xref: /linux/drivers/gpu/drm/amd/amdgpu/amdgpu_umsch_mm.c (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright 2023 Advanced Micro Devices, Inc.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be included in
13  * all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21  * OTHER DEALINGS IN THE SOFTWARE.
22  *
23  */
24 
25 #include <linux/firmware.h>
26 #include <drm/drm_exec.h>
27 
28 #include "amdgpu.h"
29 #include "amdgpu_umsch_mm.h"
30 #include "umsch_mm_v4_0.h"
31 
32 struct umsch_mm_test_ctx_data {
33 	uint8_t process_csa[PAGE_SIZE];
34 	uint8_t vpe_ctx_csa[PAGE_SIZE];
35 	uint8_t vcn_ctx_csa[PAGE_SIZE];
36 };
37 
38 struct umsch_mm_test_mqd_data {
39 	uint8_t vpe_mqd[PAGE_SIZE];
40 	uint8_t vcn_mqd[PAGE_SIZE];
41 };
42 
43 struct umsch_mm_test_ring_data {
44 	uint8_t vpe_ring[PAGE_SIZE];
45 	uint8_t vpe_ib[PAGE_SIZE];
46 	uint8_t vcn_ring[PAGE_SIZE];
47 	uint8_t vcn_ib[PAGE_SIZE];
48 };
49 
50 struct umsch_mm_test_queue_info {
51 	uint64_t mqd_addr;
52 	uint64_t csa_addr;
53 	uint32_t doorbell_offset_0;
54 	uint32_t doorbell_offset_1;
55 	enum UMSCH_SWIP_ENGINE_TYPE engine;
56 };
57 
58 struct umsch_mm_test {
59 	struct amdgpu_bo	*ctx_data_obj;
60 	uint64_t		ctx_data_gpu_addr;
61 	uint32_t		*ctx_data_cpu_addr;
62 
63 	struct amdgpu_bo	*mqd_data_obj;
64 	uint64_t		mqd_data_gpu_addr;
65 	uint32_t		*mqd_data_cpu_addr;
66 
67 	struct amdgpu_bo	*ring_data_obj;
68 	uint64_t		ring_data_gpu_addr;
69 	uint32_t		*ring_data_cpu_addr;
70 
71 
72 	struct amdgpu_vm	*vm;
73 	struct amdgpu_bo_va	*bo_va;
74 	uint32_t		pasid;
75 	uint32_t		vm_cntx_cntl;
76 	uint32_t		num_queues;
77 };
78 
79 static int map_ring_data(struct amdgpu_device *adev, struct amdgpu_vm *vm,
80 			  struct amdgpu_bo *bo, struct amdgpu_bo_va **bo_va,
81 			  uint64_t addr, uint32_t size)
82 {
83 	struct amdgpu_sync sync;
84 	struct drm_exec exec;
85 	int r;
86 
87 	amdgpu_sync_create(&sync);
88 
89 	drm_exec_init(&exec, 0, 0);
90 	drm_exec_until_all_locked(&exec) {
91 		r = drm_exec_lock_obj(&exec, &bo->tbo.base);
92 		drm_exec_retry_on_contention(&exec);
93 		if (unlikely(r))
94 			goto error_fini_exec;
95 
96 		r = amdgpu_vm_lock_pd(vm, &exec, 0);
97 		drm_exec_retry_on_contention(&exec);
98 		if (unlikely(r))
99 			goto error_fini_exec;
100 	}
101 
102 	*bo_va = amdgpu_vm_bo_add(adev, vm, bo);
103 	if (!*bo_va) {
104 		r = -ENOMEM;
105 		goto error_fini_exec;
106 	}
107 
108 	r = amdgpu_vm_bo_map(adev, *bo_va, addr, 0, size,
109 			     AMDGPU_PTE_READABLE | AMDGPU_PTE_WRITEABLE |
110 			     AMDGPU_PTE_EXECUTABLE);
111 
112 	if (r)
113 		goto error_del_bo_va;
114 
115 
116 	r = amdgpu_vm_bo_update(adev, *bo_va, false);
117 	if (r)
118 		goto error_del_bo_va;
119 
120 	amdgpu_sync_fence(&sync, (*bo_va)->last_pt_update);
121 
122 	r = amdgpu_vm_update_pdes(adev, vm, false);
123 	if (r)
124 		goto error_del_bo_va;
125 
126 	amdgpu_sync_fence(&sync, vm->last_update);
127 
128 	amdgpu_sync_wait(&sync, false);
129 	drm_exec_fini(&exec);
130 
131 	amdgpu_sync_free(&sync);
132 
133 	return 0;
134 
135 error_del_bo_va:
136 	amdgpu_vm_bo_del(adev, *bo_va);
137 	amdgpu_sync_free(&sync);
138 
139 error_fini_exec:
140 	drm_exec_fini(&exec);
141 	amdgpu_sync_free(&sync);
142 	return r;
143 }
144 
145 static int unmap_ring_data(struct amdgpu_device *adev, struct amdgpu_vm *vm,
