xref: /dflybsd-src/sys/dev/drm/i915/intel_ringbuffer.h (revision 40245fd9cad57ddb46a86213b9ecff8f6370c56b)
1 #ifndef _INTEL_RINGBUFFER_H_
2 #define _INTEL_RINGBUFFER_H_
3 
4 #include <linux/hashtable.h>
5 #include "i915_gem_batch_pool.h"
6 
7 #define I915_CMD_HASH_ORDER 9
8 
9 /* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill,
10  * but keeps the logic simple. Indeed, the whole purpose of this macro is just
11  * to give some inclination as to some of the magic values used in the various
12  * workarounds!
13  */
14 #define CACHELINE_BYTES 64
15 #define CACHELINE_DWORDS (CACHELINE_BYTES / sizeof(uint32_t))
16 
17 /*
18  * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
19  * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
20  * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
21  *
22  * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
23  * cacheline, the Head Pointer must not be greater than the Tail
24  * Pointer."
25  */
26 #define I915_RING_FREE_SPACE 64
27 
28 struct  intel_hw_status_page {
29 	u32		*page_addr;
30 	unsigned int	gfx_addr;
31 	struct		drm_i915_gem_object *obj;
32 };
33 
34 #define I915_READ_TAIL(engine) I915_READ(RING_TAIL((engine)->mmio_base))
35 #define I915_WRITE_TAIL(engine, val) I915_WRITE(RING_TAIL((engine)->mmio_base), val)
36 
37 #define I915_READ_START(engine) I915_READ(RING_START((engine)->mmio_base))
38 #define I915_WRITE_START(engine, val) I915_WRITE(RING_START((engine)->mmio_base), val)
39 
40 #define I915_READ_HEAD(engine)  I915_READ(RING_HEAD((engine)->mmio_base))
41 #define I915_WRITE_HEAD(engine, val) I915_WRITE(RING_HEAD((engine)->mmio_base), val)
42 
43 #define I915_READ_CTL(engine) I915_READ(RING_CTL((engine)->mmio_base))
44 #define I915_WRITE_CTL(engine, val) I915_WRITE(RING_CTL((engine)->mmio_base), val)
45 
46 #define I915_READ_IMR(engine) I915_READ(RING_IMR((engine)->mmio_base))
47 #define I915_WRITE_IMR(engine, val) I915_WRITE(RING_IMR((engine)->mmio_base), val)
48 
49 #define I915_READ_MODE(engine) I915_READ(RING_MI_MODE((engine)->mmio_base))
50 #define I915_WRITE_MODE(engine, val) I915_WRITE(RING_MI_MODE((engine)->mmio_base), val)
51 
52 /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
53  * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
54  */
55 #define gen8_semaphore_seqno_size sizeof(uint64_t)
56 #define GEN8_SEMAPHORE_OFFSET(__from, __to)			     \
57 	(((__from) * I915_NUM_ENGINES  + (__to)) * gen8_semaphore_seqno_size)
58 #define GEN8_SIGNAL_OFFSET(__ring, to)			     \
59 	(i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
60 	 GEN8_SEMAPHORE_OFFSET((__ring)->id, (to)))
61 #define GEN8_WAIT_OFFSET(__ring, from)			     \
62 	(i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
63 	 GEN8_SEMAPHORE_OFFSET(from, (__ring)->id))
64 
65 enum intel_ring_hangcheck_action {
66 	HANGCHECK_IDLE = 0,
67 	HANGCHECK_WAIT,
68 	HANGCHECK_ACTIVE,
69 	HANGCHECK_KICK,
70 	HANGCHECK_HUNG,
71 };
72 
73 #define HANGCHECK_SCORE_RING_HUNG 31
74 
75 struct intel_ring_hangcheck {
76 	u64 acthd;
77 	unsigned long user_interrupts;
78 	u32 seqno;
79 	int score;
80 	enum intel_ring_hangcheck_action action;
81 	int deadlock;
82 	u32 instdone[I915_NUM_INSTDONE_REG];
83 };
84 
85 struct intel_ringbuffer {
86 	struct drm_i915_gem_object *obj;
87 	void *vaddr;
88 	struct i915_vma *vma;
89 
90 	struct intel_engine_cs *engine;
91 	struct list_head link;
92 
93 	u32 head;
94 	u32 tail;
95 	int space;
96 	int size;
97 	int effective_size;
98 
99 	/** We track the position of the requests in the ring buffer, and
100 	 * when each is retired we increment last_retired_head as the GPU
101 	 * must have finished processing the request and so we know we
102 	 * can advance the ringbuffer up to that position.
