xref: /linux/drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
1 /*
2  * Copyright 2023 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 "amdgpu.h"
26 #include "amdgpu_vcn.h"
27 #include "amdgpu_pm.h"
28 #include "amdgpu_cs.h"
29 #include "soc15.h"
30 #include "soc15d.h"
31 #include "soc15_hw_ip.h"
32 #include "vcn_v2_0.h"
33 #include "mmsch_v4_0.h"
34 #include "vcn_v4_0_5.h"
35 
36 #include "vcn/vcn_4_0_5_offset.h"
37 #include "vcn/vcn_4_0_5_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
39 
40 #include <drm/drm_drv.h>
41 
42 #define mmUVD_DPG_LMA_CTL							regUVD_DPG_LMA_CTL
43 #define mmUVD_DPG_LMA_CTL_BASE_IDX					regUVD_DPG_LMA_CTL_BASE_IDX
44 #define mmUVD_DPG_LMA_DATA							regUVD_DPG_LMA_DATA
45 #define mmUVD_DPG_LMA_DATA_BASE_IDX					regUVD_DPG_LMA_DATA_BASE_IDX
46 
47 #define VCN_VID_SOC_ADDRESS_2_0						0x1fb00
48 #define VCN1_VID_SOC_ADDRESS_3_0					(0x48300 + 0x38000)
49 
50 #define VCN_HARVEST_MMSCH							0
51 
52 #define RDECODE_MSG_CREATE							0x00000000
53 #define RDECODE_MESSAGE_CREATE						0x00000001
54 
55 static int amdgpu_ih_clientid_vcns[] = {
56 	SOC15_IH_CLIENTID_VCN,
57 	SOC15_IH_CLIENTID_VCN1
58 };
59 
60 static void vcn_v4_0_5_set_unified_ring_funcs(struct amdgpu_device *adev);
61 static void vcn_v4_0_5_set_irq_funcs(struct amdgpu_device *adev);
62 static int vcn_v4_0_5_set_powergating_state(void *handle,
63 		enum amd_powergating_state state);
64 static int vcn_v4_0_5_pause_dpg_mode(struct amdgpu_device *adev,
65 		int inst_idx, struct dpg_pause_state *new_state);
66 static void vcn_v4_0_5_unified_ring_set_wptr(struct amdgpu_ring *ring);
67 
68 /**
69  * vcn_v4_0_5_early_init - set function pointers and load microcode
70  *
71  * @handle: amdgpu_device pointer
72  *
73  * Set ring and irq function pointers
74  * Load microcode from filesystem
75  */
76 static int vcn_v4_0_5_early_init(void *handle)
77 {
78 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
79 
80 	/* re-use enc ring as unified ring */
81 	adev->vcn.num_enc_rings = 1;
82 	vcn_v4_0_5_set_unified_ring_funcs(adev);
83 	vcn_v4_0_5_set_irq_funcs(adev);
84 
85 	return amdgpu_vcn_early_init(adev);
86 }
87 
88 /**
89  * vcn_v4_0_5_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_v4_0_5_sw_init(void *handle)
96 {
97 	struct amdgpu_ring *ring;
98 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
99 	int i, r;
100 
101 	r = amdgpu_vcn_sw_init(adev);
102 	if (r)
103 		return r;
104 
105 	amdgpu_vcn_setup_ucode(adev);
106 
107 	r = amdgpu_vcn_resume(adev);
108 	if (r)
109 		return r;
110 
111 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
112 		volatile struct amdgpu_vcn4_fw_shared *fw_shared;
113 
114 		if (adev->vcn.harvest_config & (1 << i))
115 			continue;
116 
117 		atomic_set(&adev->vcn.inst[i].sched_score, 0);
118 
119 		/* VCN UNIFIED TRAP */
120 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
121 				VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[i].irq);
122 		if (r)
123 			return r;
124 
125 		/* VCN POISON TRAP */
126 		r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[i],
127 				VCN_4_0__SRCID_UVD_POISON, &adev->vcn.inst[i].irq);
128 		if (r)
129 			return r;
130 
131 		ring = &adev->vcn.inst[i].ring_enc[0];
132 		ring->use_doorbell = true;
133 		if (amdgpu_sriov_vf(adev))
134 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
135 						i * (adev->vcn.num_enc_rings + 1) + 1;
136 		else
137 			ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
138 						2 + 8 * i;
139 		ring->vm_hub = AMDGPU_MMHUB0(0);
140 		sprintf(ring->name, "vcn_unified_%d", i);
141 
142 		r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[i].irq, 0,
143 				AMDGPU_RING_PRIO_0, &adev->vcn.inst[i].sched_score);
144 		if (r)
145 			return r;
146 
147 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
148 		fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_FW_SHARED_FLAG_0_UNIFIED_QUEUE);
149 		fw_shared->sq.is_enabled = 1;
150 
151 		fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_SMU_DPM_INTERFACE_FLAG);
152 		fw_shared->smu_dpm_interface.smu_interface_type = (adev->flags & AMD_IS_APU) ?
153 			AMDGPU_VCN_SMU_DPM_INTERFACE_APU : AMDGPU_VCN_SMU_DPM_INTERFACE_DGPU;
154 
155 		if (amdgpu_sriov_vf(adev))
156 			fw_shared->present_flag_0 |= cpu_to_le32(AMDGPU_VCN_VF_RB_SETUP_FLAG);
157 
158 		if (amdgpu_vcnfw_log)
159 			amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
160 	}
161 
162 	if (amdgpu_sriov_vf(adev)) {
163 		r = amdgpu_virt_alloc_mm_table(adev);
164 		if (r)
165 			return r;
166 	}
167 
168 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
169 		adev->vcn.pause_dpg_mode = vcn_v4_0_5_pause_dpg_mode;
170 
171 	return 0;
172 }
173 
174 /**
175  * vcn_v4_0_5_sw_fini - sw fini for VCN block
176  *
177  * @handle: amdgpu_device pointer
178  *
179  * VCN suspend and free up sw allocation
180  */
181 static int vcn_v4_0_5_sw_fini(void *handle)
182 {
183 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
184 	int i, r, idx;
185 
186 	if (drm_dev_enter(adev_to_drm(adev), &idx)) {
187 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
188 			volatile struct amdgpu_vcn4_fw_shared *fw_shared;
189 
190 			if (adev->vcn.harvest_config & (1 << i))
191 				continue;
192 
193 			fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
194 			fw_shared->present_flag_0 = 0;
195 			fw_shared->sq.is_enabled = 0;
196 		}
197 
198 		drm_dev_exit(idx);
199 	}
200 
201 	if (amdgpu_sriov_vf(adev))
202 		amdgpu_virt_free_mm_table(adev);
203 
204 	r = amdgpu_vcn_suspend(adev);
205 	if (r)
206 		return r;
207 
208 	r = amdgpu_vcn_sw_fini(adev);
209 
210 	return r;
211 }
212 
213 /**
214  * vcn_v4_0_5_hw_init - start and test VCN block
215  *
216  * @handle: amdgpu_device pointer
217  *
218  * Initialize the hardware, boot up the VCPU and do some testing
219  */
220 static int vcn_v4_0_5_hw_init(void *handle)
221 {
222 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
223 	struct amdgpu_ring *ring;
224 	int i, r;
225 
226 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
227 		if (adev->vcn.harvest_config & (1 << i))
228 			continue;
229 
230 		ring = &adev->vcn.inst[i].ring_enc[0];
231 
232 		adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
233 				((adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 8 * i), i);
234 
235 		r = amdgpu_ring_test_helper(ring);
236 		if (r)
237 			return r;
238 	}
239 
240 	return 0;
241 }
242 
243 /**
244  * vcn_v4_0_5_hw_fini - stop the hardware block
245  *
246  * @handle: amdgpu_device pointer
247  *
248  * Stop the VCN block, mark ring as not ready any more
249  */
250 static int vcn_v4_0_5_hw_fini(void *handle)
251 {
252 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
253 	int i;
254 
255 	cancel_delayed_work_sync(&adev->vcn.idle_work);
256 
257 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
258 		if (adev->vcn.harvest_config & (1 << i))
259 			continue;
260 		if (!amdgpu_sriov_vf(adev)) {
261 			if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
262 				(adev->vcn.cur_state != AMD_PG_STATE_GATE &&
263 				RREG32_SOC15(VCN, i, regUVD_STATUS))) {
264 				vcn_v4_0_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
265 			}
266 		}
267 	}
268 
269 	return 0;
270 }
271 
272 /**
273  * vcn_v4_0_5_suspend - suspend VCN block
274  *
275  * @handle: amdgpu_device pointer
276  *
277  * HW fini and suspend VCN block
278  */
279 static int vcn_v4_0_5_suspend(void *handle)
280 {
281 	int r;
282 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
283 
284 	r = vcn_v4_0_5_hw_fini(adev);
285 	if (r)
286 		return r;
287 
288 	r = amdgpu_vcn_suspend(adev);
289 
290 	return r;
291 }
292 
293 /**
294  * vcn_v4_0_5_resume - resume VCN block
295  *
296  * @handle: amdgpu_device pointer
297  *
298  * Resume firmware and hw init VCN block
299  */
300 static int vcn_v4_0_5_resume(void *handle)
301 {
302 	int r;
