• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 struct intel_engine_hangcheck {
77 	u64 acthd;
78 	u32 seqno;
79 	int score;
80 	enum intel_engine_hangcheck_action action;
81 	int deadlock;
82 	u32 instdone[I915_NUM_INSTDONE_REG];
83 };
84 
85 struct intel_ring {
86 	struct i915_vma *vma;
87 	void *vaddr;
88 
89 	struct intel_engine_cs *engine;
90 
91 	struct list_head request_list;
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 i915_vma *vma;
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 	enum intel_engine_hw_id {
148 		RCS_HW = 0,
149 		VCS_HW,
150 		BCS_HW,
151 		VECS_HW,
152 		VCS2_HW
153 	} hw_id;
154 	enum intel_engine_hw_id guc_id; /* XXX same as hw_id? */
155 	u64 fence_context;
156 	u32		mmio_base;
157 	unsigned int irq_shift;
158 	struct intel_ring *buffer;
159 
160 	/* Rather than have every client wait upon all user interrupts,
161 	 * with the herd waking after every interrupt and each doing the
162 	 * heavyweight seqno dance, we delegate the task (of being the
163 	 * bottom-half of the user interrupt) to the first client. After
164 	 * every interrupt, we wake up one client, who does the heavyweight
165 	 * coherent seqno read and either goes back to sleep (if incomplete),
166 	 * or wakes up all the completed clients in parallel, before then
167 	 * transferring the bottom-half status to the next client in the queue.
168 	 *
169 	 * Compared to walking the entire list of waiters in a single dedicated
170 	 * bottom-half, we reduce the latency of the first waiter by avoiding
171 	 * a context switch, but incur additional coherent seqno reads when
172 	 * following the chain of request breadcrumbs. Since it is most likely
173 	 * that we have a single client waiting on each seqno, then reducing
174 	 * the overhead of waking that client is much preferred.
175 	 */
176 	struct intel_breadcrumbs {
177 		struct task_struct __rcu *irq_seqno_bh; /* bh for interrupts */
178 		bool irq_posted;
179 
180 		spinlock_t lock; /* protects the lists of requests */
181 		struct rb_root waiters; /* sorted by retirement, priority */
182 		struct rb_root signals; /* sorted by retirement */
183 		struct intel_wait *first_wait; /* oldest waiter by retirement */
184 		struct task_struct *signaler; /* used for fence signalling */
185 		struct drm_i915_gem_request *first_signal;
186 		struct timer_list fake_irq; /* used after a missed interrupt */
187 		struct timer_list hangcheck; /* detect missed interrupts */
188 
189 		unsigned long timeout;
190 
191 		bool irq_enabled : 1;
192 		bool rpm_wakelock : 1;
193 	} breadcrumbs;
194 
195 	/*
196 	 * A pool of objects to use as shadow copies of client batch buffers
197 	 * when the command parser is enabled. Prevents the client from
198 	 * modifying the batch contents after software parsing.
199 	 */
200 	struct i915_gem_batch_pool batch_pool;
201 
202 	struct intel_hw_status_page status_page;
203 	struct i915_ctx_workarounds wa_ctx;
204 	struct i915_vma *scratch;
205 
206 	u32             irq_keep_mask; /* always keep these interrupts */
207 	u32		irq_enable_mask; /* bitmask to enable ring interrupt */
208 	void		(*irq_enable)(struct intel_engine_cs *engine);
209 	void		(*irq_disable)(struct intel_engine_cs *engine);
210 
211 	int		(*init_hw)(struct intel_engine_cs *engine);
212 	void		(*reset_hw)(struct intel_engine_cs *engine,
213 				    struct drm_i915_gem_request *req);
214 
215 	int		(*init_context)(struct drm_i915_gem_request *req);
216 
217 	int		(*emit_flush)(struct drm_i915_gem_request *request,
218 				      u32 mode);
219 #define EMIT_INVALIDATE	BIT(0)
220 #define EMIT_FLUSH	BIT(1)
221 #define EMIT_BARRIER	(EMIT_INVALIDATE | EMIT_FLUSH)
222 	int		(*emit_bb_start)(struct drm_i915_gem_request *req,
223 					 u64 offset, u32 length,
224 					 unsigned int dispatch_flags);
225 #define I915_DISPATCH_SECURE BIT(0)
226 #define I915_DISPATCH_PINNED BIT(1)
227 #define I915_DISPATCH_RS     BIT(2)
228 	int		(*emit_request)(struct drm_i915_gem_request *req);
229 
230 	/* Pass the request to the hardware queue (e.g. directly into
231 	 * the legacy ringbuffer or to the end of an execlist).
