1 #ifndef _INTEL_RINGBUFFER_H_
2 #define _INTEL_RINGBUFFER_H_
4 #include <linux/hashtable.h>
5 #include "i915_gem_batch_pool.h"
7 #define I915_CMD_HASH_ORDER 9
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
14 #define CACHELINE_BYTES 64
15 #define CACHELINE_DWORDS (CACHELINE_BYTES / sizeof(uint32_t))
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"
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
26 #define I915_RING_FREE_SPACE 64
28 struct intel_hw_status_page {
30 unsigned int gfx_addr;
31 struct drm_i915_gem_object *obj;
34 #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base))
35 #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val)
37 #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base))
38 #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val)
40 #define I915_READ_HEAD(ring) I915_READ(RING_HEAD((ring)->mmio_base))
41 #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val)
43 #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base))
44 #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val)
46 #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base))
47 #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val)
49 #define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base))
50 #define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val)
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.
55 #define i915_semaphore_seqno_size sizeof(uint64_t)
56 #define GEN8_SIGNAL_OFFSET(__ring, to) \
57 (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
58 ((__ring)->id * I915_NUM_RINGS * i915_semaphore_seqno_size) + \
59 (i915_semaphore_seqno_size * (to)))
61 #define GEN8_WAIT_OFFSET(__ring, from) \
62 (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
63 ((from) * I915_NUM_RINGS * i915_semaphore_seqno_size) + \
64 (i915_semaphore_seqno_size * (__ring)->id))
66 #define GEN8_RING_SEMAPHORE_INIT do { \
67 if (!dev_priv->semaphore_obj) { \
70 ring->semaphore.signal_ggtt[RCS] = GEN8_SIGNAL_OFFSET(ring, RCS); \
71 ring->semaphore.signal_ggtt[VCS] = GEN8_SIGNAL_OFFSET(ring, VCS); \
72 ring->semaphore.signal_ggtt[BCS] = GEN8_SIGNAL_OFFSET(ring, BCS); \
73 ring->semaphore.signal_ggtt[VECS] = GEN8_SIGNAL_OFFSET(ring, VECS); \
74 ring->semaphore.signal_ggtt[VCS2] = GEN8_SIGNAL_OFFSET(ring, VCS2); \
75 ring->semaphore.signal_ggtt[ring->id] = MI_SEMAPHORE_SYNC_INVALID; \
78 enum intel_ring_hangcheck_action {
82 HANGCHECK_ACTIVE_LOOP,
87 #define HANGCHECK_SCORE_RING_HUNG 31
89 struct intel_ring_hangcheck {
94 enum intel_ring_hangcheck_action action;
96 u32 instdone[I915_NUM_INSTDONE_REG];
99 struct intel_ringbuffer {
100 struct drm_i915_gem_object *obj;
101 void __iomem *virtual_start;
102 struct i915_vma *vma;
104 struct intel_engine_cs *ring;
105 struct list_head link;
114 bool reserved_in_use;
116 /** We track the position of the requests in the ring buffer, and
117 * when each is retired we increment last_retired_head as the GPU
118 * must have finished processing the request and so we know we
119 * can advance the ringbuffer up to that position.
121 * last_retired_head is set to -1 after the value is consumed so
122 * we can detect new retirements.
124 u32 last_retired_head;
127 struct intel_context;
128 struct drm_i915_reg_descriptor;
131 * we use a single page to load ctx workarounds so all of these
132 * values are referred in terms of dwords
134 * struct i915_wa_ctx_bb:
135 * offset: specifies batch starting position, also helpful in case
136 * if we want to have multiple batches at different offsets based on
137 * some criteria. It is not a requirement at the moment but provides
138 * an option for future use.
