xref: /linux/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c (revision bd628c1bed7902ec1f24ba0fe70758949146abbe)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  * All Rights Reserved.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the
7  * "Software"), to deal in the Software without restriction, including
8  * without limitation the rights to use, copy, modify, merge, publish,
9  * distribute, sub license, and/or sell copies of the Software, and to
10  * permit persons to whom the Software is furnished to do so, subject to
11  * the following conditions:
12  *
13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19  * USE OR OTHER DEALINGS IN THE SOFTWARE.
20  *
21  * The above copyright notice and this permission notice (including the
22  * next paragraph) shall be included in all copies or substantial portions
23  * of the Software.
24  *
25  * Authors: Christian König <christian.koenig@amd.com>
26  */
27 
28 #include <linux/firmware.h>
29 #include <drm/drmP.h>
30 #include "amdgpu.h"
31 #include "amdgpu_vce.h"
32 #include "vid.h"
33 #include "vce/vce_3_0_d.h"
34 #include "vce/vce_3_0_sh_mask.h"
35 #include "oss/oss_3_0_d.h"
36 #include "oss/oss_3_0_sh_mask.h"
37 #include "gca/gfx_8_0_d.h"
38 #include "smu/smu_7_1_2_d.h"
39 #include "smu/smu_7_1_2_sh_mask.h"
40 #include "gca/gfx_8_0_sh_mask.h"
41 #include "ivsrcid/ivsrcid_vislands30.h"
42 
43 
44 #define GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT	0x04
45 #define GRBM_GFX_INDEX__VCE_INSTANCE_MASK	0x10
46 #define GRBM_GFX_INDEX__VCE_ALL_PIPE		0x07
47 
48 #define mmVCE_LMI_VCPU_CACHE_40BIT_BAR0	0x8616
49 #define mmVCE_LMI_VCPU_CACHE_40BIT_BAR1	0x8617
50 #define mmVCE_LMI_VCPU_CACHE_40BIT_BAR2	0x8618
51 #define mmGRBM_GFX_INDEX_DEFAULT 0xE0000000
52 
53 #define VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK	0x02
54 
55 #define VCE_V3_0_FW_SIZE	(384 * 1024)
56 #define VCE_V3_0_STACK_SIZE	(64 * 1024)
57 #define VCE_V3_0_DATA_SIZE	((16 * 1024 * AMDGPU_MAX_VCE_HANDLES) + (52 * 1024))
58 
59 #define FW_52_8_3	((52 << 24) | (8 << 16) | (3 << 8))
60 
61 #define GET_VCE_INSTANCE(i)  ((i) << GRBM_GFX_INDEX__VCE_INSTANCE__SHIFT \
62 					| GRBM_GFX_INDEX__VCE_ALL_PIPE)
63 
64 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx);
65 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev);
66 static void vce_v3_0_set_irq_funcs(struct amdgpu_device *adev);
67 static int vce_v3_0_wait_for_idle(void *handle);
68 static int vce_v3_0_set_clockgating_state(void *handle,
69 					  enum amd_clockgating_state state);
70 /**
71  * vce_v3_0_ring_get_rptr - get read pointer
72  *
73  * @ring: amdgpu_ring pointer
74  *
75  * Returns the current hardware read pointer
76  */
77 static uint64_t vce_v3_0_ring_get_rptr(struct amdgpu_ring *ring)
78 {
79 	struct amdgpu_device *adev = ring->adev;
80 	u32 v;
81 
82 	mutex_lock(&adev->grbm_idx_mutex);
83 	if (adev->vce.harvest_config == 0 ||
84 		adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
85 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
86 	else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
87 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
88 
89 	if (ring->me == 0)
90 		v = RREG32(mmVCE_RB_RPTR);
91 	else if (ring->me == 1)
92 		v = RREG32(mmVCE_RB_RPTR2);
93 	else
94 		v = RREG32(mmVCE_RB_RPTR3);
95 
96 	WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
97 	mutex_unlock(&adev->grbm_idx_mutex);
98 
99 	return v;
100 }
101 
102 /**
103  * vce_v3_0_ring_get_wptr - get write pointer
104  *
105  * @ring: amdgpu_ring pointer
106  *
107  * Returns the current hardware write pointer
108  */
109 static uint64_t vce_v3_0_ring_get_wptr(struct amdgpu_ring *ring)
110 {
111 	struct amdgpu_device *adev = ring->adev;
112 	u32 v;
113 
114 	mutex_lock(&adev->grbm_idx_mutex);
115 	if (adev->vce.harvest_config == 0 ||
116 		adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