232 	 *
233 	 * This is called from an atomic context with irqs disabled; must
234 	 * be irq safe.
235 	 */
236 	void		(*submit_request)(struct drm_i915_gem_request *req);
237 
238 	/* Some chipsets are not quite as coherent as advertised and need
239 	 * an expensive kick to force a true read of the up-to-date seqno.
240 	 * However, the up-to-date seqno is not always required and the last
241 	 * seen value is good enough. Note that the seqno will always be
242 	 * monotonic, even if not coherent.
243 	 */
244 	void		(*irq_seqno_barrier)(struct intel_engine_cs *engine);
245 	void		(*cleanup)(struct intel_engine_cs *engine);
246 
247 	/* GEN8 signal/wait table - never trust comments!
248 	 *	  signal to	signal to    signal to   signal to      signal to
249 	 *	    RCS		   VCS          BCS        VECS		 VCS2
250 	 *      --------------------------------------------------------------------
251 	 *  RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
252 	 *	|-------------------------------------------------------------------
253 	 *  VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
254 	 *	|-------------------------------------------------------------------
255 	 *  BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
256 	 *	|-------------------------------------------------------------------
257 	 * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) |  NOP (0x90) | VCS2 (0x98) |
258 	 *	|-------------------------------------------------------------------
259 	 * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP  (0xc0) |
260 	 *	|-------------------------------------------------------------------
261 	 *
262 	 * Generalization:
263 	 *  f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
264 	 *  ie. transpose of g(x, y)
265 	 *
266 	 *	 sync from	sync from    sync from    sync from	sync from
267 	 *	    RCS		   VCS          BCS        VECS		 VCS2
268 	 *      --------------------------------------------------------------------
269 	 *  RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
270 	 *	|-------------------------------------------------------------------
271 	 *  VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
272 	 *	|-------------------------------------------------------------------
273 	 *  BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
274 	 *	|-------------------------------------------------------------------
275 	 * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) |  NOP (0x90) | VCS2 (0xb8) |
276 	 *	|-------------------------------------------------------------------
277 	 * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) |  NOP (0xc0) |
278 	 *	|-------------------------------------------------------------------
279 	 *
280 	 * Generalization:
281 	 *  g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
282 	 *  ie. transpose of f(x, y)
283 	 */
284 	struct {
285 		u32	sync_seqno[I915_NUM_ENGINES-1];
286 
287 		union {
288 #define GEN6_SEMAPHORE_LAST	VECS_HW
289 #define GEN6_NUM_SEMAPHORES	(GEN6_SEMAPHORE_LAST + 1)
290 #define GEN6_SEMAPHORES_MASK	GENMASK(GEN6_SEMAPHORE_LAST, 0)
291 			struct {
292 				/* our mbox written by others */
293 				u32		wait[GEN6_NUM_SEMAPHORES];
294 				/* mboxes this ring signals to */
295 				i915_reg_t	signal[GEN6_NUM_SEMAPHORES];
296 			} mbox;
297 			u64		signal_ggtt[I915_NUM_ENGINES];
298 		};
299 
300 		/* AKA wait() */
301 		int	(*sync_to)(struct drm_i915_gem_request *req,
302 				   struct drm_i915_gem_request *signal);
303 		int	(*signal)(struct drm_i915_gem_request *req);
304 	} semaphore;
305 
306 	/* Execlists */
307 	struct tasklet_struct irq_tasklet;
308 	spinlock_t execlist_lock; /* used inside tasklet, use spin_lock_bh */
309 	struct execlist_port {
310 		struct drm_i915_gem_request *request;
311 		unsigned int count;
312 	} execlist_port[2];
313 	struct list_head execlist_queue;
314 	unsigned int fw_domains;
315 	bool disable_lite_restore_wa;
316 	bool preempt_wa;
317 	u32 ctx_desc_template;
318 
319 	/**
320 	 * List of breadcrumbs associated with GPU requests currently
321 	 * outstanding.
