xref: /openbsd/sys/dev/pci/drm/radeon/radeon_vce.c (revision f005ef32)
1 /*
2  * Copyright 2013 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 <linux/module.h>
30 
31 #include <drm/drm.h>
32 
33 #include "radeon.h"
34 #include "radeon_asic.h"
35 #include "sid.h"
36 
37 /* 1 second timeout */
38 #define VCE_IDLE_TIMEOUT_MS	1000
39 
40 /* Firmware Names */
41 #define FIRMWARE_TAHITI	"radeon/TAHITI_vce.bin"
42 #define FIRMWARE_BONAIRE	"radeon/BONAIRE_vce.bin"
43 
44 MODULE_FIRMWARE(FIRMWARE_TAHITI);
45 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
46 
47 static void radeon_vce_idle_work_handler(struct work_struct *work);
48 
49 /**
50  * radeon_vce_init - allocate memory, load vce firmware
51  *
52  * @rdev: radeon_device pointer
53  *
54  * First step to get VCE online, allocate memory and load the firmware
55  */
radeon_vce_init(struct radeon_device * rdev)56 int radeon_vce_init(struct radeon_device *rdev)
57 {
58 	static const char *fw_version = "[ATI LIB=VCEFW,";
59 	static const char *fb_version = "[ATI LIB=VCEFWSTATS,";
60 	unsigned long size;
61 	const char *fw_name, *c;
62 	uint8_t start, mid, end;
63 	int i, r;
64 
65 	INIT_DELAYED_WORK(&rdev->vce.idle_work, radeon_vce_idle_work_handler);
66 
67 	switch (rdev->family) {
68 	case CHIP_TAHITI:
69 	case CHIP_PITCAIRN:
70 	case CHIP_VERDE:
71 	case CHIP_ARUBA:
72 		fw_name = FIRMWARE_TAHITI;
73 		break;
74 
75 	case CHIP_BONAIRE:
76 	case CHIP_KAVERI:
77 	case CHIP_KABINI:
78 	case CHIP_HAWAII:
79 	case CHIP_MULLINS:
80 		fw_name = FIRMWARE_BONAIRE;
81 		break;
82 
83 	default:
84 		return -EINVAL;
85 	}
86 
87 	r = request_firmware(&rdev->vce_fw, fw_name, rdev->dev);
88 	if (r) {
89 		dev_err(rdev->dev, "radeon_vce: Can't load firmware \"%s\"\n",
90 			fw_name);
91 		return r;
92 	}
93 
94 	/* search for firmware version */
95 
96 	size = rdev->vce_fw->size - strlen(fw_version) - 9;
97 	c = rdev->vce_fw->data;
98 	for (; size > 0; --size, ++c)
99 		if (strncmp(c, fw_version, strlen(fw_version)) == 0)
100 			break;
101 
102 	if (size == 0)
103 		return -EINVAL;
104 
105 	c += strlen(fw_version);
106 #ifdef notyet
107 	if (sscanf(c, "%2hhd.%2hhd.%2hhd]", &start, &mid, &end) != 3)
108 		return -EINVAL;
109 #else
110 	if (c[2] != '.') {
111 		printf("%s %s bad start value\n", rdev->self.dv_xname, __func__);
112 		return -EINVAL;
113 	}
114 	start = (10 * (c[0] - '0')) + (c[1] - '0');
115 	c += 3;
116 
117 	if (c[1] != '.') {
118 		printf("%s %s bad mid value\n", rdev->self.dv_xname, __func__);
119 		return -EINVAL;
120 	}
121 	mid = c[0] - '0';
122 	c += 2;
123 
124 	if (c[1] != ']') {
125 		printf("%s %s bad end value\n", rdev->self.dv_xname, __func__);
126 		return -EINVAL;
127 	}
128 	end = c[0] - '0';
129 #endif
130 
131 	/* search for feedback version */
132 
133 	size = rdev->vce_fw->size - strlen(fb_version) - 3;
134 	c = rdev->vce_fw->data;
135 	for (; size > 0; --size, ++c)
136 		if (strncmp(c, fb_version, strlen(fb_version)) == 0)
137 			break;
138 
139 	if (size == 0)
140 		return -EINVAL;
141 
142 	c += strlen(fb_version);
143 #ifdef notyet
144 	if (sscanf(c, "%2u]", &rdev->vce.fb_version) != 1)
145 		return -EINVAL;
146 #else
147 	if (c[2] != ']') {
148 		printf("%s %s bad fb_version value\n", rdev->self.dv_xname, __func__);
149 		return -EINVAL;
150 	}
151 	rdev->vce.fb_version = (10 * (c[0] - '0')) + (c[1] - '0');
152 #endif
153 
154 	DRM_INFO("Found VCE firmware/feedback version %d.%d.%d / %d!\n",
155 		 start, mid, end, rdev->vce.fb_version);
156 
157 	rdev->vce.fw_version = (start << 24) | (mid << 16) | (end << 8);
158 
159 	/* we can only work with this fw version for now */