117 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
118 	else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
119 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
120 
121 	if (ring->me == 0)
122 		v = RREG32(mmVCE_RB_WPTR);
123 	else if (ring->me == 1)
124 		v = RREG32(mmVCE_RB_WPTR2);
125 	else
126 		v = RREG32(mmVCE_RB_WPTR3);
127 
128 	WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
129 	mutex_unlock(&adev->grbm_idx_mutex);
130 
131 	return v;
132 }
133 
134 /**
135  * vce_v3_0_ring_set_wptr - set write pointer
136  *
137  * @ring: amdgpu_ring pointer
138  *
139  * Commits the write pointer to the hardware
140  */
141 static void vce_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
142 {
143 	struct amdgpu_device *adev = ring->adev;
144 
145 	mutex_lock(&adev->grbm_idx_mutex);
146 	if (adev->vce.harvest_config == 0 ||
147 		adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE1)
148 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
149 	else if (adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0)
150 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
151 
152 	if (ring->me == 0)
153 		WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
154 	else if (ring->me == 1)
155 		WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
156 	else
157 		WREG32(mmVCE_RB_WPTR3, lower_32_bits(ring->wptr));
158 
159 	WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
160 	mutex_unlock(&adev->grbm_idx_mutex);
161 }
162 
163 static void vce_v3_0_override_vce_clock_gating(struct amdgpu_device *adev, bool override)
164 {
165 	WREG32_FIELD(VCE_RB_ARB_CTRL, VCE_CGTT_OVERRIDE, override ? 1 : 0);
166 }
167 
168 static void vce_v3_0_set_vce_sw_clock_gating(struct amdgpu_device *adev,
169 					     bool gated)
170 {
171 	u32 data;
172 
173 	/* Set Override to disable Clock Gating */
174 	vce_v3_0_override_vce_clock_gating(adev, true);
175 
176 	/* This function enables MGCG which is controlled by firmware.
177 	   With the clocks in the gated state the core is still
178 	   accessible but the firmware will throttle the clocks on the
179 	   fly as necessary.
180 	*/
181 	if (!gated) {
182 		data = RREG32(mmVCE_CLOCK_GATING_B);
183 		data |= 0x1ff;
184 		data &= ~0xef0000;
185 		WREG32(mmVCE_CLOCK_GATING_B, data);
186 
187 		data = RREG32(mmVCE_UENC_CLOCK_GATING);
188 		data |= 0x3ff000;
189 		data &= ~0xffc00000;
190 		WREG32(mmVCE_UENC_CLOCK_GATING, data);
191 
192 		data = RREG32(mmVCE_UENC_CLOCK_GATING_2);
193 		data |= 0x2;
194 		data &= ~0x00010000;
195 		WREG32(mmVCE_UENC_CLOCK_GATING_2, data);
196 
197 		data = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
198 		data |= 0x37f;
199 		WREG32(mmVCE_UENC_REG_CLOCK_GATING, data);
200 
201 		data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL);
202 		data |= VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK |
203 			VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK |
204 			VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK  |
205 			0x8;
206 		WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data);
207 	} else {
208 		data = RREG32(mmVCE_CLOCK_GATING_B);
209 		data &= ~0x80010;
210 		data |= 0xe70008;
211 		WREG32(mmVCE_CLOCK_GATING_B, data);
212 
213 		data = RREG32(mmVCE_UENC_CLOCK_GATING);
214 		data |= 0xffc00000;
215 		WREG32(mmVCE_UENC_CLOCK_GATING, data);
216 
217 		data = RREG32(mmVCE_UENC_CLOCK_GATING_2);
218 		data |= 0x10000;
219 		WREG32(mmVCE_UENC_CLOCK_GATING_2, data);
220 
221 		data = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
222 		data &= ~0x3ff;
223 		WREG32(mmVCE_UENC_REG_CLOCK_GATING, data);
224 
225 		data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL);
226 		data &= ~(VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK |
227 			  VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK |
228 			  VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK  |
229 			  0x8);