322 	 */
323 	struct list_head request_list;
324 
325 	/**
326 	 * Seqno of request most recently submitted to request_list.
327 	 * Used exclusively by hang checker to avoid grabbing lock while
328 	 * inspecting request list.
329 	 */
330 	u32 last_submitted_seqno;
331 	u32 last_pending_seqno;
332 
333 	/* An RCU guarded pointer to the last request. No reference is
334 	 * held to the request, users must carefully acquire a reference to
335 	 * the request using i915_gem_active_get_rcu(), or hold the
336 	 * struct_mutex.
337 	 */
338 	struct i915_gem_active last_request;
339 
340 	struct i915_gem_context *last_context;
341 
342 	struct intel_engine_hangcheck hangcheck;
343 
344 	bool needs_cmd_parser;
345 
346 	/*
347 	 * Table of commands the command parser needs to know about
348 	 * for this engine.
349 	 */
350 	DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
351 
352 	/*
353 	 * Table of registers allowed in commands that read/write registers.
354 	 */
355 	const struct drm_i915_reg_table *reg_tables;
356 	int reg_table_count;
357 
358 	/*
359 	 * Returns the bitmask for the length field of the specified command.
360 	 * Return 0 for an unrecognized/invalid command.
361 	 *
362 	 * If the command parser finds an entry for a command in the engine's
363 	 * cmd_tables, it gets the command's length based on the table entry.
364 	 * If not, it calls this function to determine the per-engine length
365 	 * field encoding for the command (i.e. different opcode ranges use
366 	 * certain bits to encode the command length in the header).
367 	 */
368 	u32 (*get_cmd_length_mask)(u32 cmd_header);
369 };
370 
371 static inline bool
intel_engine_initialized(const struct intel_engine_cs * engine)372 intel_engine_initialized(const struct intel_engine_cs *engine)
373 {
374 	return engine->i915 != NULL;
375 }
376 
377 static inline unsigned
intel_engine_flag(const struct intel_engine_cs * engine)378 intel_engine_flag(const struct intel_engine_cs *engine)
379 {
380 	return 1 << engine->id;
381 }
382 
383 static inline u32
intel_engine_sync_index(struct intel_engine_cs * engine,struct intel_engine_cs * other)384 intel_engine_sync_index(struct intel_engine_cs *engine,
385 			struct intel_engine_cs *other)
386 {
387 	int idx;
388 
389 	/*
390 	 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
391 	 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
392 	 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
393 	 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
394 	 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
395 	 */
396 
397 	idx = (other - engine) - 1;
398 	if (idx < 0)
399 		idx += I915_NUM_ENGINES;
400 
401 	return idx;
402 }
403 
404 static inline void
intel_flush_status_page(struct intel_engine_cs * engine,int reg)405 intel_flush_status_page(struct intel_engine_cs *engine, int reg)
406 {
407 	mb();
408 	clflush(&engine->status_page.page_addr[reg]);
409 	mb();
410 }
411 
412 static inline u32
intel_read_status_page(struct intel_engine_cs * engine,int reg)413 intel_read_status_page(struct intel_engine_cs *engine, int reg)
414 {
415 	/* Ensure that the compiler doesn't optimize away the load. */
416 	return READ_ONCE(engine->status_page.page_addr[reg]);
417 }
418 
419 static inline void
intel_write_status_page(struct intel_engine_cs * engine,int reg,u32 value)420 intel_write_status_page(struct intel_engine_cs *engine,
421 			int reg, u32 value)
422 {
423 	engine->status_page.page_addr[reg] = value;
424 }
425 
426 /*
427  * Reads a dword out of the status page, which is written to from the command
428  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
429  * MI_STORE_DATA_IMM.