160 	if ((rdev->vce.fw_version != ((40 << 24) | (2 << 16) | (2 << 8))) &&
161 	    (rdev->vce.fw_version != ((50 << 24) | (0 << 16) | (1 << 8))) &&
162 	    (rdev->vce.fw_version != ((50 << 24) | (1 << 16) | (2 << 8))))
163 		return -EINVAL;
164 
165 	/* allocate firmware, stack and heap BO */
166 
167 	if (rdev->family < CHIP_BONAIRE)
168 		size = vce_v1_0_bo_size(rdev);
169 	else
170 		size = vce_v2_0_bo_size(rdev);
171 	r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
172 			     RADEON_GEM_DOMAIN_VRAM, 0, NULL, NULL,
173 			     &rdev->vce.vcpu_bo);
174 	if (r) {
175 		dev_err(rdev->dev, "(%d) failed to allocate VCE bo\n", r);
176 		return r;
177 	}
178 
179 	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
180 	if (r) {
181 		radeon_bo_unref(&rdev->vce.vcpu_bo);
182 		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
183 		return r;
184 	}
185 
186 	r = radeon_bo_pin(rdev->vce.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
187 			  &rdev->vce.gpu_addr);
188 	radeon_bo_unreserve(rdev->vce.vcpu_bo);
189 	if (r) {
190 		radeon_bo_unref(&rdev->vce.vcpu_bo);
191 		dev_err(rdev->dev, "(%d) VCE bo pin failed\n", r);
192 		return r;
193 	}
194 
195 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
196 		atomic_set(&rdev->vce.handles[i], 0);
197 		rdev->vce.filp[i] = NULL;
198 	}
199 
200 	return 0;
201 }
202 
203 /**
204  * radeon_vce_fini - free memory
205  *
206  * @rdev: radeon_device pointer
207  *
208  * Last step on VCE teardown, free firmware memory
209  */
radeon_vce_fini(struct radeon_device * rdev)210 void radeon_vce_fini(struct radeon_device *rdev)
211 {
212 	if (rdev->vce.vcpu_bo == NULL)
213 		return;
214 
215 	radeon_bo_unref(&rdev->vce.vcpu_bo);
216 
217 	release_firmware(rdev->vce_fw);
218 }
219 
220 /**
221  * radeon_vce_suspend - unpin VCE fw memory
222  *
223  * @rdev: radeon_device pointer
224  *
225  */
radeon_vce_suspend(struct radeon_device * rdev)226 int radeon_vce_suspend(struct radeon_device *rdev)
227 {
228 	int i;
229 
230 	if (rdev->vce.vcpu_bo == NULL)
231 		return 0;
232 
233 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
234 		if (atomic_read(&rdev->vce.handles[i]))
235 			break;
236 
237 	if (i == RADEON_MAX_VCE_HANDLES)
238 		return 0;
239 
240 	/* TODO: suspending running encoding sessions isn't supported */
241 	return -EINVAL;
242 }
243 
244 /**
245  * radeon_vce_resume - pin VCE fw memory
246  *
247  * @rdev: radeon_device pointer
248  *
249  */
radeon_vce_resume(struct radeon_device * rdev)250 int radeon_vce_resume(struct radeon_device *rdev)
251 {
252 	void *cpu_addr;
253 	int r;
254 
255 	if (rdev->vce.vcpu_bo == NULL)
256 		return -EINVAL;
257 
258 	r = radeon_bo_reserve(rdev->vce.vcpu_bo, false);
259 	if (r) {
260 		dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r);
261 		return r;
262 	}
263 
264 	r = radeon_bo_kmap(rdev->vce.vcpu_bo, &cpu_addr);
265 	if (r) {
266 		radeon_bo_unreserve(rdev->vce.vcpu_bo);
267 		dev_err(rdev->dev, "(%d) VCE map failed\n", r);
268 		return r;
269 	}
270 
271 	memset(cpu_addr, 0, radeon_bo_size(rdev->vce.vcpu_bo));
272 	if (rdev->family < CHIP_BONAIRE)
273 		r = vce_v1_0_load_fw(rdev, cpu_addr);
274 	else
275 		memcpy(cpu_addr, rdev->vce_fw->data, rdev->vce_fw->size);
276 
277 	radeon_bo_kunmap(rdev->vce.vcpu_bo);
278 
279 	radeon_bo_unreserve(rdev->vce.vcpu_bo);
280 
281 	return r;
282 }
283 
284 /**
285  * radeon_vce_idle_work_handler - power off VCE
286  *
287  * @work: pointer to work structure
288  *
289  * power of VCE when it's not used any more
290  */
radeon_vce_idle_work_handler(struct work_struct * work)291 static void radeon_vce_idle_work_handler(struct work_struct *work)
