOX_RESP = 0x%08x\n", data);
1842 		return -EBUSY;
1843 	}
1844 
1845 	return 0;
1846 }
1847 
1848 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev)
1849 {
1850 	int r;
1851 	uint32_t tmp;
1852 	struct amdgpu_ring *ring;
1853 	uint32_t offset, size;
1854 	uint32_t table_size = 0;
1855 	struct mmsch_v2_0_cmd_direct_write direct_wt = { {0} };
1856 	struct mmsch_v2_0_cmd_direct_read_modify_write direct_rd_mod_wt = { {0} };
1857 	struct mmsch_v2_0_cmd_end end = { {0} };
1858 	struct mmsch_v2_0_init_header *header;
1859 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1860 	uint8_t i = 0;
1861 
1862 	header = (struct mmsch_v2_0_init_header *)init_table;
1863 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1864 	direct_rd_mod_wt.cmd_header.command_type =
1865 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1866 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1867 
1868 	if (header->vcn_table_offset == 0 && header->vcn_table_size == 0) {
1869 		header->version = MMSCH_VERSION;
1870 		header->header_size = sizeof(struct mmsch_v2_0_init_header) >> 2;
1871 
1872 		header->vcn_table_offset = header->header_size;
1873 
1874 		init_table += header->vcn_table_offset;
1875 
1876 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[0]->size + 4);
1877 
1878 		MMSCH_V2_0_INSERT_DIRECT_RD_MOD_WT(
1879 			SOC15_REG_OFFSET(UVD, i, mmUVD_STATUS),
1880 			0xFFFFFFFF, 0x00000004);
1881 
1882 		/* mc resume*/
1883 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1884 			MMSCH_V2_0_INSERT_DIRECT_WT(
1885 				SOC15_REG_OFFSET(UVD, i,
1886 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1887 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo);
1888 			MMSCH_V2_0_INSERT_DIRECT_WT(
1889 				SOC15_REG_OFFSET(UVD, i,
1890 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1891 				adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi);
1892 			offset = 0;
1893 		} else {
1894 			MMSCH_V2_0_INSERT_DIRECT_WT(
1895 				SOC15_REG_OFFSET(UVD, i,
1896 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1897 				lower_32_bits(adev->vcn.inst->gpu_addr));
1898 			MMSCH_V2_0_INSERT_DIRECT_WT(
1899 				SOC15_REG_OFFSET(UVD, i,
1900 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1901 				upper_32_bits(adev->vcn.inst->gpu_addr));
1902 			offset = size;
1903 		}
1904 
1905 		MMSCH_V2_0_INSERT_DIRECT_WT(
1906 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0),
1907 			0);
1908 		MMSCH_V2_0_INSERT_DIRECT_WT(
1909 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE0),
1910 			size);
1911 
1912 		MMSCH_V2_0_INSERT_DIRECT_WT(
1913 			SOC15_REG_OFFSET(UVD, i,
1914 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1915 			lower_32_bits(adev->vcn.inst->gpu_addr + offset));
1916 		MMSCH_V2_0_INSERT_DIRECT_WT(
1917 			SOC15_REG_OFFSET(UVD, i,
1918 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1919 			upper_32_bits(adev->vcn.inst->gpu_addr + offset));
1920 		MMSCH_V2_0_INSERT_DIRECT_WT(
1921 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET1),
1922 			0);
1923 		MMSCH_V2_0_INSERT_DIRECT_WT(
1924 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE1),
1925 			AMDGPU_VCN_STACK_SIZE);
1926 
1927 		MMSCH_V2_0_INSERT_DIRECT_WT(
1928 			SOC15_REG_OFFSET(UVD, i,
1929 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1930 			lower_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,
1934 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1935 			upper_32_bits(adev->vcn.inst->gpu_addr + offset +
1936 				AMDGPU_VCN_STACK_SIZE));
1937 		MMSCH_V2_0_INSERT_DIRECT_WT(
1938 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET2),
1939 			0);
1940 		MMSCH_V2_0_INSERT_DIRECT_WT(
1941 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE2),
1942 			AMDGPU_VCN_CONTEXT_SIZE);
1943 
1944 		for (r = 0; r < adev->vcn.num_enc_rings; ++r) {
1945 			ring = &adev->vcn.inst->ring_enc[r];
1946 			ring->wptr = 0;
1947 			MMSCH_V2_0_INSERT_DIRECT_WT(
1948 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_LO),
1949 				lower_32_bits(ring->gpu_addr));
1950 			MMSCH_V2_0_INSERT_DIRECT_WT(
1951 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_HI),
1952 				upper_32_bits(ring->gpu_addr));
1953 			MMSCH_V2_0_INSERT_DIRECT_WT(
1954 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_SIZE),
1955 				ring->ring_size / 4);
1956 		}
1957 
1958 		ring = &adev->vcn.inst->ring_dec;
1959 		ring->wptr = 0;
1960 		MMSCH_V2_0_INSERT_DIRECT_WT(
1961 			SOC15_REG_OFFSET(UVD, i,
1962 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1963 			lower_32_bits(ring->gpu_addr));
1964 		MMSCH_V2_0_INSERT_DIRECT_WT(
1965 			SOC15_REG_OFFSET(UVD, i,
1966 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1967 			upper_32_bits(ring->gpu_addr));
1968 		/* force RBC into idle state */
1969 		tmp = order_base_2(ring->ring_size);
1970 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1971 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1972 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1973 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1974 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1975 		MMSCH_V2_0_INSERT_DIRECT_WT(