146 			    struct amdgpu_bo *bo, struct amdgpu_bo_va *bo_va,
147 			    uint64_t addr)
148 {
149 	struct drm_exec exec;
150 	long r;
151 
152 	drm_exec_init(&exec, 0, 0);
153 	drm_exec_until_all_locked(&exec) {
154 		r = drm_exec_lock_obj(&exec, &bo->tbo.base);
155 		drm_exec_retry_on_contention(&exec);
156 		if (unlikely(r))
157 			goto out_unlock;
158 
159 		r = amdgpu_vm_lock_pd(vm, &exec, 0);
160 		drm_exec_retry_on_contention(&exec);
161 		if (unlikely(r))
162 			goto out_unlock;
163 	}
164 
165 
166 	r = amdgpu_vm_bo_unmap(adev, bo_va, addr);
167 	if (r)
168 		goto out_unlock;
169 
170 	amdgpu_vm_bo_del(adev, bo_va);
171 
172 out_unlock:
173 	drm_exec_fini(&exec);
174 
175 	return r;
176 }
177 
178 static void setup_vpe_queue(struct amdgpu_device *adev,
179 			    struct umsch_mm_test *test,
180 			    struct umsch_mm_test_queue_info *qinfo)
181 {
182 	struct MQD_INFO *mqd = (struct MQD_INFO *)test->mqd_data_cpu_addr;
183 	uint64_t ring_gpu_addr = test->ring_data_gpu_addr;
184 
185 	mqd->rb_base_lo = (ring_gpu_addr >> 8);
186 	mqd->rb_base_hi = (ring_gpu_addr >> 40);
187 	mqd->rb_size = PAGE_SIZE / 4;
188 	mqd->wptr_val = 0;
189 	mqd->rptr_val = 0;
190 	mqd->unmapped = 1;
191 
192 	qinfo->mqd_addr = test->mqd_data_gpu_addr;
193 	qinfo->csa_addr = test->ctx_data_gpu_addr +
194 		offsetof(struct umsch_mm_test_ctx_data, vpe_ctx_csa);
195 	qinfo->doorbell_offset_0 = (adev->doorbell_index.vpe_ring + 1) << 1;
196 	qinfo->doorbell_offset_1 = 0;
197 }
198 
199 static void setup_vcn_queue(struct amdgpu_device *adev,
200 			    struct umsch_mm_test *test,
201 			    struct umsch_mm_test_queue_info *qinfo)
202 {
203 }
204 
205 static int add_test_queue(struct amdgpu_device *adev,
206 			  struct umsch_mm_test *test,
207 			  struct umsch_mm_test_queue_info *qinfo)
208 {
209 	struct umsch_mm_add_queue_input queue_input = {};
210 	int r;
211 
212 	queue_input.process_id = test->pasid;
213 	queue_input.page_table_base_addr = amdgpu_gmc_pd_addr(test->vm->root.bo);
214 
215 	queue_input.process_va_start = 0;
216 	queue_input.process_va_end = (adev->vm_manager.max_pfn - 1) << AMDGPU_GPU_PAGE_SHIFT;
217 
218 	queue_input.process_quantum = 100000; /* 10ms */
219 	queue_input.process_csa_addr = test->ctx_data_gpu_addr +
220 				       offsetof(struct umsch_mm_test_ctx_data, process_csa);
221 
222 	queue_input.context_quantum = 10000; /* 1ms */
223 	queue_input.context_csa_addr = qinfo->csa_addr;
224 
225 	queue_input.inprocess_context_priority = CONTEXT_PRIORITY_LEVEL_NORMAL;
226 	queue_input.context_global_priority_level = CONTEXT_PRIORITY_LEVEL_NORMAL;
227 	queue_input.doorbell_offset_0 = qinfo->doorbell_offset_0;
228 	queue_input.doorbell_offset_1 = qinfo->doorbell_offset_1;
229 
230 	queue_input.engine_type = qinfo->engine;
231 	queue_input.mqd_addr = qinfo->mqd_addr;
232 	queue_input.vm_context_cntl = test->vm_cntx_cntl;
233 
234 	amdgpu_umsch_mm_lock(&adev->umsch_mm);
235 	r = adev->umsch_mm.funcs->add_queue(&adev->umsch_mm, &queue_input);
236 	amdgpu_umsch_mm_unlock(&adev->umsch_mm);
237 	if (r)
238 		return r;
239 
240 	return 0;
241 }
242 
243 static int remove_test_queue(struct amdgpu_device *adev,
244 			     struct umsch_mm_test *test,
245 			     struct umsch_mm_test_queue_info *qinfo)