303 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
304 
305 	r = amdgpu_vcn_resume(adev);
306 	if (r)
307 		return r;
308 
309 	r = vcn_v4_0_5_hw_init(adev);
310 
311 	return r;
312 }
313 
314 /**
315  * vcn_v4_0_5_mc_resume - memory controller programming
316  *
317  * @adev: amdgpu_device pointer
318  * @inst: instance number
319  *
320  * Let the VCN memory controller know it's offsets
321  */
322 static void vcn_v4_0_5_mc_resume(struct amdgpu_device *adev, int inst)
323 {
324 	uint32_t offset, size;
325 	const struct common_firmware_header *hdr;
326 
327 	hdr = (const struct common_firmware_header *)adev->vcn.fw[inst]->data;
328 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
329 
330 	/* cache window 0: fw */
331 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
332 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
333 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_lo));
334 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
335 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst].tmr_mc_addr_hi));
336 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, 0);
337 		offset = 0;
338 	} else {
339 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
340 			lower_32_bits(adev->vcn.inst[inst].gpu_addr));
341 		WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
342 			upper_32_bits(adev->vcn.inst[inst].gpu_addr));
343 		offset = size;
344 		WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET0, AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
345 	}
346 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE0, size);
347 
348 	/* cache window 1: stack */
349 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
350 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
351 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
352 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset));
353 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET1, 0);
354 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
355 
356 	/* cache window 2: context */
357 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
358 		lower_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
359 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
360 		upper_32_bits(adev->vcn.inst[inst].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
361 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_OFFSET2, 0);
362 	WREG32_SOC15(VCN, inst, regUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
363 
364 	/* non-cache window */
365 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
366 		lower_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
367 	WREG32_SOC15(VCN, inst, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
368 		upper_32_bits(adev->vcn.inst[inst].fw_shared.gpu_addr));
369 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_OFFSET0, 0);
370 	WREG32_SOC15(VCN, inst, regUVD_VCPU_NONCACHE_SIZE0,
371 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)));
372 }
373 
374 /**
375  * vcn_v4_0_5_mc_resume_dpg_mode - memory controller programming for dpg mode
376  *
377  * @adev: amdgpu_device pointer
378  * @inst_idx: instance number index
379  * @indirect: indirectly write sram
380  *
381  * Let the VCN memory controller know it's offsets with dpg mode
382  */
383 static void vcn_v4_0_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
384 {
385 	uint32_t offset, size;
386 	const struct common_firmware_header *hdr;
387 
388 	hdr = (const struct common_firmware_header *)adev->vcn.fw[inst_idx]->data;
389 	size = AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
390 
391 	/* cache window 0: fw */
392 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
393 		if (!indirect) {
394 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
395 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
396 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo),
397 			0, indirect);
398 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
399 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
400 			(adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi),
401 			0, indirect);
402 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
403 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
404 		} else {
405 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
406 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
407 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
408 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
409 			WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
410 				VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
411 		}
412 		offset = 0;
413 	} else {
414 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
415 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
416 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
417 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
418 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
419 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
420 		offset = size;
421 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
422 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET0),
423 			AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
424 	}
425 
426 	if (!indirect)
427 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
428 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
429 	else
430 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
431 			VCN, inst_idx, regUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
432 
433 	/* cache window 1: stack */
434 	if (!indirect) {
435 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
436 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
437 			lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
438 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
439 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
440 			upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
441 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
442 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
443 	} else {
444 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
445 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
446 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
447 			VCN, inst_idx, regUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
448 		WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
449 			VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
450 	}
451 
452 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
453 		VCN, inst_idx, regUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
454 
455 	/* cache window 2: context */
456 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
457 		VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
458 		lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE),
459 		0, indirect);
460 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
461 		VCN, inst_idx, regUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
462 		upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE),
463 		0, indirect);
464 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
465 		VCN, inst_idx, regUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
466 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
467 		VCN, inst_idx, regUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
468 
469 	/* non-cache window */
470 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
471 		VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
472 		lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
473 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
474 		VCN, inst_idx, regUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
475 		upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
476 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
477 		VCN, inst_idx, regUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
478 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