103 	 *
104 	 * last_retired_head is set to -1 after the value is consumed so
105 	 * we can detect new retirements.
106 	 */
107 	u32 last_retired_head;
108 };
109 
110 struct i915_gem_context;
111 struct drm_i915_reg_table;
112 
113 /*
114  * we use a single page to load ctx workarounds so all of these
115  * values are referred in terms of dwords
116  *
117  * struct i915_wa_ctx_bb:
118  *  offset: specifies batch starting position, also helpful in case
119  *    if we want to have multiple batches at different offsets based on
120  *    some criteria. It is not a requirement at the moment but provides
121  *    an option for future use.
122  *  size: size of the batch in DWORDS
123  */
124 struct  i915_ctx_workarounds {
125 	struct i915_wa_ctx_bb {
126 		u32 offset;
127 		u32 size;
128 	} indirect_ctx, per_ctx;
129 	struct drm_i915_gem_object *obj;
130 };
131 
132 struct drm_i915_gem_request;
133 
134 struct intel_engine_cs {
135 	struct drm_i915_private *i915;
136 	const char	*name;
137 	enum intel_engine_id {
138 		RCS = 0,
139 		BCS,
140 		VCS,
141 		VCS2,	/* Keep instances of the same type engine together. */
142 		VECS
143 	} id;
144 #define I915_NUM_ENGINES 5
145 #define _VCS(n) (VCS + (n))
146 	unsigned int exec_id;
147 	unsigned int hw_id;
148 	unsigned int guc_id; /* XXX same as hw_id? */
149 	u64 fence_context;
150 	u32		mmio_base;
151 	unsigned int irq_shift;
152 	struct intel_ringbuffer *buffer;
153 	struct list_head buffers;
154 
155 	/* Rather than have every client wait upon all user interrupts,
156 	 * with the herd waking after every interrupt and each doing the
157 	 * heavyweight seqno dance, we delegate the task (of being the
158 	 * bottom-half of the user interrupt) to the first client. After
159 	 * every interrupt, we wake up one client, who does the heavyweight
160 	 * coherent seqno read and either goes back to sleep (if incomplete),
161 	 * or wakes up all the completed clients in parallel, before then
162 	 * transferring the bottom-half status to the next client in the queue.
163 	 *
164 	 * Compared to walking the entire list of waiters in a single dedicated
165 	 * bottom-half, we reduce the latency of the first waiter by avoiding
166 	 * a context switch, but incur additional coherent seqno reads when
167 	 * following the chain of request breadcrumbs. Since it is most likely
168 	 * that we have a single client waiting on each seqno, then reducing
169 	 * the overhead of waking that client is much preferred.
170 	 */
171 	struct intel_breadcrumbs {
172 		struct task_struct *irq_seqno_bh; /* bh for user interrupts */
173 		unsigned long irq_wakeups;
174 		bool irq_posted;
175 
176 		struct lock lock; /* protects the lists of requests */
177 		struct rb_root waiters; /* sorted by retirement, priority */
178 		struct rb_root signals; /* sorted by retirement */
179 		struct intel_wait *first_wait; /* oldest waiter by retirement */
180 		struct task_struct *signaler; /* used for fence signalling */
181 		struct drm_i915_gem_request *first_signal;
182 		struct timer_list fake_irq; /* used after a missed interrupt */
183 
184 		bool irq_enabled : 1;
185 		bool rpm_wakelock : 1;
186 	} breadcrumbs;
187 
188 	/*
189 	 * A pool of objects to use as shadow copies of client batch buffers
190 	 * when the command parser is enabled. Prevents the client from
191 	 * modifying the batch contents after software parsing.
