xref: /dragonfly/sys/dev/drm/i915/intel_ringbuffer.h (revision 6a3cbbc2)
1 #ifndef _INTEL_RINGBUFFER_H_
2 #define _INTEL_RINGBUFFER_H_
3 
4 #include <linux/hashtable.h>
5 #include "i915_gem_batch_pool.h"
6 #include "i915_gem_request.h"
7 
8 #define I915_CMD_HASH_ORDER 9
9 
10 /* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill,
11  * but keeps the logic simple. Indeed, the whole purpose of this macro is just
12  * to give some inclination as to some of the magic values used in the various
13  * workarounds!
14  */
15 #define CACHELINE_BYTES 64
16 #define CACHELINE_DWORDS (CACHELINE_BYTES / sizeof(uint32_t))
17 
18 /*
19  * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
20  * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
21  * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
22  *
23  * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
24  * cacheline, the Head Pointer must not be greater than the Tail
25  * Pointer."
26  */
27 #define I915_RING_FREE_SPACE 64
28 
29 struct intel_hw_status_page {
30 	struct i915_vma *vma;
31 	u32 *page_addr;
32 	u32 ggtt_offset;
33 };
34 
35 #define I915_READ_TAIL(engine) I915_READ(RING_TAIL((engine)->mmio_base))
36 #define I915_WRITE_TAIL(engine, val) I915_WRITE(RING_TAIL((engine)->mmio_base), val)
37 
38 #define I915_READ_START(engine) I915_READ(RING_START((engine)->mmio_base))
39 #define I915_WRITE_START(engine, val) I915_WRITE(RING_START((engine)->mmio_base), val)
40 
41 #define I915_READ_HEAD(engine)  I915_READ(RING_HEAD((engine)->mmio_base))
42 #define I915_WRITE_HEAD(engine, val) I915_WRITE(RING_HEAD((engine)->mmio_base), val)
43 
44 #define I915_READ_CTL(engine) I915_READ(RING_CTL((engine)->mmio_base))
45 #define I915_WRITE_CTL(engine, val) I915_WRITE(RING_CTL((engine)->mmio_base), val)
46 
47 #define I915_READ_IMR(engine) I915_READ(RING_IMR((engine)->mmio_base))
48 #define I915_WRITE_IMR(engine, val) I915_WRITE(RING_IMR((engine)->mmio_base), val)
49 
50 #define I915_READ_MODE(engine) I915_READ(RING_MI_MODE((engine)->mmio_base))
51 #define I915_WRITE_MODE(engine, val) I915_WRITE(RING_MI_MODE((engine)->mmio_base), val)
52 
53 /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
54  * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
55  */
56 #define gen8_semaphore_seqno_size sizeof(uint64_t)
57 #define GEN8_SEMAPHORE_OFFSET(__from, __to)			     \
58 	(((__from) * I915_NUM_ENGINES  + (__to)) * gen8_semaphore_seqno_size)
59 #define GEN8_SIGNAL_OFFSET(__ring, to)			     \
60 	(dev_priv->semaphore->node.start + \
61 	 GEN8_SEMAPHORE_OFFSET((__ring)->id, (to)))
62 #define GEN8_WAIT_OFFSET(__ring, from)			     \
63 	(dev_priv->semaphore->node.start + \
64 	 GEN8_SEMAPHORE_OFFSET(from, (__ring)->id))
65 
66 enum intel_engine_hangcheck_action {
67 	HANGCHECK_IDLE = 0,
68 	HANGCHECK_WAIT,
69 	HANGCHECK_ACTIVE,
70 	HANGCHECK_KICK,
71 	HANGCHECK_HUNG,
72 };
73 
74 #define HANGCHECK_SCORE_RING_HUNG 31
75 
76 #define I915_MAX_SLICES	3
77 #define I915_MAX_SUBSLICES 3
78 
79 #define instdone_slice_mask(dev_priv__) \
80 	(INTEL_GEN(dev_priv__) == 7 ? \
81 	 1 : INTEL_INFO(dev_priv__)->sseu.slice_mask)
82 
83 #define instdone_subslice_mask(dev_priv__) \
84 	(INTEL_GEN(dev_priv__) == 7 ? \
85 	 1 : INTEL_INFO(dev_priv__)->sseu.subslice_mask)
86 
87 #define for_each_instdone_slice_subslice(dev_priv__, slice__, subslice__) \
88 	for ((slice__) = 0, (subslice__) = 0; \