230 		WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data);
231 	}
232 	vce_v3_0_override_vce_clock_gating(adev, false);
233 }
234 
235 static int vce_v3_0_firmware_loaded(struct amdgpu_device *adev)
236 {
237 	int i, j;
238 
239 	for (i = 0; i < 10; ++i) {
240 		for (j = 0; j < 100; ++j) {
241 			uint32_t status = RREG32(mmVCE_STATUS);
242 
243 			if (status & VCE_STATUS_VCPU_REPORT_FW_LOADED_MASK)
244 				return 0;
245 			mdelay(10);
246 		}
247 
248 		DRM_ERROR("VCE not responding, trying to reset the ECPU!!!\n");
249 		WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 1);
250 		mdelay(10);
251 		WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 0);
252 		mdelay(10);
253 	}
254 
255 	return -ETIMEDOUT;
256 }
257 
258 /**
259  * vce_v3_0_start - start VCE block
260  *
261  * @adev: amdgpu_device pointer
262  *
263  * Setup and start the VCE block
264  */
265 static int vce_v3_0_start(struct amdgpu_device *adev)
266 {
267 	struct amdgpu_ring *ring;
268 	int idx, r;
269 
270 	mutex_lock(&adev->grbm_idx_mutex);
271 	for (idx = 0; idx < 2; ++idx) {
272 		if (adev->vce.harvest_config & (1 << idx))
273 			continue;
274 
275 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(idx));
276 
277 		/* Program instance 0 reg space for two instances or instance 0 case
278 		program instance 1 reg space for only instance 1 available case */
279 		if (idx != 1 || adev->vce.harvest_config == AMDGPU_VCE_HARVEST_VCE0) {
280 			ring = &adev->vce.ring[0];
281 			WREG32(mmVCE_RB_RPTR, lower_32_bits(ring->wptr));
282 			WREG32(mmVCE_RB_WPTR, lower_32_bits(ring->wptr));
283 			WREG32(mmVCE_RB_BASE_LO, ring->gpu_addr);
284 			WREG32(mmVCE_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
285 			WREG32(mmVCE_RB_SIZE, ring->ring_size / 4);
286 
287 			ring = &adev->vce.ring[1];
288 			WREG32(mmVCE_RB_RPTR2, lower_32_bits(ring->wptr));
289 			WREG32(mmVCE_RB_WPTR2, lower_32_bits(ring->wptr));
290 			WREG32(mmVCE_RB_BASE_LO2, ring->gpu_addr);
291 			WREG32(mmVCE_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
292 			WREG32(mmVCE_RB_SIZE2, ring->ring_size / 4);
293 
294 			ring = &adev->vce.ring[2];
295 			WREG32(mmVCE_RB_RPTR3, lower_32_bits(ring->wptr));
296 			WREG32(mmVCE_RB_WPTR3, lower_32_bits(ring->wptr));
297 			WREG32(mmVCE_RB_BASE_LO3, ring->gpu_addr);
298 			WREG32(mmVCE_RB_BASE_HI3, upper_32_bits(ring->gpu_addr));
299 			WREG32(mmVCE_RB_SIZE3, ring->ring_size / 4);
300 		}
301 
302 		vce_v3_0_mc_resume(adev, idx);
303 		WREG32_FIELD(VCE_STATUS, JOB_BUSY, 1);
304 
305 		if (adev->asic_type >= CHIP_STONEY)
306 			WREG32_P(mmVCE_VCPU_CNTL, 1, ~0x200001);
307 		else
308 			WREG32_FIELD(VCE_VCPU_CNTL, CLK_EN, 1);
309 
310 		WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 0);
311 		mdelay(100);
312 
313 		r = vce_v3_0_firmware_loaded(adev);
314 
315 		/* clear BUSY flag */
316 		WREG32_FIELD(VCE_STATUS, JOB_BUSY, 0);
317 
318 		if (r) {
319 			DRM_ERROR("VCE not responding, giving up!!!\n");
320 			mutex_unlock(&adev->grbm_idx_mutex);
321 			return r;
322 		}
323 	}
324 
325 	WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
326 	mutex_unlock(&adev->grbm_idx_mutex);
327 
328 	return 0;
329 }
330 
331 static int vce_v3_0_stop(struct amdgpu_device *adev)
332 {
333 	int idx;
334 
335 	mutex_lock(&adev->grbm_idx_mutex);
336 	for (idx = 0; idx < 2; ++idx) {
337 		if (adev->vce.harvest_config & (1 << idx))
338 			continue;
339 
340 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(idx));
341 
342 		if (adev->asic_type >= CHIP_STONEY)
343 			WREG32_P(mmVCE_VCPU_CNTL, 0, ~0x200001);
344 		else
345 			WREG32_FIELD(VCE_VCPU_CNTL, CLK_EN, 0);
346 
347 		/* hold on ECPU */
348 		WREG32_FIELD(VCE_SOFT_RESET, ECPU_SOFT_RESET, 1);
349 
350 		/* clear VCE STATUS */
351 		WREG32(mmVCE_STATUS, 0);
352 	}
353 
354 	WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
355 	mutex_unlock(&adev->grbm_idx_mutex);
356 
357 	return 0;