246 {
247 	struct umsch_mm_remove_queue_input queue_input = {};
248 	int r;
249 
250 	queue_input.doorbell_offset_0 = qinfo->doorbell_offset_0;
251 	queue_input.doorbell_offset_1 = qinfo->doorbell_offset_1;
252 	queue_input.context_csa_addr = qinfo->csa_addr;
253 
254 	amdgpu_umsch_mm_lock(&adev->umsch_mm);
255 	r = adev->umsch_mm.funcs->remove_queue(&adev->umsch_mm, &queue_input);
256 	amdgpu_umsch_mm_unlock(&adev->umsch_mm);
257 	if (r)
258 		return r;
259 
260 	return 0;
261 }
262 
263 static int submit_vpe_queue(struct amdgpu_device *adev, struct umsch_mm_test *test)
264 {
265 	struct MQD_INFO *mqd = (struct MQD_INFO *)test->mqd_data_cpu_addr;
266 	uint32_t *ring = test->ring_data_cpu_addr +
267 		offsetof(struct umsch_mm_test_ring_data, vpe_ring) / 4;
268 	uint32_t *ib = test->ring_data_cpu_addr +
269 		offsetof(struct umsch_mm_test_ring_data, vpe_ib) / 4;
270 	uint64_t ib_gpu_addr = test->ring_data_gpu_addr +
271 		offsetof(struct umsch_mm_test_ring_data, vpe_ib);
272 	uint32_t *fence = ib + 2048 / 4;
273 	uint64_t fence_gpu_addr = ib_gpu_addr + 2048;
274 	const uint32_t test_pattern = 0xdeadbeef;
275 	int i;
276 
277 	ib[0] = VPE_CMD_HEADER(VPE_CMD_OPCODE_FENCE, 0);
278 	ib[1] = lower_32_bits(fence_gpu_addr);
279 	ib[2] = upper_32_bits(fence_gpu_addr);
280 	ib[3] = test_pattern;
281 
282 	ring[0] = VPE_CMD_HEADER(VPE_CMD_OPCODE_INDIRECT, 0);
283 	ring[1] = (ib_gpu_addr & 0xffffffe0);
284 	ring[2] = upper_32_bits(ib_gpu_addr);
285 	ring[3] = 4;
286 	ring[4] = 0;
287 	ring[5] = 0;
288 
289 	mqd->wptr_val = (6 << 2);
290 	// WDOORBELL32(adev->umsch_mm.agdb_index[CONTEXT_PRIORITY_LEVEL_NORMAL], mqd->wptr_val);
291 
292 	for (i = 0; i < adev->usec_timeout; i++) {
293 		if (*fence == test_pattern)
294 			return 0;
295 		udelay(1);
296 	}
297 
298 	dev_err(adev->dev, "vpe queue submission timeout\n");
299 
300 	return -ETIMEDOUT;
301 }
302 
303 static int submit_vcn_queue(struct amdgpu_device *adev, struct umsch_mm_test *test)
304 {
305 	return 0;
306 }
307 
308 static int setup_umsch_mm_test(struct amdgpu_device *adev,
309 			  struct umsch_mm_test *test)
310 {
311 	struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB0(0)];
312 	int r;
313 
314 	test->vm_cntx_cntl = hub->vm_cntx_cntl;
315 
316 	test->vm = kzalloc(sizeof(*test->vm), GFP_KERNEL);
317 	if (!test->vm) {
318 		r = -ENOMEM;
319 		return r;
320 	}
321 
322 	r = amdgpu_vm_init(adev, test->vm, -1);
323 	if (r)
324 		goto error_free_vm;
325 
326 	r = amdgpu_pasid_alloc(16);
327 	if (r < 0)
328 		goto error_fini_vm;
329 	test->pasid = r;
330 
331 	r = amdgpu_bo_create_kernel(adev, sizeof(struct umsch_mm_test_ctx_data),
332 				    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
333 				    &test->ctx_data_obj,
334 				    &test->ctx_data_gpu_addr,
335 				    (void **)&test->ctx_data_cpu_addr);
336 	if (r)
337 		goto error_free_pasid;
338 
339 	memset(test->ctx_data_cpu_addr, 0, sizeof(struct umsch_mm_test_ctx_data));
340 
341 	r = amdgpu_bo_create_kernel(adev, PAGE_SIZE,
342 				    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
343 				    &test->mqd_data_obj,
344 				    &test->mqd_data_gpu_addr,
345 				    (void **)&test->mqd_data_cpu_addr);
346 	if (r)
347 		goto error_free_ctx_data_obj;
348 
349 	memset(test->mqd_data_cpu_addr, 0, PAGE_SIZE);