139 * size: size of the batch in DWORDS
141 struct i915_ctx_workarounds {
142 struct i915_wa_ctx_bb {
145 } indirect_ctx, per_ctx;
146 struct drm_i915_gem_object *obj;
149 struct intel_engine_cs {
155 VCS2, /* Keep instances of the same type engine together. */
158 #define I915_NUM_RINGS 5
159 #define _VCS(n) (VCS + (n))
160 unsigned int exec_id;
163 struct drm_device *dev;
164 struct intel_ringbuffer *buffer;
165 struct list_head buffers;
168 * A pool of objects to use as shadow copies of client batch buffers
169 * when the command parser is enabled. Prevents the client from
170 * modifying the batch contents after software parsing.
172 struct i915_gem_batch_pool batch_pool;
174 struct intel_hw_status_page status_page;
175 struct i915_ctx_workarounds wa_ctx;
177 unsigned irq_refcount; /* protected by dev_priv->irq_lock */
178 u32 irq_enable_mask; /* bitmask to enable ring interrupt */
179 struct drm_i915_gem_request *trace_irq_req;
180 bool __must_check (*irq_get)(struct intel_engine_cs *ring);
181 void (*irq_put)(struct intel_engine_cs *ring);
183 int (*init_hw)(struct intel_engine_cs *ring);
185 int (*init_context)(struct drm_i915_gem_request *req);
187 void (*write_tail)(struct intel_engine_cs *ring,
189 int __must_check (*flush)(struct drm_i915_gem_request *req,
190 u32 invalidate_domains,
192 int (*add_request)(struct drm_i915_gem_request *req);
193 /* Some chipsets are not quite as coherent as advertised and need
194 * an expensive kick to force a true read of the up-to-date seqno.
195 * However, the up-to-date seqno is not always required and the last
196 * seen value is good enough. Note that the seqno will always be
197 * monotonic, even if not coherent.
199 u32 (*get_seqno)(struct intel_engine_cs *ring,
200 bool lazy_coherency);
201 void (*set_seqno)(struct intel_engine_cs *ring,
203 int (*dispatch_execbuffer)(struct drm_i915_gem_request *req,
204 u64 offset, u32 length,
205 unsigned dispatch_flags);
206 #define I915_DISPATCH_SECURE 0x1
207 #define I915_DISPATCH_PINNED 0x2
208 #define I915_DISPATCH_RS 0x4
209 void (*cleanup)(struct intel_engine_cs *ring);
211 /* GEN8 signal/wait table - never trust comments!
212 * signal to signal to signal to signal to signal to
213 * RCS VCS BCS VECS VCS2
214 * --------------------------------------------------------------------
215 * RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
216 * |-------------------------------------------------------------------
217 * VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
218 * |-------------------------------------------------------------------
219 * BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
220 * |-------------------------------------------------------------------
221 * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) | NOP (0x90) | VCS2 (0x98) |
222 * |-------------------------------------------------------------------
223 * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP (0xc0) |
224 * |-------------------------------------------------------------------
227 * f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
228 * ie. transpose of g(x, y)
230 * sync from sync from sync from sync from sync from
231 * RCS VCS BCS VECS VCS2
232 * --------------------------------------------------------------------
233 * RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
234 * |-------------------------------------------------------------------
235 * VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
236 * |-------------------------------------------------------------------
237 * BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
238 * |-------------------------------------------------------------------
239 * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) | NOP (0x90) | VCS2 (0xb8) |
240 * |-------------------------------------------------------------------
241 * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) | NOP (0xc0) |
242 * |-------------------------------------------------------------------
245 * g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
246 * ie. transpose of f(x, y)
249 u32 sync_seqno[I915_NUM_RINGS-1];
253 /* our mbox written by others */
254 u32 wait[I915_NUM_RINGS];
255 /* mboxes this ring signals to */
256 i915_reg_t signal[I915_NUM_RINGS];
258 u64 signal_ggtt[I915_NUM_RINGS];
262 int (*sync_to)(struct drm_i915_gem_request *to_req,
263 struct intel_engine_cs *from,
265 int (*signal)(struct drm_i915_gem_request *signaller_req,
266 /* num_dwords needed by caller */
267 unsigned int num_dwords);
271 spinlock_t execlist_lock;
272 struct list_head execlist_queue;
273 struct list_head execlist_retired_req_list;
274 unsigned int next_context_status_buffer;
275 unsigned int idle_lite_restore_wa;
276 bool disable_lite_restore_wa;
277 u32 ctx_desc_template;
278 u32 irq_keep_mask; /* bitmask for interrupts that should not be masked */
279 int (*emit_request)(struct drm_i915_gem_request *request);
280 int (*emit_flush)(struct drm_i915_gem_request *request,
281 u32 invalidate_domains,
283 int (*emit_bb_start)(struct drm_i915_gem_request *req,
284 u64 offset, unsigned dispatch_flags);
287 * List of objects currently involved in rendering from the
290 * Includes buffers having the contents of their GPU caches
291 * flushed, not necessarily primitives. last_read_req
292 * represents when the rendering involved will be completed.