292 {
293 	struct radeon_device *rdev =
294 		container_of(work, struct radeon_device, vce.idle_work.work);
295 
296 	if ((radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE1_INDEX) == 0) &&
297 	    (radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE2_INDEX) == 0)) {
298 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
299 			radeon_dpm_enable_vce(rdev, false);
300 		} else {
301 			radeon_set_vce_clocks(rdev, 0, 0);
302 		}
303 	} else {
304 		schedule_delayed_work(&rdev->vce.idle_work,
305 				      msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
306 	}
307 }
308 
309 /**
310  * radeon_vce_note_usage - power up VCE
311  *
312  * @rdev: radeon_device pointer
313  *
314  * Make sure VCE is powerd up when we want to use it
315  */
radeon_vce_note_usage(struct radeon_device * rdev)316 void radeon_vce_note_usage(struct radeon_device *rdev)
317 {
318 	bool streams_changed = false;
319 	bool set_clocks = !cancel_delayed_work_sync(&rdev->vce.idle_work);
320 	set_clocks &= schedule_delayed_work(&rdev->vce.idle_work,
321 					    msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS));
322 
323 	if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
324 		/* XXX figure out if the streams changed */
325 		streams_changed = false;
326 	}
327 
328 	if (set_clocks || streams_changed) {
329 		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
330 			radeon_dpm_enable_vce(rdev, true);
331 		} else {
332 			radeon_set_vce_clocks(rdev, 53300, 40000);
333 		}
334 	}
335 }
336 
337 /**
338  * radeon_vce_free_handles - free still open VCE handles
339  *
340  * @rdev: radeon_device pointer
341  * @filp: drm file pointer
342  *
343  * Close all VCE handles still open by this file pointer
344  */
radeon_vce_free_handles(struct radeon_device * rdev,struct drm_file * filp)345 void radeon_vce_free_handles(struct radeon_device *rdev, struct drm_file *filp)
346 {
347 	int i, r;
348 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
349 		uint32_t handle = atomic_read(&rdev->vce.handles[i]);
350 		if (!handle || rdev->vce.filp[i] != filp)
351 			continue;
352 
353 		radeon_vce_note_usage(rdev);
354 
355 		r = radeon_vce_get_destroy_msg(rdev, TN_RING_TYPE_VCE1_INDEX,
356 					       handle, NULL);
357 		if (r)
358 			DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
359 
360 		rdev->vce.filp[i] = NULL;
361 		atomic_set(&rdev->vce.handles[i], 0);
362 	}
363 }
364 
365 /**
366  * radeon_vce_get_create_msg - generate a VCE create msg
367  *
368  * @rdev: radeon_device pointer
369  * @ring: ring we should submit the msg to
370  * @handle: VCE session handle to use
371  * @fence: optional fence to return
372  *
373  * Open up a stream for HW test
374  */
radeon_vce_get_create_msg(struct radeon_device * rdev,int ring,uint32_t handle,struct radeon_fence ** fence)375 int radeon_vce_get_create_msg(struct radeon_device *rdev, int ring,
376 			      uint32_t handle, struct radeon_fence **fence)
377 {
378 	const unsigned ib_size_dw = 1024;
379 	struct radeon_ib ib;
380 	uint64_t dummy;
381 	int i, r;
382 
383 	r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4);
384 	if (r) {
385 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
386 		return r;
387 	}
388 
389 	dummy = ib.gpu_addr + 1024;
390 
391 	/* stitch together an VCE create msg */
392 	ib.length_dw = 0;
393 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000c); /* len */
394 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001); /* session cmd */
395 	ib.ptr[ib.length_dw++] = cpu_to_le32(handle);
396 