192 	 */
193 	struct i915_gem_batch_pool batch_pool;
194 
195 	struct intel_hw_status_page status_page;
196 	struct i915_ctx_workarounds wa_ctx;
197 
198 	u32             irq_keep_mask; /* always keep these interrupts */
199 	u32		irq_enable_mask; /* bitmask to enable ring interrupt */
200 	void		(*irq_enable)(struct intel_engine_cs *engine);
201 	void		(*irq_disable)(struct intel_engine_cs *engine);
202 
203 	int		(*init_hw)(struct intel_engine_cs *engine);
204 
205 	int		(*init_context)(struct drm_i915_gem_request *req);
206 
207 	void		(*write_tail)(struct intel_engine_cs *engine,
208 				      u32 value);
209 	int __must_check (*flush)(struct drm_i915_gem_request *req,
210 				  u32	invalidate_domains,
211 				  u32	flush_domains);
212 	int		(*add_request)(struct drm_i915_gem_request *req);
213 	/* Some chipsets are not quite as coherent as advertised and need
214 	 * an expensive kick to force a true read of the up-to-date seqno.
215 	 * However, the up-to-date seqno is not always required and the last
216 	 * seen value is good enough. Note that the seqno will always be
217 	 * monotonic, even if not coherent.
218 	 */
219 	void		(*irq_seqno_barrier)(struct intel_engine_cs *engine);
220 	int		(*dispatch_execbuffer)(struct drm_i915_gem_request *req,
221 					       u64 offset, u32 length,
222 					       unsigned dispatch_flags);
223 #define I915_DISPATCH_SECURE 0x1
224 #define I915_DISPATCH_PINNED 0x2
225 #define I915_DISPATCH_RS     0x4
226 	void		(*cleanup)(struct intel_engine_cs *engine);
227 
228 	/* GEN8 signal/wait table - never trust comments!
229 	 *	  signal to	signal to    signal to   signal to      signal to
230 	 *	    RCS		   VCS          BCS        VECS		 VCS2
231 	 *      --------------------------------------------------------------------
232 	 *  RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
233 	 *	|-------------------------------------------------------------------
234 	 *  VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
235 	 *	|-------------------------------------------------------------------
236 	 *  BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
237 	 *	|-------------------------------------------------------------------
238 	 * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) |  NOP (0x90) | VCS2 (0x98) |
239 	 *	|-------------------------------------------------------------------
240 	 * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP  (0xc0) |
241 	 *	|-------------------------------------------------------------------
242 	 *
243 	 * Generalization:
244 	 *  f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
245 	 *  ie. transpose of g(x, y)
246 	 *
247 	 *	 sync from	sync from    sync from    sync from	sync from
248 	 *	    RCS		   VCS          BCS        VECS		 VCS2
249 	 *      --------------------------------------------------------------------
250 	 *  RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
251 	 *	|-------------------------------------------------------------------
252 	 *  VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
253 	 *	|-------------------------------------------------------------------
254 	 *  BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
255 	 *	|-------------------------------------------------------------------
256 	 * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) |  NOP (0x90) | VCS2 (0xb8) |
257 	 *	|-------------------------------------------------------------------
