xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c (revision 453431a5)
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 
26 #include "amdgpu.h"
27 #include "amdgpu_vcn.h"
28 #include "soc15.h"
29 #include "soc15d.h"
30 #include "amdgpu_pm.h"
31 #include "amdgpu_psp.h"
32 #include "mmsch_v2_0.h"
33 #include "vcn_v2_0.h"
34 
35 #include "vcn/vcn_2_0_0_offset.h"
36 #include "vcn/vcn_2_0_0_sh_mask.h"
37 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
38 
39 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET			0x1fd
40 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET			0x503
41 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET			0x504
42 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET			0x505
43 #define mmUVD_NO_OP_INTERNAL_OFFSET				0x53f
44 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET			0x54a
45 #define mmUVD_SCRATCH9_INTERNAL_OFFSET				0xc01d
46 
47 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET			0x1e1
48 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 	0x5a6
49 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET		0x5a7
50 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET			0x1e2
51 
52 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev);
53 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev);
54 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev);
55 static int vcn_v2_0_set_powergating_state(void *handle,
56 				enum amd_powergating_state state);
57 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
58 				int inst_idx, struct dpg_pause_state *new_state);
59 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev);
60 /**
61  * vcn_v2_0_early_init - set function pointers
62  *
63  * @handle: amdgpu_device pointer
64  *
65  * Set ring and irq function pointers
66  */
67 static int vcn_v2_0_early_init(void *handle)
68 {
69 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
70 
71 	adev->vcn.num_vcn_inst = 1;
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 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
119 		const struct common_firmware_header *hdr;
120 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
121 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
122 		adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
123 		adev->firmware.fw_size +=
124 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
125 		DRM_INFO("PSP loading VCN firmware\n");
126 	}
127 
128 	r = amdgpu_vcn_resume(adev);
129 	if (r)
130 		return r;
131 
132 	ring = &adev->vcn.inst->ring_dec;
133 
134 	ring->use_doorbell = true;
135 	ring->doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
136 
137 	sprintf(ring->name, "vcn_dec");
138 	r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0,
139 			     AMDGPU_RING_PRIO_DEFAULT);
140 	if (r)
141 		return r;
142 
143 	adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
144 	adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
145 	adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
146 	adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
147 	adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
148 	adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
149 
150 	adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
151 	adev->vcn.inst->external.scratch9 = SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9);
152 	adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
153 	adev->vcn.inst->external.data0 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0);
154 	adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
155 	adev->vcn.inst->external.data1 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1);
156 	adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
157 	adev->vcn.inst->external.cmd = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD);
158 	adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
159 	adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP);
160 
161 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
162 		ring = &adev->vcn.inst->ring_enc[i];
163 		ring->use_doorbell = true;
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 				     AMDGPU_RING_PRIO_DEFAULT);
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 	return 0;
184 }
185 
186 /**
187  * vcn_v2_0_sw_fini - sw fini for VCN block
188  *
189  * @handle: amdgpu_device pointer
190  *
191  * VCN suspend and free up sw allocation
192  */
193 static int vcn_v2_0_sw_fini(void *handle)
194 {
195 	int r;
196 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
197 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
198 
199 	fw_shared->present_flag_0 = 0;
200 
201 	amdgpu_virt_free_mm_table(adev);
202 
203 	r = amdgpu_vcn_suspend(adev);
204 	if (r)
205 		return r;
206 
207 	r = amdgpu_vcn_sw_fini(adev);
208 
209 	return r;
210 }
211 
212 /**
213  * vcn_v2_0_hw_init - start and test VCN block
214  *
215  * @handle: amdgpu_device pointer
216  *
217  * Initialize the hardware, boot up the VCPU and do some testing
218  */
219 static int vcn_v2_0_hw_init(void *handle)
220 {
221 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
222 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
223 	int i, r;
224 
225 	adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
226 					     ring->doorbell_index, 0);
227 
228 	if (amdgpu_sriov_vf(adev))
229 		vcn_v2_0_start_sriov(adev);
230 
231 	r = amdgpu_ring_test_helper(ring);
232 	if (r)
233 		goto done;
234 
235 	//Disable vcn decode for sriov
236 	if (amdgpu_sriov_vf(adev))
237 		ring->sched.ready = false;
238 
239 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
240 		ring = &adev->vcn.inst->ring_enc[i];
241 		r = amdgpu_ring_test_helper(ring);
242 		if (r)
243 			goto done;
244 	}
245 
246 done:
247 	if (!r)
248 		DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
249 			(adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
250 
251 	return r;
252 }
253 
254 /**
255  * vcn_v2_0_hw_fini - stop the hardware block
256  *
257  * @handle: amdgpu_device pointer
258  *
259  * Stop the VCN block, mark ring as not ready any more
260  */
261 static int vcn_v2_0_hw_fini(void *handle)
262 {
263 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
264 
265 	if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
266 	    (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
267 	      RREG32_SOC15(VCN, 0, mmUVD_STATUS)))
268 		vcn_v2_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
269 
270 	return 0;
271 }
272 
273 /**
274  * vcn_v2_0_suspend - suspend VCN block
275  *
276  * @handle: amdgpu_device pointer
277  *
278  * HW fini and suspend VCN block
279  */
280 static int vcn_v2_0_suspend(void *handle)
281 {
282 	int r;
283 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
284 
285 	r = vcn_v2_0_hw_fini(adev);
286 	if (r)
287 		return r;
288 
289 	r = amdgpu_vcn_suspend(adev);
290 
291 	return r;
292 }
293 
294 /**
295  * vcn_v2_0_resume - resume VCN block
296  *
297  * @handle: amdgpu_device pointer
298  *
299  * Resume firmware and hw init VCN block
300  */
301 static int vcn_v2_0_resume(void *handle)
302 {
303 	int r;
304 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
305 
306 	r = amdgpu_vcn_resume(adev);
307 	if (r)
308 		return r;
309 
310 	r = vcn_v2_0_hw_init(adev);
311 
312 	return r;
313 }
314 
315 /**
316  * vcn_v2_0_mc_resume - memory controller programming
317  *
318  * @adev: amdgpu_device pointer
319  *
320  * Let the VCN memory controller know it's offsets
321  */
322 static void vcn_v2_0_mc_resume(struct amdgpu_device *adev)
323 {
324 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
325 	uint32_t offset;
326 
327 	if (amdgpu_sriov_vf(adev))
328 		return;
329 
330 	/* cache window 0: fw */
331 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
332 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
333 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo));
334 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
335 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi));
336 		WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0);
337 		offset = 0;
338 	} else {
339 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
340 			lower_32_bits(adev->vcn.inst->gpu_addr));
341 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
342 			upper_32_bits(adev->vcn.inst->gpu_addr));
343 		offset = size;
344 		WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0,
345 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
346 	}
347 
348 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size);
349 
350 	/* cache window 1: stack */
351 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
352 		lower_32_bits(adev->vcn.inst->gpu_addr + offset));
353 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
354 		upper_32_bits(adev->vcn.inst->gpu_addr + offset));
355 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0);
356 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
357 
358 	/* cache window 2: context */
359 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
360 		lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
361 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
362 		upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
363 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0);
364 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
365 
366 	/* non-cache window */
367 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
368 		lower_32_bits(adev->vcn.inst->fw_shared_gpu_addr));
369 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
370 		upper_32_bits(adev->vcn.inst->fw_shared_gpu_addr));
371 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
372 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0,
373 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
374 
375 	WREG32_SOC15(UVD, 0, mmUVD_GFX10_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
376 }
377 
378 static void vcn_v2_0_mc_resume_dpg_mode(struct amdgpu_device *adev, bool indirect)
379 {
380 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
381 	uint32_t offset;
382 
383 	/* cache window 0: fw */
384 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
385 		if (!indirect) {
386 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
387 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
388 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 0, indirect);
389 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
390 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