89 	     (slice__) < I915_MAX_SLICES; \
90 	     (subslice__) = ((subslice__) + 1) < I915_MAX_SUBSLICES ? (subslice__) + 1 : 0, \
91 	       (slice__) += ((subslice__) == 0)) \
92 		for_each_if((BIT(slice__) & instdone_slice_mask(dev_priv__)) && \
93 			    (BIT(subslice__) & instdone_subslice_mask(dev_priv__)))
94 
95 struct intel_instdone {
96 	u32 instdone;
97 	/* The following exist only in the RCS engine */
98 	u32 slice_common;
99 	u32 sampler[I915_MAX_SLICES][I915_MAX_SUBSLICES];
100 	u32 row[I915_MAX_SLICES][I915_MAX_SUBSLICES];
101 };
102 
103 struct intel_engine_hangcheck {
104 	u64 acthd;
105 	u32 seqno;
106 	int score;
107 	enum intel_engine_hangcheck_action action;
108 	int deadlock;
109 	struct intel_instdone instdone;
110 };
111 
112 struct intel_ring {
113 	struct i915_vma *vma;
114 	void *vaddr;
115 
116 	struct intel_engine_cs *engine;
117 
118 	struct list_head request_list;
119 
120 	u32 head;
121 	u32 tail;
122 	int space;
123 	int size;
124 	int effective_size;
125 
126 	/** We track the position of the requests in the ring buffer, and
127 	 * when each is retired we increment last_retired_head as the GPU
128 	 * must have finished processing the request and so we know we
129 	 * can advance the ringbuffer up to that position.
130 	 *
131 	 * last_retired_head is set to -1 after the value is consumed so
132 	 * we can detect new retirements.
133 	 */
134 	u32 last_retired_head;
135 };
136 
137 struct i915_gem_context;
138 struct drm_i915_reg_table;
139 
140 /*
141  * we use a single page to load ctx workarounds so all of these
142  * values are referred in terms of dwords
143  *
144  * struct i915_wa_ctx_bb:
145  *  offset: specifies batch starting position, also helpful in case
146  *    if we want to have multiple batches at different offsets based on
147  *    some criteria. It is not a requirement at the moment but provides
148  *    an option for future use.
149  *  size: size of the batch in DWORDS
150  */
151 struct i915_ctx_workarounds {
152 	struct i915_wa_ctx_bb {
153 		u32 offset;
154 		u32 size;
155 	} indirect_ctx, per_ctx;
156 	struct i915_vma *vma;
157 };
158 
159 struct drm_i915_gem_request;
160 
161 struct intel_engine_cs {
162 	struct drm_i915_private *i915;
163 	const char	*name;
164 	enum intel_engine_id {
165 		RCS = 0,
166 		BCS,
167 		VCS,
168 		VCS2,	/* Keep instances of the same type engine together. */
169 		VECS
170 	} id;
171 #define I915_NUM_ENGINES 5
172 #define _VCS(n) (VCS + (n))
173 	unsigned int exec_id;
174 	enum intel_engine_hw_id {
175 		RCS_HW = 0,
176 		VCS_HW,
177 		BCS_HW,
178 		VECS_HW,
179 		VCS2_HW
180 	} hw_id;
181 	enum intel_engine_hw_id guc_id; /* XXX same as hw_id? */
182 	u64 fence_context;
183 	u32		mmio_base;
184 	unsigned int irq_shift;
185 	struct intel_ring *buffer;
186 
187 	/* Rather than have every client wait upon all user interrupts,
188 	 * with the herd waking after every interrupt and each doing the
189 	 * heavyweight seqno dance, we delegate the task (of being the
190 	 * bottom-half of the user interrupt) to the first client. After
191 	 * every interrupt, we wake up one client, who does the heavyweight
192 	 * coherent seqno read and either goes back to sleep (if incomplete),
193 	 * or wakes up all the completed clients in parallel, before then
194 	 * transferring the bottom-half status to the next client in the queue.