1976 			SOC15_REG_OFFSET(UVD, i, mmUVD_RBC_RB_CNTL), tmp);
1977 
1978 		/* add end packet */
1979 		tmp = sizeof(struct mmsch_v2_0_cmd_end);
1980 		memcpy((void *)init_table, &end, tmp);
1981 		table_size += (tmp / 4);
1982 		header->vcn_table_size = table_size;
1983 
1984 	}
1985 	return vcn_v2_0_start_mmsch(adev, &adev->virt.mm_table);
1986 }
1987 
1988 static const struct amd_ip_funcs vcn_v2_0_ip_funcs = {
1989 	.name = "vcn_v2_0",
1990 	.early_init = vcn_v2_0_early_init,
1991 	.late_init = NULL,
1992 	.sw_init = vcn_v2_0_sw_init,
1993 	.sw_fini = vcn_v2_0_sw_fini,
1994 	.hw_init = vcn_v2_0_hw_init,
1995 	.hw_fini = vcn_v2_0_hw_fini,
1996 	.suspend = vcn_v2_0_suspend,
1997 	.resume = vcn_v2_0_resume,
1998 	.is_idle = vcn_v2_0_is_idle,
1999 	.wait_for_idle = vcn_v2_0_wait_for_idle,
2000 	.check_soft_reset = NULL,
2001 	.pre_soft_reset = NULL,
2002 	.soft_reset = NULL,
2003 	.post_soft_reset = NULL,
2004 	.set_clockgating_state = vcn_v2_0_set_clockgating_state,
2005 	.set_powergating_state = vcn_v2_0_set_powergating_state,
2006 	.dump_ip_state = NULL,
2007 	.print_ip_state = NULL,
2008 };
2009 
2010 static const struct amdgpu_ring_funcs vcn_v2_0_dec_ring_vm_funcs = {
2011 	.type = AMDGPU_RING_TYPE_VCN_DEC,
2012 	.align_mask = 0xf,
2013 	.secure_submission_supported = true,
2014 	.get_rptr = vcn_v2_0_dec_ring_get_rptr,
2015 	.get_wptr = vcn_v2_0_dec_ring_get_wptr,
2016 	.set_wptr = vcn_v2_0_dec_ring_set_wptr,
2017 	.emit_frame_size =
2018 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
2019 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
2020 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
2021 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
2022 		6,
2023 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
2024 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
2025 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
2026 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
2027 	.test_ring = vcn_v2_0_dec_ring_test_ring,
2028 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
2029 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
2030 	.insert_start = vcn_v2_0_dec_ring_insert_start,
2031 	.insert_end = vcn_v2_0_dec_ring_insert_end,
2032 	.pad_ib = amdgpu_ring_generic_pad_ib,
2033 	.begin_use = amdgpu_vcn_ring_begin_use,
2034 	.end_use = amdgpu_vcn_ring_end_use,
2035 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
2036 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
2037 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2038 };
2039 
2040 static const struct amdgpu_ring_funcs vcn_v2_0_enc_ring_vm_funcs = {
2041 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2042 	.align_mask = 0x3f,
2043 	.nop = VCN_ENC_CMD_NO_OP,
2044 	.get_rptr = vcn_v2_0_enc_ring_get_rptr,
2045 	.get_wptr = vcn_v2_0_enc_ring_get_wptr,
2046 	.set_wptr = vcn_v2_0_enc_ring_set_wptr,
2047 	.emit_frame_size =
2048 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2049 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2050 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2051 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2052 		1, /* vcn_v2_0_enc_ring_insert_end */
2053 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2054 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2055 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2056 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2057 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2058 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2059 	.insert_nop = amdgpu_ring_insert_nop,
2060 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2061 	.pad_ib = amdgpu_ring_generic_pad_ib,
2062 	.begin_use = amdgpu_vcn_ring_begin_use,
2063 	.end_use = amdgpu_vcn_ring_end_use,
2064 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2065 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2066 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2067 };
2068 
2069 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2070 {
2071 	adev->vcn.inst->ring_dec.funcs = &vcn_v2_0_dec_ring_vm_funcs;
2072 }
2073 
2074 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2075 {
2076 	int i;
2077 
2078 	for (i = 0; i < adev->vcn.num_enc_rings; ++i)
2079 		adev->vcn.inst->ring_enc[i].funcs = &vcn_v2_0_enc_ring_vm_funcs;
2080 }
2081 
2082 static const struct amdgpu_irq_src_funcs vcn_v2_0_irq_funcs = {
2083 	.set = vcn_v2_0_set_interrupt_state,
2084 	.process = vcn_v2_0_process_interrupt,
2085 };
2086 
2087 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev)
2088 {
2089 	adev->vcn.inst->irq.num_types = adev->vcn.num_enc_rings + 1;
2090 	adev->vcn.inst->irq.funcs = &vcn_v2_0_irq_funcs;
2091 }
2092 
2093 const struct amdgpu_ip_block_version vcn_v2_0_ip_block =
2094 {
2095 		.type = AMD_IP_BLOCK_TYPE_VCN,
2096 		.major = 2,
2097 		.minor = 0,
2098 		.rev = 0,
2099 		.funcs = &vcn_v2_0_ip_funcs,
2100 };
2101