258 	 * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) |  NOP (0xc0) |
259 	 *	|-------------------------------------------------------------------
260 	 *
261 	 * Generalization:
262 	 *  g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
263 	 *  ie. transpose of f(x, y)
264 	 */
265 	struct {
266 		u32	sync_seqno[I915_NUM_ENGINES-1];
267 
268 		union {
269 			struct {
270 				/* our mbox written by others */
271 				u32		wait[I915_NUM_ENGINES];
272 				/* mboxes this ring signals to */
273 				i915_reg_t	signal[I915_NUM_ENGINES];
274 			} mbox;
275 			u64		signal_ggtt[I915_NUM_ENGINES];
276 		};
277 
278 		/* AKA wait() */
279 		int	(*sync_to)(struct drm_i915_gem_request *to_req,
280 				   struct intel_engine_cs *from,
281 				   u32 seqno);
282 		int	(*signal)(struct drm_i915_gem_request *signaller_req,
283 				  /* num_dwords needed by caller */
284 				  unsigned int num_dwords);
285 	} semaphore;
286 
287 	/* Execlists */
288 	struct tasklet_struct irq_tasklet;
289 	struct lock execlist_lock;	/* used inside tasklet, use spin_lock_bh */
290 	struct list_head execlist_queue;
291 	unsigned int fw_domains;
292 	unsigned int next_context_status_buffer;
293 	unsigned int idle_lite_restore_wa;
294 	bool disable_lite_restore_wa;
295 	u32 ctx_desc_template;
296 	int		(*emit_request)(struct drm_i915_gem_request *request);
297 	int		(*emit_flush)(struct drm_i915_gem_request *request,
298 				      u32 invalidate_domains,
299 				      u32 flush_domains);
300 	int		(*emit_bb_start)(struct drm_i915_gem_request *req,
301 					 u64 offset, unsigned dispatch_flags);
302 
303 	/**
304 	 * List of objects currently involved in rendering from the
305 	 * ringbuffer.
306 	 *
307 	 * Includes buffers having the contents of their GPU caches
308 	 * flushed, not necessarily primitives.  last_read_req
309 	 * represents when the rendering involved will be completed.
310 	 *
311 	 * A reference is held on the buffer while on this list.
312 	 */
313 	struct list_head active_list;
314 
315 	/**
316 	 * List of breadcrumbs associated with GPU requests currently
317 	 * outstanding.
318 	 */
319 	struct list_head request_list;
320 
321 	/**
322 	 * Seqno of request most recently submitted to request_list.
323 	 * Used exclusively by hang checker to avoid grabbing lock while
324 	 * inspecting request list.
325 	 */
326 	u32 last_submitted_seqno;
327 
328 	bool gpu_caches_dirty;
329 
330 	struct i915_gem_context *last_context;
331 
332 	struct intel_ring_hangcheck hangcheck;
333 
334 	struct {
335 		struct drm_i915_gem_object *obj;
336 		u32 gtt_offset;
337 	} scratch;
338 
339 	bool needs_cmd_parser;
340 
341 	/*
342 	 * Table of commands the command parser needs to know about
343 	 * for this ring.
344 	 */
345 	DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
346 
347 	/*
348 	 * Table of registers allowed in commands that read/write registers.
349 	 */
350 	const struct drm_i915_reg_table *reg_tables;
351 	int reg_table_count;
352 
353 	/*
354 	 * Returns the bitmask for the length field of the specified command.
355 	 * Return 0 for an unrecognized/invalid command.
356 	 *
357 	 * If the command parser finds an entry for a command in the ring's
358 	 * cmd_tables, it gets the command's length based on the table entry.
359 	 * If not, it calls this function to determine the per-ring length field
360 	 * encoding for the command (i.e. certain opcode ranges use certain bits
361 	 * to encode the command length in the header).