195 	 *
196 	 * Compared to walking the entire list of waiters in a single dedicated
197 	 * bottom-half, we reduce the latency of the first waiter by avoiding
198 	 * a context switch, but incur additional coherent seqno reads when
199 	 * following the chain of request breadcrumbs. Since it is most likely
200 	 * that we have a single client waiting on each seqno, then reducing
201 	 * the overhead of waking that client is much preferred.
202 	 */
203 	struct intel_breadcrumbs {
204 		struct task_struct __rcu *irq_seqno_bh; /* bh for interrupts */
205 		bool irq_posted;
206 
207 		spinlock_t lock; /* protects the lists of requests */
208 		struct rb_root waiters; /* sorted by retirement, priority */
209 		struct rb_root signals; /* sorted by retirement */
210 		struct intel_wait *first_wait; /* oldest waiter by retirement */
211 		struct task_struct *signaler; /* used for fence signalling */
212 		struct drm_i915_gem_request *first_signal;
213 		struct timer_list fake_irq; /* used after a missed interrupt */
214 		struct timer_list hangcheck; /* detect missed interrupts */
215 
216 		unsigned long timeout;
217 
218 		bool irq_enabled : 1;
219 		bool rpm_wakelock : 1;
220 	} breadcrumbs;
221 
222 	/*
223 	 * A pool of objects to use as shadow copies of client batch buffers
224 	 * when the command parser is enabled. Prevents the client from
225 	 * modifying the batch contents after software parsing.
226 	 */
227 	struct i915_gem_batch_pool batch_pool;
228 
229 	struct intel_hw_status_page status_page;
230 	struct i915_ctx_workarounds wa_ctx;
231 	struct i915_vma *scratch;
232 
233 	u32             irq_keep_mask; /* always keep these interrupts */
234 	u32		irq_enable_mask; /* bitmask to enable ring interrupt */
235 	void		(*irq_enable)(struct intel_engine_cs *engine);
236 	void		(*irq_disable)(struct intel_engine_cs *engine);
237 
238 	int		(*init_hw)(struct intel_engine_cs *engine);
239 	void		(*reset_hw)(struct intel_engine_cs *engine,
240 				    struct drm_i915_gem_request *req);
241 
242 	int		(*init_context)(struct drm_i915_gem_request *req);
243 
244 	int		(*emit_flush)(struct drm_i915_gem_request *request,
245 				      u32 mode);
246 #define EMIT_INVALIDATE	BIT(0)
247 #define EMIT_FLUSH	BIT(1)
248 #define EMIT_BARRIER	(EMIT_INVALIDATE | EMIT_FLUSH)
249 	int		(*emit_bb_start)(struct drm_i915_gem_request *req,
250 					 u64 offset, u32 length,
251 					 unsigned int dispatch_flags);
252 #define I915_DISPATCH_SECURE BIT(0)
253 #define I915_DISPATCH_PINNED BIT(1)
254 #define I915_DISPATCH_RS     BIT(2)
255 	int		(*emit_request)(struct drm_i915_gem_request *req);
256 
257 	/* Pass the request to the hardware queue (e.g. directly into
258 	 * the legacy ringbuffer or to the end of an execlist).
259 	 *
260 	 * This is called from an atomic context with irqs disabled; must
261 	 * be irq safe.
262 	 */
263 	void		(*submit_request)(struct drm_i915_gem_request *req);
264 
265 	/* Some chipsets are not quite as coherent as advertised and need
266 	 * an expensive kick to force a true read of the up-to-date seqno.
267 	 * However, the up-to-date seqno is not always required and the last
268 	 * seen value is good enough. Note that the seqno will always be
269 	 * monotonic, even if not coherent.
270 	 */
271 	void		(*irq_seqno_barrier)(struct intel_engine_cs *engine);
272 	void		(*cleanup)(struct intel_engine_cs *engine);
273 
274 	/* GEN8 signal/wait table - never trust comments!