350 
351 	r = amdgpu_bo_create_kernel(adev, sizeof(struct umsch_mm_test_ring_data),
352 				    PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
353 				    &test->ring_data_obj,
354 				    NULL,
355 				    (void **)&test->ring_data_cpu_addr);
356 	if (r)
357 		goto error_free_mqd_data_obj;
358 
359 	memset(test->ring_data_cpu_addr, 0, sizeof(struct umsch_mm_test_ring_data));
360 
361 	test->ring_data_gpu_addr = AMDGPU_VA_RESERVED_BOTTOM;
362 	r = map_ring_data(adev, test->vm, test->ring_data_obj, &test->bo_va,
363 			  test->ring_data_gpu_addr, sizeof(struct umsch_mm_test_ring_data));
364 	if (r)
365 		goto error_free_ring_data_obj;
366 
367 	return 0;
368 
369 error_free_ring_data_obj:
370 	amdgpu_bo_free_kernel(&test->ring_data_obj, NULL,
371 			      (void **)&test->ring_data_cpu_addr);
372 error_free_mqd_data_obj:
373 	amdgpu_bo_free_kernel(&test->mqd_data_obj, &test->mqd_data_gpu_addr,
374 			      (void **)&test->mqd_data_cpu_addr);
375 error_free_ctx_data_obj:
376 	amdgpu_bo_free_kernel(&test->ctx_data_obj, &test->ctx_data_gpu_addr,
377 			      (void **)&test->ctx_data_cpu_addr);
378 error_free_pasid:
379 	amdgpu_pasid_free(test->pasid);
380 error_fini_vm:
381 	amdgpu_vm_fini(adev, test->vm);
382 error_free_vm:
383 	kfree(test->vm);
384 
385 	return r;
386 }
387 
388 static void cleanup_umsch_mm_test(struct amdgpu_device *adev,
389 				  struct umsch_mm_test *test)
390 {
391 	unmap_ring_data(adev, test->vm, test->ring_data_obj,
392 			test->bo_va, test->ring_data_gpu_addr);
393 	amdgpu_bo_free_kernel(&test->mqd_data_obj, &test->mqd_data_gpu_addr,
394 			      (void **)&test->mqd_data_cpu_addr);
395 	amdgpu_bo_free_kernel(&test->ring_data_obj, NULL,
396 			      (void **)&test->ring_data_cpu_addr);
397 	amdgpu_bo_free_kernel(&test->ctx_data_obj, &test->ctx_data_gpu_addr,
398 			       (void **)&test->ctx_data_cpu_addr);
399 	amdgpu_pasid_free(test->pasid);
400 	amdgpu_vm_fini(adev, test->vm);
401 	kfree(test->vm);
402 }
403 
404 static int setup_test_queues(struct amdgpu_device *adev,
405 			     struct umsch_mm_test *test,
406 			     struct umsch_mm_test_queue_info *qinfo)
407 {
408 	int i, r;
409 
410 	for (i = 0; i < test->num_queues; i++) {
411 		if (qinfo[i].engine == UMSCH_SWIP_ENGINE_TYPE_VPE)
412 			setup_vpe_queue(adev, test, &qinfo[i]);
413 		else
414 			setup_vcn_queue(adev, test, &qinfo[i]);
415 
416 		r = add_test_queue(adev, test, &qinfo[i]);
417 		if (r)
418 			return r;
419 	}
420 
421 	return 0;
422 }
423 
424 static int submit_test_queues(struct amdgpu_device *adev,
425 			      struct umsch_mm_test *test,
426 			      struct umsch_mm_test_queue_info *qinfo)
427 {
428 	int i, r;
429 
430 	for (i = 0; i < test->num_queues; i++) {
431 		if (qinfo[i].engine == UMSCH_SWIP_ENGINE_TYPE_VPE)
432 			r = submit_vpe_queue(adev, test);
433 		else
434 			r = submit_vcn_queue(adev, test);
435 		if (r)
436 			return r;
437 	}
438 
439 	return 0;
440 }
441 
442 static void cleanup_test_queues(struct amdgpu_device *adev,
443 			      struct umsch_mm_test *test,
444 			      struct umsch_mm_test_queue_info *qinfo)
445 {
446 	int i;
447 
448 	for (i = 0; i < test->num_queues; i++)
449 		remove_test_queue(adev, test, &qinfo[i]);
450 }
451 
452 static int umsch_mm_test(struct amdgpu_device *adev)