391 				(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 0, indirect);
392 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
393 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
394 		} else {
395 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
396 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
397 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
398 				UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
399 			WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
400 				UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
401 		}
402 		offset = 0;
403 	} else {
404 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
405 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
406 			lower_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
407 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
408 			UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
409 			upper_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
410 		offset = size;
411 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
412 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
413 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
414 	}
415 
416 	if (!indirect)
417 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
418 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
419 	else
420 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
421 			UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
422 
423 	/* cache window 1: stack */
424 	if (!indirect) {
425 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
426 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
427 			lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
428 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
429 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
430 			upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
431 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
432 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
433 	} else {
434 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
435 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
436 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
437 			UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
438 		WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
439 			UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
440 	}
441 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
442 		UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
443 
444 	/* cache window 2: context */
445 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
446 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
447 		lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
448 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
449 		UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
450 		upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
451 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
452 		UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
453 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
454 		UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
455 
456 	/* non-cache window */
457 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
458 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
459 		lower_32_bits(adev->vcn.inst->fw_shared_gpu_addr), 0, indirect);
460 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
461 		UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
462 		upper_32_bits(adev->vcn.inst->fw_shared_gpu_addr), 0, indirect);
463 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
464 		UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
465 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
466 		UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0),
467 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
468 
469 	/* VCN global tiling registers */
470 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
471 		UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
472 }
473 
474 /**
475  * vcn_v2_0_disable_clock_gating - disable VCN clock gating
476  *
477  * @adev: amdgpu_device pointer
478  * @sw: enable SW clock gating
479  *
480  * Disable clock gating for VCN block
481  */
482 static void vcn_v2_0_disable_clock_gating(struct amdgpu_device *adev)
483 {
484 	uint32_t data;
485 
486 	if (amdgpu_sriov_vf(adev))
487 		return;
488 
489 	/* UVD disable CGC */
490 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
491 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
492 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
493 	else
494 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
495 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
496 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
497 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
498 
499 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_GATE);
500 	data &= ~(UVD_CGC_GATE__SYS_MASK
501 		| UVD_CGC_GATE__UDEC_MASK
502 		| UVD_CGC_GATE__MPEG2_MASK
503 		| UVD_CGC_GATE__REGS_MASK
504 		| UVD_CGC_GATE__RBC_MASK
505 		| UVD_CGC_GATE__LMI_MC_MASK
506 		| UVD_CGC_GATE__LMI_UMC_MASK
507 		| UVD_CGC_GATE__IDCT_MASK
508 		| UVD_CGC_GATE__MPRD_MASK
509 		| UVD_CGC_GATE__MPC_MASK
510 		| UVD_CGC_GATE__LBSI_MASK
511 		| UVD_CGC_GATE__LRBBM_MASK
512 		| UVD_CGC_GATE__UDEC_RE_MASK
513 		| UVD_CGC_GATE__UDEC_CM_MASK
514 		| UVD_CGC_GATE__UDEC_IT_MASK
515 		| UVD_CGC_GATE__UDEC_DB_MASK
516 		| UVD_CGC_GATE__UDEC_MP_MASK
517 		| UVD_CGC_GATE__WCB_MASK
518 		| UVD_CGC_GATE__VCPU_MASK
519 		| UVD_CGC_GATE__SCPU_MASK);
520 	WREG32_SOC15(VCN, 0, mmUVD_CGC_GATE, data);
521 
522 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
523 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
524 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
525 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
526 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
527 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
528 		| UVD_CGC_CTRL__SYS_MODE_MASK
529 		| UVD_CGC_CTRL__UDEC_MODE_MASK
530 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
531 		| UVD_CGC_CTRL__REGS_MODE_MASK
532 		| UVD_CGC_CTRL__RBC_MODE_MASK
533 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
534 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
535 		| UVD_CGC_CTRL__IDCT_MODE_MASK
536 		| UVD_CGC_CTRL__MPRD_MODE_MASK
537 		| UVD_CGC_CTRL__MPC_MODE_MASK
538 		| UVD_CGC_CTRL__LBSI_MODE_MASK
539 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
540 		| UVD_CGC_CTRL__WCB_MODE_MASK
541 		| UVD_CGC_CTRL__VCPU_MODE_MASK
542 		| UVD_CGC_CTRL__SCPU_MODE_MASK);
543 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
544 
545 	/* turn on */
546 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE);
547 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
548 		| UVD_SUVD_CGC_GATE__SIT_MASK
549 		| UVD_SUVD_CGC_GATE__SMP_MASK
550 		| UVD_SUVD_CGC_GATE__SCM_MASK
551 		| UVD_SUVD_CGC_GATE__SDB_MASK
552 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
553 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
554 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
555 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
556 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
557 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
558 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
559 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
560 		| UVD_SUVD_CGC_GATE__SCLR_MASK
561 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
562 		| UVD_SUVD_CGC_GATE__ENT_MASK
563 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
564 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
565 		| UVD_SUVD_CGC_GATE__SITE_MASK
566 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
567 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
568 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
569 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
570 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
571 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE, data);
572 
573 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
574 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
575 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
576 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
577 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
578 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
579 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
580 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
581 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
582 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
583 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
584 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
585 }
586 
587 static void vcn_v2_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
588 		uint8_t sram_sel, uint8_t indirect)
589 {
590 	uint32_t reg_data = 0;
591 
592 	/* enable sw clock gating control */
593 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
594 		reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
595 	else
596 		reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
597 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
598 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
599 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
600 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
601 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
602 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
603 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
604 		 UVD_CGC_CTRL__SYS_MODE_MASK |
605 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
606 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
607 		 UVD_CGC_CTRL__REGS_MODE_MASK |
608 		 UVD_CGC_CTRL__RBC_MODE_MASK |
609 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
610 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
611 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
612 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
613 		 UVD_CGC_CTRL__MPC_MODE_MASK |
614 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
615 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
616 		 UVD_CGC_CTRL__WCB_MODE_MASK |
617 		 UVD_CGC_CTRL__VCPU_MODE_MASK |
618 		 UVD_CGC_CTRL__SCPU_MODE_MASK);
619 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
620 		UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
621 
622 	/* turn off clock gating */