358 }
359 
360 #define ixVCE_HARVEST_FUSE_MACRO__ADDRESS     0xC0014074
361 #define VCE_HARVEST_FUSE_MACRO__SHIFT       27
362 #define VCE_HARVEST_FUSE_MACRO__MASK        0x18000000
363 
364 static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev)
365 {
366 	u32 tmp;
367 
368 	if ((adev->asic_type == CHIP_FIJI) ||
369 	    (adev->asic_type == CHIP_STONEY))
370 		return AMDGPU_VCE_HARVEST_VCE1;
371 
372 	if (adev->flags & AMD_IS_APU)
373 		tmp = (RREG32_SMC(ixVCE_HARVEST_FUSE_MACRO__ADDRESS) &
374 		       VCE_HARVEST_FUSE_MACRO__MASK) >>
375 			VCE_HARVEST_FUSE_MACRO__SHIFT;
376 	else
377 		tmp = (RREG32_SMC(ixCC_HARVEST_FUSES) &
378 		       CC_HARVEST_FUSES__VCE_DISABLE_MASK) >>
379 			CC_HARVEST_FUSES__VCE_DISABLE__SHIFT;
380 
381 	switch (tmp) {
382 	case 1:
383 		return AMDGPU_VCE_HARVEST_VCE0;
384 	case 2:
385 		return AMDGPU_VCE_HARVEST_VCE1;
386 	case 3:
387 		return AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1;
388 	default:
389 		if ((adev->asic_type == CHIP_POLARIS10) ||
390 		    (adev->asic_type == CHIP_POLARIS11) ||
391 		    (adev->asic_type == CHIP_POLARIS12) ||
392 		    (adev->asic_type == CHIP_VEGAM))
393 			return AMDGPU_VCE_HARVEST_VCE1;
394 
395 		return 0;
396 	}
397 }
398 
399 static int vce_v3_0_early_init(void *handle)
400 {
401 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
402 
403 	adev->vce.harvest_config = vce_v3_0_get_harvest_config(adev);
404 
405 	if ((adev->vce.harvest_config &
406 	     (AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1)) ==
407 	    (AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1))
408 		return -ENOENT;
409 
410 	adev->vce.num_rings = 3;
411 
412 	vce_v3_0_set_ring_funcs(adev);
413 	vce_v3_0_set_irq_funcs(adev);
414 
415 	return 0;
416 }
417 
418 static int vce_v3_0_sw_init(void *handle)
419 {
420 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
421 	struct amdgpu_ring *ring;
422 	int r, i;
423 
424 	/* VCE */
425 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_VCE_TRAP, &adev->vce.irq);
426 	if (r)
427 		return r;
428 
429 	r = amdgpu_vce_sw_init(adev, VCE_V3_0_FW_SIZE +
430 		(VCE_V3_0_STACK_SIZE + VCE_V3_0_DATA_SIZE) * 2);
431 	if (r)
432 		return r;
433 
434 	/* 52.8.3 required for 3 ring support */
435 	if (adev->vce.fw_version < FW_52_8_3)
436 		adev->vce.num_rings = 2;
437 
438 	r = amdgpu_vce_resume(adev);
439 	if (r)
440 		return r;
441 
442 	for (i = 0; i < adev->vce.num_rings; i++) {
443 		ring = &adev->vce.ring[i];
444 		sprintf(ring->name, "vce%d", i);
445 		r = amdgpu_ring_init(adev, ring, 512, &adev->vce.irq, 0);
446 		if (r)
447 			return r;
448 	}
449 
450 	r = amdgpu_vce_entity_init(adev);
451 
452 	return r;
453 }
454 
455 static int vce_v3_0_sw_fini(void *handle)
456 {
457 	int r;
458 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
459 
460 	r = amdgpu_vce_suspend(adev);
461 	if (r)
462 		return r;
463 
464 	return amdgpu_vce_sw_fini(adev);
465 }
466 
467 static int vce_v3_0_hw_init(void *handle)
468 {
469 	int r, i;
470 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
471 
472 	vce_v3_0_override_vce_clock_gating(adev, true);
473 
474 	amdgpu_asic_set_vce_clocks(adev, 10000, 10000);
475 
476 	for (i = 0; i < adev->vce.num_rings; i++) {
477 		r = amdgpu_ring_test_helper(&adev->vce.ring[i]);
478 		if (r)
479 			return r;
480 	}
481 
482 	DRM_INFO("VCE initialized successfully.\n");
483 
484 	return 0;
485 }
486 
487 static int vce_v3_0_hw_fini(void *handle)
488 {
489 	int r;
490 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
491 
492 	r = vce_v3_0_wait_for_idle(handle);
493 	if (r)
494 		return r;
495 
496 	vce_v3_0_stop(adev);
497 	return vce_v3_0_set_clockgating_state(adev, AMD_CG_STATE_GATE);
498 }
499 
500 static int vce_v3_0_suspend(void *handle)
501 {
502 	int r;
503 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
504 
505 	r = vce_v3_0_hw_fini(adev);
506 	if (r)
507 		return r;