275 	 *	  signal to	signal to    signal to   signal to      signal to
276 	 *	    RCS		   VCS          BCS        VECS		 VCS2
277 	 *      --------------------------------------------------------------------
278 	 *  RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
279 	 *	|-------------------------------------------------------------------
280 	 *  VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
281 	 *	|-------------------------------------------------------------------
282 	 *  BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
283 	 *	|-------------------------------------------------------------------
284 	 * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) |  NOP (0x90) | VCS2 (0x98) |
285 	 *	|-------------------------------------------------------------------
286 	 * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP  (0xc0) |
287 	 *	|-------------------------------------------------------------------
288 	 *
289 	 * Generalization:
290 	 *  f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
291 	 *  ie. transpose of g(x, y)
292 	 *
293 	 *	 sync from	sync from    sync from    sync from	sync from
294 	 *	    RCS		   VCS          BCS        VECS		 VCS2
295 	 *      --------------------------------------------------------------------
296 	 *  RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
297 	 *	|-------------------------------------------------------------------
298 	 *  VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
299 	 *	|-------------------------------------------------------------------
300 	 *  BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
301 	 *	|-------------------------------------------------------------------
302 	 * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) |  NOP (0x90) | VCS2 (0xb8) |
303 	 *	|-------------------------------------------------------------------
304 	 * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) |  NOP (0xc0) |
305 	 *	|-------------------------------------------------------------------
306 	 *
307 	 * Generalization:
308 	 *  g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
309 	 *  ie. transpose of f(x, y)
310 	 */
311 	struct {
312 		u32	sync_seqno[I915_NUM_ENGINES-1];
313 
314 		union {
315 #define GEN6_SEMAPHORE_LAST	VECS_HW
316 #define GEN6_NUM_SEMAPHORES	(GEN6_SEMAPHORE_LAST + 1)
317 #define GEN6_SEMAPHORES_MASK	GENMASK(GEN6_SEMAPHORE_LAST, 0)
318 			struct {
319 				/* our mbox written by others */
320 				u32		wait[GEN6_NUM_SEMAPHORES];
321 				/* mboxes this ring signals to */
322 				i915_reg_t	signal[GEN6_NUM_SEMAPHORES];
323 			} mbox;
324 			u64		signal_ggtt[I915_NUM_ENGINES];
325 		};
326 
327 		/* AKA wait() */
328 		int	(*sync_to)(struct drm_i915_gem_request *req,
329 				   struct drm_i915_gem_request *signal);
330 		int	(*signal)(struct drm_i915_gem_request *req);
331 	} semaphore;
332 
333 	/* Execlists */
334 	struct tasklet_struct irq_tasklet;
335 	spinlock_t execlist_lock; /* used inside tasklet, use spin_lock_bh */
336 	struct execlist_port {
337 		struct drm_i915_gem_request *request;
338 		unsigned int count;
339 	} execlist_port[2];
340 	struct list_head execlist_queue;
341 	unsigned int fw_domains;
342 	bool disable_lite_restore_wa;
343 	bool preempt_wa;
344 	u32 ctx_desc_template;
345 
346 	/**
347 	 * List of breadcrumbs associated with GPU requests currently
348 	 * outstanding.
349 	 */
350 	struct list_head request_list;
351 
352 	/**
353 	 * Seqno of request most recently submitted to request_list.
354 	 * Used exclusively by hang checker to avoid grabbing lock while
355 	 * inspecting request list.
356 	 */
357 	u32 last_submitted_seqno;
358 	u32 last_pending_seqno;
359 
360 	/* An RCU guarded pointer to the last request. No reference is
361 	 * held to the request, users must carefully acquire a reference to
362 	 * the request using i915_gem_active_get_rcu(), or hold the
363 	 * struct_mutex.
364 	 */
365 	struct i915_gem_active last_request;
366 
367 	struct i915_gem_context *last_context;
368 
369 	struct intel_engine_hangcheck hangcheck;
370 
371 	bool needs_cmd_parser;
372 
373 	/*
374 	 * Table of commands the command parser needs to know about
375 	 * for this engine.
376 	 */
377 	DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
378 
379 	/*
380 	 * Table of registers allowed in commands that read/write registers.
381 	 */
382 	const struct drm_i915_reg_table *reg_tables;
383 	int reg_table_count;
384 
385 	/*
386 	 * Returns the bitmask for the length field of the specified command.
387 	 * Return 0 for an unrecognized/invalid command.
388 	 *
389 	 * If the command parser finds an entry for a command in the engine's
390 	 * cmd_tables, it gets the command's length based on the table entry.
391 	 * If not, it calls this function to determine the per-engine length
392 	 * field encoding for the command (i.e. different opcode ranges use
393 	 * certain bits to encode the command length in the header).