623 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
624 		UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
625 
626 	/* turn on SUVD clock gating */
627 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
628 		UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
629 
630 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
631 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
632 		UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
633 }
634 
635 /**
636  * vcn_v2_0_enable_clock_gating - enable VCN clock gating
637  *
638  * @adev: amdgpu_device pointer
639  * @sw: enable SW clock gating
640  *
641  * Enable clock gating for VCN block
642  */
643 static void vcn_v2_0_enable_clock_gating(struct amdgpu_device *adev)
644 {
645 	uint32_t data = 0;
646 
647 	if (amdgpu_sriov_vf(adev))
648 		return;
649 
650 	/* enable UVD CGC */
651 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
652 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
653 		data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
654 	else
655 		data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
656 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
657 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
658 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
659 
660 	data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
661 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
662 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
663 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
664 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
665 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
666 		| UVD_CGC_CTRL__SYS_MODE_MASK
667 		| UVD_CGC_CTRL__UDEC_MODE_MASK
668 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
669 		| UVD_CGC_CTRL__REGS_MODE_MASK
670 		| UVD_CGC_CTRL__RBC_MODE_MASK
671 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
672 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
673 		| UVD_CGC_CTRL__IDCT_MODE_MASK
674 		| UVD_CGC_CTRL__MPRD_MODE_MASK
675 		| UVD_CGC_CTRL__MPC_MODE_MASK
676 		| UVD_CGC_CTRL__LBSI_MODE_MASK
677 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
678 		| UVD_CGC_CTRL__WCB_MODE_MASK
679 		| UVD_CGC_CTRL__VCPU_MODE_MASK
680 		| UVD_CGC_CTRL__SCPU_MODE_MASK);
681 	WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
682 
683 	data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
684 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
685 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
686 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
687 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
688 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
689 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
690 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
691 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
692 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
693 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
694 	WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
695 }
696 
697 static void vcn_v2_0_disable_static_power_gating(struct amdgpu_device *adev)
698 {
699 	uint32_t data = 0;
700 
701 	if (amdgpu_sriov_vf(adev))
702 		return;
703 
704 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
705 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
706 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
707 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
708 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
709 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
710 			| 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
711 			| 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
712 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
713 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
714 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
715 
716 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
717 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS,
718 			UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON_2_0, 0xFFFFF);
719 	} else {
720 		data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
721 			| 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
722 			| 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
723 			| 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
724 			| 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
725 			| 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
726 			| 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
727 			| 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
728 			| 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
729 			| 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
730 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
731 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0,  0xFFFFF);
732 	}
733 
734 	/* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS,
735 	 * UVDU_PWR_STATUS are 0 (power on) */
736 
737 	data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
738 	data &= ~0x103;
739 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
740 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
741 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
742 
743 	WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
744 }
745 
746 static void vcn_v2_0_enable_static_power_gating(struct amdgpu_device *adev)
747 {
748 	uint32_t data = 0;
749 
750 	if (amdgpu_sriov_vf(adev))
751 		return;
752 
753 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
754 		/* Before power off, this indicator has to be turned on */
755 		data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
756 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
757 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
758 		WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
759 
760 
761 		data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
762 			| 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
763 			| 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
764 			| 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
765 			| 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
766 			| 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
767 			| 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
768 			| 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
769 			| 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
770 			| 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
771 
772 		WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
773 
774 		data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
775 			| 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
776 			| 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
777 			| 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
778 			| 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
779 			| 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT
780 			| 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT
781 			| 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
782 			| 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
783 			| 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT);
784 		SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFF);
785 	}
786 }
787 
788 static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
789 {
790 	volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
791 	struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
792 	uint32_t rb_bufsz, tmp;
793 
794 	vcn_v2_0_enable_static_power_gating(adev);
795 
796 	/* enable dynamic power gating mode */
797 	tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS);
798 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
799 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
800 	WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp);
801 
802 	if (indirect)
803 		adev->vcn.inst->dpg_sram_curr_addr = (uint32_t*)adev->vcn.inst->dpg_sram_cpu_addr;
804 
805 	/* enable clock gating */
806 	vcn_v2_0_clock_gating_dpg_mode(adev, 0, indirect);
807 
808 	/* enable VCPU clock */
809 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
810 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
811 	tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK;
812 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
813 		UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
814 
815 	/* disable master interupt */
816 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
817 		UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
818 
819 	/* setup mmUVD_LMI_CTRL */
820 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
821 		UVD_LMI_CTRL__REQ_MODE_MASK |
822 		UVD_LMI_CTRL__CRC_RESET_MASK |
823 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
824 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
825 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
826 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
827 		0x00100000L);
828 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
829 		UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
830 
831 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
832 		UVD, 0, mmUVD_MPC_CNTL),
833 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
834 
835 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
836 		UVD, 0, mmUVD_MPC_SET_MUXA0),
837 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
838 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
839 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
840 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
841 
842 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
843 		UVD, 0, mmUVD_MPC_SET_MUXB0),
844 		((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
845 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
846 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
847 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
848 
849 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
850 		UVD, 0, mmUVD_MPC_SET_MUX),
851 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
852 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
853 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
854 
855 	vcn_v2_0_mc_resume_dpg_mode(adev, indirect);
856 
857 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
858 		UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
859 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
860 		UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