453 {
454 	struct umsch_mm_test_queue_info qinfo[] = {
455 		{ .engine = UMSCH_SWIP_ENGINE_TYPE_VPE },
456 	};
457 	struct umsch_mm_test test = { .num_queues = ARRAY_SIZE(qinfo) };
458 	int r;
459 
460 	r = setup_umsch_mm_test(adev, &test);
461 	if (r)
462 		return r;
463 
464 	r = setup_test_queues(adev, &test, qinfo);
465 	if (r)
466 		goto cleanup;
467 
468 	r = submit_test_queues(adev, &test, qinfo);
469 	if (r)
470 		goto cleanup;
471 
472 	cleanup_test_queues(adev, &test, qinfo);
473 	cleanup_umsch_mm_test(adev, &test);
474 
475 	return 0;
476 
477 cleanup:
478 	cleanup_test_queues(adev, &test, qinfo);
479 	cleanup_umsch_mm_test(adev, &test);
480 	return r;
481 }
482 
483 int amdgpu_umsch_mm_submit_pkt(struct amdgpu_umsch_mm *umsch, void *pkt, int ndws)
484 {
485 	struct amdgpu_ring *ring = &umsch->ring;
486 
487 	if (amdgpu_ring_alloc(ring, ndws))
488 		return -ENOMEM;
489 
490 	amdgpu_ring_write_multiple(ring, pkt, ndws);
491 	amdgpu_ring_commit(ring);
492 
493 	return 0;
494 }
495 
496 int amdgpu_umsch_mm_query_fence(struct amdgpu_umsch_mm *umsch)
497 {
498 	struct amdgpu_ring *ring = &umsch->ring;
499 	struct amdgpu_device *adev = ring->adev;
500 	int r;
501 
502 	r = amdgpu_fence_wait_polling(ring, ring->fence_drv.sync_seq, adev->usec_timeout);
503 	if (r < 1) {
504 		dev_err(adev->dev, "ring umsch timeout, emitted fence %u\n",
505 			ring->fence_drv.sync_seq);
506 		return -ETIMEDOUT;
507 	}
508 
509 	return 0;
510 }
511 
512 static void umsch_mm_ring_set_wptr(struct amdgpu_ring *ring)
513 {
514 	struct amdgpu_umsch_mm *umsch = (struct amdgpu_umsch_mm *)ring;
515 	struct amdgpu_device *adev = ring->adev;
516 
517 	if (ring->use_doorbell)
518 		WDOORBELL32(ring->doorbell_index, ring->wptr << 2);
519 	else
520 		WREG32(umsch->rb_wptr, ring->wptr << 2);
521 }
522 
523 static u64 umsch_mm_ring_get_rptr(struct amdgpu_ring *ring)
524 {
525 	struct amdgpu_umsch_mm *umsch = (struct amdgpu_umsch_mm *)ring;
526 	struct amdgpu_device *adev = ring->adev;
527 
528 	return RREG32(umsch->rb_rptr);
529 }
530 
531 static u64 umsch_mm_ring_get_wptr(struct amdgpu_ring *ring)
532 {
533 	struct amdgpu_umsch_mm *umsch = (struct amdgpu_umsch_mm *)ring;
534 	struct amdgpu_device *adev = ring->adev;
535 
536 	return RREG32(umsch->rb_wptr);
537 }
538 
539 static const struct amdgpu_ring_funcs umsch_v4_0_ring_funcs = {
540 	.type = AMDGPU_RING_TYPE_UMSCH_MM,
541 	.align_mask = 0,
542 	.nop = 0,
543 	.support_64bit_ptrs = false,
544 	.get_rptr = umsch_mm_ring_get_rptr,
545 	.get_wptr = umsch_mm_ring_get_wptr,
546 	.set_wptr = umsch_mm_ring_set_wptr,
547 	.insert_nop = amdgpu_ring_insert_nop,
548 };
549 
550 int amdgpu_umsch_mm_ring_init(struct amdgpu_umsch_mm *umsch)
551 {
552 	struct amdgpu_device *adev = container_of(umsch, struct amdgpu_device, umsch_mm);
553 	struct amdgpu_ring *ring = &umsch->ring;
554 
555 	ring->vm_hub = AMDGPU_MMHUB0(0);
556 	ring->use_doorbell = true;
557 	ring->no_scheduler = true;
558 	ring->doorbell_index = (AMDGPU_NAVI10_DOORBELL64_VCN0_1 << 1) + 6;
559 
560 	snprintf(ring->name, sizeof(ring->name), "umsch");
561 
562 	return amdgpu_ring_init(adev, ring, 1024, NULL, 0, AMDGPU_RING_PRIO_DEFAULT, NULL);
563 }
564 
565 int amdgpu_umsch_mm_init_microcode(struct amdgpu_umsch_mm *umsch)