397 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000030); /* len */
398 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x01000001); /* create cmd */
399 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000000);
400 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000042);
401 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000a);
402 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001);
403 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000080);
404 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000060);
405 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000100);
406 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000100);
407 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000c);
408 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000000);
409 
410 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000014); /* len */
411 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x05000005); /* feedback buffer */
412 	ib.ptr[ib.length_dw++] = cpu_to_le32(upper_32_bits(dummy));
413 	ib.ptr[ib.length_dw++] = cpu_to_le32(dummy);
414 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001);
415 
416 	for (i = ib.length_dw; i < ib_size_dw; ++i)
417 		ib.ptr[i] = cpu_to_le32(0x0);
418 
419 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
420 	if (r)
421 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
422 
423 
424 	if (fence)
425 		*fence = radeon_fence_ref(ib.fence);
426 
427 	radeon_ib_free(rdev, &ib);
428 
429 	return r;
430 }
431 
432 /**
433  * radeon_vce_get_destroy_msg - generate a VCE destroy msg
434  *
435  * @rdev: radeon_device pointer
436  * @ring: ring we should submit the msg to
437  * @handle: VCE session handle to use
438  * @fence: optional fence to return
439  *
440  * Close up a stream for HW test or if userspace failed to do so
441  */
radeon_vce_get_destroy_msg(struct radeon_device * rdev,int ring,uint32_t handle,struct radeon_fence ** fence)442 int radeon_vce_get_destroy_msg(struct radeon_device *rdev, int ring,
443 			       uint32_t handle, struct radeon_fence **fence)
444 {
445 	const unsigned ib_size_dw = 1024;
446 	struct radeon_ib ib;
447 	uint64_t dummy;
448 	int i, r;
449 
450 	r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4);
451 	if (r) {
452 		DRM_ERROR("radeon: failed to get ib (%d).\n", r);
453 		return r;
454 	}
455 
456 	dummy = ib.gpu_addr + 1024;
457 
458 	/* stitch together an VCE destroy msg */
459 	ib.length_dw = 0;
460 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x0000000c); /* len */
461 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001); /* session cmd */
462 	ib.ptr[ib.length_dw++] = cpu_to_le32(handle);
463 
464 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000014); /* len */
465 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x05000005); /* feedback buffer */
466 	ib.ptr[ib.length_dw++] = cpu_to_le32(upper_32_bits(dummy));
467 	ib.ptr[ib.length_dw++] = cpu_to_le32(dummy);
468 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000001);
469 
470 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x00000008); /* len */
471 	ib.ptr[ib.length_dw++] = cpu_to_le32(0x02000001); /* destroy cmd */
472 
473 	for (i = ib.length_dw; i < ib_size_dw; ++i)
474 		ib.ptr[i] = cpu_to_le32(0x0);
475 
476 	r = radeon_ib_schedule(rdev, &ib, NULL, false);
477 	if (r) {
478 		DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
479 	}
480 
481 	if (fence)
482 		*fence = radeon_fence_ref(ib.fence);
483 
484 	radeon_ib_free(rdev, &ib);
485 
486 	return r;
487 }
488 
489 /**
490  * radeon_vce_cs_reloc - command submission relocation
491  *
492  * @p: parser context