394 	 */
395 	u32 (*get_cmd_length_mask)(u32 cmd_header);
396 };
397 
398 static inline unsigned
399 intel_engine_flag(const struct intel_engine_cs *engine)
400 {
401 	return 1 << engine->id;
402 }
403 
404 static inline u32
405 intel_engine_sync_index(struct intel_engine_cs *engine,
406 			struct intel_engine_cs *other)
407 {
408 	int idx;
409 
410 	/*
411 	 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
412 	 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
413 	 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
414 	 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
415 	 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
416 	 */
417 
418 	idx = (other->id - engine->id) - 1;
419 	if (idx < 0)
420 		idx += I915_NUM_ENGINES;
421 
422 	return idx;
423 }
424 
425 static inline void
426 intel_flush_status_page(struct intel_engine_cs *engine, int reg)
427 {
428 	mb();
429 	linux_clflush(&engine->status_page.page_addr[reg]);
430 	mb();
431 }
432 
433 static inline u32
434 intel_read_status_page(struct intel_engine_cs *engine, int reg)
435 {
436 	/* Ensure that the compiler doesn't optimize away the load. */
437 	return READ_ONCE(engine->status_page.page_addr[reg]);
438 }
439 
440 static inline void
441 intel_write_status_page(struct intel_engine_cs *engine,
442 			int reg, u32 value)
443 {
444 	engine->status_page.page_addr[reg] = value;
445 }
446 
447 /*
448  * Reads a dword out of the status page, which is written to from the command
449  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
450  * MI_STORE_DATA_IMM.
451  *
452  * The following dwords have a reserved meaning:
453  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
454  * 0x04: ring 0 head pointer
455  * 0x05: ring 1 head pointer (915-class)
456  * 0x06: ring 2 head pointer (915-class)
457  * 0x10-0x1b: Context status DWords (GM45)
458  * 0x1f: Last written status offset. (GM45)
459  * 0x20-0x2f: Reserved (Gen6+)
460  *
461  * The area from dword 0x30 to 0x3ff is available for driver usage.
462  */
463 #define I915_GEM_HWS_INDEX		0x30
464 #define I915_GEM_HWS_INDEX_ADDR (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
465 #define I915_GEM_HWS_SCRATCH_INDEX	0x40
466 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
467 
468 struct intel_ring *
469 intel_engine_create_ring(struct intel_engine_cs *engine, int size);
470 int intel_ring_pin(struct intel_ring *ring);
471 void intel_ring_unpin(struct intel_ring *ring);
472 void intel_ring_free(struct intel_ring *ring);
473 
474 void intel_engine_stop(struct intel_engine_cs *engine);
475 void intel_engine_cleanup(struct intel_engine_cs *engine);
476 
477 void intel_legacy_submission_resume(struct drm_i915_private *dev_priv);
478 
479 int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
480 
481 int __must_check intel_ring_begin(struct drm_i915_gem_request *req, int n);
482 int __must_check intel_ring_cacheline_align(struct drm_i915_gem_request *req);
483 
484 static inline void intel_ring_emit(struct intel_ring *ring, u32 data)
485 {
486 	*(uint32_t *)(ring->vaddr + ring->tail) = data;
487 	ring->tail += 4;
488 }
489 
490 static inline void intel_ring_emit_reg(struct intel_ring *ring, i915_reg_t reg)
491 {
492 	intel_ring_emit(ring, i915_mmio_reg_offset(reg));
493 }
494 
495 static inline void intel_ring_advance(struct intel_ring *ring)
496 {
497 	/* Dummy function.
498 	 *
499 	 * This serves as a placeholder in the code so that the reader
500 	 * can compare against the preceding intel_ring_begin() and
501 	 * check that the number of dwords emitted matches the space
502 	 * reserved for the command packet (i.e. the value passed to
503 	 * intel_ring_begin()).