566 {
567 	const struct umsch_mm_firmware_header_v1_0 *umsch_mm_hdr;
568 	struct amdgpu_device *adev = umsch->ring.adev;
569 	const char *fw_name = NULL;
570 	int r;
571 
572 	switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
573 	case IP_VERSION(4, 0, 5):
574 		fw_name = "amdgpu/umsch_mm_4_0_0.bin";
575 		break;
576 	default:
577 		break;
578 	}
579 
580 	r = amdgpu_ucode_request(adev, &adev->umsch_mm.fw, fw_name);
581 	if (r) {
582 		release_firmware(adev->umsch_mm.fw);
583 		adev->umsch_mm.fw = NULL;
584 		return r;
585 	}
586 
587 	umsch_mm_hdr = (const struct umsch_mm_firmware_header_v1_0 *)adev->umsch_mm.fw->data;
588 
589 	adev->umsch_mm.ucode_size = le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_size_bytes);
590 	adev->umsch_mm.data_size = le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_data_size_bytes);
591 
592 	adev->umsch_mm.irq_start_addr =
593 		le32_to_cpu(umsch_mm_hdr->umsch_mm_irq_start_addr_lo) |
594 		((uint64_t)(le32_to_cpu(umsch_mm_hdr->umsch_mm_irq_start_addr_hi)) << 32);
595 	adev->umsch_mm.uc_start_addr =
596 		le32_to_cpu(umsch_mm_hdr->umsch_mm_uc_start_addr_lo) |
597 		((uint64_t)(le32_to_cpu(umsch_mm_hdr->umsch_mm_uc_start_addr_hi)) << 32);
598 	adev->umsch_mm.data_start_addr =
599 		le32_to_cpu(umsch_mm_hdr->umsch_mm_data_start_addr_lo) |
600 		((uint64_t)(le32_to_cpu(umsch_mm_hdr->umsch_mm_data_start_addr_hi)) << 32);
601 
602 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
603 		struct amdgpu_firmware_info *info;
604 
605 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_UMSCH_MM_UCODE];
606 		info->ucode_id = AMDGPU_UCODE_ID_UMSCH_MM_UCODE;
607 		info->fw = adev->umsch_mm.fw;
608 		adev->firmware.fw_size +=
609 			ALIGN(le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_size_bytes), PAGE_SIZE);
610 
611 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_UMSCH_MM_DATA];
612 		info->ucode_id = AMDGPU_UCODE_ID_UMSCH_MM_DATA;
613 		info->fw = adev->umsch_mm.fw;
614 		adev->firmware.fw_size +=
615 			ALIGN(le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_data_size_bytes), PAGE_SIZE);
616 	}
617 
618 	return 0;
619 }
620 
621 int amdgpu_umsch_mm_allocate_ucode_buffer(struct amdgpu_umsch_mm *umsch)
622 {
623 	const struct umsch_mm_firmware_header_v1_0 *umsch_mm_hdr;
624 	struct amdgpu_device *adev = umsch->ring.adev;
625 	const __le32 *fw_data;
626 	uint32_t fw_size;
627 	int r;
628 
629 	umsch_mm_hdr = (const struct umsch_mm_firmware_header_v1_0 *)
630 		       adev->umsch_mm.fw->data;
631 
632 	fw_data = (const __le32 *)(adev->umsch_mm.fw->data +
633 		  le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_offset_bytes));
634 	fw_size = le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_size_bytes);
635 
636 	r = amdgpu_bo_create_reserved(adev, fw_size,
637 				      4 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
638 				      &adev->umsch_mm.ucode_fw_obj,
639 				      &adev->umsch_mm.ucode_fw_gpu_addr,
640 				      (void **)&adev->umsch_mm.ucode_fw_ptr);
641 	if (r) {
642 		dev_err(adev->dev, "(%d) failed to create umsch_mm fw ucode bo\n", r);
643 		return r;
644 	}
645 
646 	memcpy(adev->umsch_mm.ucode_fw_ptr, fw_data, fw_size);
647 
648 	amdgpu_bo_kunmap(adev->umsch_mm.ucode_fw_obj);
649 	amdgpu_bo_unreserve(adev->umsch_mm.ucode_fw_obj);