493  * @lo: address of lower dword
494  * @hi: address of higher dword
495  * @size: size of checker for relocation buffer
496  *
497  * Patch relocation inside command stream with real buffer address
498  */
radeon_vce_cs_reloc(struct radeon_cs_parser * p,int lo,int hi,unsigned size)499 int radeon_vce_cs_reloc(struct radeon_cs_parser *p, int lo, int hi,
500 			unsigned size)
501 {
502 	struct radeon_cs_chunk *relocs_chunk;
503 	struct radeon_bo_list *reloc;
504 	uint64_t start, end, offset;
505 	unsigned idx;
506 
507 	relocs_chunk = p->chunk_relocs;
508 	offset = radeon_get_ib_value(p, lo);
509 	idx = radeon_get_ib_value(p, hi);
510 
511 	if (idx >= relocs_chunk->length_dw) {
512 		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
513 			  idx, relocs_chunk->length_dw);
514 		return -EINVAL;
515 	}
516 
517 	reloc = &p->relocs[(idx / 4)];
518 	start = reloc->gpu_offset;
519 	end = start + radeon_bo_size(reloc->robj);
520 	start += offset;
521 
522 	p->ib.ptr[lo] = start & 0xFFFFFFFF;
523 	p->ib.ptr[hi] = start >> 32;
524 
525 	if (end <= start) {
526 		DRM_ERROR("invalid reloc offset %llX!\n", offset);
527 		return -EINVAL;
528 	}
529 	if ((end - start) < size) {
530 		DRM_ERROR("buffer to small (%d / %d)!\n",
531 			(unsigned)(end - start), size);
532 		return -EINVAL;
533 	}
534 
535 	return 0;
536 }
537 
538 /**
539  * radeon_vce_validate_handle - validate stream handle
540  *
541  * @p: parser context
542  * @handle: handle to validate
543  * @allocated: allocated a new handle?
544  *
545  * Validates the handle and return the found session index or -EINVAL
546  * we don't have another free session index.
547  */
radeon_vce_validate_handle(struct radeon_cs_parser * p,uint32_t handle,bool * allocated)548 static int radeon_vce_validate_handle(struct radeon_cs_parser *p,
549 				      uint32_t handle, bool *allocated)
550 {
551 	unsigned i;
552 
553 	*allocated = false;
554 
555 	/* validate the handle */
556 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
557 		if (atomic_read(&p->rdev->vce.handles[i]) == handle) {
558 			if (p->rdev->vce.filp[i] != p->filp) {
559 				DRM_ERROR("VCE handle collision detected!\n");
560 				return -EINVAL;
561 			}
562 			return i;
563 		}
564 	}
565 
566 	/* handle not found try to alloc a new one */
567 	for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) {
568 		if (!atomic_cmpxchg(&p->rdev->vce.handles[i], 0, handle)) {
569 			p->rdev->vce.filp[i] = p->filp;
570 			p->rdev->vce.img_size[i] = 0;
571 			*allocated = true;
572 			return i;
573 		}
574 	}
575 
576 	DRM_ERROR("No more free VCE handles!\n");
577 	return -EINVAL;
578 }
579 
580 /**
581  * radeon_vce_cs_parse - parse and validate the command stream
582  *
583  * @p: parser context
584  *
585  */
radeon_vce_cs_parse(struct radeon_cs_parser * p)586 int radeon_vce_cs_parse(struct radeon_cs_parser *p)
587 {
588 	int session_idx = -1;
589 	bool destroyed = false, created = false, allocated = false;
590 	uint32_t tmp, handle = 0;
591 	uint32_t *size = &tmp;
592 	int i, r = 0;
593 
594 	while (p->idx < p->chunk_ib->length_dw) {
595 		uint32_t len = radeon_get_ib_value(p, p->idx);
596 		uint32_t cmd = radeon_get_ib_value(p, p->idx + 1);
597 
598 		if ((len < 8) || (len & 3)) {
599 			DRM_ERROR("invalid VCE command length (%d)!\n", len);
600 			r = -EINVAL;
601 			goto out;
602 		}
603 
604 		if (destroyed) {
605 			DRM_ERROR("No other command allowed after destroy!\n");
606 			r = -EINVAL;
607 			goto out;
608 		}
609 
610 		switch (cmd) {
611 		case 0x00000001: // session
612 			handle = radeon_get_ib_value(p, p->idx + 2);