861 
862 	/* release VCPU reset to boot */
863 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
864 		UVD, 0, mmUVD_SOFT_RESET), 0, 0, indirect);
865 
866 	/* enable LMI MC and UMC channels */
867 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
868 		UVD, 0, mmUVD_LMI_CTRL2),
869 		0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 0, indirect);
870 
871 	/* enable master interrupt */
872 	WREG32_SOC15_DPG_MODE(0, SOC15_DPG_MODE_OFFSET(
873 		UVD, 0, mmUVD_MASTINT_EN),
874 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
875 
876 	if (indirect)
877 		psp_update_vcn_sram(adev, 0, adev->vcn.inst->dpg_sram_gpu_addr,
878 				    (uint32_t)((uintptr_t)adev->vcn.inst->dpg_sram_curr_addr -
879 					       (uintptr_t)adev->vcn.inst->dpg_sram_cpu_addr));
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 	/* programm 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 	/* programm 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 	uint32_t tmp;
1101 
1102 	/* Wait for power status to be 1 */
1103 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1104 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1105 
1106 	/* wait for read ptr to be equal to write ptr */
1107 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1108 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1109 
1110 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1111 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1112 
1113 	tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1114 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1115 
1116 	SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1117 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1118 
1119 	/* disable dynamic power gating mode */
1120 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0,
1121 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1122 
1123 	return 0;
1124 }
1125 
1126 static int vcn_v2_0_stop(struct amdgpu_device *adev)
1127 {
1128 	uint32_t tmp;
1129 	int r;
1130 
1131 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1132 		r = vcn_v2_0_stop_dpg_mode(adev);
1133 		if (r)
1134 			return r;
1135 		goto power_off;
1136 	}
1137 
1138 	/* wait for uvd idle */
1139 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1140 	if (r)
1141 		return r;
1142 
1143 	tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1144 		UVD_LMI_STATUS__READ_CLEAN_MASK |
1145 		UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1146 		UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1147 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp);
1148 	if (r)
1149 		return r;
1150 
1151 	/* stall UMC channel */
1152 	tmp = RREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2);
1153 	tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1154 	WREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2, tmp);
1155 
1156 	tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1157 		UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1158 	r = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp);
1159 	if (r)
1160 		return r;
1161 
1162 	/* disable VCPU clock */
1163 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0,
1164 		~(UVD_VCPU_CNTL__CLK_EN_MASK));
1165 
1166 	/* reset LMI UMC */
1167 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1168 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK,
1169 		~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
1170 
1171 	/* reset LMI */
1172 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1173 		UVD_SOFT_RESET__LMI_SOFT_RESET_MASK,
1174 		~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
1175 
1176 	/* reset VCPU */
1177 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1178 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
1179 		~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1180 
1181 	/* clear status */
1182 	WREG32_SOC15(VCN, 0, mmUVD_STATUS, 0);
1183 
1184 	vcn_v2_0_enable_clock_gating(adev);
1185 	vcn_v2_0_enable_static_power_gating(adev);
1186 
1187 power_off:
1188 	if (adev->pm.dpm_enabled)
1189 		amdgpu_dpm_enable_uvd(adev, false);
1190 
1191 	return 0;
1192 }
1193 
1194 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
1195 				int inst_idx, struct dpg_pause_state *new_state)
1196 {
1197 	struct amdgpu_ring *ring;
1198 	uint32_t reg_data = 0;
1199 	int ret_code;
1200 
1201 	/* pause/unpause if state is changed */
1202 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1203 		DRM_DEBUG("dpg pause state changed %d -> %d",
1204 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1205 		reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) &
1206 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1207 
1208 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1209 			ret_code = SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 0x1,
1210 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1211 
1212 			if (!ret_code) {
1213 				volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;
1214 				/* pause DPG */
1215 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1216 				WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1217 
1218 				/* wait for ACK */
1219 				SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE,
1220 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1221 					   UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1222 
1223 				/* Stall DPG before WPTR/RPTR reset */
1224 				WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
1225 					   UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1226 					   ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1227 				/* Restore */
1228 				fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1229 				ring = &adev->vcn.inst->ring_enc[0];
1230 				ring->wptr = 0;
1231 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1232 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1233 				WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1234 				WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1235 				WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1236 				fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1237 
1238 				fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1239 				ring = &adev->vcn.inst->ring_enc[1];
1240 				ring->wptr = 0;
1241 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1242 				WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1243 				WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1244 				WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1245 				WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1246 				fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1247 
1248 				fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1249 				WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1250 					   RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF);
1251 				fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1252 				/* Unstall DPG */
1253 				WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS),
1254 					   0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1255 
1256 				SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS,
1257 					   UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON,
1258 					   UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1259 			}
1260 		} else {
1261 			/* unpause dpg, no need to wait */
1262 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1263 			WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1264 		}
1265 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1266 	}
1267 
1268 	return 0;
1269 }
1270 
1271 static bool vcn_v2_0_is_idle(void *handle)
1272 {
1273 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1274 
1275 	return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE);
1276 }
1277 
1278 static int vcn_v2_0_wait_for_idle(void *handle)
1279 {
1280 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1281 	int ret;
1282 
1283 	ret = SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE,
1284 		UVD_STATUS__IDLE);
1285 
1286 	return ret;
1287 }
1288 
1289 static int vcn_v2_0_set_clockgating_state(void *handle,
1290 					  enum amd_clockgating_state state)
1291 {
1292 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1293 	bool enable = (state == AMD_CG_STATE_GATE);
1294 
1295 	if (amdgpu_sriov_vf(adev))
1296 		return 0;
1297 
1298 	if (enable) {
1299 		/* wait for STATUS to clear */
1300 		if (!vcn_v2_0_is_idle(handle))
1301 			return -EBUSY;
1302 		vcn_v2_0_enable_clock_gating(adev);
1303 	} else {
1304 		/* disable HW gating and enable Sw gating */
1305 		vcn_v2_0_disable_clock_gating(adev);
1306 	}
1307 	return 0;
1308 }
1309 
1310 /**
1311  * vcn_v2_0_dec_ring_get_rptr - get read pointer
1312  *
1313  * @ring: amdgpu_ring pointer
1314  *
1315  * Returns the current hardware read pointer
1316  */
1317 static uint64_t vcn_v2_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1318 {
1319 	struct amdgpu_device *adev = ring->adev;
1320 
1321 	return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1322 }
1323 
1324 /**
1325  * vcn_v2_0_dec_ring_get_wptr - get write pointer
1326  *
1327  * @ring: amdgpu_ring pointer
1328  *
1329  * Returns the current hardware write pointer
1330  */
1331 static uint64_t vcn_v2_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1332 {
1333 	struct amdgpu_device *adev = ring->adev;
1334 
1335 	if (ring->use_doorbell)
1336 		return adev->wb.wb[ring->wptr_offs];
1337 	else
1338 		return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR);
1339 }
1340 
1341 /**
1342  * vcn_v2_0_dec_ring_set_wptr - set write pointer
1343  *
1344  * @ring: amdgpu_ring pointer
1345  *
1346  * Commits the write pointer to the hardware
1347  */
1348 static void vcn_v2_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1349 {
1350 	struct amdgpu_device *adev = ring->adev;
1351 
1352 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
1353 		WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2,
1354 			lower_32_bits(ring->wptr) | 0x80000000);
1355 
1356 	if (ring->use_doorbell) {
1357 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1358 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1359 	} else {
1360 		WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1361 	}
1362 }
1363 
1364 /**
1365  * vcn_v2_0_dec_ring_insert_start - insert a start command
1366  *
1367  * @ring: amdgpu_ring pointer
1368  *
1369  * Write a start command to the ring.