479 		VCN, inst_idx, regUVD_VCPU_NONCACHE_SIZE0),
480 		AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_vcn4_fw_shared)), 0, indirect);
481 
482 	/* VCN global tiling registers */
483 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
484 		VCN, inst_idx, regUVD_GFX10_ADDR_CONFIG),
485 		adev->gfx.config.gb_addr_config, 0, indirect);
486 }
487 
488 /**
489  * vcn_v4_0_5_disable_static_power_gating - disable VCN static power gating
490  *
491  * @adev: amdgpu_device pointer
492  * @inst: instance number
493  *
494  * Disable static power gating for VCN block
495  */
496 static void vcn_v4_0_5_disable_static_power_gating(struct amdgpu_device *adev, int inst)
497 {
498 	uint32_t data = 0;
499 
500 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
501 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
502 					1 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT);
503 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS, 0,
504 					UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK);
505 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
506 					2 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT);
507 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
508 					1 << UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS__SHIFT,
509 					UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK);
510 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
511 					2 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT);
512 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
513 					1 << UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS__SHIFT,
514 					UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK);
515 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
516 					2 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT);
517 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
518 					1 << UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS__SHIFT,
519 					UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK);
520 	} else {
521 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
522 			1 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT);
523 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
524 			0, UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK);
525 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
526 			1 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT);
527 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
528 			0, UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK);
529 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
530 			1 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT);
531 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
532 			0, UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK);
533 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
534 			1 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT);
535 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
536 			0, UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK);
537 	}
538 
539 	data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
540 	data &= ~0x103;
541 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
542 		data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
543 			UVD_POWER_STATUS__UVD_PG_EN_MASK;
544 	WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
545 }
546 
547 /**
548  * vcn_v4_0_5_enable_static_power_gating - enable VCN static power gating
549  *
550  * @adev: amdgpu_device pointer
551  * @inst: instance number
552  *
553  * Enable static power gating for VCN block
554  */
555 static void vcn_v4_0_5_enable_static_power_gating(struct amdgpu_device *adev, int inst)
556 {
557 	uint32_t data;
558 
559 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
560 		/* Before power off, this indicator has to be turned on */
561 		data = RREG32_SOC15(VCN, inst, regUVD_POWER_STATUS);
562 		data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
563 		data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
564 		WREG32_SOC15(VCN, inst, regUVD_POWER_STATUS, data);
565 
566 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
567 			2 << UVD_IPX_DLDO_CONFIG__ONO5_PWR_CONFIG__SHIFT);
568 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
569 			1 << UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS__SHIFT,
570 			UVD_IPX_DLDO_STATUS__ONO5_PWR_STATUS_MASK);
571 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
572 			2 << UVD_IPX_DLDO_CONFIG__ONO4_PWR_CONFIG__SHIFT);
573 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
574 			1 << UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS__SHIFT,
575 			UVD_IPX_DLDO_STATUS__ONO4_PWR_STATUS_MASK);
576 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
577 			2 << UVD_IPX_DLDO_CONFIG__ONO3_PWR_CONFIG__SHIFT);
578 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
579 			1 << UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS__SHIFT,
580 			UVD_IPX_DLDO_STATUS__ONO3_PWR_STATUS_MASK);
581 		WREG32_SOC15(VCN, inst, regUVD_IPX_DLDO_CONFIG,
582 			2 << UVD_IPX_DLDO_CONFIG__ONO2_PWR_CONFIG__SHIFT);
583 		SOC15_WAIT_ON_RREG(VCN, inst, regUVD_IPX_DLDO_STATUS,
584 			1 << UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS__SHIFT,
585 			UVD_IPX_DLDO_STATUS__ONO2_PWR_STATUS_MASK);
586 	}
587 }
588 
589 /**
590  * vcn_v4_0_5_disable_clock_gating - disable VCN clock gating
591  *
592  * @adev: amdgpu_device pointer
593  * @inst: instance number
594  *
595  * Disable clock gating for VCN block
596  */
597 static void vcn_v4_0_5_disable_clock_gating(struct amdgpu_device *adev, int inst)
598 {
599 	uint32_t data;
600 
601 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
602 		return;
603 
604 	/* VCN disable CGC */
605 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
606 	data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
607 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
608 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
609 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
610 
611 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_GATE);
612 	data &= ~(UVD_CGC_GATE__SYS_MASK
613 		| UVD_CGC_GATE__UDEC_MASK
614 		| UVD_CGC_GATE__MPEG2_MASK
615 		| UVD_CGC_GATE__REGS_MASK
616 		| UVD_CGC_GATE__RBC_MASK
617 		| UVD_CGC_GATE__LMI_MC_MASK
618 		| UVD_CGC_GATE__LMI_UMC_MASK
619 		| UVD_CGC_GATE__IDCT_MASK
620 		| UVD_CGC_GATE__MPRD_MASK
621 		| UVD_CGC_GATE__MPC_MASK
622 		| UVD_CGC_GATE__LBSI_MASK
623 		| UVD_CGC_GATE__LRBBM_MASK
624 		| UVD_CGC_GATE__UDEC_RE_MASK
625 		| UVD_CGC_GATE__UDEC_CM_MASK
626 		| UVD_CGC_GATE__UDEC_IT_MASK
627 		| UVD_CGC_GATE__UDEC_DB_MASK
628 		| UVD_CGC_GATE__UDEC_MP_MASK
629 		| UVD_CGC_GATE__WCB_MASK
630 		| UVD_CGC_GATE__VCPU_MASK
631 		| UVD_CGC_GATE__MMSCH_MASK);
632 
633 	WREG32_SOC15(VCN, inst, regUVD_CGC_GATE, data);
634 	SOC15_WAIT_ON_RREG(VCN, inst, regUVD_CGC_GATE, 0,  0xFFFFFFFF);
635 
636 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
637 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
638 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
639 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
640 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
641 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
642 		| UVD_CGC_CTRL__SYS_MODE_MASK
643 		| UVD_CGC_CTRL__UDEC_MODE_MASK
644 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
645 		| UVD_CGC_CTRL__REGS_MODE_MASK
646 		| UVD_CGC_CTRL__RBC_MODE_MASK
647 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
648 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
649 		| UVD_CGC_CTRL__IDCT_MODE_MASK
650 		| UVD_CGC_CTRL__MPRD_MODE_MASK
651 		| UVD_CGC_CTRL__MPC_MODE_MASK
652 		| UVD_CGC_CTRL__LBSI_MODE_MASK
653 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
654 		| UVD_CGC_CTRL__WCB_MODE_MASK
655 		| UVD_CGC_CTRL__VCPU_MODE_MASK
656 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