430  *
431  * The following dwords have a reserved meaning:
432  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
433  * 0x04: ring 0 head pointer
434  * 0x05: ring 1 head pointer (915-class)
435  * 0x06: ring 2 head pointer (915-class)
436  * 0x10-0x1b: Context status DWords (GM45)
437  * 0x1f: Last written status offset. (GM45)
438  * 0x20-0x2f: Reserved (Gen6+)
439  *
440  * The area from dword 0x30 to 0x3ff is available for driver usage.
441  */
442 #define I915_GEM_HWS_INDEX		0x30
443 #define I915_GEM_HWS_INDEX_ADDR (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
444 #define I915_GEM_HWS_SCRATCH_INDEX	0x40
445 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
446 
447 struct intel_ring *
448 intel_engine_create_ring(struct intel_engine_cs *engine, int size);
449 int intel_ring_pin(struct intel_ring *ring);
450 void intel_ring_unpin(struct intel_ring *ring);
451 void intel_ring_free(struct intel_ring *ring);
452 
453 void intel_engine_stop(struct intel_engine_cs *engine);
454 void intel_engine_cleanup(struct intel_engine_cs *engine);
455 
456 void intel_legacy_submission_resume(struct drm_i915_private *dev_priv);
457 
458 int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
459 
460 int __must_check intel_ring_begin(struct drm_i915_gem_request *req, int n);
461 int __must_check intel_ring_cacheline_align(struct drm_i915_gem_request *req);
462 
intel_ring_emit(struct intel_ring * ring,u32 data)463 static inline void intel_ring_emit(struct intel_ring *ring, u32 data)
464 {
465 	*(uint32_t *)(ring->vaddr + ring->tail) = data;
466 	ring->tail += 4;
467 }
468 
intel_ring_emit_reg(struct intel_ring * ring,i915_reg_t reg)469 static inline void intel_ring_emit_reg(struct intel_ring *ring, i915_reg_t reg)
470 {
471 	intel_ring_emit(ring, i915_mmio_reg_offset(reg));
472 }
473 
intel_ring_advance(struct intel_ring * ring)474 static inline void intel_ring_advance(struct intel_ring *ring)
475 {
476 	/* Dummy function.
477 	 *
478 	 * This serves as a placeholder in the code so that the reader
479 	 * can compare against the preceding intel_ring_begin() and
480 	 * check that the number of dwords emitted matches the space
481 	 * reserved for the command packet (i.e. the value passed to
482 	 * intel_ring_begin()).
483 	 */
484 }
485 
intel_ring_offset(struct intel_ring * ring,u32 value)486 static inline u32 intel_ring_offset(struct intel_ring *ring, u32 value)
487 {
488 	/* Don't write ring->size (equivalent to 0) as that hangs some GPUs. */
489 	return value & (ring->size - 1);
490 }
491 
492 int __intel_ring_space(int head, int tail, int size);
493 void intel_ring_update_space(struct intel_ring *ring);
494 
495 void intel_engine_init_seqno(struct intel_engine_cs *engine, u32 seqno);
496 
497 void intel_engine_setup_common(struct intel_engine_cs *engine);
498 int intel_engine_init_common(struct intel_engine_cs *engine);
499 int intel_engine_create_scratch(struct intel_engine_cs *engine, int size);
500 void intel_engine_cleanup_common(struct intel_engine_cs *engine);
501 
intel_engine_idle(struct intel_engine_cs * engine,unsigned int flags)502 static inline int intel_engine_idle(struct intel_engine_cs *engine,
503 				    unsigned int flags)
504 {
505 	/* Wait upon the last request to be completed */
506 	return i915_gem_active_wait_unlocked(&engine->last_request,
507 					     flags, NULL, NULL);
508 }
509 
510 int intel_init_render_ring_buffer(struct intel_engine_cs *engine);
511 int intel_init_bsd_ring_buffer(struct intel_engine_cs *engine);
512 int intel_init_bsd2_ring_buffer(struct intel_engine_cs *engine);
513 int intel_init_blt_ring_buffer(struct intel_engine_cs *engine);
514 int intel_init_vebox_ring_buffer(struct intel_engine_cs *engine);
515 
516 u64 intel_engine_get_active_head(struct intel_engine_cs *engine);
intel_engine_get_seqno(struct intel_engine_cs * engine)517 static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine)
518 {
519 	return intel_read_status_page(engine, I915_GEM_HWS_INDEX);
520 }
521 
522 int init_workarounds_ring(struct intel_engine_cs *engine);
523 
524 /*
525  * Arbitrary size for largest possible 'add request' sequence. The code paths
526  * are complex and variable. Empirical measurement shows that the worst case
527  * is BDW at 192 bytes (6 + 6 + 36 dwords), then ILK at 136 bytes. However,
528  * we need to allocate double the largest single packet within that emission
529  * to account for tail wraparound (so 6 + 6 + 72 dwords for BDW).