1370  */
1371 void vcn_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring)
1372 {
1373 	struct amdgpu_device *adev = ring->adev;
1374 
1375 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1376 	amdgpu_ring_write(ring, 0);
1377 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1378 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1379 }
1380 
1381 /**
1382  * vcn_v2_0_dec_ring_insert_end - insert a end command
1383  *
1384  * @ring: amdgpu_ring pointer
1385  *
1386  * Write a end command to the ring.
1387  */
1388 void vcn_v2_0_dec_ring_insert_end(struct amdgpu_ring *ring)
1389 {
1390 	struct amdgpu_device *adev = ring->adev;
1391 
1392 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1393 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_END << 1));
1394 }
1395 
1396 /**
1397  * vcn_v2_0_dec_ring_insert_nop - insert a nop command
1398  *
1399  * @ring: amdgpu_ring pointer
1400  *
1401  * Write a nop command to the ring.
1402  */
1403 void vcn_v2_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
1404 {
1405 	struct amdgpu_device *adev = ring->adev;
1406 	int i;
1407 
1408 	WARN_ON(ring->wptr % 2 || count % 2);
1409 
1410 	for (i = 0; i < count / 2; i++) {
1411 		amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.nop, 0));
1412 		amdgpu_ring_write(ring, 0);
1413 	}
1414 }
1415 
1416 /**
1417  * vcn_v2_0_dec_ring_emit_fence - emit an fence & trap command
1418  *
1419  * @ring: amdgpu_ring pointer
1420  * @fence: fence to emit
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  * @ib: indirect buffer to execute
1458  *
1459  * Write ring commands to execute the indirect buffer
1460  */
1461 void vcn_v2_0_dec_ring_emit_ib(struct amdgpu_ring *ring,
1462 			       struct amdgpu_job *job,
1463 			       struct amdgpu_ib *ib,
1464 			       uint32_t flags)
1465 {
1466 	struct amdgpu_device *adev = ring->adev;
1467 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1468 
1469 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_vmid, 0));
1470 	amdgpu_ring_write(ring, vmid);
1471 
1472 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_bar_low, 0));
1473 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1474 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_bar_high, 0));
1475 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1476 	amdgpu_ring_write(ring,	PACKET0(adev->vcn.internal.ib_size, 0));
1477 	amdgpu_ring_write(ring, ib->length_dw);
1478 }
1479 
1480 void vcn_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1481 				uint32_t val, uint32_t mask)
1482 {
1483 	struct amdgpu_device *adev = ring->adev;
1484 
1485 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1486 	amdgpu_ring_write(ring, reg << 2);
1487 
1488 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1489 	amdgpu_ring_write(ring, val);
1490 
1491 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.gp_scratch8, 0));
1492 	amdgpu_ring_write(ring, mask);
1493 
1494 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1495 
1496 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_REG_READ_COND_WAIT << 1));
1497 }
1498 
1499 void vcn_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
1500 				unsigned vmid, uint64_t pd_addr)
1501 {
1502 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1503 	uint32_t data0, data1, mask;
1504 
1505 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1506 
1507 	/* wait for register write */
1508 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
1509 	data1 = lower_32_bits(pd_addr);
1510 	mask = 0xffffffff;
1511 	vcn_v2_0_dec_ring_emit_reg_wait(ring, data0, data1, mask);
1512 }
1513 
1514 void vcn_v2_0_dec_ring_emit_wreg(struct amdgpu_ring *ring,
1515 				uint32_t reg, uint32_t val)
1516 {
1517 	struct amdgpu_device *adev = ring->adev;
1518 
1519 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1520 	amdgpu_ring_write(ring, reg << 2);
1521 
1522 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1523 	amdgpu_ring_write(ring, val);
1524 
1525 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1526 
1527 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_WRITE_REG << 1));
1528 }
1529 
1530 /**
1531  * vcn_v2_0_enc_ring_get_rptr - get enc read pointer
1532  *
1533  * @ring: amdgpu_ring pointer
1534  *
1535  * Returns the current hardware enc read pointer
1536  */
1537 static uint64_t vcn_v2_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1538 {
1539 	struct amdgpu_device *adev = ring->adev;
1540 
1541 	if (ring == &adev->vcn.inst->ring_enc[0])
1542 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR);
1543 	else
1544 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2);
1545 }
1546 
1547  /**
1548  * vcn_v2_0_enc_ring_get_wptr - get enc write pointer
1549  *
1550  * @ring: amdgpu_ring pointer
1551  *
1552  * Returns the current hardware enc write pointer
1553  */
1554 static uint64_t vcn_v2_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1555 {
1556 	struct amdgpu_device *adev = ring->adev;
1557 
1558 	if (ring == &adev->vcn.inst->ring_enc[0]) {
1559 		if (ring->use_doorbell)
1560 			return adev->wb.wb[ring->wptr_offs];
1561 		else
1562 			return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1563 	} else {
1564 		if (ring->use_doorbell)
1565 			return adev->wb.wb[ring->wptr_offs];
1566 		else
1567 			return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1568 	}
1569 }
1570 
1571  /**
1572  * vcn_v2_0_enc_ring_set_wptr - set enc write pointer
1573  *
1574  * @ring: amdgpu_ring pointer
1575  *
1576  * Commits the enc write pointer to the hardware
1577  */
1578 static void vcn_v2_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1579 {
1580 	struct amdgpu_device *adev = ring->adev;
1581 
1582 	if (ring == &adev->vcn.inst->ring_enc[0]) {
1583 		if (ring->use_doorbell) {
1584 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1585 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1586 		} else {
1587 			WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1588 		}
1589 	} else {
1590 		if (ring->use_doorbell) {
1591 			adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1592 			WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1593 		} else {
1594 			WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1595 		}
1596 	}
1597 }
1598 
1599 /**
1600  * vcn_v2_0_enc_ring_emit_fence - emit an enc fence & trap command
1601  *
1602  * @ring: amdgpu_ring pointer
1603  * @fence: fence to emit
1604  *
1605  * Write enc a fence and a trap command to the ring.