508 
509 	return amdgpu_vce_suspend(adev);
510 }
511 
512 static int vce_v3_0_resume(void *handle)
513 {
514 	int r;
515 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
516 
517 	r = amdgpu_vce_resume(adev);
518 	if (r)
519 		return r;
520 
521 	return vce_v3_0_hw_init(adev);
522 }
523 
524 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
525 {
526 	uint32_t offset, size;
527 
528 	WREG32_P(mmVCE_CLOCK_GATING_A, 0, ~(1 << 16));
529 	WREG32_P(mmVCE_UENC_CLOCK_GATING, 0x1FF000, ~0xFF9FF000);
530 	WREG32_P(mmVCE_UENC_REG_CLOCK_GATING, 0x3F, ~0x3F);
531 	WREG32(mmVCE_CLOCK_GATING_B, 0x1FF);
532 
533 	WREG32(mmVCE_LMI_CTRL, 0x00398000);
534 	WREG32_P(mmVCE_LMI_CACHE_CTRL, 0x0, ~0x1);
535 	WREG32(mmVCE_LMI_SWAP_CNTL, 0);
536 	WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
537 	WREG32(mmVCE_LMI_VM_CTRL, 0);
538 	WREG32_OR(mmVCE_VCPU_CNTL, 0x00100000);
539 
540 	if (adev->asic_type >= CHIP_STONEY) {
541 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8));
542 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8));
543 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8));
544 	} else
545 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
546 	offset = AMDGPU_VCE_FIRMWARE_OFFSET;
547 	size = VCE_V3_0_FW_SIZE;
548 	WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
549 	WREG32(mmVCE_VCPU_CACHE_SIZE0, size);
550 
551 	if (idx == 0) {
552 		offset += size;
553 		size = VCE_V3_0_STACK_SIZE;
554 		WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0x7fffffff);
555 		WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
556 		offset += size;
557 		size = VCE_V3_0_DATA_SIZE;
558 		WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0x7fffffff);
559 		WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
560 	} else {
561 		offset += size + VCE_V3_0_STACK_SIZE + VCE_V3_0_DATA_SIZE;
562 		size = VCE_V3_0_STACK_SIZE;
563 		WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0xfffffff);
564 		WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
565 		offset += size;
566 		size = VCE_V3_0_DATA_SIZE;
567 		WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0xfffffff);
568 		WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
569 	}
570 
571 	WREG32_P(mmVCE_LMI_CTRL2, 0x0, ~0x100);
572 	WREG32_FIELD(VCE_SYS_INT_EN, VCE_SYS_INT_TRAP_INTERRUPT_EN, 1);
573 }
574 
575 static bool vce_v3_0_is_idle(void *handle)
576 {
577 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
578 	u32 mask = 0;
579 
580 	mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_STATUS2__VCE0_BUSY_MASK;
581 	mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_STATUS2__VCE1_BUSY_MASK;
582 
583 	return !(RREG32(mmSRBM_STATUS2) & mask);
584 }
585 
586 static int vce_v3_0_wait_for_idle(void *handle)
587 {
588 	unsigned i;
589 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
590 
591 	for (i = 0; i < adev->usec_timeout; i++)
592 		if (vce_v3_0_is_idle(handle))
593 			return 0;
594 
595 	return -ETIMEDOUT;
596 }
597 
598 #define  VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK  0x00000008L   /* AUTO_BUSY */
599 #define  VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK   0x00000010L   /* RB0_BUSY */
600 #define  VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK   0x00000020L   /* RB1_BUSY */
601 #define  AMDGPU_VCE_STATUS_BUSY_MASK (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK | \
602 				      VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK)
603 
604 static bool vce_v3_0_check_soft_reset(void *handle)
605 {
606 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
607 	u32 srbm_soft_reset = 0;
608 
609 	/* According to VCE team , we should use VCE_STATUS instead
610 	 * SRBM_STATUS.VCE_BUSY bit for busy status checking.
611 	 * GRBM_GFX_INDEX.INSTANCE_INDEX is used to specify which VCE
612 	 * instance's registers are accessed
613 	 * (0 for 1st instance, 10 for 2nd instance).
