i915_request.h 12.0 KB
Newer Older
1
/*
2
 * Copyright © 2008-2018 Intel Corporation
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
 *
 * 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 (including the next
 * paragraph) 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 AUTHORS OR COPYRIGHT HOLDERS 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.
 *
 */

25 26
#ifndef I915_REQUEST_H
#define I915_REQUEST_H
27

28
#include <linux/dma-fence.h>
29 30

#include "i915_gem.h"
31
#include "i915_scheduler.h"
32
#include "i915_selftest.h"
33
#include "i915_sw_fence.h"
34

35 36
#include <uapi/drm/i915_drm.h>

J
Joonas Lahtinen 已提交
37 38
struct drm_file;
struct drm_i915_gem_object;
39
struct i915_request;
40
struct i915_timeline;
J
Joonas Lahtinen 已提交
41

42 43
struct i915_capture_list {
	struct i915_capture_list *next;
44 45 46
	struct i915_vma *vma;
};

47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69
enum {
	/*
	 * I915_FENCE_FLAG_ACTIVE - this request is currently submitted to HW.
	 *
	 * Set by __i915_request_submit() on handing over to HW, and cleared
	 * by __i915_request_unsubmit() if we preempt this request.
	 *
	 * Finally cleared for consistency on retiring the request, when
	 * we know the HW is no longer running this request.
	 *
	 * See i915_request_is_active()
	 */
	I915_FENCE_FLAG_ACTIVE = DMA_FENCE_FLAG_USER_BITS,

	/*
	 * I915_FENCE_FLAG_SIGNAL - this request is currently on signal_list
	 *
	 * Internal bookkeeping used by the breadcrumb code to track when
	 * a request is on the various signal_list.
	 */
	I915_FENCE_FLAG_SIGNAL,
};

70 71 72 73 74 75 76 77 78 79 80
/**
 * Request queue structure.
 *
 * The request queue allows us to note sequence numbers that have been emitted
 * and may be associated with active buffers to be retired.
 *
 * By keeping this list, we can avoid having to do questionable sequence
 * number comparisons on buffer last_read|write_seqno. It also allows an
 * emission time to be associated with the request for tracking how far ahead
 * of the GPU the submission is.
 *
81 82 83 84 85 86 87
 * When modifying this structure be very aware that we perform a lockless
 * RCU lookup of it that may race against reallocation of the struct
 * from the slab freelist. We intentionally do not zero the structure on
 * allocation so that the lookup can use the dangling pointers (and is
 * cogniscent that those pointers may be wrong). Instead, everything that
 * needs to be initialised must be done so explicitly.
 *
88
 * The requests are reference counted.
89
 */
90
struct i915_request {
91
	struct dma_fence fence;
92
	spinlock_t lock;
93 94 95 96 97 98 99 100 101 102 103

	/** On Which ring this request was generated */
	struct drm_i915_private *i915;

	/**
	 * Context and ring buffer related to this request
	 * Contexts are refcounted, so when this request is associated with a
	 * context, we must increment the context's refcount, to guarantee that
	 * it persists while any request is linked to it. Requests themselves
	 * are also refcounted, so the request will only be freed when the last
	 * reference to it is dismissed, and the code in
104
	 * i915_request_free() will then decrement the refcount on the
105 106
	 * context.
	 */
C
Chris Wilson 已提交
107
	struct i915_gem_context *gem_context;
108
	struct intel_engine_cs *engine;
109
	struct intel_context *hw_context;
110
	struct intel_ring *ring;
111
	struct i915_timeline *timeline;
112
	struct list_head signal_link;
113

114 115 116 117 118 119 120 121
	/*
	 * The rcu epoch of when this request was allocated. Used to judiciously
	 * apply backpressure on future allocations to ensure that under
	 * mempressure there is sufficient RCU ticks for us to reclaim our
	 * RCU protected slabs.
	 */
	unsigned long rcustate;

122 123
	/*
	 * Fences for the various phases in the request's lifetime.
124 125 126 127 128
	 *
	 * The submit fence is used to await upon all of the request's
	 * dependencies. When it is signaled, the request is ready to run.
	 * It is used by the driver to then queue the request for execution.
	 */
129
	struct i915_sw_fence submit;
130
	wait_queue_entry_t submitq;
131

