amdgpu_ib.c 11.0 KB
Newer Older
A
Alex Deucher 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
/*
 * Copyright 2008 Advanced Micro Devices, Inc.
 * Copyright 2008 Red Hat Inc.
 * Copyright 2009 Jerome Glisse.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 *
 * Authors: Dave Airlie
 *          Alex Deucher
 *          Jerome Glisse
 *          Christian König
 */
#include <linux/seq_file.h>
#include <linux/slab.h>
#include <drm/drmP.h>
#include <drm/amdgpu_drm.h>
#include "amdgpu.h"
#include "atom.h"

36
#define AMDGPU_IB_TEST_TIMEOUT	msecs_to_jiffies(1000)
37

A
Alex Deucher 已提交
38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59
/*
 * IB
 * IBs (Indirect Buffers) and areas of GPU accessible memory where
 * commands are stored.  You can put a pointer to the IB in the
 * command ring and the hw will fetch the commands from the IB
 * and execute them.  Generally userspace acceleration drivers
 * produce command buffers which are send to the kernel and
 * put in IBs for execution by the requested ring.
 */
static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev);

/**
 * amdgpu_ib_get - request an IB (Indirect Buffer)
 *
 * @ring: ring index the IB is associated with
 * @size: requested IB size
 * @ib: IB object returned
 *
 * Request an IB (all asics).  IBs are allocated using the
 * suballocator.
 * Returns 0 on success, error on failure.
 */
60
int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm,
A
Alex Deucher 已提交
61 62 63 64 65
		  unsigned size, struct amdgpu_ib *ib)
{
	int r;

	if (size) {
66
		r = amdgpu_sa_bo_new(&adev->ring_tmp_bo,
A
Alex Deucher 已提交
67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
				      &ib->sa_bo, size, 256);
		if (r) {
			dev_err(adev->dev, "failed to get a new IB (%d)\n", r);
			return r;
		}

		ib->ptr = amdgpu_sa_bo_cpu_addr(ib->sa_bo);

		if (!vm)
			ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
	}

	return 0;
}

/**
 * amdgpu_ib_free - free an IB (Indirect Buffer)
 *
 * @adev: amdgpu_device pointer
 * @ib: IB object to free
87
 * @f: the fence SA bo need wait on for the ib alloation
A
Alex Deucher 已提交
88 89 90
 *
 * Free an IB (all asics).
 */
91
void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib,
92
		    struct dma_fence *f)
A
Alex Deucher 已提交
93
{
94
	amdgpu_sa_bo_free(adev, &ib->sa_bo, f);
A
Alex Deucher 已提交
95 96 97 98 99 100 101 102
}

/**
 * amdgpu_ib_schedule - schedule an IB (Indirect Buffer) on the ring
 *
 * @adev: amdgpu_device pointer
 * @num_ibs: number of IBs to schedule
 * @ibs: IB objects to schedule
103
 * @f: fence created during this submission
A
Alex Deucher 已提交
104 105 106 107 108 109 110 111 112 113 114 115 116 117
 *
 * Schedule an IB on the associated ring (all asics).
 * Returns 0 on success, error on failure.
 *
 * On SI, there are two parallel engines fed from the primary ring,
 * the CE (Constant Engine) and the DE (Drawing Engine).  Since
 * resource descriptors have moved to memory, the CE allows you to
 * prime the caches while the DE is updating register state so that
 * the resource descriptors will be already in cache when the draw is
 * processed.  To accomplish this, the userspace driver submits two
 * IBs, one for the CE and one for the DE.  If there is a CE IB (called
 * a CONST_IB), it will be put on the ring prior to the DE IB.  Prior
 * to SI there was just a DE IB.
 */
118
int amdgpu_ib_schedule(struct amdgpu_ring *ring, unsigned num_ibs,
119 120
		       struct amdgpu_ib *ibs, struct amdgpu_job *job,
		       struct dma_fence **f)
A
Alex Deucher 已提交
121
{
122
	struct amdgpu_device *adev = ring->adev;
A
Alex Deucher 已提交
123
	struct amdgpu_ib *ib = &ibs[0];
124
	struct dma_fence *tmp = NULL;
125
	bool skip_preamble, need_ctx_switch;
126 127
	unsigned patch_offset = ~0;
	struct amdgpu_vm *vm;
128
	uint64_t fence_ctx;
129
	uint32_t status = 0, alloc_size;
130
	unsigned fence_flags = 0;
M
Monk Liu 已提交
131