530  */
531 #define MIN_SPACE_FOR_ADD_REQUEST 336
532 
intel_hws_seqno_address(struct intel_engine_cs * engine)533 static inline u32 intel_hws_seqno_address(struct intel_engine_cs *engine)
534 {
535 	return engine->status_page.ggtt_offset + I915_GEM_HWS_INDEX_ADDR;
536 }
537 
538 /* intel_breadcrumbs.c -- user interrupt bottom-half for waiters */
539 int intel_engine_init_breadcrumbs(struct intel_engine_cs *engine);
540 
intel_wait_init(struct intel_wait * wait,u32 seqno)541 static inline void intel_wait_init(struct intel_wait *wait, u32 seqno)
542 {
543 	wait->tsk = current;
544 	wait->seqno = seqno;
545 }
546 
intel_wait_complete(const struct intel_wait * wait)547 static inline bool intel_wait_complete(const struct intel_wait *wait)
548 {
549 	return RB_EMPTY_NODE(&wait->node);
550 }
551 
552 bool intel_engine_add_wait(struct intel_engine_cs *engine,
553 			   struct intel_wait *wait);
554 void intel_engine_remove_wait(struct intel_engine_cs *engine,
555 			      struct intel_wait *wait);
556 void intel_engine_enable_signaling(struct drm_i915_gem_request *request);
557 
intel_engine_has_waiter(const struct intel_engine_cs * engine)558 static inline bool intel_engine_has_waiter(const struct intel_engine_cs *engine)
559 {
560 	return rcu_access_pointer(engine->breadcrumbs.irq_seqno_bh);
561 }
562 
intel_engine_wakeup(const struct intel_engine_cs * engine)563 static inline bool intel_engine_wakeup(const struct intel_engine_cs *engine)
564 {
565 	bool wakeup = false;
566 
567 	/* Note that for this not to dangerously chase a dangling pointer,
568 	 * we must hold the rcu_read_lock here.
569 	 *
570 	 * Also note that tsk is likely to be in !TASK_RUNNING state so an
571 	 * early test for tsk->state != TASK_RUNNING before wake_up_process()
572 	 * is unlikely to be beneficial.
573 	 */
574 	if (intel_engine_has_waiter(engine)) {
575 		struct task_struct *tsk;
576 
577 		rcu_read_lock();
578 		tsk = rcu_dereference(engine->breadcrumbs.irq_seqno_bh);
579 		if (tsk)
580 			wakeup = wake_up_process(tsk);
581 		rcu_read_unlock();
582 	}
583 
584 	return wakeup;
585 }
586 
587 void intel_engine_reset_breadcrumbs(struct intel_engine_cs *engine);
588 void intel_engine_fini_breadcrumbs(struct intel_engine_cs *engine);
589 unsigned int intel_kick_waiters(struct drm_i915_private *i915);
590 unsigned int intel_kick_signalers(struct drm_i915_private *i915);
591 
intel_engine_is_active(struct intel_engine_cs * engine)592 static inline bool intel_engine_is_active(struct intel_engine_cs *engine)
593 {
594 	return i915_gem_active_isset(&engine->last_request);
595 }
596 
597 #endif /* _INTEL_RINGBUFFER_H_ */
598