132 133
	/*
	 * A list of everyone we wait upon, and everyone who waits upon us.
134 135 136 137 138 139 140
	 * Even though we will not be submitted to the hardware before the
	 * submit fence is signaled (it waits for all external events as well
	 * as our own requests), the scheduler still needs to know the
	 * dependency tree for the lifetime of the request (from execbuf
	 * to retirement), i.e. bidirectional dependency information for the
	 * request not tied to individual fences.
	 */
141
	struct i915_sched_node sched;
142 143
	struct i915_dependency dep;

144 145 146 147 148 149 150
	/*
	 * A convenience pointer to the current breadcrumb value stored in
	 * the HW status page (or our timeline's local equivalent). The full
	 * path would be rq->hw_context->ring->timeline->hwsp_seqno.
	 */
	const u32 *hwsp_seqno;

151
	/** Position in the ring of the start of the request */
152 153
	u32 head;

154 155 156
	/** Position in the ring of the start of the user packets */
	u32 infix;

157
	/**
158 159 160
	 * Position in the ring of the start of the postfix.
	 * This is required to calculate the maximum available ring space
	 * without overwriting the postfix.
161 162 163
	 */
	u32 postfix;

164
	/** Position in the ring of the end of the whole request */
165 166
	u32 tail;

167 168 169 170
	/** Position in the ring of the end of any workarounds after the tail */
	u32 wa_tail;

	/** Preallocate space in the ring for the emitting the request */
171 172 173 174 175
	u32 reserved_space;

	/** Batch buffer related to this request if any (used for
	 * error state dump only).
	 */
C
Chris Wilson 已提交
176
	struct i915_vma *batch;
177 178
	/**
	 * Additional buffers requested by userspace to be captured upon
179 180 181 182
	 * a GPU hang. The vma/obj on this list are protected by their
	 * active reference - all objects on this list must also be
	 * on the active_list (of their final request).
	 */
183
	struct i915_capture_list *capture_list;
184
	struct list_head active_list;
185 186 187 188

	/** Time at which this request was emitted, in jiffies. */
	unsigned long emitted_jiffies;

189 190
	bool waitboost;

191 192
	/** engine->request_list entry for this request */
	struct list_head link;
193

194 195 196
	/** ring->request_list entry for this request */
	struct list_head ring_link;

197 198
	struct drm_i915_file_private *file_priv;
	/** file_priv list entry for this request */
199
	struct list_head client_link;
200 201 202 203 204

	I915_SELFTEST_DECLARE(struct {
		struct list_head link;
		unsigned long delay;
	} mock;)
205 206
};

207 208
#define I915_FENCE_GFP (GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_NOWARN)

209
extern const struct dma_fence_ops i915_fence_ops;
210

211
static inline bool dma_fence_is_i915(const struct dma_fence *fence)
212 213 214 215
{
	return fence->ops == &i915_fence_ops;
}

216 217 218 219
struct i915_request * __must_check
i915_request_alloc(struct intel_engine_cs *engine,
		   struct i915_gem_context *ctx);
void i915_request_retire_upto(struct i915_request *rq);
220

221
static inline struct i915_request *
222
to_request(struct dma_fence *fence)
223 224
{
	/* We assume that NULL fence/request are interoperable */
225
	BUILD_BUG_ON(offsetof(struct i915_request, fence) != 0);
226
	GEM_BUG_ON(fence && !dma_fence_is_i915(fence));
227
	return container_of(fence, struct i915_request, fence);
228 229
}

230 231
static inline struct i915_request *
i915_request_get(struct i915_request *rq)
232
{
233
	return to_request(dma_fence_get(&rq->fence));
234 235
}

236 237
static inline struct i915_request *
i915_request_get_rcu(struct i915_request *rq)
238
{
239
	return to_request(dma_fence_get_rcu(&rq->fence));
240 241
}

242
static inline void
243
i915_request_put(struct i915_request *rq)
244
{
245
	dma_fence_put(&rq->fence);
246 247
}

248
int i915_request_await_object(struct i915_request *to,
249 250
			      struct drm_i915_gem_object *obj,
			      bool write);
251 252
int i915_request_await_dma_fence(struct i915_request *rq,
				 struct dma_fence *fence);
253

254
void i915_request_add(struct i915_request *rq);
255

256 257
void __i915_request_submit(struct i915_request *request);
void i915_request_submit(struct i915_request *request);
258

259 260
void i915_request_skip(struct i915_request *request, int error);