650 	return 0;
651 }
652 
653 int amdgpu_umsch_mm_allocate_ucode_data_buffer(struct amdgpu_umsch_mm *umsch)
654 {
655 	const struct umsch_mm_firmware_header_v1_0 *umsch_mm_hdr;
656 	struct amdgpu_device *adev = umsch->ring.adev;
657 	const __le32 *fw_data;
658 	uint32_t fw_size;
659 	int r;
660 
661 	umsch_mm_hdr = (const struct umsch_mm_firmware_header_v1_0 *)
662 		       adev->umsch_mm.fw->data;
663 
664 	fw_data = (const __le32 *)(adev->umsch_mm.fw->data +
665 		  le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_data_offset_bytes));
666 	fw_size = le32_to_cpu(umsch_mm_hdr->umsch_mm_ucode_data_size_bytes);
667 
668 	r = amdgpu_bo_create_reserved(adev, fw_size,
669 				      64 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
670 				      &adev->umsch_mm.data_fw_obj,
671 				      &adev->umsch_mm.data_fw_gpu_addr,
672 				      (void **)&adev->umsch_mm.data_fw_ptr);
673 	if (r) {
674 		dev_err(adev->dev, "(%d) failed to create umsch_mm fw data bo\n", r);
675 		return r;
676 	}
677 
678 	memcpy(adev->umsch_mm.data_fw_ptr, fw_data, fw_size);
679 
680 	amdgpu_bo_kunmap(adev->umsch_mm.data_fw_obj);
681 	amdgpu_bo_unreserve(adev->umsch_mm.data_fw_obj);
682 	return 0;
683 }
684 
685 int amdgpu_umsch_mm_psp_execute_cmd_buf(struct amdgpu_umsch_mm *umsch)
686 {
687 	struct amdgpu_device *adev = umsch->ring.adev;
688 	struct amdgpu_firmware_info ucode = {
689 		.ucode_id = AMDGPU_UCODE_ID_UMSCH_MM_CMD_BUFFER,
690 		.mc_addr = adev->umsch_mm.cmd_buf_gpu_addr,
691 		.ucode_size = ((uintptr_t)adev->umsch_mm.cmd_buf_curr_ptr -
692 			      (uintptr_t)adev->umsch_mm.cmd_buf_ptr),
693 	};
694 
695 	return psp_execute_ip_fw_load(&adev->psp, &ucode);
696 }
697 
698 static void umsch_mm_agdb_index_init(struct amdgpu_device *adev)
699 {
700 	uint32_t umsch_mm_agdb_start;
701 	int i;
702 
703 	umsch_mm_agdb_start = adev->doorbell_index.max_assignment + 1;
704 	umsch_mm_agdb_start = roundup(umsch_mm_agdb_start, 1024);
705 	umsch_mm_agdb_start += (AMDGPU_NAVI10_DOORBELL64_VCN0_1 << 1);
706 
707 	for (i = 0; i < CONTEXT_PRIORITY_NUM_LEVELS; i++)
708 		adev->umsch_mm.agdb_index[i] = umsch_mm_agdb_start + i;
709 }
710 
711 static int umsch_mm_init(struct amdgpu_device *adev)
712 {
713 	int r;
714 
715 	adev->umsch_mm.vmid_mask_mm_vpe = 0xf00;
716 	adev->umsch_mm.engine_mask = (1 << UMSCH_SWIP_ENGINE_TYPE_VPE);
717 	adev->umsch_mm.vpe_hqd_mask = 0xfe;
718 
719 	r = amdgpu_device_wb_get(adev, &adev->umsch_mm.wb_index);
720 	if (r) {
721 		dev_err(adev->dev, "failed to alloc wb for umsch: %d\n", r);
722 		return r;
723 	}
724 
725 	adev->umsch_mm.sch_ctx_gpu_addr = adev->wb.gpu_addr +
726 					  (adev->umsch_mm.wb_index * 4);
727 
728 	r = amdgpu_bo_create_kernel(adev, PAGE_SIZE, PAGE_SIZE,
729 				    AMDGPU_GEM_DOMAIN_GTT,
730 				    &adev->umsch_mm.cmd_buf_obj,
731 				    &adev->umsch_mm.cmd_buf_gpu_addr,
732 				    (void **)&adev->umsch_mm.cmd_buf_ptr);
733 	if (r) {
734 		dev_err(adev->dev, "failed to allocate cmdbuf bo %d\n", r);
735 		amdgpu_device_wb_free(adev, adev->umsch_mm.wb_index);
736 		return r;
737 	}
738 
739 	mutex_init(&adev->umsch_mm.mutex_hidden);
740 
741 	umsch_mm_agdb_index_init(adev);
742 
743 	return 0;
744 }
745 
746 
747 static int umsch_mm_early_init(void *handle)
748 {