657 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
658 
659 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE);
660 	data |= (UVD_SUVD_CGC_GATE__SRE_MASK
661 		| UVD_SUVD_CGC_GATE__SIT_MASK
662 		| UVD_SUVD_CGC_GATE__SMP_MASK
663 		| UVD_SUVD_CGC_GATE__SCM_MASK
664 		| UVD_SUVD_CGC_GATE__SDB_MASK
665 		| UVD_SUVD_CGC_GATE__SRE_H264_MASK
666 		| UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
667 		| UVD_SUVD_CGC_GATE__SIT_H264_MASK
668 		| UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
669 		| UVD_SUVD_CGC_GATE__SCM_H264_MASK
670 		| UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
671 		| UVD_SUVD_CGC_GATE__SDB_H264_MASK
672 		| UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
673 		| UVD_SUVD_CGC_GATE__SCLR_MASK
674 		| UVD_SUVD_CGC_GATE__UVD_SC_MASK
675 		| UVD_SUVD_CGC_GATE__ENT_MASK
676 		| UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
677 		| UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
678 		| UVD_SUVD_CGC_GATE__SITE_MASK
679 		| UVD_SUVD_CGC_GATE__SRE_VP9_MASK
680 		| UVD_SUVD_CGC_GATE__SCM_VP9_MASK
681 		| UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
682 		| UVD_SUVD_CGC_GATE__SDB_VP9_MASK
683 		| UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
684 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_GATE, data);
685 
686 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
687 	data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
688 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
689 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
690 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
691 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
692 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
693 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
694 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
695 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
696 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
697 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
698 }
699 
700 /**
701  * vcn_v4_0_5_disable_clock_gating_dpg_mode - disable VCN clock gating dpg mode
702  *
703  * @adev: amdgpu_device pointer
704  * @sram_sel: sram select
705  * @inst_idx: instance number index
706  * @indirect: indirectly write sram
707  *
708  * Disable clock gating for VCN block with dpg mode
709  */
710 static void vcn_v4_0_5_disable_clock_gating_dpg_mode(struct amdgpu_device *adev, uint8_t sram_sel,
711 		int inst_idx, uint8_t indirect)
712 {
713 	uint32_t reg_data = 0;
714 
715 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
716 		return;
717 
718 	/* enable sw clock gating control */
719 	reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
720 	reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
721 	reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
722 	reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
723 		 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
724 		 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
725 		 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
726 		 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
727 		 UVD_CGC_CTRL__SYS_MODE_MASK |
728 		 UVD_CGC_CTRL__UDEC_MODE_MASK |
729 		 UVD_CGC_CTRL__MPEG2_MODE_MASK |
730 		 UVD_CGC_CTRL__REGS_MODE_MASK |
731 		 UVD_CGC_CTRL__RBC_MODE_MASK |
732 		 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
733 		 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
734 		 UVD_CGC_CTRL__IDCT_MODE_MASK |
735 		 UVD_CGC_CTRL__MPRD_MODE_MASK |
736 		 UVD_CGC_CTRL__MPC_MODE_MASK |
737 		 UVD_CGC_CTRL__LBSI_MODE_MASK |
738 		 UVD_CGC_CTRL__LRBBM_MODE_MASK |
739 		 UVD_CGC_CTRL__WCB_MODE_MASK |
740 		 UVD_CGC_CTRL__VCPU_MODE_MASK);
741 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
742 		VCN, inst_idx, regUVD_CGC_CTRL), reg_data, sram_sel, indirect);
743 
744 	/* turn off clock gating */
745 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
746 		VCN, inst_idx, regUVD_CGC_GATE), 0, sram_sel, indirect);
747 
748 	/* turn on SUVD clock gating */
749 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
750 		VCN, inst_idx, regUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
751 
752 	/* turn on sw mode in UVD_SUVD_CGC_CTRL */
753 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
754 		VCN, inst_idx, regUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
755 }
756 
757 /**
758  * vcn_v4_0_5_enable_clock_gating - enable VCN clock gating
759  *
760  * @adev: amdgpu_device pointer
761  * @inst: instance number
762  *
763  * Enable clock gating for VCN block
764  */
765 static void vcn_v4_0_5_enable_clock_gating(struct amdgpu_device *adev, int inst)
766 {
767 	uint32_t data;
768 
769 	if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
770 		return;
771 
772 	/* enable VCN CGC */
773 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
774 	data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
775 	data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
776 	data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
777 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
778 
779 	data = RREG32_SOC15(VCN, inst, regUVD_CGC_CTRL);
780 	data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
781 		| UVD_CGC_CTRL__UDEC_CM_MODE_MASK
782 		| UVD_CGC_CTRL__UDEC_IT_MODE_MASK
783 		| UVD_CGC_CTRL__UDEC_DB_MODE_MASK
784 		| UVD_CGC_CTRL__UDEC_MP_MODE_MASK
785 		| UVD_CGC_CTRL__SYS_MODE_MASK
786 		| UVD_CGC_CTRL__UDEC_MODE_MASK
787 		| UVD_CGC_CTRL__MPEG2_MODE_MASK
788 		| UVD_CGC_CTRL__REGS_MODE_MASK
789 		| UVD_CGC_CTRL__RBC_MODE_MASK
790 		| UVD_CGC_CTRL__LMI_MC_MODE_MASK
791 		| UVD_CGC_CTRL__LMI_UMC_MODE_MASK
792 		| UVD_CGC_CTRL__IDCT_MODE_MASK
793 		| UVD_CGC_CTRL__MPRD_MODE_MASK
794 		| UVD_CGC_CTRL__MPC_MODE_MASK
795 		| UVD_CGC_CTRL__LBSI_MODE_MASK
796 		| UVD_CGC_CTRL__LRBBM_MODE_MASK
797 		| UVD_CGC_CTRL__WCB_MODE_MASK
798 		| UVD_CGC_CTRL__VCPU_MODE_MASK
799 		| UVD_CGC_CTRL__MMSCH_MODE_MASK);
800 	WREG32_SOC15(VCN, inst, regUVD_CGC_CTRL, data);
801 
802 	data = RREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL);
803 	data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
804 		| UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
805 		| UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
806 		| UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
807 		| UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
808 		| UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
809 		| UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
810 		| UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
811 		| UVD_SUVD_CGC_CTRL__IME_MODE_MASK
812 		| UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
813 	WREG32_SOC15(VCN, inst, regUVD_SUVD_CGC_CTRL, data);
814 }
815 
816 /**
817  * vcn_v4_0_5_start_dpg_mode - VCN start with dpg mode
818  *
819  * @adev: amdgpu_device pointer
820  * @inst_idx: instance number index
821  * @indirect: indirectly write sram
822  *
823  * Start VCN block with dpg mode
824  */
825 static int vcn_v4_0_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
826 {
827 	volatile struct amdgpu_vcn4_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
828 	struct amdgpu_ring *ring;
829 	uint32_t tmp;
830 
831 	/* disable register anti-hang mechanism */
832 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 1,
833 		~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
834 	/* enable dynamic power gating mode */
835 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS);
836 	tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
837 	tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
838 	WREG32_SOC15(VCN, inst_idx, regUVD_POWER_STATUS, tmp);
839 
840 	if (indirect)
841 		adev->vcn.inst[inst_idx].dpg_sram_curr_addr =
842 					(uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
843 
844 	/* enable clock gating */