614 	 *
615 	 *VCE_STATUS
616 	 *|UENC|ACPI|AUTO ACTIVE|RB1 |RB0 |RB2 |          |FW_LOADED|JOB |
617 	 *|----+----+-----------+----+----+----+----------+---------+----|
618 	 *|bit8|bit7|    bit6   |bit5|bit4|bit3|   bit2   |  bit1   |bit0|
619 	 *
620 	 * VCE team suggest use bit 3--bit 6 for busy status check
621 	 */
622 	mutex_lock(&adev->grbm_idx_mutex);
623 	WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
624 	if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
625 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
626 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
627 	}
628 	WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
629 	if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
630 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
631 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
632 	}
633 	WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
634 	mutex_unlock(&adev->grbm_idx_mutex);
635 
636 	if (srbm_soft_reset) {
637 		adev->vce.srbm_soft_reset = srbm_soft_reset;
638 		return true;
639 	} else {
640 		adev->vce.srbm_soft_reset = 0;
641 		return false;
642 	}
643 }
644 
645 static int vce_v3_0_soft_reset(void *handle)
646 {
647 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
648 	u32 srbm_soft_reset;
649 
650 	if (!adev->vce.srbm_soft_reset)
651 		return 0;
652 	srbm_soft_reset = adev->vce.srbm_soft_reset;
653 
654 	if (srbm_soft_reset) {
655 		u32 tmp;
656 
657 		tmp = RREG32(mmSRBM_SOFT_RESET);
658 		tmp |= srbm_soft_reset;
659 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
660 		WREG32(mmSRBM_SOFT_RESET, tmp);
661 		tmp = RREG32(mmSRBM_SOFT_RESET);
662 
663 		udelay(50);
664 
665 		tmp &= ~srbm_soft_reset;
666 		WREG32(mmSRBM_SOFT_RESET, tmp);
667 		tmp = RREG32(mmSRBM_SOFT_RESET);
668 
669 		/* Wait a little for things to settle down */
670 		udelay(50);
671 	}
672 
673 	return 0;
674 }
675 
676 static int vce_v3_0_pre_soft_reset(void *handle)
677 {
678 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
679 
680 	if (!adev->vce.srbm_soft_reset)
681 		return 0;
682 
683 	mdelay(5);
684 
685 	return vce_v3_0_suspend(adev);
686 }
687 
688 
689 static int vce_v3_0_post_soft_reset(void *handle)
690 {
691 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
692 
693 	if (!adev->vce.srbm_soft_reset)
694 		return 0;
695 
696 	mdelay(5);
697 
698 	return vce_v3_0_resume(adev);
699 }
700 
701 static int vce_v3_0_set_interrupt_state(struct amdgpu_device *adev,
702 					struct amdgpu_irq_src *source,
703 					unsigned type,
704 					enum amdgpu_interrupt_state state)
705 {
706 	uint32_t val = 0;
707 
708 	if (state == AMDGPU_IRQ_STATE_ENABLE)
709 		val |= VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK;
710 
711 	WREG32_P(mmVCE_SYS_INT_EN, val, ~VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK);
712 	return 0;
713 }
714 
715 static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
716 				      struct amdgpu_irq_src *source,
717 				      struct amdgpu_iv_entry *entry)
718 {
719 	DRM_DEBUG("IH: VCE\n");
720 
721 	WREG32_FIELD(VCE_SYS_INT_STATUS, VCE_SYS_INT_TRAP_INTERRUPT_INT, 1);
722 
723 	switch (entry->src_data[0]) {
724 	case 0:
725 	case 1:
726 	case 2:
727 		amdgpu_fence_process(&adev->vce.ring[entry->src_data[0]]);
728 		break;
729 	default:
730 		DRM_ERROR("Unhandled interrupt: %d %d\n",
731 			  entry->src_id, entry->src_data[0]);
732 		break;
733 	}
734 
735 	return 0;
736 }
737 
738 static int vce_v3_0_set_clockgating_state(void *handle,
739 					  enum amd_clockgating_state state)
740 {
741 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
742 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
743 	int i;
744 
745 	if (!(adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG))
746 		return 0;
747 
748 	mutex_lock(&adev->grbm_idx_mutex);
749 	for (i = 0; i < 2; i++) {
750 		/* Program VCE Instance 0 or 1 if not harvested */
751 		if (adev->vce.harvest_config & (1 << i))
752 			continue;
753 
754 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(i));
755 
756 		if (!enable) {
757 			/* initialize VCE_CLOCK_GATING_A: Clock ON/OFF delay */
758 			uint32_t data = RREG32(mmVCE_CLOCK_GATING_A);
759 			data &= ~(0xf | 0xff0);
760 			data |= ((0x0 << 0) | (0x04 << 4));
761 			WREG32(mmVCE_CLOCK_GATING_A, data);
762 
763 			/* initialize VCE_UENC_CLOCK_GATING: Clock ON/OFF delay */
764 			data = RREG32(mmVCE_UENC_CLOCK_GATING);
765 			data &= ~(0xf | 0xff0);
766 			data |= ((0x0 << 0) | (0x04 << 4));
767 			WREG32(mmVCE_UENC_CLOCK_GATING, data);
768 		}
769 
770 		vce_v3_0_set_vce_sw_clock_gating(adev, enable);
771 	}
772 
773 	WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
774 	mutex_unlock(&adev->grbm_idx_mutex);
775 
776 	return 0;
777 }
778 
779 static int vce_v3_0_set_powergating_state(void *handle,
780 					  enum amd_powergating_state state)
781 {
782 	/* This doesn't actually powergate the VCE block.