749 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
750 
751 	switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
752 	case IP_VERSION(4, 0, 5):
753 		umsch_mm_v4_0_set_funcs(&adev->umsch_mm);
754 		break;
755 	default:
756 		return -EINVAL;
757 	}
758 
759 	adev->umsch_mm.ring.funcs = &umsch_v4_0_ring_funcs;
760 	umsch_mm_set_regs(&adev->umsch_mm);
761 
762 	return 0;
763 }
764 
765 static int umsch_mm_late_init(void *handle)
766 {
767 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
768 
769 	return umsch_mm_test(adev);
770 }
771 
772 static int umsch_mm_sw_init(void *handle)
773 {
774 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
775 	int r;
776 
777 	r = umsch_mm_init(adev);
778 	if (r)
779 		return r;
780 
781 	r = umsch_mm_ring_init(&adev->umsch_mm);
782 	if (r)
783 		return r;
784 
785 	r = umsch_mm_init_microcode(&adev->umsch_mm);
786 	if (r)
787 		return r;
788 
789 	return 0;
790 }
791 
792 static int umsch_mm_sw_fini(void *handle)
793 {
794 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
795 
796 	release_firmware(adev->umsch_mm.fw);
797 	adev->umsch_mm.fw = NULL;
798 
799 	amdgpu_ring_fini(&adev->umsch_mm.ring);
800 
801 	mutex_destroy(&adev->umsch_mm.mutex_hidden);
802 
803 	amdgpu_bo_free_kernel(&adev->umsch_mm.cmd_buf_obj,
804 			      &adev->umsch_mm.cmd_buf_gpu_addr,
805 			      (void **)&adev->umsch_mm.cmd_buf_ptr);
806 
807 	amdgpu_device_wb_free(adev, adev->umsch_mm.wb_index);
808 
809 	return 0;
810 }
811 
812 static int umsch_mm_hw_init(void *handle)
813 {
814 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
815 	int r;
816 
817 	r = umsch_mm_load_microcode(&adev->umsch_mm);
818 	if (r)
819 		return r;
820 
821 	umsch_mm_ring_start(&adev->umsch_mm);
822 
823 	r = umsch_mm_set_hw_resources(&adev->umsch_mm);
824 	if (r)
825 		return r;
826 
827 	return 0;
828 }
829 
830 static int umsch_mm_hw_fini(void *handle)
831 {
832 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
833 
834 	umsch_mm_ring_stop(&adev->umsch_mm);
835 
836 	amdgpu_bo_free_kernel(&adev->umsch_mm.data_fw_obj,
837 			      &adev->umsch_mm.data_fw_gpu_addr,
838 			      (void **)&adev->umsch_mm.data_fw_ptr);
839 
840 	amdgpu_bo_free_kernel(&adev->umsch_mm.ucode_fw_obj,
841 			      &adev->umsch_mm.ucode_fw_gpu_addr,
842 			      (void **)&adev->umsch_mm.ucode_fw_ptr);
843 	return 0;
844 }
845 
846 static int umsch_mm_suspend(void *handle)
847 {
848 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
849 
850 	return umsch_mm_hw_fini(adev);
851 }
852 
853 static int umsch_mm_resume(void *handle)
854 {
855 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
856 
857 	return umsch_mm_hw_init(adev);
858 }
859 
860 static const struct amd_ip_funcs umsch_mm_v4_0_ip_funcs = {
861 	.name = "umsch_mm_v4_0",
862 	.early_init = umsch_mm_early_init,
863 	.late_init = umsch_mm_late_init,
864 	.sw_init = umsch_mm_sw_init,
865 	.sw_fini = umsch_mm_sw_fini,
866 	.hw_init = umsch_mm_hw_init,
867 	.hw_fini = umsch_mm_hw_fini,
868 	.suspend = umsch_mm_suspend,
869 	.resume = umsch_mm_resume,
870 };
871 
872 const struct amdgpu_ip_block_version umsch_mm_v4_0_ip_block = {
873 	.type = AMD_IP_BLOCK_TYPE_UMSCH_MM,
874 	.major = 4,
875 	.minor = 0,
876 	.rev = 0,
877 	.funcs = &umsch_mm_v4_0_ip_funcs,
878 };
879