132
	unsigned i;
A
Alex Deucher 已提交
133
	int r = 0;
M
Monk Liu 已提交
134
	bool need_pipe_sync = false;
A
Alex Deucher 已提交
135 136 137 138

	if (num_ibs == 0)
		return -EINVAL;

139 140
	/* ring tests don't use a job */
	if (job) {
141
		vm = job->vm;
142
		fence_ctx = job->base.s_fence->scheduled.context;
143 144
	} else {
		vm = NULL;
145
		fence_ctx = 0;
146
	}
147

A
Alex Deucher 已提交
148
	if (!ring->ready) {
149
		dev_err(adev->dev, "couldn't schedule ib on ring <%s>\n", ring->name);
A
Alex Deucher 已提交
150 151
		return -EINVAL;
	}
152

153
	if (vm && !job->vmid) {
154 155 156 157
		dev_err(adev->dev, "VM IB without ID\n");
		return -EINVAL;
	}

158 159
	alloc_size = ring->funcs->emit_frame_size + num_ibs *
		ring->funcs->emit_ib_size;
160 161

	r = amdgpu_ring_alloc(ring, alloc_size);
A
Alex Deucher 已提交
162 163 164 165
	if (r) {
		dev_err(adev->dev, "scheduling IB failed (%d).\n", r);
		return r;
	}
166

167
	need_ctx_switch = ring->current_ctx != fence_ctx;
168
	if (ring->funcs->emit_pipeline_sync && job &&
169
	    ((tmp = amdgpu_sync_get_fence(&job->sched_sync, NULL)) ||
170
	     (amdgpu_sriov_vf(adev) && need_ctx_switch) ||
171
	     amdgpu_vm_need_pipeline_sync(ring, job))) {
M
Monk Liu 已提交
172
		need_pipe_sync = true;
173 174
		dma_fence_put(tmp);
	}
A
Alex Deucher 已提交
175

176 177 178
	if (ring->funcs->insert_start)
		ring->funcs->insert_start(ring);

179
	if (job) {
M
Monk Liu 已提交
180
		r = amdgpu_vm_flush(ring, job, need_pipe_sync);
181 182 183 184
		if (r) {
			amdgpu_ring_undo(ring);
			return r;
		}
185
	}
186

187
	if (job && ring->funcs->init_cond_exec)
188 189
		patch_offset = amdgpu_ring_init_cond_exec(ring);

190
#ifdef CONFIG_X86_64
191
	if (!(adev->flags & AMD_IS_APU))
192
#endif
193 194 195 196 197 198
	{
		if (ring->funcs->emit_hdp_flush)
			amdgpu_ring_emit_hdp_flush(ring);
		else
			amdgpu_asic_flush_hdp(adev, ring);
	}
199

200
	skip_preamble = ring->current_ctx == fence_ctx;
201 202 203 204
	if (job && ring->funcs->emit_cntxcntl) {
		if (need_ctx_switch)
			status |= AMDGPU_HAVE_CTX_SWITCH;
		status |= job->preamble_status;
205

206 207 208
		amdgpu_ring_emit_cntxcntl(ring, status);
	}

A
Alex Deucher 已提交
209
	for (i = 0; i < num_ibs; ++i) {
210
		ib = &ibs[i];
211 212