783 	 * That's done in the dpm code via the SMC.  This
784 	 * just re-inits the block as necessary.  The actual
785 	 * gating still happens in the dpm code.  We should
786 	 * revisit this when there is a cleaner line between
787 	 * the smc and the hw blocks
788 	 */
789 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
790 	int ret = 0;
791 
792 	if (state == AMD_PG_STATE_GATE) {
793 		ret = vce_v3_0_stop(adev);
794 		if (ret)
795 			goto out;
796 	} else {
797 		ret = vce_v3_0_start(adev);
798 		if (ret)
799 			goto out;
800 	}
801 
802 out:
803 	return ret;
804 }
805 
806 static void vce_v3_0_get_clockgating_state(void *handle, u32 *flags)
807 {
808 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
809 	int data;
810 
811 	mutex_lock(&adev->pm.mutex);
812 
813 	if (adev->flags & AMD_IS_APU)
814 		data = RREG32_SMC(ixCURRENT_PG_STATUS_APU);
815 	else
816 		data = RREG32_SMC(ixCURRENT_PG_STATUS);
817 
818 	if (data & CURRENT_PG_STATUS__VCE_PG_STATUS_MASK) {
819 		DRM_INFO("Cannot get clockgating state when VCE is powergated.\n");
820 		goto out;
821 	}
822 
823 	WREG32_FIELD(GRBM_GFX_INDEX, VCE_INSTANCE, 0);
824 
825 	/* AMD_CG_SUPPORT_VCE_MGCG */
826 	data = RREG32(mmVCE_CLOCK_GATING_A);
827 	if (data & (0x04 << 4))
828 		*flags |= AMD_CG_SUPPORT_VCE_MGCG;
829 
830 out:
831 	mutex_unlock(&adev->pm.mutex);
832 }
833 
834 static void vce_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
835 				  struct amdgpu_job *job,
836 				  struct amdgpu_ib *ib,
837 				  bool ctx_switch)
838 {
839 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
840 
841 	amdgpu_ring_write(ring, VCE_CMD_IB_VM);
842 	amdgpu_ring_write(ring, vmid);
843 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
844 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
845 	amdgpu_ring_write(ring, ib->length_dw);
846 }
847 
848 static void vce_v3_0_emit_vm_flush(struct amdgpu_ring *ring,
849 				   unsigned int vmid, uint64_t pd_addr)
850 {
851 	amdgpu_ring_write(ring, VCE_CMD_UPDATE_PTB);
852 	amdgpu_ring_write(ring, vmid);
853 	amdgpu_ring_write(ring, pd_addr >> 12);
854 
855 	amdgpu_ring_write(ring, VCE_CMD_FLUSH_TLB);
856 	amdgpu_ring_write(ring, vmid);
857 	amdgpu_ring_write(ring, VCE_CMD_END);
858 }
859 
860 static void vce_v3_0_emit_pipeline_sync(struct amdgpu_ring *ring)
861 {
862 	uint32_t seq = ring->fence_drv.sync_seq;
863 	uint64_t addr = ring->fence_drv.gpu_addr;
864 
865 	amdgpu_ring_write(ring, VCE_CMD_WAIT_GE);
866 	amdgpu_ring_write(ring, lower_32_bits(addr));
867 	amdgpu_ring_write(ring, upper_32_bits(addr));
868 	amdgpu_ring_write(ring, seq);
869 }
870 
871 static const struct amd_ip_funcs vce_v3_0_ip_funcs = {
872 	.name = "vce_v3_0",
873 	.early_init = vce_v3_0_early_init,
874 	.late_init = NULL,
875 	.sw_init = vce_v3_0_sw_init,
876 	.sw_fini = vce_v3_0_sw_fini,
877 	.hw_init = vce_v3_0_hw_init,
878 	.hw_fini = vce_v3_0_hw_fini,
879 	.suspend = vce_v3_0_suspend,
880 	.resume = vce_v3_0_resume,
881 	.is_idle = vce_v3_0_is_idle,
882 	.wait_for_idle = vce_v3_0_wait_for_idle,
883 	.check_soft_reset = vce_v3_0_check_soft_reset,
884 	.pre_soft_reset = vce_v3_0_pre_soft_reset,
885 	.soft_reset = vce_v3_0_soft_reset,
886 	.post_soft_reset = vce_v3_0_post_soft_reset,
887 	.set_clockgating_state = vce_v3_0_set_clockgating_state,
888 	.set_powergating_state = vce_v3_0_set_powergating_state,
889 	.get_clockgating_state = vce_v3_0_get_clockgating_state,
890 };
