xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/vce_v3_0.c (revision 9dbbc3b9)
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 
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 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
447 		if (r)
448 			return r;
449 	}
450 
451 	r = amdgpu_vce_entity_init(adev);
452 
453 	return r;
454 }
455 
456 static int vce_v3_0_sw_fini(void *handle)
457 {
458 	int r;
459 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
460 
461 	r = amdgpu_vce_suspend(adev);
462 	if (r)
463 		return r;
464 
465 	return amdgpu_vce_sw_fini(adev);
466 }
467 
468 static int vce_v3_0_hw_init(void *handle)
469 {
470 	int r, i;
471 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
472 
473 	vce_v3_0_override_vce_clock_gating(adev, true);
474 
475 	amdgpu_asic_set_vce_clocks(adev, 10000, 10000);
476 
477 	for (i = 0; i < adev->vce.num_rings; i++) {
478 		r = amdgpu_ring_test_helper(&adev->vce.ring[i]);
479 		if (r)
480 			return r;
481 	}
482 
483 	DRM_INFO("VCE initialized successfully.\n");
484 
485 	return 0;
486 }
487 
488 static int vce_v3_0_hw_fini(void *handle)
489 {
490 	int r;
491 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
492 
493 	r = vce_v3_0_wait_for_idle(handle);
494 	if (r)
495 		return r;
496 
497 	vce_v3_0_stop(adev);
498 	return vce_v3_0_set_clockgating_state(adev, AMD_CG_STATE_GATE);
499 }
500 
501 static int vce_v3_0_suspend(void *handle)
502 {
503 	int r;
504 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
505 
506 	r = vce_v3_0_hw_fini(adev);
507 	if (r)
508 		return r;
509 
510 	return amdgpu_vce_suspend(adev);
511 }
512 
513 static int vce_v3_0_resume(void *handle)
514 {
515 	int r;
516 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
517 
518 	r = amdgpu_vce_resume(adev);
519 	if (r)
520 		return r;
521 
522 	return vce_v3_0_hw_init(adev);
523 }
524 
525 static void vce_v3_0_mc_resume(struct amdgpu_device *adev, int idx)
526 {
527 	uint32_t offset, size;
528 
529 	WREG32_P(mmVCE_CLOCK_GATING_A, 0, ~(1 << 16));
530 	WREG32_P(mmVCE_UENC_CLOCK_GATING, 0x1FF000, ~0xFF9FF000);
531 	WREG32_P(mmVCE_UENC_REG_CLOCK_GATING, 0x3F, ~0x3F);
532 	WREG32(mmVCE_CLOCK_GATING_B, 0x1FF);
533 
534 	WREG32(mmVCE_LMI_CTRL, 0x00398000);
535 	WREG32_P(mmVCE_LMI_CACHE_CTRL, 0x0, ~0x1);
536 	WREG32(mmVCE_LMI_SWAP_CNTL, 0);
537 	WREG32(mmVCE_LMI_SWAP_CNTL1, 0);
538 	WREG32(mmVCE_LMI_VM_CTRL, 0);
539 	WREG32_OR(mmVCE_VCPU_CNTL, 0x00100000);
540 
541 	if (adev->asic_type >= CHIP_STONEY) {
542 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR0, (adev->vce.gpu_addr >> 8));
543 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR1, (adev->vce.gpu_addr >> 8));
544 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR2, (adev->vce.gpu_addr >> 8));
545 	} else
546 		WREG32(mmVCE_LMI_VCPU_CACHE_40BIT_BAR, (adev->vce.gpu_addr >> 8));
547 	offset = AMDGPU_VCE_FIRMWARE_OFFSET;
548 	size = VCE_V3_0_FW_SIZE;
549 	WREG32(mmVCE_VCPU_CACHE_OFFSET0, offset & 0x7fffffff);
550 	WREG32(mmVCE_VCPU_CACHE_SIZE0, size);
551 
552 	if (idx == 0) {
553 		offset += size;
554 		size = VCE_V3_0_STACK_SIZE;
555 		WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0x7fffffff);
556 		WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
557 		offset += size;
558 		size = VCE_V3_0_DATA_SIZE;
559 		WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0x7fffffff);
560 		WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
561 	} else {