845 	vcn_v4_0_5_disable_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
846 
847 	/* enable VCPU clock */
848 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
849 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK | UVD_VCPU_CNTL__BLK_RST_MASK;
850 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
851 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
852 
853 	/* disable master interrupt */
854 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
855 		VCN, inst_idx, regUVD_MASTINT_EN), 0, 0, indirect);
856 
857 	/* setup regUVD_LMI_CTRL */
858 	tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
859 		UVD_LMI_CTRL__REQ_MODE_MASK |
860 		UVD_LMI_CTRL__CRC_RESET_MASK |
861 		UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
862 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
863 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
864 		(8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
865 		0x00100000L);
866 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
867 		VCN, inst_idx, regUVD_LMI_CTRL), tmp, 0, indirect);
868 
869 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
870 		VCN, inst_idx, regUVD_MPC_CNTL),
871 		0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
872 
873 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
874 		VCN, inst_idx, regUVD_MPC_SET_MUXA0),
875 		((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
876 		 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
877 		 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
878 		 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
879 
880 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
881 		VCN, inst_idx, regUVD_MPC_SET_MUXB0),
882 		 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
883 		 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
884 		 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
885 		 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
886 
887 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
888 		VCN, inst_idx, regUVD_MPC_SET_MUX),
889 		((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
890 		 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
891 		 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
892 
893 	vcn_v4_0_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
894 
895 	tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
896 	tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
897 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
898 		VCN, inst_idx, regUVD_VCPU_CNTL), tmp, 0, indirect);
899 
900 	/* enable LMI MC and UMC channels */
901 	tmp = 0x1f << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT;
902 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
903 		VCN, inst_idx, regUVD_LMI_CTRL2), tmp, 0, indirect);
904 
905 	/* enable master interrupt */
906 	WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
907 		VCN, inst_idx, regUVD_MASTINT_EN),
908 		UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
909 
910 	if (indirect)
911 		amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
912 
913 	ring = &adev->vcn.inst[inst_idx].ring_enc[0];
914 
915 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_LO, ring->gpu_addr);
916 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
917 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_SIZE, ring->ring_size / 4);
918 
919 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
920 	tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
921 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
922 	fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
923 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR, 0);
924 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, 0);
925 
926 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_RPTR);
927 	WREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR, tmp);
928 	ring->wptr = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
929 
930 	tmp = RREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE);
931 	tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
932 	WREG32_SOC15(VCN, inst_idx, regVCN_RB_ENABLE, tmp);
933 	fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
934 
935 	WREG32_SOC15(VCN, inst_idx, regVCN_RB1_DB_CTRL,
936 			ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
937 			VCN_RB1_DB_CTRL__EN_MASK);
938 
939 	return 0;
940 }
941 
942 
943 /**
944  * vcn_v4_0_5_start - VCN start
945  *
946  * @adev: amdgpu_device pointer
947  *
948  * Start VCN block
949  */
950 static int vcn_v4_0_5_start(struct amdgpu_device *adev)
951 {
952 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
953 	struct amdgpu_ring *ring;
954 	uint32_t tmp;
955 	int i, j, k, r;
956 
957 	if (adev->pm.dpm_enabled)
958 		amdgpu_dpm_enable_uvd(adev, true);
959 
960 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
961 		if (adev->vcn.harvest_config & (1 << i))
962 			continue;
963 
964 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
965 
966 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
967 			r = vcn_v4_0_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
968 			continue;
969 		}
970 
971 		/* disable VCN power gating */
972 		vcn_v4_0_5_disable_static_power_gating(adev, i);
973 
974 		/* set VCN status busy */
975 		tmp = RREG32_SOC15(VCN, i, regUVD_STATUS) | UVD_STATUS__UVD_BUSY;
976 		WREG32_SOC15(VCN, i, regUVD_STATUS, tmp);
977 
978 		/*SW clock gating */
979 		vcn_v4_0_5_disable_clock_gating(adev, i);
980 
981 		/* enable VCPU clock */
982 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
983 				UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
984 
985 		/* disable master interrupt */
986 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN), 0,
987 				~UVD_MASTINT_EN__VCPU_EN_MASK);
988 
989 		/* enable LMI MC and UMC channels */
990 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_LMI_CTRL2), 0,
991 				~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
992 
993 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
994 		tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
995 		tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
996 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
997 
998 		/* setup regUVD_LMI_CTRL */
999 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL);
1000 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL, tmp |
1001 				UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1002 				UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1003 				UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1004 				UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1005 
1006 		/* setup regUVD_MPC_CNTL */
1007 		tmp = RREG32_SOC15(VCN, i, regUVD_MPC_CNTL);
1008 		tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1009 		tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1010 		WREG32_SOC15(VCN, i, regUVD_MPC_CNTL, tmp);
1011 
1012 		/* setup UVD_MPC_SET_MUXA0 */
1013 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXA0,
1014 				((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1015 				 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1016 				 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1017 				 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1018 
1019 		/* setup UVD_MPC_SET_MUXB0 */
1020 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUXB0,
1021 				((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1022 				 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1023 				 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1024 				 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1025 
1026 		/* setup UVD_MPC_SET_MUX */
1027 		WREG32_SOC15(VCN, i, regUVD_MPC_SET_MUX,
1028 				((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1029 				 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1030 				 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1031 