504 	 */
505 }
506 
507 static inline u32 intel_ring_offset(struct intel_ring *ring, u32 value)
508 {
509 	/* Don't write ring->size (equivalent to 0) as that hangs some GPUs. */
510 	return value & (ring->size - 1);
511 }
512 
513 int __intel_ring_space(int head, int tail, int size);
514 void intel_ring_update_space(struct intel_ring *ring);
515 
516 void intel_engine_init_seqno(struct intel_engine_cs *engine, u32 seqno);
517 
518 void intel_engine_setup_common(struct intel_engine_cs *engine);
519 int intel_engine_init_common(struct intel_engine_cs *engine);
520 int intel_engine_create_scratch(struct intel_engine_cs *engine, int size);
521 void intel_engine_cleanup_common(struct intel_engine_cs *engine);
522 
523 static inline int intel_engine_idle(struct intel_engine_cs *engine,
524 				    unsigned int flags)
525 {
526 	/* Wait upon the last request to be completed */
527 	return i915_gem_active_wait_unlocked(&engine->last_request,
528 					     flags, NULL, NULL);
529 }
530 
531 int intel_init_render_ring_buffer(struct intel_engine_cs *engine);
532 int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine);
533 int intel_init_bsd2_ring_buffer(struct intel_engine_cs *engine);
534 int intel_init_blt_ring_buffer(struct intel_engine_cs *engine);
535 int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine);
536 
537 u64 intel_engine_get_active_head(struct intel_engine_cs *engine);
538 u64 intel_engine_get_last_batch_head(struct intel_engine_cs *engine);
539 
540 static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine)
541 {
542 	return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
543 }
544 
545 int init_workarounds_ring(struct intel_engine_cs *engine);
546 
547 void intel_engine_get_instdone(struct intel_engine_cs *engine,
548 			       struct intel_instdone *instdone);
549 
550 /*
551  * Arbitrary size for largest possible 'add request' sequence. The code paths
552  * are complex and variable. Empirical measurement shows that the worst case
553  * is BDW at 192 bytes (6 + 6 + 36 dwords), then ILK at 136 bytes. However,
554  * we need to allocate double the largest single packet within that emission
555  * to account for tail wraparound (so 6 + 6 + 72 dwords for BDW).
556  */
557 #define MIN_SPACE_FOR_ADD_REQUEST 336
558 
559 static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
560 {
561 	return engine->status_page.ggtt_offset + I915_GEM_HWS_INDEX_ADDR;
562 }
563 
564 /* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
565 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
566 
567 static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
568 {
569 	wait->tsk = current;
570 	wait->seqno = seqno;
571 }
572 
573 static inline bool intel_wait_complete(const struct intel_wait *wait)
574 {
575 	return RB_EMPTY_NODE(&wait->node);
576 }
577 
578 bool intel_engine_add_wait(struct intel_engine_cs *engine,
579 			   struct intel_wait *wait);
580 void intel_engine_remove_wait(struct intel_engine_cs *engine,
581 			      struct intel_wait *wait);
582 void intel_engine_enable_signaling(struct drm_i915_gem_request *request);
583 
584 static inline bool intel_engine_has_waiter(const struct intel_engine_cs *engine)
585 {
586 	return rcu_access_pointer(engine->breadcrumbs.irq_seqno_bh);
587 }
588 
589 static inline bool intel_engine_wakeup(const struct intel_engine_cs *engine)
590 {
591 	bool wakeup = false;
592 
593 	/* Note that for this not to dangerously chase a dangling pointer,
594 	 * we must hold the rcu_read_lock here.
595 	 *
596 	 * Also note that tsk is likely to be in !TASK_RUNNING state so an
597 	 * early test for tsk->state != TASK_RUNNING before wake_up_process()
598 	 * is unlikely to be beneficial.
599 	 */
600 	if (intel_engine_has_waiter(engine)) {
601 		struct task_struct *tsk;
602 
603 		rcu_read_lock();
604 		tsk = rcu_dereference(engine->breadcrumbs.irq_seqno_bh);
605 		if (tsk)
606 			wakeup = wake_up_process(tsk);
607 		rcu_read_unlock();
608 	}
609 
610 	return wakeup;
611 }
612 
613 void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine);
614 void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
615 unsigned int intel_kick_waiters(struct drm_i915_private *i915);
616 unsigned int intel_kick_signalers(struct drm_i915_private *i915);
617 
618 static inline bool intel_engine_is_active(struct intel_engine_cs *engine)
619 {
620 	return i915_gem_active_isset(&engine->last_request);
621 }
622 
623 #endif /* _INTEL_RINGBUFFER_H_ */
624