562 		offset += size + VCE_V3_0_STACK_SIZE + VCE_V3_0_DATA_SIZE;
563 		size = VCE_V3_0_STACK_SIZE;
564 		WREG32(mmVCE_VCPU_CACHE_OFFSET1, offset & 0xfffffff);
565 		WREG32(mmVCE_VCPU_CACHE_SIZE1, size);
566 		offset += size;
567 		size = VCE_V3_0_DATA_SIZE;
568 		WREG32(mmVCE_VCPU_CACHE_OFFSET2, offset & 0xfffffff);
569 		WREG32(mmVCE_VCPU_CACHE_SIZE2, size);
570 	}
571 
572 	WREG32_P(mmVCE_LMI_CTRL2, 0x0, ~0x100);
573 	WREG32_FIELD(VCE_SYS_INT_EN, VCE_SYS_INT_TRAP_INTERRUPT_EN, 1);
574 }
575 
576 static bool vce_v3_0_is_idle(void *handle)
577 {
578 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
579 	u32 mask = 0;
580 
581 	mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_STATUS2__VCE0_BUSY_MASK;
582 	mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_STATUS2__VCE1_BUSY_MASK;
583 
584 	return !(RREG32(mmSRBM_STATUS2) & mask);
585 }
586 
587 static int vce_v3_0_wait_for_idle(void *handle)
588 {
589 	unsigned i;
590 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
591 
592 	for (i = 0; i < adev->usec_timeout; i++)
593 		if (vce_v3_0_is_idle(handle))
594 			return 0;
595 
596 	return -ETIMEDOUT;
597 }
598 
599 #define  VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK  0x00000008L   /* AUTO_BUSY */
600 #define  VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK   0x00000010L   /* RB0_BUSY */
601 #define  VCE_STATUS_VCPU_REPORT_RB1_BUSY_MASK   0x00000020L   /* RB1_BUSY */
602 #define  AMDGPU_VCE_STATUS_BUSY_MASK (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK | \
603 				      VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK)
604 
605 static bool vce_v3_0_check_soft_reset(void *handle)
606 {
607 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
608 	u32 srbm_soft_reset = 0;
609 
610 	/* According to VCE team , we should use VCE_STATUS instead
611 	 * SRBM_STATUS.VCE_BUSY bit for busy status checking.
612 	 * GRBM_GFX_INDEX.INSTANCE_INDEX is used to specify which VCE
613 	 * instance's registers are accessed
614 	 * (0 for 1st instance, 10 for 2nd instance).
615 	 *
616 	 *VCE_STATUS
617 	 *|UENC|ACPI|AUTO ACTIVE|RB1 |RB0 |RB2 |          |FW_LOADED|JOB |
618 	 *|----+----+-----------+----+----+----+----------+---------+----|
619 	 *|bit8|bit7|    bit6   |bit5|bit4|bit3|   bit2   |  bit1   |bit0|
620 	 *
621 	 * VCE team suggest use bit 3--bit 6 for busy status check
622 	 */
623 	mutex_lock(&adev->grbm_idx_mutex);
624 	WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
625 	if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
626 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
627 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
628 	}
629 	WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(1));
630 	if (RREG32(mmVCE_STATUS) & AMDGPU_VCE_STATUS_BUSY_MASK) {
631 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE0, 1);
632 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET, SOFT_RESET_VCE1, 1);
633 	}
634 	WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(0));
635 	mutex_unlock(&adev->grbm_idx_mutex);
636 
637 	if (srbm_soft_reset) {
638 		adev->vce.srbm_soft_reset = srbm_soft_reset;
639 		return true;
640 	} else {
641 		adev->vce.srbm_soft_reset = 0;
642 		return false;
643 	}
644 }
645 
646 static int vce_v3_0_soft_reset(void *handle)
647 {
648 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
649 	u32 srbm_soft_reset;
650 
651 	if (!adev->vce.srbm_soft_reset)
652 		return 0;
653 	srbm_soft_reset = adev->vce.srbm_soft_reset;
654 