1606  */
1607 void vcn_v2_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1608 				u64 seq, unsigned flags)
1609 {
1610 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1611 
1612 	amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE);
1613 	amdgpu_ring_write(ring, addr);
1614 	amdgpu_ring_write(ring, upper_32_bits(addr));
1615 	amdgpu_ring_write(ring, seq);
1616 	amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP);
1617 }
1618 
1619 void vcn_v2_0_enc_ring_insert_end(struct amdgpu_ring *ring)
1620 {
1621 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
1622 }
1623 
1624 /**
1625  * vcn_v2_0_enc_ring_emit_ib - enc execute indirect buffer
1626  *
1627  * @ring: amdgpu_ring pointer
1628  * @ib: indirect buffer to execute
1629  *
1630  * Write enc ring commands to execute the indirect buffer
1631  */
1632 void vcn_v2_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
1633 			       struct amdgpu_job *job,
1634 			       struct amdgpu_ib *ib,
1635 			       uint32_t flags)
1636 {
1637 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1638 
1639 	amdgpu_ring_write(ring, VCN_ENC_CMD_IB);
1640 	amdgpu_ring_write(ring, vmid);
1641 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1642 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1643 	amdgpu_ring_write(ring, ib->length_dw);
1644 }
1645 
1646 void vcn_v2_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1647 				uint32_t val, uint32_t mask)
1648 {
1649 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
1650 	amdgpu_ring_write(ring, reg << 2);
1651 	amdgpu_ring_write(ring, mask);
1652 	amdgpu_ring_write(ring, val);
1653 }
1654 
1655 void vcn_v2_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
1656 				unsigned int vmid, uint64_t pd_addr)
1657 {
1658 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1659 
1660 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1661 
1662 	/* wait for reg writes */
1663 	vcn_v2_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 +
1664 					vmid * hub->ctx_addr_distance,
1665 					lower_32_bits(pd_addr), 0xffffffff);
1666 }
1667 
1668 void vcn_v2_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1669 {
1670 	amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
1671 	amdgpu_ring_write(ring,	reg << 2);
1672 	amdgpu_ring_write(ring, val);
1673 }
1674 
1675 static int vcn_v2_0_set_interrupt_state(struct amdgpu_device *adev,
1676 					struct amdgpu_irq_src *source,
1677 					unsigned type,
1678 					enum amdgpu_interrupt_state state)
1679 {
1680 	return 0;
1681 }
1682 
1683 static int vcn_v2_0_process_interrupt(struct amdgpu_device *adev,
1684 				      struct amdgpu_irq_src *source,
1685 				      struct amdgpu_iv_entry *entry)
1686 {
1687 	DRM_DEBUG("IH: VCN TRAP\n");
1688 
1689 	switch (entry->src_id) {
1690 	case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1691 		amdgpu_fence_process(&adev->vcn.inst->ring_dec);
1692 		break;
1693 	case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1694 		amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]);
1695 		break;
1696 	case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1697 		amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]);
1698 		break;
1699 	default:
1700 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1701 			  entry->src_id, entry->src_data[0]);
1702 		break;
1703 	}
1704 
1705 	return 0;
1706 }
1707 
1708 int vcn_v2_0_dec_ring_test_ring(struct amdgpu_ring *ring)
1709 {
1710 	struct amdgpu_device *adev = ring->adev;
1711 	uint32_t tmp = 0;
1712 	unsigned i;
1713 	int r;
1714 
1715 	if (amdgpu_sriov_vf(adev))
1716 		return 0;
1717 
1718 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
1719 	r = amdgpu_ring_alloc(ring, 4);
1720 	if (r)
1721 		return r;
1722 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1723 	amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1724 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
1725 	amdgpu_ring_write(ring, 0xDEADBEEF);
1726 	amdgpu_ring_commit(ring);
1727 	for (i = 0; i < adev->usec_timeout; i++) {
1728 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
1729 		if (tmp == 0xDEADBEEF)
1730 			break;
1731 		udelay(1);
1732 	}
1733 
1734 	if (i >= adev->usec_timeout)
1735 		r = -ETIMEDOUT;
1736 
1737 	return r;
1738 }
1739 
1740 
1741 static int vcn_v2_0_set_powergating_state(void *handle,
1742 					  enum amd_powergating_state state)
1743 {
1744 	/* This doesn't actually powergate the VCN block.
1745 	 * That's done in the dpm code via the SMC.  This
1746 	 * just re-inits the block as necessary.  The actual
1747 	 * gating still happens in the dpm code.  We should
1748 	 * revisit this when there is a cleaner line between
1749 	 * the smc and the hw blocks
1750 	 */
1751 	int ret;
1752 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1753 
1754 	if (amdgpu_sriov_vf(adev)) {
1755 		adev->vcn.cur_state = AMD_PG_STATE_UNGATE;
1756 		return 0;
1757 	}
1758 
1759 	if (state == adev->vcn.cur_state)
1760 		return 0;
1761 
1762 	if (state == AMD_PG_STATE_GATE)
1763 		ret = vcn_v2_0_stop(adev);
1764 	else
1765 		ret = vcn_v2_0_start(adev);
1766 
1767 	if (!ret)
1768 		adev->vcn.cur_state = state;
1769 	return ret;
1770 }
1771 
1772 static int vcn_v2_0_start_mmsch(struct amdgpu_device *adev,
1773 				struct amdgpu_mm_table *table)
1774 {
1775 	uint32_t data = 0, loop;
1776 	uint64_t addr = table->gpu_addr;
1777 	struct mmsch_v2_0_init_header *header;
1778 	uint32_t size;
1779 	int i;
1780 
1781 	header = (struct mmsch_v2_0_init_header *)table->cpu_addr;
1782 	size = header->header_size + header->vcn_table_size;
1783 
1784 	/* 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr
1785 	 * of memory descriptor location
1786 	 */
1787 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1788 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1789 