613 			session_idx = radeon_vce_validate_handle(p, handle,
614 								 &allocated);
615 			if (session_idx < 0)
616 				return session_idx;
617 			size = &p->rdev->vce.img_size[session_idx];
618 			break;
619 
620 		case 0x00000002: // task info
621 			break;
622 
623 		case 0x01000001: // create
624 			created = true;
625 			if (!allocated) {
626 				DRM_ERROR("Handle already in use!\n");
627 				r = -EINVAL;
628 				goto out;
629 			}
630 
631 			*size = radeon_get_ib_value(p, p->idx + 8) *
632 				radeon_get_ib_value(p, p->idx + 10) *
633 				8 * 3 / 2;
634 			break;
635 
636 		case 0x04000001: // config extension
637 		case 0x04000002: // pic control
638 		case 0x04000005: // rate control
639 		case 0x04000007: // motion estimation
640 		case 0x04000008: // rdo
641 		case 0x04000009: // vui
642 			break;
643 
644 		case 0x03000001: // encode
645 			r = radeon_vce_cs_reloc(p, p->idx + 10, p->idx + 9,
646 						*size);
647 			if (r)
648 				goto out;
649 
650 			r = radeon_vce_cs_reloc(p, p->idx + 12, p->idx + 11,
651 						*size / 3);
652 			if (r)
653 				goto out;
654 			break;
655 
656 		case 0x02000001: // destroy
657 			destroyed = true;
658 			break;
659 
660 		case 0x05000001: // context buffer
661 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
662 						*size * 2);
663 			if (r)
664 				goto out;
665 			break;
666 
667 		case 0x05000004: // video bitstream buffer
668 			tmp = radeon_get_ib_value(p, p->idx + 4);
669 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
670 						tmp);
671 			if (r)
672 				goto out;
673 			break;
674 
675 		case 0x05000005: // feedback buffer
676 			r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2,
677 						4096);
678 			if (r)
679 				goto out;
680 			break;
681 
682 		default:
683 			DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
684 			r = -EINVAL;
685 			goto out;
686 		}
687 
688 		if (session_idx == -1) {
689 			DRM_ERROR("no session command at start of IB\n");
690 			r = -EINVAL;
691 			goto out;
692 		}
693 
694 		p->idx += len / 4;
695 	}
696 
697 	if (allocated && !created) {
698 		DRM_ERROR("New session without create command!\n");
699 		r = -ENOENT;
700 	}
701 
702 out:
703 	if ((!r && destroyed) || (r && allocated)) {
704 		/*
705 		 * IB contains a destroy msg or we have allocated an
706 		 * handle and got an error, anyway free the handle
707 		 */
708 		for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i)
709 			atomic_cmpxchg(&p->rdev->vce.handles[i], handle, 0);
710 	}
711 
712 	return r;
713 }
714 
715 /**
716  * radeon_vce_semaphore_emit - emit a semaphore command
717  *
718  * @rdev: radeon_device pointer
719  * @ring: engine to use
720  * @semaphore: address of semaphore
721  * @emit_wait: true=emit wait, false=emit signal
722  *
723  */
radeon_vce_semaphore_emit(struct radeon_device * rdev,struct radeon_ring * ring,struct radeon_semaphore * semaphore,bool emit_wait)724 bool radeon_vce_semaphore_emit(struct radeon_device *rdev,
725 			       struct radeon_ring *ring,
726 			       struct radeon_semaphore *semaphore,
727 			       bool emit_wait)
728 {
729 	uint64_t addr = semaphore->gpu_addr;
730 
731 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_SEMAPHORE));
732 	radeon_ring_write(ring, cpu_to_le32((addr >> 3) & 0x000FFFFF));
733 	radeon_ring_write(ring, cpu_to_le32((addr >> 23) & 0x000FFFFF));
734 	radeon_ring_write(ring, cpu_to_le32(0x01003000 | (emit_wait ? 1 : 0)));
735 	if (!emit_wait)
736 		radeon_ring_write(ring, cpu_to_le32(VCE_CMD_END));
737 
738 	return true;
739 }
740 
741 /**
742  * radeon_vce_ib_execute - execute indirect buffer
743  *