891 
892 static const struct amdgpu_ring_funcs vce_v3_0_ring_phys_funcs = {
893 	.type = AMDGPU_RING_TYPE_VCE,
894 	.align_mask = 0xf,
895 	.nop = VCE_CMD_NO_OP,
896 	.support_64bit_ptrs = false,
897 	.get_rptr = vce_v3_0_ring_get_rptr,
898 	.get_wptr = vce_v3_0_ring_get_wptr,
899 	.set_wptr = vce_v3_0_ring_set_wptr,
900 	.parse_cs = amdgpu_vce_ring_parse_cs,
901 	.emit_frame_size =
902 		4 + /* vce_v3_0_emit_pipeline_sync */
903 		6, /* amdgpu_vce_ring_emit_fence x1 no user fence */
904 	.emit_ib_size = 4, /* amdgpu_vce_ring_emit_ib */
905 	.emit_ib = amdgpu_vce_ring_emit_ib,
906 	.emit_fence = amdgpu_vce_ring_emit_fence,
907 	.test_ring = amdgpu_vce_ring_test_ring,
908 	.test_ib = amdgpu_vce_ring_test_ib,
909 	.insert_nop = amdgpu_ring_insert_nop,
910 	.pad_ib = amdgpu_ring_generic_pad_ib,
911 	.begin_use = amdgpu_vce_ring_begin_use,
912 	.end_use = amdgpu_vce_ring_end_use,
913 };
914 
915 static const struct amdgpu_ring_funcs vce_v3_0_ring_vm_funcs = {
916 	.type = AMDGPU_RING_TYPE_VCE,
917 	.align_mask = 0xf,
918 	.nop = VCE_CMD_NO_OP,
919 	.support_64bit_ptrs = false,
920 	.get_rptr = vce_v3_0_ring_get_rptr,
921 	.get_wptr = vce_v3_0_ring_get_wptr,
922 	.set_wptr = vce_v3_0_ring_set_wptr,
923 	.parse_cs = amdgpu_vce_ring_parse_cs_vm,
924 	.emit_frame_size =
925 		6 + /* vce_v3_0_emit_vm_flush */
926 		4 + /* vce_v3_0_emit_pipeline_sync */
927 		6 + 6, /* amdgpu_vce_ring_emit_fence x2 vm fence */
928 	.emit_ib_size = 5, /* vce_v3_0_ring_emit_ib */
929 	.emit_ib = vce_v3_0_ring_emit_ib,
930 	.emit_vm_flush = vce_v3_0_emit_vm_flush,
931 	.emit_pipeline_sync = vce_v3_0_emit_pipeline_sync,
932 	.emit_fence = amdgpu_vce_ring_emit_fence,
933 	.test_ring = amdgpu_vce_ring_test_ring,
934 	.test_ib = amdgpu_vce_ring_test_ib,
935 	.insert_nop = amdgpu_ring_insert_nop,
936 	.pad_ib = amdgpu_ring_generic_pad_ib,
937 	.begin_use = amdgpu_vce_ring_begin_use,
938 	.end_use = amdgpu_vce_ring_end_use,
939 };
940 
941 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev)
942 {
943 	int i;
944 
945 	if (adev->asic_type >= CHIP_STONEY) {
946 		for (i = 0; i < adev->vce.num_rings; i++) {
947 			adev->vce.ring[i].funcs = &vce_v3_0_ring_vm_funcs;
948 			adev->vce.ring[i].me = i;
949 		}
950 		DRM_INFO("VCE enabled in VM mode\n");
951 	} else {
952 		for (i = 0; i < adev->vce.num_rings; i++) {
953 			adev->vce.ring[i].funcs = &vce_v3_0_ring_phys_funcs;
954 			adev->vce.ring[i].me = i;
955 		}
956 		DRM_INFO("VCE enabled in physical mode\n");
957 	}
958 }
959 
960 static const struct amdgpu_irq_src_funcs vce_v3_0_irq_funcs = {
961 	.set = vce_v3_0_set_interrupt_state,
962 	.process = vce_v3_0_process_interrupt,
963 };
964 
965 static void vce_v3_0_set_irq_funcs(struct amdgpu_device *adev)
966 {
967 	adev->vce.irq.num_types = 1;
968 	adev->vce.irq.funcs = &vce_v3_0_irq_funcs;
969 };
970 
971 const struct amdgpu_ip_block_version vce_v3_0_ip_block =
972 {
973 	.type = AMD_IP_BLOCK_TYPE_VCE,
974 	.major = 3,
975 	.minor = 0,
976 	.rev = 0,
977 	.funcs = &vce_v3_0_ip_funcs,
978 };
979 
980 const struct amdgpu_ip_block_version vce_v3_1_ip_block =
981 {
982 	.type = AMD_IP_BLOCK_TYPE_VCE,
983 	.major = 3,
984 	.minor = 1,
985 	.rev = 0,
986 	.funcs = &vce_v3_0_ip_funcs,
987 };
988 
989 const struct amdgpu_ip_block_version vce_v3_4_ip_block =
990 {
991 	.type = AMD_IP_BLOCK_TYPE_VCE,
992 	.major = 3,
993 	.minor = 4,
994 	.rev = 0,
995 	.funcs = &vce_v3_0_ip_funcs,
996 };
997