362 	 */
363 	u32 (*get_cmd_length_mask)(u32 cmd_header);
364 };
365 
366 static inline bool
367 intel_engine_initialized(const struct intel_engine_cs *engine)
368 {
369 	return engine->i915 != NULL;
370 }
371 
372 static inline unsigned
373 intel_engine_flag(const struct intel_engine_cs *engine)
374 {
375 	return 1 << engine->id;
376 }
377 
378 static inline u32
379 intel_ring_sync_index(struct intel_engine_cs *engine,
380 		      struct intel_engine_cs *other)
381 {
382 	int idx;
383 
384 	/*
385 	 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
386 	 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
387 	 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
388 	 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
389 	 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
390 	 */
391 
392 	idx = (other - engine) - 1;
393 	if (idx < 0)
394 		idx += I915_NUM_ENGINES;
395 
396 	return idx;
397 }
398 
399 static inline void
400 intel_flush_status_page(struct intel_engine_cs *engine, int reg)
401 {
402 	mb();
403 	linux_clflush(&engine->status_page.page_addr[reg]);
404 	mb();
405 }
406 
407 static inline u32
408 intel_read_status_page(struct intel_engine_cs *engine, int reg)
409 {
410 	/* Ensure that the compiler doesn't optimize away the load. */
411 	return READ_ONCE(engine->status_page.page_addr[reg]);
412 }
413 
414 static inline void
415 intel_write_status_page(struct intel_engine_cs *engine,
416 			int reg, u32 value)
417 {
418 	engine->status_page.page_addr[reg] = value;
419 }
420 
421 /*
422  * Reads a dword out of the status page, which is written to from the command
423  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
424  * MI_STORE_DATA_IMM.
425  *
426  * The following dwords have a reserved meaning:
427  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
428  * 0x04: ring 0 head pointer
429  * 0x05: ring 1 head pointer (915-class)
430  * 0x06: ring 2 head pointer (915-class)
431  * 0x10-0x1b: Context status DWords (GM45)
432  * 0x1f: Last written status offset. (GM45)
433  * 0x20-0x2f: Reserved (Gen6+)
434  *
435  * The area from dword 0x30 to 0x3ff is available for driver usage.
436  */
437 #define I915_GEM_HWS_INDEX		0x30
438 #define I915_GEM_HWS_INDEX_ADDR (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
439 #define I915_GEM_HWS_SCRATCH_INDEX	0x40
440 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
441 
442 struct intel_ringbuffer *
443 intel_engine_create_ringbuffer(struct intel_engine_cs *engine, int size);
444 int intel_pin_and_map_ringbuffer_obj(struct drm_i915_private *dev_priv,
445 				     struct intel_ringbuffer *ringbuf);
446 void intel_unpin_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
447 void intel_ringbuffer_free(struct intel_ringbuffer *ring);
448 
449 void intel_stop_engine(struct intel_engine_cs *engine);
450 void intel_cleanup_engine(struct intel_engine_cs *engine);
451 
452 int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
453 
454 int __must_check intel_ring_begin(struct drm_i915_gem_request *req, int n);
455 int __must_check intel_ring_cacheline_align(struct drm_i915_gem_request *req);
456 
457 static inline void __intel_ringbuffer_emit(struct intel_ringbuffer *rb,
458 					   u32 data)
459 {
460 	*(uint32_t *)(rb->vaddr + rb->tail) = data;
461 	rb->tail += 4;
462 }
463 
464 static inline void __intel_ringbuffer_advance(struct intel_ringbuffer *rb)
465 {
466 	rb->tail &= rb->size - 1;
467 }
468 
469 static inline void intel_ring_emit(struct intel_engine_cs *engine, u32 data)
470 {
471 	__intel_ringbuffer_emit(engine->buffer, data);
472 }
473 
474 static inline void intel_ring_emit_reg(struct intel_engine_cs *engine,
475 				       i915_reg_t reg)
476 {
477 	intel_ring_emit(engine, i915_mmio_reg_offset(reg));
478 }
479 
480 static inline void intel_ring_advance(struct intel_engine_cs *engine)
481 {
482 	__intel_ringbuffer_advance(engine->buffer);
483 }
484 
485 int __intel_ring_space(int head, int tail, int size);
486 void intel_ring_update_space(struct intel_ringbuffer *ringbuf);