655 	if (srbm_soft_reset) {
656 		u32 tmp;
657 
658 		tmp = RREG32(mmSRBM_SOFT_RESET);
659 		tmp |= srbm_soft_reset;
660 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
661 		WREG32(mmSRBM_SOFT_RESET, tmp);
662 		tmp = RREG32(mmSRBM_SOFT_RESET);
663 
664 		udelay(50);
665 
666 		tmp &= ~srbm_soft_reset;
667 		WREG32(mmSRBM_SOFT_RESET, tmp);
668 		tmp = RREG32(mmSRBM_SOFT_RESET);
669 
670 		/* Wait a little for things to settle down */
671 		udelay(50);
672 	}
673 
674 	return 0;
675 }
676 
677 static int vce_v3_0_pre_soft_reset(void *handle)
678 {
679 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
680 
681 	if (!adev->vce.srbm_soft_reset)
682 		return 0;
683 
684 	mdelay(5);
685 
686 	return vce_v3_0_suspend(adev);
687 }
688 
689 
690 static int vce_v3_0_post_soft_reset(void *handle)
691 {
692 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
693 
694 	if (!adev->vce.srbm_soft_reset)
695 		return 0;
696 
697 	mdelay(5);
698 
699 	return vce_v3_0_resume(adev);
700 }
701 
702 static int vce_v3_0_set_interrupt_state(struct amdgpu_device *adev,
703 					struct amdgpu_irq_src *source,
704 					unsigned type,
705 					enum amdgpu_interrupt_state state)
706 {
707 	uint32_t val = 0;
708 
709 	if (state == AMDGPU_IRQ_STATE_ENABLE)
710 		val |= VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK;
711 
712 	WREG32_P(mmVCE_SYS_INT_EN, val, ~VCE_SYS_INT_EN__VCE_SYS_INT_TRAP_INTERRUPT_EN_MASK);
713 	return 0;
714 }
715 
716 static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
717 				      struct amdgpu_irq_src *source,
718 				      struct amdgpu_iv_entry *entry)
719 {
720 	DRM_DEBUG("IH: VCE\n");
721 
722 	WREG32_FIELD(VCE_SYS_INT_STATUS, VCE_SYS_INT_TRAP_INTERRUPT_INT, 1);
723 
724 	switch (entry->src_data[0]) {
725 	case 0:
726 	case 1:
727 	case 2:
728 		amdgpu_fence_process(&adev->vce.ring[entry->src_data[0]]);
729 		break;
730 	default:
731 		DRM_ERROR("Unhandled interrupt: %d %d\n",
732 			  entry->src_id, entry->src_data[0]);
733 		break;
734 	}
735 
736 	return 0;
737 }
738 
739 static int vce_v3_0_set_clockgating_state(void *handle,
740 					  enum amd_clockgating_state state)
741 {
742 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
743 	bool enable = (state == AMD_CG_STATE_GATE);
744 	int i;
745 
746 	if (!(adev->cg_flags & AMD_CG_SUPPORT_VCE_MGCG))
747 		return 0;
748 
749 	mutex_lock(&adev->grbm_idx_mutex);
750 	for (i = 0; i < 2; i++) {
751 		/* Program VCE Instance 0 or 1 if not harvested */
752 		if (adev->vce.harvest_config & (1 << i))
753 			continue;
754 
755 		WREG32(mmGRBM_GFX_INDEX, GET_VCE_INSTANCE(i));
756 
757 		if (!enable) {
758 			/* initialize VCE_CLOCK_GATING_A: Clock ON/OFF delay */
759 			uint32_t data = RREG32(mmVCE_CLOCK_GATING_A);
760 			data &= ~(0xf | 0xff0);
761 			data |= ((0x0 << 0) | (0x04 << 4));
762 			WREG32(mmVCE_CLOCK_GATING_A, data);
763 
764 			/* initialize VCE_UENC_CLOCK_GATING: Clock ON/OFF delay */
765 			data = RREG32(mmVCE_UENC_CLOCK_GATING);
766 			data &= ~(0xf | 0xff0);
767 			data |= ((0x0 << 0) | (0x04 << 4));
768 			WREG32(mmVCE_UENC_CLOCK_GATING, data);
769 		}
770 
771 		vce_v3_0_set_vce_sw_clock_gating(adev, enable);
772 	}
773 
774 	WREG32(mmGRBM_GFX_INDEX, mmGRBM_GFX_INDEX_DEFAULT);
775 	mutex_unlock(&adev->grbm_idx_mutex);
776 
777 	return 0;
778 }
779 
780 static int vce_v3_0_set_powergating_state(void *handle,
781 					  enum amd_powergating_state state)
782 {
783 	/* This doesn't actually powergate the VCE block.