1032 		vcn_v4_0_5_mc_resume(adev, i);
1033 
1034 		/* VCN global tiling registers */
1035 		WREG32_SOC15(VCN, i, regUVD_GFX10_ADDR_CONFIG,
1036 				adev->gfx.config.gb_addr_config);
1037 
1038 		/* unblock VCPU register access */
1039 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL), 0,
1040 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1041 
1042 		/* release VCPU reset to boot */
1043 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1044 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1045 
1046 		for (j = 0; j < 10; ++j) {
1047 			uint32_t status;
1048 
1049 			for (k = 0; k < 100; ++k) {
1050 				status = RREG32_SOC15(VCN, i, regUVD_STATUS);
1051 				if (status & 2)
1052 					break;
1053 				mdelay(10);
1054 				if (amdgpu_emu_mode == 1)
1055 					msleep(1);
1056 			}
1057 
1058 			if (amdgpu_emu_mode == 1) {
1059 				r = -1;
1060 				if (status & 2) {
1061 					r = 0;
1062 					break;
1063 				}
1064 			} else {
1065 				r = 0;
1066 				if (status & 2)
1067 					break;
1068 
1069 				dev_err(adev->dev,
1070 					"VCN[%d] is not responding, trying to reset VCPU!!!\n", i);
1071 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1072 							UVD_VCPU_CNTL__BLK_RST_MASK,
1073 							~UVD_VCPU_CNTL__BLK_RST_MASK);
1074 				mdelay(10);
1075 				WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1076 						~UVD_VCPU_CNTL__BLK_RST_MASK);
1077 
1078 				mdelay(10);
1079 				r = -1;
1080 			}
1081 		}
1082 
1083 		if (r) {
1084 			dev_err(adev->dev, "VCN[%d] is not responding, giving up!!!\n", i);
1085 			return r;
1086 		}
1087 
1088 		/* enable master interrupt */
1089 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_MASTINT_EN),
1090 				UVD_MASTINT_EN__VCPU_EN_MASK,
1091 				~UVD_MASTINT_EN__VCPU_EN_MASK);
1092 
1093 		/* clear the busy bit of VCN_STATUS */
1094 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_STATUS), 0,
1095 				~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1096 
1097 		ring = &adev->vcn.inst[i].ring_enc[0];
1098 		WREG32_SOC15(VCN, i, regVCN_RB1_DB_CTRL,
1099 				ring->doorbell_index << VCN_RB1_DB_CTRL__OFFSET__SHIFT |
1100 				VCN_RB1_DB_CTRL__EN_MASK);
1101 
1102 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_LO, ring->gpu_addr);
1103 		WREG32_SOC15(VCN, i, regUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1104 		WREG32_SOC15(VCN, i, regUVD_RB_SIZE, ring->ring_size / 4);
1105 
1106 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1107 		tmp &= ~(VCN_RB_ENABLE__RB1_EN_MASK);
1108 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1109 		fw_shared->sq.queue_mode |= FW_QUEUE_RING_RESET;
1110 		WREG32_SOC15(VCN, i, regUVD_RB_RPTR, 0);
1111 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, 0);
1112 
1113 		tmp = RREG32_SOC15(VCN, i, regUVD_RB_RPTR);
1114 		WREG32_SOC15(VCN, i, regUVD_RB_WPTR, tmp);
1115 		ring->wptr = RREG32_SOC15(VCN, i, regUVD_RB_WPTR);
1116 
1117 		tmp = RREG32_SOC15(VCN, i, regVCN_RB_ENABLE);
1118 		tmp |= VCN_RB_ENABLE__RB1_EN_MASK;
1119 		WREG32_SOC15(VCN, i, regVCN_RB_ENABLE, tmp);
1120 		fw_shared->sq.queue_mode &= ~(FW_QUEUE_RING_RESET | FW_QUEUE_DPG_HOLD_OFF);
1121 	}
1122 
1123 	return 0;
1124 }
1125 
1126 /**
1127  * vcn_v4_0_5_stop_dpg_mode - VCN stop with dpg mode
1128  *
1129  * @adev: amdgpu_device pointer
1130  * @inst_idx: instance number index
1131  *
1132  * Stop VCN block with dpg mode
1133  */
1134 static void vcn_v4_0_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1135 {
1136 	uint32_t tmp;
1137 
1138 	/* Wait for power status to be 1 */
1139 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1140 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1141 
1142 	/* wait for read ptr to be equal to write ptr */
1143 	tmp = RREG32_SOC15(VCN, inst_idx, regUVD_RB_WPTR);
1144 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1145 
1146 	SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 1,
1147 		UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1148 
1149 	/* disable dynamic power gating mode */
1150 	WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, regUVD_POWER_STATUS), 0,
1151 		~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1152 }
1153 
1154 /**
1155  * vcn_v4_0_5_stop - VCN stop
1156  *
1157  * @adev: amdgpu_device pointer
1158  *
1159  * Stop VCN block
1160  */
1161 static int vcn_v4_0_5_stop(struct amdgpu_device *adev)
1162 {
1163 	volatile struct amdgpu_vcn4_fw_shared *fw_shared;
1164 	uint32_t tmp;
1165 	int i, r = 0;
1166 
1167 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1168 		if (adev->vcn.harvest_config & (1 << i))
1169 			continue;
1170 
1171 		fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1172 		fw_shared->sq.queue_mode |= FW_QUEUE_DPG_HOLD_OFF;
1173 
1174 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1175 			vcn_v4_0_5_stop_dpg_mode(adev, i);
1176 			continue;
1177 		}
1178 
1179 		/* wait for vcn idle */
1180 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1181 		if (r)
1182 			return r;
1183 
1184 		tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1185 			UVD_LMI_STATUS__READ_CLEAN_MASK |
1186 			UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1187 			UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1188 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1189 		if (r)
1190 			return r;
1191 
1192 		/* disable LMI UMC channel */
1193 		tmp = RREG32_SOC15(VCN, i, regUVD_LMI_CTRL2);
1194 		tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1195 		WREG32_SOC15(VCN, i, regUVD_LMI_CTRL2, tmp);
1196 		tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK |
1197 			UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1198 		r = SOC15_WAIT_ON_RREG(VCN, i, regUVD_LMI_STATUS, tmp, tmp);
1199 		if (r)
1200 			return r;
1201 
1202 		/* block VCPU register access */
1203 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_RB_ARB_CTRL),
1204 				UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1205 				~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1206 
1207 		/* reset VCPU */
1208 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL),
1209 				UVD_VCPU_CNTL__BLK_RST_MASK,
1210 				~UVD_VCPU_CNTL__BLK_RST_MASK);
1211 
1212 		/* disable VCPU clock */
1213 		WREG32_P(SOC15_REG_OFFSET(VCN, i, regUVD_VCPU_CNTL), 0,
1214 				~(UVD_VCPU_CNTL__CLK_EN_MASK));
1215 
1216 		/* apply soft reset */
1217 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1218 		tmp |= UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
1219 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1220 		tmp = RREG32_SOC15(VCN, i, regUVD_SOFT_RESET);
1221 		tmp |= UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
1222 		WREG32_SOC15(VCN, i, regUVD_SOFT_RESET, tmp);
1223 
1224 		/* clear status */
1225 		WREG32_SOC15(VCN, i, regUVD_STATUS, 0);
1226 
1227 		/* apply HW clock gating */
1228 		vcn_v4_0_5_enable_clock_gating(adev, i);
1229 
1230 		/* enable VCN power gating */
1231 		vcn_v4_0_5_enable_static_power_gating(adev, i);
1232 	}
1233 
1234 	if (adev->pm.dpm_enabled)
1235 		amdgpu_dpm_enable_uvd(adev, false);
1236 
1237 	return 0;
1238 }
1239 
1240 /**
1241  * vcn_v4_0_5_pause_dpg_mode - VCN pause with dpg mode
1242  *
1243  * @adev: amdgpu_device pointer
1244  * @inst_idx: instance number index
1245  * @new_state: pause state
1246  *
1247  * Pause dpg mode for VCN block
1248  */
1249 static int vcn_v4_0_5_pause_dpg_mode(struct amdgpu_device *adev, int inst_idx,
1250 		struct dpg_pause_state *new_state)
1251 {
1252 	uint32_t reg_data = 0;
1253 	int ret_code;
1254 
1255 	/* pause/unpause if state is changed */
1256 	if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1257 		DRM_DEV_DEBUG(adev->dev, "dpg pause state changed %d -> %d",
1258 			adev->vcn.inst[inst_idx].pause_state.fw_based,	new_state->fw_based);
1259 		reg_data = RREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE) &
1260 			(~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1261 