294 * A reference is held on the buffer while on this list.
296 struct list_head active_list;
299 * List of breadcrumbs associated with GPU requests currently
302 struct list_head request_list;
305 * Seqno of request most recently submitted to request_list.
306 * Used exclusively by hang checker to avoid grabbing lock while
307 * inspecting request list.
309 u32 last_submitted_seqno;
311 bool gpu_caches_dirty;
313 wait_queue_head_t irq_queue;
315 struct intel_context *last_context;
317 struct intel_ring_hangcheck hangcheck;
320 struct drm_i915_gem_object *obj;
322 volatile u32 *cpu_page;
325 bool needs_cmd_parser;
328 * Table of commands the command parser needs to know about
331 DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
334 * Table of registers allowed in commands that read/write registers.
336 const struct drm_i915_reg_descriptor *reg_table;
340 * Table of registers allowed in commands that read/write registers, but
341 * only from the DRM master.
343 const struct drm_i915_reg_descriptor *master_reg_table;
344 int master_reg_count;
347 * Returns the bitmask for the length field of the specified command.
348 * Return 0 for an unrecognized/invalid command.
350 * If the command parser finds an entry for a command in the ring's
351 * cmd_tables, it gets the command's length based on the table entry.
352 * If not, it calls this function to determine the per-ring length field
353 * encoding for the command (i.e. certain opcode ranges use certain bits
354 * to encode the command length in the header).
356 u32 (*get_cmd_length_mask)(u32 cmd_header);
360 intel_ring_initialized(struct intel_engine_cs *ring)
362 return ring->dev != NULL;
365 static inline unsigned
366 intel_ring_flag(struct intel_engine_cs *ring)
368 return 1 << ring->id;
372 intel_ring_sync_index(struct intel_engine_cs *ring,
373 struct intel_engine_cs *other)
378 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
379 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
380 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
381 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
382 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
385 idx = (other - ring) - 1;
387 idx += I915_NUM_RINGS;
393 intel_flush_status_page(struct intel_engine_cs *ring, int reg)
395 drm_clflush_virt_range(&ring->status_page.page_addr[reg],
400 intel_read_status_page(struct intel_engine_cs *ring,
403 /* Ensure that the compiler doesn't optimize away the load. */
405 return ring->status_page.page_addr[reg];
409 intel_write_status_page(struct intel_engine_cs *ring,
412 ring->status_page.page_addr[reg] = value;
416 * Reads a dword out of the status page, which is written to from the command
417 * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
420 * The following dwords have a reserved meaning:
421 * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
422 * 0x04: ring 0 head pointer
423 * 0x05: ring 1 head pointer (915-class)
424 * 0x06: ring 2 head pointer (915-class)
425 * 0x10-0x1b: Context status DWords (GM45)
426 * 0x1f: Last written status offset. (GM45)
427 * 0x20-0x2f: Reserved (Gen6+)
429 * The area from dword 0x30 to 0x3ff is available for driver usage.