784 	 * That's done in the dpm code via the SMC.  This
785 	 * just re-inits the block as necessary.  The actual
786 	 * gating still happens in the dpm code.  We should
787 	 * revisit this when there is a cleaner line between
788 	 * the smc and the hw blocks
789 	 */
790 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
791 	int ret = 0;
792 
793 	if (state == AMD_PG_STATE_GATE) {
794 		ret = vce_v3_0_stop(adev);
795 		if (ret)
796 			goto out;
797 	} else {
798 		ret = vce_v3_0_start(adev);
799 		if (ret)
800 			goto out;
801 	}
802 
803 out:
804 	return ret;
805 }
806 
807 static void vce_v3_0_get_clockgating_state(void *handle, u32 *flags)
808 {
809 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
810 	int data;
811 
812 	mutex_lock(&adev->pm.mutex);
813 
814 	if (adev->flags & AMD_IS_APU)
815 		data = RREG32_SMC(ixCURRENT_PG_STATUS_APU);
816 	else
817 		data = RREG32_SMC(ixCURRENT_PG_STATUS);
818 
819 	if (data & CURRENT_PG_STATUS__VCE_PG_STATUS_MASK) {
820 		DRM_INFO("Cannot get clockgating state when VCE is powergated.\n");
821 		goto out;
822 	}
823 
824 	WREG32_FIELD(GRBM_GFX_INDEX, VCE_INSTANCE, 0);
825 
826 	/* AMD_CG_SUPPORT_VCE_MGCG */
827 	data = RREG32(mmVCE_CLOCK_GATING_A);
828 	if (data & (0x04 << 4))
829 		*flags |= AMD_CG_SUPPORT_VCE_MGCG;
830 
831 out:
832 	mutex_unlock(&adev->pm.mutex);
833 }
834 
835 static void vce_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
836 				  struct amdgpu_job *job,
837 				  struct amdgpu_ib *ib,
838 				  uint32_t flags)
839 {
840 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
841 
842 	amdgpu_ring_write(ring, VCE_CMD_IB_VM);
843 	amdgpu_ring_write(ring, vmid);
844 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
845 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
846 	amdgpu_ring_write(ring, ib->length_dw);
847 }
848 
849 static void vce_v3_0_emit_vm_flush(struct amdgpu_ring *ring,
850 				   unsigned int vmid, uint64_t pd_addr)
851 {
852 	amdgpu_ring_write(ring, VCE_CMD_UPDATE_PTB);
853 	amdgpu_ring_write(ring, vmid);
854 	amdgpu_ring_write(ring, pd_addr >> 12);
855 
856 	amdgpu_ring_write(ring, VCE_CMD_FLUSH_TLB);
857 	amdgpu_ring_write(ring, vmid);
858 	amdgpu_ring_write(ring, VCE_CMD_END);
859 }
860 
861 static void vce_v3_0_emit_pipeline_sync(struct amdgpu_ring *ring)
862 {
863 	uint32_t seq = ring->fence_drv.sync_seq;
864 	uint64_t addr = ring->fence_drv.gpu_addr;
865 
866 	amdgpu_ring_write(ring, VCE_CMD_WAIT_GE);
867 	amdgpu_ring_write(ring, lower_32_bits(addr));
868 	amdgpu_ring_write(ring, upper_32_bits(addr));
869 	amdgpu_ring_write(ring, seq);
870 }
871 
872 static const struct amd_ip_funcs vce_v3_0_ip_funcs = {
873 	.name = "vce_v3_0",
874 	.early_init = vce_v3_0_early_init,
875 	.late_init = NULL,
876 	.sw_init = vce_v3_0_sw_init,
877 	.sw_fini = vce_v3_0_sw_fini,
878 	.hw_init = vce_v3_0_hw_init,
879 	.hw_fini = vce_v3_0_hw_fini,
880 	.suspend = vce_v3_0_suspend,
881 	.resume = vce_v3_0_resume,
882 	.is_idle = vce_v3_0_is_idle,
883 	.wait_for_idle = vce_v3_0_wait_for_idle,
884 	.check_soft_reset = vce_v3_0_check_soft_reset,
885 	.pre_soft_reset = vce_v3_0_pre_soft_reset,
886 	.soft_reset = vce_v3_0_soft_reset,
887 	.post_soft_reset = vce_v3_0_post_soft_reset,
888 	.set_clockgating_state = vce_v3_0_set_clockgating_state,
889 	.set_powergating_state = vce_v3_0_set_powergating_state,
890 	.get_clockgating_state = vce_v3_0_get_clockgating_state,
891 };
892 