1262 		if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1263 			ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS, 0x1,
1264 				UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1265 
1266 			if (!ret_code) {
1267 				/* pause DPG */
1268 				reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1269 				WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1270 
1271 				/* wait for ACK */
1272 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_DPG_PAUSE,
1273 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1274 					UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1275 
1276 				SOC15_WAIT_ON_RREG(VCN, inst_idx, regUVD_POWER_STATUS,
1277 					UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON,
1278 					UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1279 			}
1280 		} else {
1281 			/* unpause dpg, no need to wait */
1282 			reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1283 			WREG32_SOC15(VCN, inst_idx, regUVD_DPG_PAUSE, reg_data);
1284 		}
1285 		adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1286 	}
1287 
1288 	return 0;
1289 }
1290 
1291 /**
1292  * vcn_v4_0_5_unified_ring_get_rptr - get unified read pointer
1293  *
1294  * @ring: amdgpu_ring pointer
1295  *
1296  * Returns the current hardware unified read pointer
1297  */
1298 static uint64_t vcn_v4_0_5_unified_ring_get_rptr(struct amdgpu_ring *ring)
1299 {
1300 	struct amdgpu_device *adev = ring->adev;
1301 
1302 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1303 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1304 
1305 	return RREG32_SOC15(VCN, ring->me, regUVD_RB_RPTR);
1306 }
1307 
1308 /**
1309  * vcn_v4_0_5_unified_ring_get_wptr - get unified write pointer
1310  *
1311  * @ring: amdgpu_ring pointer
1312  *
1313  * Returns the current hardware unified write pointer
1314  */
1315 static uint64_t vcn_v4_0_5_unified_ring_get_wptr(struct amdgpu_ring *ring)
1316 {
1317 	struct amdgpu_device *adev = ring->adev;
1318 
1319 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1320 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1321 
1322 	if (ring->use_doorbell)
1323 		return *ring->wptr_cpu_addr;
1324 	else
1325 		return RREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR);
1326 }
1327 
1328 /**
1329  * vcn_v4_0_5_unified_ring_set_wptr - set enc write pointer
1330  *
1331  * @ring: amdgpu_ring pointer
1332  *
1333  * Commits the enc write pointer to the hardware
1334  */
1335 static void vcn_v4_0_5_unified_ring_set_wptr(struct amdgpu_ring *ring)
1336 {
1337 	struct amdgpu_device *adev = ring->adev;
1338 
1339 	if (ring != &adev->vcn.inst[ring->me].ring_enc[0])
1340 		DRM_ERROR("wrong ring id is identified in %s", __func__);
1341 
1342 	if (ring->use_doorbell) {
1343 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1344 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1345 	} else {
1346 		WREG32_SOC15(VCN, ring->me, regUVD_RB_WPTR, lower_32_bits(ring->wptr));
1347 	}
1348 }
1349 
1350 static int vcn_v4_0_5_limit_sched(struct amdgpu_cs_parser *p,
1351 				struct amdgpu_job *job)
1352 {
1353 	struct drm_gpu_scheduler **scheds;
1354 
1355 	/* The create msg must be in the first IB submitted */
1356 	if (atomic_read(&job->base.entity->fence_seq))
1357 		return -EINVAL;
1358 
1359 	/* if VCN0 is harvested, we can't support AV1 */
1360 	if (p->adev->vcn.harvest_config & AMDGPU_VCN_HARVEST_VCN0)
1361 		return -EINVAL;
1362 
1363 	scheds = p->adev->gpu_sched[AMDGPU_HW_IP_VCN_ENC]
1364 		[AMDGPU_RING_PRIO_0].sched;
1365 	drm_sched_entity_modify_sched(job->base.entity, scheds, 1);
1366 	return 0;
1367 }
1368 
1369 static int vcn_v4_0_5_dec_msg(struct amdgpu_cs_parser *p, struct amdgpu_job *job,
1370 			    uint64_t addr)
1371 {
1372 	struct ttm_operation_ctx ctx = { false, false };
1373 	struct amdgpu_bo_va_mapping *map;
1374 	uint32_t *msg, num_buffers;
1375 	struct amdgpu_bo *bo;
1376 	uint64_t start, end;
1377 	unsigned int i;
1378 	void *ptr;
1379 	int r;
1380 
1381 	addr &= AMDGPU_GMC_HOLE_MASK;
1382 	r = amdgpu_cs_find_mapping(p, addr, &bo, &map);
1383 	if (r) {
1384 		DRM_ERROR("Can't find BO for addr 0x%08llx\n", addr);
1385 		return r;
1386 	}
1387 
1388 	start = map->start * AMDGPU_GPU_PAGE_SIZE;
1389 	end = (map->last + 1) * AMDGPU_GPU_PAGE_SIZE;
1390 	if (addr & 0x7) {
1391 		DRM_ERROR("VCN messages must be 8 byte aligned!\n");
1392 		return -EINVAL;
1393 	}
1394 
1395 	bo->flags |= AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
1396 	amdgpu_bo_placement_from_domain(bo, bo->allowed_domains);
1397 	r = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
1398 	if (r) {
1399 		DRM_ERROR("Failed validating the VCN message BO (%d)!\n", r);
1400 		return r;
1401 	}
1402 
1403 	r = amdgpu_bo_kmap(bo, &ptr);
1404 	if (r) {
1405 		DRM_ERROR("Failed mapping the VCN message (%d)!\n", r);
1406 		return r;
1407 	}
1408 
1409 	msg = ptr + addr - start;
1410 
1411 	/* Check length */
1412 	if (msg[1] > end - addr) {
1413 		r = -EINVAL;
1414 		goto out;
1415 	}
1416 
1417 	if (msg[3] != RDECODE_MSG_CREATE)
1418 		goto out;
1419 
1420 	num_buffers = msg[2];
1421 	for (i = 0, msg = &msg[6]; i < num_buffers; ++i, msg += 4) {
1422 		uint32_t offset, size, *create;
1423 
1424 		if (msg[0] != RDECODE_MESSAGE_CREATE)
1425 			continue;
1426 
1427 		offset = msg[1];
1428 		size = msg[2];
1429 
1430 		if (offset + size > end) {
1431 			r = -EINVAL;
1432 			goto out;
1433 		}
1434 
1435 		create = ptr + addr + offset - start;
1436 
1437 		/* H264, HEVC and VP9 can run on any instance */
1438 		if (create[0] == 0x7 || create[0] == 0x10 || create[0] == 0x11)
1439 			continue;
1440 
1441 		r = vcn_v4_0_5_limit_sched(p, job);
1442 		if (r)
1443 			goto out;
1444 	}
1445 
1446 out:
1447 	amdgpu_bo_kunmap(bo);
1448 	return r;
1449 }
1450 
1451 #define RADEON_VCN_ENGINE_TYPE_ENCODE			(0x00000002)
1452 #define RADEON_VCN_ENGINE_TYPE_DECODE			(0x00000003)
1453 
1454 #define RADEON_VCN_ENGINE_INFO				(0x30000001)
1455 #define RADEON_VCN_ENGINE_INFO_MAX_OFFSET		16
1456 
1457 #define RENCODE_ENCODE_STANDARD_AV1			2
1458 #define RENCODE_IB_PARAM_SESSION_INIT			0x00000003
1459 #define RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET	64
1460 
1461 /* return the offset in ib if id is found, -1 otherwise
1462  * to speed up the searching we only search upto max_offset
1463  */
1464 static int vcn_v4_0_5_enc_find_ib_param(struct amdgpu_ib *ib, uint32_t id, int max_offset)
1465 {
1466 	int i;
1467 
1468 	for (i = 0; i < ib->length_dw && i < max_offset && ib->ptr[i] >= 8; i += ib->ptr[i]/4) {
1469 		if (ib->ptr[i + 1] == id)
1470 			return i;
1471 	}
1472 	return -1;
1473 }
1474 
1475 static int vcn_v4_0_5_ring_patch_cs_in_place(struct amdgpu_cs_parser *p,
1476 					   struct amdgpu_job *job,
1477 					   struct amdgpu_ib *ib)
1478 {
1479 	struct amdgpu_ring *ring = amdgpu_job_ring(job);
1480 	struct amdgpu_vcn_decode_buffer *decode_buffer;
1481 	uint64_t addr;
1482 	uint32_t val;
1483 	int idx;
1484 
1485 	/* The first instance can decode anything */
1486 	if (!ring->me)
1487 		return 0;
1488 
1489 	/* RADEON_VCN_ENGINE_INFO is at the top of ib block */
1490 	idx = vcn_v4_0_5_enc_find_ib_param(ib, RADEON_VCN_ENGINE_INFO,
1491 			RADEON_VCN_ENGINE_INFO_MAX_OFFSET);
1492 	if (idx < 0) /* engine info is missing */
1493 		return 0;
1494 
1495 	val = amdgpu_ib_get_value(ib, idx + 2); /* RADEON_VCN_ENGINE_TYPE */
1496 	if (val == RADEON_VCN_ENGINE_TYPE_DECODE) {
1497 		decode_buffer = (struct amdgpu_vcn_decode_buffer *)&ib->ptr[idx + 6];
1498 
1499 		if (!(decode_buffer->valid_buf_flag  & 0x1))
1500 			return 0;
1501 
1502 		addr = ((u64)decode_buffer->msg_buffer_address_hi) << 32 |
1503 			decode_buffer->msg_buffer_address_lo;
1504 		return vcn_v4_0_5_dec_msg(p, job, addr);
1505 	} else if (val == RADEON_VCN_ENGINE_TYPE_ENCODE) {
1506 		idx = vcn_v4_0_5_enc_find_ib_param(ib, RENCODE_IB_PARAM_SESSION_INIT,