487 
488 int __must_check intel_engine_idle(struct intel_engine_cs *engine);
489 void intel_ring_init_seqno(struct intel_engine_cs *engine, u32 seqno);
490 int intel_ring_flush_all_caches(struct drm_i915_gem_request *req);
491 int intel_ring_invalidate_all_caches(struct drm_i915_gem_request *req);
492 
493 int intel_init_pipe_control(struct intel_engine_cs *engine, int size);
494 void intel_fini_pipe_control(struct intel_engine_cs *engine);
495 
496 void intel_engine_setup_common(struct intel_engine_cs *engine);
497 int intel_engine_init_common(struct intel_engine_cs *engine);
498 
499 int intel_init_render_ring_buffer(struct intel_engine_cs *engine);
500 int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine);
501 int intel_init_bsd2_ring_buffer(struct intel_engine_cs *engine);
502 int intel_init_blt_ring_buffer(struct intel_engine_cs *engine);
503 int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine);
504 
505 u64 intel_ring_get_active_head(struct intel_engine_cs *engine);
506 static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine)
507 {
508 	return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
509 }
510 
511 int init_workarounds_ring(struct intel_engine_cs *engine);
512 
513 static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf)
514 {
515 	return ringbuf->tail;
516 }
517 
518 /*
519  * Arbitrary size for largest possible 'add request' sequence. The code paths
520  * are complex and variable. Empirical measurement shows that the worst case
521  * is BDW at 192 bytes (6 + 6 + 36 dwords), then ILK at 136 bytes. However,
522  * we need to allocate double the largest single packet within that emission
523  * to account for tail wraparound (so 6 + 6 + 72 dwords for BDW).
524  */
525 #define MIN_SPACE_FOR_ADD_REQUEST 336
526 
527 static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
528 {
529 	return engine->status_page.gfx_addr + I915_GEM_HWS_INDEX_ADDR;
530 }
531 
532 /* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
533 struct intel_wait {
534 	struct rb_node node;
535 	struct task_struct *tsk;
536 	u32 seqno;
537 };
538 
539 struct intel_signal_node {
540 	struct rb_node node;
541 	struct intel_wait wait;
542 };
543 
544 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
545 
546 static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
547 {
548 	wait->tsk = current;
549 	wait->seqno = seqno;
550 }
551 
552 static inline bool intel_wait_complete(const struct intel_wait *wait)
553 {
554 	return RB_EMPTY_NODE(&wait->node);
555 }
556 
557 bool intel_engine_add_wait(struct intel_engine_cs *engine,
558 			   struct intel_wait *wait);
559 void intel_engine_remove_wait(struct intel_engine_cs *engine,
560 			      struct intel_wait *wait);
561 void intel_engine_enable_signaling(struct drm_i915_gem_request *request);
562 
563 static inline bool intel_engine_has_waiter(struct intel_engine_cs *engine)
564 {
565 	return READ_ONCE(engine->breadcrumbs.irq_seqno_bh);
566 }
567 
568 static inline bool intel_engine_wakeup(struct intel_engine_cs *engine)
569 {
570 	bool wakeup = false;
571 	struct task_struct *tsk = READ_ONCE(engine->breadcrumbs.irq_seqno_bh);
572 	/* Note that for this not to dangerously chase a dangling pointer,
573 	 * the caller is responsible for ensure that the task remain valid for
574 	 * wake_up_process() i.e. that the RCU grace period cannot expire.
575 	 *
576 	 * Also note that tsk is likely to be in !TASK_RUNNING state so an
577 	 * early test for tsk->state != TASK_RUNNING before wake_up_process()
578 	 * is unlikely to be beneficial.
579 	 */
580 	if (tsk)
581 		wakeup = wake_up_process(tsk);
582 	return wakeup;
583 }
584 
585 void intel_engine_enable_fake_irq(struct intel_engine_cs *engine);
586 void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
587 unsigned int intel_kick_waiters(struct drm_i915_private *i915);
588 unsigned int intel_kick_signalers(struct drm_i915_private *i915);
589 
590 #endif /* _INTEL_RINGBUFFER_H_ */
591