431 #define I915_GEM_HWS_INDEX 0x30
432 #define I915_GEM_HWS_INDEX_ADDR (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
433 #define I915_GEM_HWS_SCRATCH_INDEX 0x40
434 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
436 struct intel_ringbuffer *
437 intel_engine_create_ringbuffer(struct intel_engine_cs *engine, int size);
438 int intel_pin_and_map_ringbuffer_obj(struct drm_device *dev,
439 struct intel_ringbuffer *ringbuf);
440 void intel_unpin_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
441 void intel_ringbuffer_free(struct intel_ringbuffer *ring);
443 void intel_stop_ring_buffer(struct intel_engine_cs *ring);
444 void intel_cleanup_ring_buffer(struct intel_engine_cs *ring);
446 int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
448 int __must_check intel_ring_begin(struct drm_i915_gem_request *req, int n);
449 int __must_check intel_ring_cacheline_align(struct drm_i915_gem_request *req);
450 static inline void intel_ring_emit(struct intel_engine_cs *ring,
453 struct intel_ringbuffer *ringbuf = ring->buffer;
454 iowrite32(data, ringbuf->virtual_start + ringbuf->tail);
457 static inline void intel_ring_emit_reg(struct intel_engine_cs *ring,
460 intel_ring_emit(ring, i915_mmio_reg_offset(reg));
462 static inline void intel_ring_advance(struct intel_engine_cs *ring)
464 struct intel_ringbuffer *ringbuf = ring->buffer;
465 ringbuf->tail &= ringbuf->size - 1;
467 int __intel_ring_space(int head, int tail, int size);
468 void intel_ring_update_space(struct intel_ringbuffer *ringbuf);
469 int intel_ring_space(struct intel_ringbuffer *ringbuf);
470 bool intel_ring_stopped(struct intel_engine_cs *ring);
472 int __must_check intel_ring_idle(struct intel_engine_cs *ring);
473 void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno);
474 int intel_ring_flush_all_caches(struct drm_i915_gem_request *req);
475 int intel_ring_invalidate_all_caches(struct drm_i915_gem_request *req);
477 void intel_fini_pipe_control(struct intel_engine_cs *ring);
478 int intel_init_pipe_control(struct intel_engine_cs *ring);
480 int intel_init_render_ring_buffer(struct drm_device *dev);
481 int intel_init_bsd_ring_buffer(struct drm_device *dev);
482 int intel_init_bsd2_ring_buffer(struct drm_device *dev);
483 int intel_init_blt_ring_buffer(struct drm_device *dev);
484 int intel_init_vebox_ring_buffer(struct drm_device *dev);
486 u64 intel_ring_get_active_head(struct intel_engine_cs *ring);
488 int init_workarounds_ring(struct intel_engine_cs *ring);
490 static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf)
492 return ringbuf->tail;
496 * Arbitrary size for largest possible 'add request' sequence. The code paths
497 * are complex and variable. Empirical measurement shows that the worst case
498 * is ILK at 136 words. Reserving too much is better than reserving too little
499 * as that allows for corner cases that might have been missed. So the figure
500 * has been rounded up to 160 words.
502 #define MIN_SPACE_FOR_ADD_REQUEST 160
505 * Reserve space in the ring to guarantee that the i915_add_request() call
506 * will always have sufficient room to do its stuff. The request creation
507 * code calls this automatically.
509 void intel_ring_reserved_space_reserve(struct intel_ringbuffer *ringbuf, int size);
510 /* Cancel the reservation, e.g. because the request is being discarded. */
511 void intel_ring_reserved_space_cancel(struct intel_ringbuffer *ringbuf);
512 /* Use the reserved space - for use by i915_add_request() only. */
513 void intel_ring_reserved_space_use(struct intel_ringbuffer *ringbuf);
514 /* Finish with the reserved space - for use by i915_add_request() only. */
515 void intel_ring_reserved_space_end(struct intel_ringbuffer *ringbuf);
517 /* Legacy ringbuffer specific portion of reservation code: */
518 int intel_ring_reserve_space(struct drm_i915_gem_request *request);
520 #endif /* _INTEL_RINGBUFFER_H_ */