893 static const struct amdgpu_ring_funcs vce_v3_0_ring_phys_funcs = {
894 	.type = AMDGPU_RING_TYPE_VCE,
895 	.align_mask = 0xf,
896 	.nop = VCE_CMD_NO_OP,
897 	.support_64bit_ptrs = false,
898 	.no_user_fence = true,
899 	.get_rptr = vce_v3_0_ring_get_rptr,
900 	.get_wptr = vce_v3_0_ring_get_wptr,
901 	.set_wptr = vce_v3_0_ring_set_wptr,
902 	.parse_cs = amdgpu_vce_ring_parse_cs,
903 	.emit_frame_size =
904 		4 + /* vce_v3_0_emit_pipeline_sync */
905 		6, /* amdgpu_vce_ring_emit_fence x1 no user fence */
906 	.emit_ib_size = 4, /* amdgpu_vce_ring_emit_ib */
907 	.emit_ib = amdgpu_vce_ring_emit_ib,
908 	.emit_fence = amdgpu_vce_ring_emit_fence,
909 	.test_ring = amdgpu_vce_ring_test_ring,
910 	.test_ib = amdgpu_vce_ring_test_ib,
911 	.insert_nop = amdgpu_ring_insert_nop,
912 	.pad_ib = amdgpu_ring_generic_pad_ib,
913 	.begin_use = amdgpu_vce_ring_begin_use,
914 	.end_use = amdgpu_vce_ring_end_use,
915 };
916 
917 static const struct amdgpu_ring_funcs vce_v3_0_ring_vm_funcs = {
918 	.type = AMDGPU_RING_TYPE_VCE,
919 	.align_mask = 0xf,
920 	.nop = VCE_CMD_NO_OP,
921 	.support_64bit_ptrs = false,
922 	.no_user_fence = true,
923 	.get_rptr = vce_v3_0_ring_get_rptr,
924 	.get_wptr = vce_v3_0_ring_get_wptr,
925 	.set_wptr = vce_v3_0_ring_set_wptr,
926 	.parse_cs = amdgpu_vce_ring_parse_cs_vm,
927 	.emit_frame_size =
928 		6 + /* vce_v3_0_emit_vm_flush */
929 		4 + /* vce_v3_0_emit_pipeline_sync */
930 		6 + 6, /* amdgpu_vce_ring_emit_fence x2 vm fence */
931 	.emit_ib_size = 5, /* vce_v3_0_ring_emit_ib */
932 	.emit_ib = vce_v3_0_ring_emit_ib,
933 	.emit_vm_flush = vce_v3_0_emit_vm_flush,
934 	.emit_pipeline_sync = vce_v3_0_emit_pipeline_sync,
935 	.emit_fence = amdgpu_vce_ring_emit_fence,
936 	.test_ring = amdgpu_vce_ring_test_ring,
937 	.test_ib = amdgpu_vce_ring_test_ib,
938 	.insert_nop = amdgpu_ring_insert_nop,
939 	.pad_ib = amdgpu_ring_generic_pad_ib,
940 	.begin_use = amdgpu_vce_ring_begin_use,
941 	.end_use = amdgpu_vce_ring_end_use,
942 };
943 
944 static void vce_v3_0_set_ring_funcs(struct amdgpu_device *adev)
945 {
946 	int i;
947 
948 	if (adev->asic_type >= CHIP_STONEY) {
949 		for (i = 0; i < adev->vce.num_rings; i++) {
950 			adev->vce.ring[i].funcs = &vce_v3_0_ring_vm_funcs;
951 			adev->vce.ring[i].me = i;
952 		}
953 		DRM_INFO("VCE enabled in VM mode\n");
954 	} else {
955 		for (i = 0; i < adev->vce.num_rings; i++) {
956 			adev->vce.ring[i].funcs = &vce_v3_0_ring_phys_funcs;
957 			adev->vce.ring[i].me = i;
958 		}
959 		DRM_INFO("VCE enabled in physical mode\n");
960 	}
961 }
962 
963 static const struct amdgpu_irq_src_funcs vce_v3_0_irq_funcs = {
964 	.set = vce_v3_0_set_interrupt_state,
965 	.process = vce_v3_0_process_interrupt,
966 };
967 
968 static void vce_v3_0_set_irq_funcs(struct amdgpu_device *adev)
969 {
970 	adev->vce.irq.num_types = 1;
971 	adev->vce.irq.funcs = &vce_v3_0_irq_funcs;
972 };
973 
974 const struct amdgpu_ip_block_version vce_v3_0_ip_block =
975 {
976 	.type = AMD_IP_BLOCK_TYPE_VCE,
977 	.major = 3,
978 	.minor = 0,
979 	.rev = 0,
980 	.funcs = &vce_v3_0_ip_funcs,
981 };
982 
983 const struct amdgpu_ip_block_version vce_v3_1_ip_block =
984 {
985 	.type = AMD_IP_BLOCK_TYPE_VCE,
986 	.major = 3,
987 	.minor = 1,
988 	.rev = 0,
989 	.funcs = &vce_v3_0_ip_funcs,
990 };
991 
992 const struct amdgpu_ip_block_version vce_v3_4_ip_block =
993 {
994 	.type = AMD_IP_BLOCK_TYPE_VCE,
995 	.major = 3,
996 	.minor = 4,
997 	.rev = 0,
998 	.funcs = &vce_v3_0_ip_funcs,
999 };
1000