261 262
void __i915_request_unsubmit(struct i915_request *request);
void i915_request_unsubmit(struct i915_request *request);
263

264 265 266 267
/* Note: part of the intel_breadcrumbs family */
bool i915_request_enable_breadcrumb(struct i915_request *request);
void i915_request_cancel_breadcrumb(struct i915_request *request);

268
long i915_request_wait(struct i915_request *rq,
269 270
		       unsigned int flags,
		       long timeout)
271
	__attribute__((nonnull(1)));
272 273
#define I915_WAIT_INTERRUPTIBLE	BIT(0)
#define I915_WAIT_LOCKED	BIT(1) /* struct_mutex held, handle GPU reset */
274 275 276
#define I915_WAIT_PRIORITY	BIT(2) /* small priority bump for the request */
#define I915_WAIT_ALL		BIT(3) /* used by i915_gem_object_wait() */
#define I915_WAIT_FOR_IDLE_BOOST BIT(4)
277

278 279
static inline bool i915_request_signaled(const struct i915_request *rq)
{
280
	/* The request may live longer than its HWSP, so check flags first! */
281 282 283
	return test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &rq->fence.flags);
}

284 285 286 287 288
static inline bool i915_request_is_active(const struct i915_request *rq)
{
	return test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags);
}

289 290 291 292 293 294 295 296
/**
 * Returns true if seq1 is later than seq2.
 */
static inline bool i915_seqno_passed(u32 seq1, u32 seq2)
{
	return (s32)(seq1 - seq2) >= 0;
}

297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
static inline u32 __hwsp_seqno(const struct i915_request *rq)
{
	return READ_ONCE(*rq->hwsp_seqno);
}

/**
 * hwsp_seqno - the current breadcrumb value in the HW status page
 * @rq: the request, to chase the relevant HW status page
 *
 * The emphasis in naming here is that hwsp_seqno() is not a property of the
 * request, but an indication of the current HW state (associated with this
 * request). Its value will change as the GPU executes more requests.
 *
 * Returns the current breadcrumb value in the associated HW status page (or
 * the local timeline's equivalent) for this request. The request itself
 * has the associated breadcrumb value of rq->fence.seqno, when the HW
 * status page has that breadcrumb or later, this request is complete.
 */
static inline u32 hwsp_seqno(const struct i915_request *rq)
{
	u32 seqno;

	rcu_read_lock(); /* the HWSP may be freed at runtime */
	seqno = __hwsp_seqno(rq);
	rcu_read_unlock();

	return seqno;
}

326 327 328 329 330
static inline bool __i915_request_has_started(const struct i915_request *rq)
{
	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno - 1);
}

331 332 333 334 335 336 337 338 339 340 341
/**
 * i915_request_started - check if the request has begun being executed
 * @rq: the request
 *
 * Returns true if the request has been submitted to hardware, and the hardware
 * has advanced passed the end of the previous request and so should be either
 * currently processing the request (though it may be preempted and so
 * not necessarily the next request to complete) or have completed the request.
 */
static inline bool i915_request_started(const struct i915_request *rq)
{
342 343
	if (i915_request_signaled(rq))
		return true;
344

345
	/* Remember: started but may have since been preempted! */
346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362
	return __i915_request_has_started(rq);
}

/**
 * i915_request_is_running - check if the request may actually be executing
 * @rq: the request
 *
 * Returns true if the request is currently submitted to hardware, has passed
 * its start point (i.e. the context is setup and not busywaiting). Note that
 * it may no longer be running by the time the function returns!
 */
static inline bool i915_request_is_running(const struct i915_request *rq)
{
	if (!i915_request_is_active(rq))
		return false;

	return __i915_request_has_started(rq);
363 364
}

365
static inline bool i915_request_completed(const struct i915_request *rq)
366
{
367 368
	if (i915_request_signaled(rq))
		return true;
369

370 371
	return i915_seqno_passed(hwsp_seqno(rq), rq->fence.seqno);
}
372

373 374 375
static inline void i915_request_mark_complete(struct i915_request *rq)
{
	rq->hwsp_seqno = (u32 *)&rq->fence.seqno; /* decouple from HWSP */
376 377
}

378 379
void i915_retire_requests(struct drm_i915_private *i915);

380 381 382 383
int i915_global_request_init(void);
void i915_global_request_shrink(void);
void i915_global_request_exit(void);

384
#endif /* I915_REQUEST_H */