1507 			RENCODE_IB_PARAM_SESSION_INIT_MAX_OFFSET);
1508 		if (idx >= 0 && ib->ptr[idx + 2] == RENCODE_ENCODE_STANDARD_AV1)
1509 			return vcn_v4_0_5_limit_sched(p, job);
1510 	}
1511 	return 0;
1512 }
1513 
1514 static const struct amdgpu_ring_funcs vcn_v4_0_5_unified_ring_vm_funcs = {
1515 	.type = AMDGPU_RING_TYPE_VCN_ENC,
1516 	.align_mask = 0x3f,
1517 	.nop = VCN_ENC_CMD_NO_OP,
1518 	.get_rptr = vcn_v4_0_5_unified_ring_get_rptr,
1519 	.get_wptr = vcn_v4_0_5_unified_ring_get_wptr,
1520 	.set_wptr = vcn_v4_0_5_unified_ring_set_wptr,
1521 	.patch_cs_in_place = vcn_v4_0_5_ring_patch_cs_in_place,
1522 	.emit_frame_size =
1523 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1524 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1525 		4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1526 		5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1527 		1, /* vcn_v2_0_enc_ring_insert_end */
1528 	.emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1529 	.emit_ib = vcn_v2_0_enc_ring_emit_ib,
1530 	.emit_fence = vcn_v2_0_enc_ring_emit_fence,
1531 	.emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1532 	.test_ring = amdgpu_vcn_enc_ring_test_ring,
1533 	.test_ib = amdgpu_vcn_unified_ring_test_ib,
1534 	.insert_nop = amdgpu_ring_insert_nop,
1535 	.insert_end = vcn_v2_0_enc_ring_insert_end,
1536 	.pad_ib = amdgpu_ring_generic_pad_ib,
1537 	.begin_use = amdgpu_vcn_ring_begin_use,
1538 	.end_use = amdgpu_vcn_ring_end_use,
1539 	.emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1540 	.emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1541 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1542 };
1543 
1544 /**
1545  * vcn_v4_0_5_set_unified_ring_funcs - set unified ring functions
1546  *
1547  * @adev: amdgpu_device pointer
1548  *
1549  * Set unified ring functions
1550  */
1551 static void vcn_v4_0_5_set_unified_ring_funcs(struct amdgpu_device *adev)
1552 {
1553 	int i;
1554 
1555 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1556 		if (adev->vcn.harvest_config & (1 << i))
1557 			continue;
1558 
1559 		adev->vcn.inst[i].ring_enc[0].funcs = &vcn_v4_0_5_unified_ring_vm_funcs;
1560 		adev->vcn.inst[i].ring_enc[0].me = i;
1561 	}
1562 }
1563 
1564 /**
1565  * vcn_v4_0_5_is_idle - check VCN block is idle
1566  *
1567  * @handle: amdgpu_device pointer
1568  *
1569  * Check whether VCN block is idle
1570  */
1571 static bool vcn_v4_0_5_is_idle(void *handle)
1572 {
1573 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1574 	int i, ret = 1;
1575 
1576 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1577 		if (adev->vcn.harvest_config & (1 << i))
1578 			continue;
1579 
1580 		ret &= (RREG32_SOC15(VCN, i, regUVD_STATUS) == UVD_STATUS__IDLE);
1581 	}
1582 
1583 	return ret;
1584 }
1585 
1586 /**
1587  * vcn_v4_0_5_wait_for_idle - wait for VCN block idle
1588  *
1589  * @handle: amdgpu_device pointer
1590  *
1591  * Wait for VCN block idle
1592  */
1593 static int vcn_v4_0_5_wait_for_idle(void *handle)
1594 {
1595 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1596 	int i, ret = 0;
1597 
1598 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1599 		if (adev->vcn.harvest_config & (1 << i))
1600 			continue;
1601 
1602 		ret = SOC15_WAIT_ON_RREG(VCN, i, regUVD_STATUS, UVD_STATUS__IDLE,
1603 			UVD_STATUS__IDLE);
1604 		if (ret)
1605 			return ret;
1606 	}
1607 
1608 	return ret;
1609 }
1610 
1611 /**
1612  * vcn_v4_0_5_set_clockgating_state - set VCN block clockgating state
1613  *
1614  * @handle: amdgpu_device pointer
1615  * @state: clock gating state
1616  *
1617  * Set VCN block clockgating state
1618  */
1619 static int vcn_v4_0_5_set_clockgating_state(void *handle, enum amd_clockgating_state state)
1620 {
1621 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1622 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1623 	int i;
1624 
1625 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1626 		if (adev->vcn.harvest_config & (1 << i))
1627 			continue;
1628 
1629 		if (enable) {
1630 			if (RREG32_SOC15(VCN, i, regUVD_STATUS) != UVD_STATUS__IDLE)
1631 				return -EBUSY;
1632 			vcn_v4_0_5_enable_clock_gating(adev, i);
1633 		} else {
1634 			vcn_v4_0_5_disable_clock_gating(adev, i);
1635 		}
1636 	}
1637 
1638 	return 0;
1639 }
1640 
1641 /**
1642  * vcn_v4_0_5_set_powergating_state - set VCN block powergating state
1643  *
1644  * @handle: amdgpu_device pointer
1645  * @state: power gating state
1646  *
1647  * Set VCN block powergating state
1648  */
1649 static int vcn_v4_0_5_set_powergating_state(void *handle, enum amd_powergating_state state)
1650 {
1651 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1652 	int ret;
1653 
1654 	if (state == adev->vcn.cur_state)
1655 		return 0;
1656 
1657 	if (state == AMD_PG_STATE_GATE)
1658 		ret = vcn_v4_0_5_stop(adev);
1659 	else
1660 		ret = vcn_v4_0_5_start(adev);
1661 
1662 	if (!ret)
1663 		adev->vcn.cur_state = state;
1664 
1665 	return ret;
1666 }
1667 
1668 /**
1669  * vcn_v4_0_5_process_interrupt - process VCN block interrupt
1670  *
1671  * @adev: amdgpu_device pointer
1672  * @source: interrupt sources
1673  * @entry: interrupt entry from clients and sources
1674  *
1675  * Process VCN block interrupt
1676  */
1677 static int vcn_v4_0_5_process_interrupt(struct amdgpu_device *adev, struct amdgpu_irq_src *source,
1678 		struct amdgpu_iv_entry *entry)
1679 {
1680 	uint32_t ip_instance;
1681 
1682 	switch (entry->client_id) {
1683 	case SOC15_IH_CLIENTID_VCN:
1684 		ip_instance = 0;
1685 		break;
1686 	case SOC15_IH_CLIENTID_VCN1:
1687 		ip_instance = 1;
1688 		break;
1689 	default:
1690 		DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1691 		return 0;
1692 	}
1693 
1694 	DRM_DEBUG("IH: VCN TRAP\n");
1695 
1696 	switch (entry->src_id) {
1697 	case VCN_4_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1698 		amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1699 		break;
1700 	case VCN_4_0__SRCID_UVD_POISON:
1701 		amdgpu_vcn_process_poison_irq(adev, source, entry);
1702 		break;
1703 	default:
1704 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1705 			  entry->src_id, entry->src_data[0]);
1706 		break;
1707 	}
1708 
1709 	return 0;
1710 }
1711 
1712 static const struct amdgpu_irq_src_funcs vcn_v4_0_5_irq_funcs = {
1713 	.process = vcn_v4_0_5_process_interrupt,
1714 };
1715 
1716 /**
1717  * vcn_v4_0_5_set_irq_funcs - set VCN block interrupt irq functions
1718  *
1719  * @adev: amdgpu_device pointer
1720  *
1721  * Set VCN block interrupt irq functions
1722  */
1723 static void vcn_v4_0_5_set_irq_funcs(struct amdgpu_device *adev)
1724 {
1725 	int i;
1726 
1727 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1728 		if (adev->vcn.harvest_config & (1 << i))
1729 			continue;
1730 
1731 		adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1732 		adev->vcn.inst[i].irq.funcs = &vcn_v4_0_5_irq_funcs;
1733 	}
1734 }
1735 
1736 static const struct amd_ip_funcs vcn_v4_0_5_ip_funcs = {
1737 	.name = "vcn_v4_0_5",
1738 	.early_init = vcn_v4_0_5_early_init,
1739 	.late_init = NULL,
1740 	.sw_init = vcn_v4_0_5_sw_init,
1741 	.sw_fini = vcn_v4_0_5_sw_fini,
1742 	.hw_init = vcn_v4_0_5_hw_init,
1743 	.hw_fini = vcn_v4_0_5_hw_fini,
1744 	.suspend = vcn_v4_0_5_suspend,
1745 	.resume = vcn_v4_0_5_resume,
1746 	.is_idle = vcn_v4_0_5_is_idle,
1747 	.wait_for_idle = vcn_v4_0_5_wait_for_idle,
1748 	.check_soft_reset = NULL,
1749 	.pre_soft_reset = NULL,
1750 	.soft_reset = NULL,
1751 	.post_soft_reset = NULL,
1752 	.set_clockgating_state = vcn_v4_0_5_set_clockgating_state,
1753 	.set_powergating_state = vcn_v4_0_5_set_powergating_state,
1754 	.dump_ip_state = NULL,
1755 	.print_ip_state = NULL,
1756 };
1757 
1758 const struct amdgpu_ip_block_version vcn_v4_0_5_ip_block = {
1759 	.type = AMD_IP_BLOCK_TYPE_VCN,
1760 	.major = 4,
1761 	.minor = 0,
1762 	.rev = 5,
1763 	.funcs = &vcn_v4_0_5_ip_funcs,
1764 };
1765