744  * @rdev: radeon_device pointer
745  * @ib: the IB to execute
746  *
747  */
radeon_vce_ib_execute(struct radeon_device * rdev,struct radeon_ib * ib)748 void radeon_vce_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
749 {
750 	struct radeon_ring *ring = &rdev->ring[ib->ring];
751 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_IB));
752 	radeon_ring_write(ring, cpu_to_le32(ib->gpu_addr));
753 	radeon_ring_write(ring, cpu_to_le32(upper_32_bits(ib->gpu_addr)));
754 	radeon_ring_write(ring, cpu_to_le32(ib->length_dw));
755 }
756 
757 /**
758  * radeon_vce_fence_emit - add a fence command to the ring
759  *
760  * @rdev: radeon_device pointer
761  * @fence: the fence
762  *
763  */
radeon_vce_fence_emit(struct radeon_device * rdev,struct radeon_fence * fence)764 void radeon_vce_fence_emit(struct radeon_device *rdev,
765 			   struct radeon_fence *fence)
766 {
767 	struct radeon_ring *ring = &rdev->ring[fence->ring];
768 	uint64_t addr = rdev->fence_drv[fence->ring].gpu_addr;
769 
770 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_FENCE));
771 	radeon_ring_write(ring, cpu_to_le32(addr));
772 	radeon_ring_write(ring, cpu_to_le32(upper_32_bits(addr)));
773 	radeon_ring_write(ring, cpu_to_le32(fence->seq));
774 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_TRAP));
775 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_END));
776 }
777 
778 /**
779  * radeon_vce_ring_test - test if VCE ring is working
780  *
781  * @rdev: radeon_device pointer
782  * @ring: the engine to test on
783  *
784  */
radeon_vce_ring_test(struct radeon_device * rdev,struct radeon_ring * ring)785 int radeon_vce_ring_test(struct radeon_device *rdev, struct radeon_ring *ring)
786 {
787 	uint32_t rptr = vce_v1_0_get_rptr(rdev, ring);
788 	unsigned i;
789 	int r;
790 
791 	r = radeon_ring_lock(rdev, ring, 16);
792 	if (r) {
793 		DRM_ERROR("radeon: vce failed to lock ring %d (%d).\n",
794 			  ring->idx, r);
795 		return r;
796 	}
797 	radeon_ring_write(ring, cpu_to_le32(VCE_CMD_END));
798 	radeon_ring_unlock_commit(rdev, ring, false);
799 
800 	for (i = 0; i < rdev->usec_timeout; i++) {
801 		if (vce_v1_0_get_rptr(rdev, ring) != rptr)
802 			break;
803 		udelay(1);
804 	}
805 
806 	if (i < rdev->usec_timeout) {
807 		DRM_INFO("ring test on %d succeeded in %d usecs\n",
808 			 ring->idx, i);
809 	} else {
810 		DRM_ERROR("radeon: ring %d test failed\n",
811 			 ring->idx);
812 		r = -ETIMEDOUT;
813 	}
814 
815 	return r;
816 }
817 
818 /**
819  * radeon_vce_ib_test - test if VCE IBs are working
820  *
821  * @rdev: radeon_device pointer
822  * @ring: the engine to test on
823  *
824  */
radeon_vce_ib_test(struct radeon_device * rdev,struct radeon_ring * ring)825 int radeon_vce_ib_test(struct radeon_device *rdev, struct radeon_ring *ring)
826 {
827 	struct radeon_fence *fence = NULL;
828 	int r;
829 
830 	r = radeon_vce_get_create_msg(rdev, ring->idx, 1, NULL);
831 	if (r) {
832 		DRM_ERROR("radeon: failed to get create msg (%d).\n", r);
833 		goto error;
834 	}
835 
836 	r = radeon_vce_get_destroy_msg(rdev, ring->idx, 1, &fence);
837 	if (r) {
838 		DRM_ERROR("radeon: failed to get destroy ib (%d).\n", r);
839 		goto error;
840 	}
841 
842 	r = radeon_fence_wait_timeout(fence, false, usecs_to_jiffies(
843 		RADEON_USEC_IB_TEST_TIMEOUT));
844 	if (r < 0) {
845 		DRM_ERROR("radeon: fence wait failed (%d).\n", r);
846 	} else if (r == 0) {
847 		DRM_ERROR("radeon: fence wait timed out.\n");
848 		r = -ETIMEDOUT;
849 	} else {
850 		DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
851 		r = 0;
852 	}
853 error:
854 	radeon_fence_unref(&fence);
855 	return r;
856 }
857