		/* drop preamble IBs if we don't have a context switch */
213 214
		if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) &&
			skip_preamble &&
215 216
			!(status & AMDGPU_PREAMBLE_IB_PRESENT_FIRST) &&
			!amdgpu_sriov_vf(adev)) /* for SRIOV preemption, Preamble CE ib must be inserted anyway */
217 218
			continue;

219
		amdgpu_ring_emit_ib(ring, ib, job ? job->vmid : 0,
220
				    need_ctx_switch);
221
		need_ctx_switch = false;
A
Alex Deucher 已提交
222 223
	}

224 225 226
	if (ring->funcs->emit_tmz)
		amdgpu_ring_emit_tmz(ring, false);

227
#ifdef CONFIG_X86_64
228
	if (!(adev->flags & AMD_IS_APU))
229
#endif
230
		amdgpu_asic_invalidate_hdp(adev, ring);
231

232 233 234
	if (ib->flags & AMDGPU_IB_FLAG_TC_WB_NOT_INVALIDATE)
		fence_flags |= AMDGPU_FENCE_FLAG_TC_WB_ONLY;

235 236 237 238 239 240
	/* wrap the last IB with fence */
	if (job && job->uf_addr) {
		amdgpu_ring_emit_fence(ring, job->uf_addr, job->uf_sequence,
				       fence_flags | AMDGPU_FENCE_FLAG_64BIT);
	}

241
	r = amdgpu_fence_emit(ring, f, fence_flags);
A
Alex Deucher 已提交
242 243
	if (r) {
		dev_err(adev->dev, "failed to emit fence (%d)\n", r);
244 245
		if (job && job->vmid)
			amdgpu_vmid_reset(adev, ring->funcs->vmhub, job->vmid);
246
		amdgpu_ring_undo(ring);
A
Alex Deucher 已提交
247 248 249
		return r;
	}

250 251 252
	if (ring->funcs->insert_end)
		ring->funcs->insert_end(ring);

M
Monk Liu 已提交
253 254 255
	if (patch_offset != ~0 && ring->funcs->patch_cond_exec)
		amdgpu_ring_patch_cond_exec(ring, patch_offset);

256
	ring->current_ctx = fence_ctx;
257
	if (vm && ring->funcs->emit_switch_buffer)
258
		amdgpu_ring_emit_switch_buffer(ring);
259
	amdgpu_ring_commit(ring);
A
Alex Deucher 已提交
260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322
	return 0;
}

/**
 * amdgpu_ib_pool_init - Init the IB (Indirect Buffer) pool
 *
 * @adev: amdgpu_device pointer
 *
 * Initialize the suballocator to manage a pool of memory
 * for use as IBs (all asics).
 * Returns 0 on success, error on failure.
 */
int amdgpu_ib_pool_init(struct amdgpu_device *adev)
{
	int r;

	if (adev->ib_pool_ready) {
		return 0;
	}
	r = amdgpu_sa_bo_manager_init(adev, &adev->ring_tmp_bo,
				      AMDGPU_IB_POOL_SIZE*64*1024,
				      AMDGPU_GPU_PAGE_SIZE,
				      AMDGPU_GEM_DOMAIN_GTT);
	if (r) {
		return r;
	}

	adev->ib_pool_ready = true;
	if (amdgpu_debugfs_sa_init(adev)) {
		dev_err(adev->dev, "failed to register debugfs file for SA\n");
	}
	return 0;
}

/**
 * amdgpu_ib_pool_fini - Free the IB (Indirect Buffer) pool
 *
 * @adev: amdgpu_device pointer
 *
 * Tear down the suballocator managing the pool of memory
 * for use as IBs (all asics).
 */
void amdgpu_ib_pool_fini(struct amdgpu_device *adev)
{
	if (adev->ib_pool_ready) {
		amdgpu_sa_bo_manager_fini(adev, &adev->ring_tmp_bo);
		adev->ib_pool_ready = false;
	}
}