1790 	/* 2, update vmid of descriptor */
1791 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID);
1792 	data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1793 	/* use domain0 for MM scheduler */
1794 	data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1795 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_VMID, data);
1796 
1797 	/* 3, notify mmsch about the size of this descriptor */
1798 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_CTX_SIZE, size);
1799 
1800 	/* 4, set resp to zero */
1801 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1802 
1803 	adev->vcn.inst->ring_dec.wptr = 0;
1804 	adev->vcn.inst->ring_dec.wptr_old = 0;
1805 	vcn_v2_0_dec_ring_set_wptr(&adev->vcn.inst->ring_dec);
1806 
1807 	for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1808 		adev->vcn.inst->ring_enc[i].wptr = 0;
1809 		adev->vcn.inst->ring_enc[i].wptr_old = 0;
1810 		vcn_v2_0_enc_ring_set_wptr(&adev->vcn.inst->ring_enc[i]);
1811 	}
1812 
1813 	/* 5, kick off the initialization and wait until
1814 	 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1815 	 */
1816 	WREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1817 
1818 	data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1819 	loop = 1000;
1820 	while ((data & 0x10000002) != 0x10000002) {
1821 		udelay(10);
1822 		data = RREG32_SOC15(UVD, 0, mmMMSCH_VF_MAILBOX_RESP);
1823 		loop--;
1824 		if (!loop)
1825 			break;
1826 	}
1827 
1828 	if (!loop) {
1829 		DRM_ERROR("failed to init MMSCH, " \
1830 			"mmMMSCH_VF_MAILBOX_RESP = 0x%08x\n", data);
1831 		return -EBUSY;
1832 	}
1833 
1834 	return 0;
1835 }
1836 
1837 static int vcn_v2_0_start_sriov(struct amdgpu_device *adev)
1838 {
1839 	int r;
1840 	uint32_t tmp;
1841 	struct amdgpu_ring *ring;
1842 	uint32_t offset, size;
1843 	uint32_t table_size = 0;
1844 	struct mmsch_v2_0_cmd_direct_write direct_wt = { {0} };
1845 	struct mmsch_v2_0_cmd_direct_read_modify_write direct_rd_mod_wt = { {0} };
1846 	struct mmsch_v2_0_cmd_end end = { {0} };
1847 	struct mmsch_v2_0_init_header *header;
1848 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1849 	uint8_t i = 0;
1850 
1851 	header = (struct mmsch_v2_0_init_header *)init_table;
1852 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1853 	direct_rd_mod_wt.cmd_header.command_type =
1854 		MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1855 	end.cmd_header.command_type = MMSCH_COMMAND__END;
1856 
1857 	if (header->vcn_table_offset == 0 && header->vcn_table_size == 0) {
1858 		header->version = MMSCH_VERSION;
1859 		header->header_size = sizeof(struct mmsch_v2_0_init_header) >> 2;
1860 
1861 		header->vcn_table_offset = header->header_size;
1862 
1863 		init_table += header->vcn_table_offset;
1864 
1865 		size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
1866 
1867 		MMSCH_V2_0_INSERT_DIRECT_RD_MOD_WT(
1868 			SOC15_REG_OFFSET(UVD, i, mmUVD_STATUS),
1869 			0xFFFFFFFF, 0x00000004);
1870 
1871 		/* mc resume*/
1872 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1873 			tmp = AMDGPU_UCODE_ID_VCN;
1874 			MMSCH_V2_0_INSERT_DIRECT_WT(
1875 				SOC15_REG_OFFSET(UVD, i,
1876 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1877 				adev->firmware.ucode[tmp].tmr_mc_addr_lo);
1878 			MMSCH_V2_0_INSERT_DIRECT_WT(
1879 				SOC15_REG_OFFSET(UVD, i,
1880 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1881 				adev->firmware.ucode[tmp].tmr_mc_addr_hi);
1882 			offset = 0;
1883 		} else {
1884 			MMSCH_V2_0_INSERT_DIRECT_WT(
1885 				SOC15_REG_OFFSET(UVD, i,
1886 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1887 				lower_32_bits(adev->vcn.inst->gpu_addr));
1888 			MMSCH_V2_0_INSERT_DIRECT_WT(
1889 				SOC15_REG_OFFSET(UVD, i,
1890 					mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1891 				upper_32_bits(adev->vcn.inst->gpu_addr));
1892 			offset = size;
1893 		}
1894 
1895 		MMSCH_V2_0_INSERT_DIRECT_WT(
1896 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET0),
1897 			0);
1898 		MMSCH_V2_0_INSERT_DIRECT_WT(
1899 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE0),
1900 			size);
1901 
1902 		MMSCH_V2_0_INSERT_DIRECT_WT(
1903 			SOC15_REG_OFFSET(UVD, i,
1904 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1905 			lower_32_bits(adev->vcn.inst->gpu_addr + offset));
1906 		MMSCH_V2_0_INSERT_DIRECT_WT(
1907 			SOC15_REG_OFFSET(UVD, i,
1908 				mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1909 			upper_32_bits(adev->vcn.inst->gpu_addr + offset));
1910 		MMSCH_V2_0_INSERT_DIRECT_WT(
1911 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_OFFSET1),
1912 			0);
1913 		MMSCH_V2_0_INSERT_DIRECT_WT(
1914 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE1),
1915 			AMDGPU_VCN_STACK_SIZE);
1916 
1917 		MMSCH_V2_0_INSERT_DIRECT_WT(
1918 			SOC15_REG_OFFSET(UVD, i,
1919 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1920 			lower_32_bits(adev->vcn.inst->gpu_addr + offset +
1921 				AMDGPU_VCN_STACK_SIZE));
1922 		MMSCH_V2_0_INSERT_DIRECT_WT(
1923 			SOC15_REG_OFFSET(UVD, i,
1924 				mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1925 			upper_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, mmUVD_VCPU_CACHE_OFFSET2),
1929 			0);
1930 		MMSCH_V2_0_INSERT_DIRECT_WT(
1931 			SOC15_REG_OFFSET(UVD, i, mmUVD_VCPU_CACHE_SIZE2),
1932 			AMDGPU_VCN_CONTEXT_SIZE);
1933 
1934 		for (r = 0; r < adev->vcn.num_enc_rings; ++r) {
1935 			ring = &adev->vcn.inst->ring_enc[r];
1936 			ring->wptr = 0;
1937 			MMSCH_V2_0_INSERT_DIRECT_WT(
1938 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_LO),
1939 				lower_32_bits(ring->gpu_addr));
1940 			MMSCH_V2_0_INSERT_DIRECT_WT(
1941 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_BASE_HI),
1942 				upper_32_bits(ring->gpu_addr));
1943 			MMSCH_V2_0_INSERT_DIRECT_WT(
1944 				SOC15_REG_OFFSET(UVD, i, mmUVD_RB_SIZE),