/**
 * amdgpu_ib_ring_tests - test IBs on the rings
 *
 * @adev: amdgpu_device pointer
 *
 * Test an IB (Indirect Buffer) on each ring.
 * If the test fails, disable the ring.
 * Returns 0 on success, error if the primary GFX ring
 * IB test fails.
 */
int amdgpu_ib_ring_tests(struct amdgpu_device *adev)
{
	unsigned i;
323
	int r, ret = 0;
324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343
	long tmo_gfx, tmo_mm;

	tmo_mm = tmo_gfx = AMDGPU_IB_TEST_TIMEOUT;
	if (amdgpu_sriov_vf(adev)) {
		/* for MM engines in hypervisor side they are not scheduled together
		 * with CP and SDMA engines, so even in exclusive mode MM engine could
		 * still running on other VF thus the IB TEST TIMEOUT for MM engines
		 * under SR-IOV should be set to a long time. 8 sec should be enough
		 * for the MM comes back to this VF.
		 */
		tmo_mm = 8 * AMDGPU_IB_TEST_TIMEOUT;
	}

	if (amdgpu_sriov_runtime(adev)) {
		/* for CP & SDMA engines since they are scheduled together so
		 * need to make the timeout width enough to cover the time
		 * cost waiting for it coming back under RUNTIME only
		*/
		tmo_gfx = 8 * AMDGPU_IB_TEST_TIMEOUT;
	}
A
Alex Deucher 已提交
344 345 346

	for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
		struct amdgpu_ring *ring = adev->rings[i];
347
		long tmo;
A
Alex Deucher 已提交
348 349 350 351

		if (!ring || !ring->ready)
			continue;

352 353 354 355 356
		/* MM engine need more time */
		if (ring->funcs->type == AMDGPU_RING_TYPE_UVD ||
			ring->funcs->type == AMDGPU_RING_TYPE_VCE ||
			ring->funcs->type == AMDGPU_RING_TYPE_UVD_ENC ||
			ring->funcs->type == AMDGPU_RING_TYPE_VCN_DEC ||
357 358
			ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC ||
			ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG)
359 360 361 362 363
			tmo = tmo_mm;
		else
			tmo = tmo_gfx;

		r = amdgpu_ring_test_ib(ring, tmo);
A
Alex Deucher 已提交
364 365 366 367 368 369 370 371 372 373 374 375
		if (r) {
			ring->ready = false;

			if (ring == &adev->gfx.gfx_ring[0]) {
				/* oh, oh, that's really bad */
				DRM_ERROR("amdgpu: failed testing IB on GFX ring (%d).\n", r);
				adev->accel_working = false;
				return r;

			} else {
				/* still not good, but we can live with it */
				DRM_ERROR("amdgpu: failed testing IB on ring %d (%d).\n", i, r);
376
				ret = r;
A
Alex Deucher 已提交
377 378 379
			}
		}
	}
380
	return ret;
A
Alex Deucher 已提交
381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
}

/*
 * Debugfs info
 */
#if defined(CONFIG_DEBUG_FS)

static int amdgpu_debugfs_sa_info(struct seq_file *m, void *data)
{
	struct drm_info_node *node = (struct drm_info_node *) m->private;
	struct drm_device *dev = node->minor->dev;
	struct amdgpu_device *adev = dev->dev_private;

	amdgpu_sa_bo_dump_debug_info(&adev->ring_tmp_bo, m);

	return 0;

}

400
static const struct drm_info_list amdgpu_debugfs_sa_list[] = {
A
Alex Deucher 已提交
401 402 403 404 405 406 407 408 409 410 411 412 413
	{"amdgpu_sa_info", &amdgpu_debugfs_sa_info, 0, NULL},
};

#endif

static int amdgpu_debugfs_sa_init(struct amdgpu_device *adev)
{
#if defined(CONFIG_DEBUG_FS)
	return amdgpu_debugfs_add_files(adev, amdgpu_debugfs_sa_list, 1);
#else
	return 0;
#endif
}