1945 				ring->ring_size / 4);
1946 		}
1947 
1948 		ring = &adev->vcn.inst->ring_dec;
1949 		ring->wptr = 0;
1950 		MMSCH_V2_0_INSERT_DIRECT_WT(
1951 			SOC15_REG_OFFSET(UVD, i,
1952 				mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1953 			lower_32_bits(ring->gpu_addr));
1954 		MMSCH_V2_0_INSERT_DIRECT_WT(
1955 			SOC15_REG_OFFSET(UVD, i,
1956 				mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1957 			upper_32_bits(ring->gpu_addr));
1958 		/* force RBC into idle state */
1959 		tmp = order_base_2(ring->ring_size);
1960 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, tmp);
1961 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1962 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1963 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1964 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1965 		MMSCH_V2_0_INSERT_DIRECT_WT(
1966 			SOC15_REG_OFFSET(UVD, i, mmUVD_RBC_RB_CNTL), tmp);
1967 
1968 		/* add end packet */
1969 		tmp = sizeof(struct mmsch_v2_0_cmd_end);
1970 		memcpy((void *)init_table, &end, tmp);
1971 		table_size += (tmp / 4);
1972 		header->vcn_table_size = table_size;
1973 
1974 	}
1975 	return vcn_v2_0_start_mmsch(adev, &adev->virt.mm_table);
1976 }
1977 
1978 static const struct amd_ip_funcs vcn_v2_0_ip_funcs = {
1979 	.name = "vcn_v2_0",
1980 	.early_init = vcn_v2_0_early_init,
1981 	.late_init = NULL,
1982 	.sw_init = vcn_v2_0_sw_init,
1983 	.sw_fini = vcn_v2_0_sw_fini,
1984 	.hw_init = vcn_v2_0_hw_init,
1985 	.hw_fini = vcn_v2_0_hw_fini,
1986 	.suspend = vcn_v2_0_suspend,
1987 	.resume = vcn_v2_0_resume,
1988 	.is_idle = vcn_v2_0_is_idle,
1989 	.wait_for_idle = vcn_v2_0_wait_for_idle,
1990 	.check_soft_reset = NULL,
1991 	.pre_soft_reset = NULL,
1992 	.soft_reset = NULL,
1993 	.post_soft_reset = NULL,
1994 	.set_clockgating_state = vcn_v2_0_set_clockgating_state,
1995 	.set_powergating_state = vcn_v2_0_set_powergating_state,
1996 };
1997 
1998 static const struct amdgpu_ring_funcs vcn_v2_0_dec_ring_vm_funcs = {
1999 	.type = AMDGPU_RING_TYPE_VCN_DEC,
2000 	.align_mask = 0xf,
2001 	.vmhub = AMDGPU_MMHUB_0,
2002 	.get_rptr = vcn_v2_0_dec_ring_get_rptr,
2003 	.get_wptr = vcn_v2_0_dec_ring_get_wptr,
2004 	.set_wptr = vcn_v2_0_dec_ring_set_wptr,
2005 	.emit_frame_size =
2006 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
2007 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
2008 		8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
2009 		14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
2010 		6,
2011 	.emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
2012 	.emit_ib = vcn_v2_0_dec_ring_emit_ib,
2013 	.emit_fence = vcn_v2_0_dec_ring_emit_fence,
2014 	.emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
2015 	.test_ring = vcn_v2_0_dec_ring_test_ring,
2016 	.test_ib = amdgpu_vcn_dec_ring_test_ib,
2017 	.insert_nop = vcn_v2_0_dec_ring_insert_nop,
2018 	.insert_start = vcn_v2_0_dec_ring_insert_start,
2019 	.insert_end = vcn_v2_0_dec_ring_insert_end,
2020 	.pad_ib = amdgpu_ring_generic_pad_ib,
2021 	.begin_use = amdgpu_vcn_ring_begin_use,
2022 	.end_use = amdgpu_vcn_ring_end_use,
2023 	.emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
2024 	.emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
2025 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2026 };
2027 
2028 static const struct amdgpu_ring_funcs vcn_v2_0_enc_ring_vm_funcs = {
2029 	.type = AMDGPU_RING_TYPE_VCN_ENC,
2030 	.align_mask = 0x3f,
2031 	.nop = VCN_ENC_CMD_NO_OP,
2032 	.vmhub = AMDGPU_MMHUB_0,
2033 	.get_rptr = vcn_v2_0_enc_ring_get_rptr,
2034 	.get_wptr = vcn_v2_0_enc_ring_get_wptr,
2035 	.set_wptr = vcn_v2_0_enc_ring_set_wptr,
2036 	.emit_frame_size =
2037 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2038 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
2039 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
2040 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
2041 		1, /* vcn_v2_0_enc_ring_insert_end */
2042 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
2043 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
2044 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
2045 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
2046 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
2047 	.test_ib = amdgpu_vcn_enc_ring_test_ib,
2048 	.insert_nop = amdgpu_ring_insert_nop,
2049 	.insert_end = vcn_v2_0_enc_ring_insert_end,
2050 	.pad_ib = amdgpu_ring_generic_pad_ib,
2051 	.begin_use = amdgpu_vcn_ring_begin_use,
2052 	.end_use = amdgpu_vcn_ring_end_use,
2053 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
2054 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
2055 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2056 };
2057 
2058 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev)
2059 {
2060 	adev->vcn.inst->ring_dec.funcs = &vcn_v2_0_dec_ring_vm_funcs;
2061 	DRM_INFO("VCN decode is enabled in VM mode\n");
2062 }
2063 
2064 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev)
2065 {
2066 	int i;
2067 
2068 	for (i = 0; i < adev->vcn.num_enc_rings; ++i)
2069 		adev->vcn.inst->ring_enc[i].funcs = &vcn_v2_0_enc_ring_vm_funcs;
2070 
2071 	DRM_INFO("VCN encode is enabled in VM mode\n");
2072 }
2073 
2074 static const struct amdgpu_irq_src_funcs vcn_v2_0_irq_funcs = {
2075 	.set = vcn_v2_0_set_interrupt_state,
2076 	.process = vcn_v2_0_process_interrupt,
2077 };
2078 
2079 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev)
2080 {
2081 	adev->vcn.inst->irq.num_types = adev->vcn.num_enc_rings + 1;
2082 	adev->vcn.inst->irq.funcs = &vcn_v2_0_irq_funcs;
2083 }
2084 
2085 const struct amdgpu_ip_block_version vcn_v2_0_ip_block =
2086 {
2087 		.type = AMD_IP_BLOCK_TYPE_VCN,
2088 		.major = 2,
2089 		.minor = 0,
2090 		.rev = 0,
2091 		.funcs = &vcn_v2_0_ip_funcs,
2092 };
2093