1 #ifndef _INTEL_RINGBUFFER_H_
2 #define _INTEL_RINGBUFFER_H_
4 #include <linux/hashtable.h>
6 #define I915_CMD_HASH_ORDER 9
8 /* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill,
9 * but keeps the logic simple. Indeed, the whole purpose of this macro is just
10 * to give some inclination as to some of the magic values used in the various
13 #define CACHELINE_BYTES 64
16 * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
17 * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
18 * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
20 * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
21 * cacheline, the Head Pointer must not be greater than the Tail
24 #define I915_RING_FREE_SPACE 64
26 struct intel_hw_status_page
{
28 unsigned int gfx_addr
;
29 struct drm_i915_gem_object
*obj
;
32 #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base))
33 #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val)
35 #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base))
36 #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val)
38 #define I915_READ_HEAD(ring) I915_READ(RING_HEAD((ring)->mmio_base))
39 #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val)
41 #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base))
42 #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val)
44 #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base))
45 #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val)
47 #define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base))
48 #define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val)
50 /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
51 * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
53 #define i915_semaphore_seqno_size sizeof(uint64_t)
54 #define GEN8_SIGNAL_OFFSET(__ring, to) \
55 (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
56 ((__ring)->id * I915_NUM_RINGS * i915_semaphore_seqno_size) + \
57 (i915_semaphore_seqno_size * (to)))
59 #define GEN8_WAIT_OFFSET(__ring, from) \
60 (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
61 ((from) * I915_NUM_RINGS * i915_semaphore_seqno_size) + \
62 (i915_semaphore_seqno_size * (__ring)->id))
64 #define GEN8_RING_SEMAPHORE_INIT do { \
65 if (!dev_priv->semaphore_obj) { \
68 ring->semaphore.signal_ggtt[RCS] = GEN8_SIGNAL_OFFSET(ring, RCS); \
69 ring->semaphore.signal_ggtt[VCS] = GEN8_SIGNAL_OFFSET(ring, VCS); \
70 ring->semaphore.signal_ggtt[BCS] = GEN8_SIGNAL_OFFSET(ring, BCS); \
71 ring->semaphore.signal_ggtt[VECS] = GEN8_SIGNAL_OFFSET(ring, VECS); \
72 ring->semaphore.signal_ggtt[VCS2] = GEN8_SIGNAL_OFFSET(ring, VCS2); \
73 ring->semaphore.signal_ggtt[ring->id] = MI_SEMAPHORE_SYNC_INVALID; \
76 enum intel_ring_hangcheck_action
{
80 HANGCHECK_ACTIVE_LOOP
,
85 #define HANGCHECK_SCORE_RING_HUNG 31
87 struct intel_ring_hangcheck
{
92 enum intel_ring_hangcheck_action action
;
96 struct intel_ringbuffer
{
97 struct drm_i915_gem_object
*obj
;
98 void __iomem
*virtual_start
;
100 struct intel_engine_cs
*ring
;
103 * FIXME: This backpointer is an artifact of the history of how the
104 * execlist patches came into being. It will get removed once the basic
107 struct intel_context
*FIXME_lrc_ctx
;
115 /** We track the position of the requests in the ring buffer, and
116 * when each is retired we increment last_retired_head as the GPU
117 * must have finished processing the request and so we know we
118 * can advance the ringbuffer up to that position.
120 * last_retired_head is set to -1 after the value is consumed so
121 * we can detect new retirements.
123 u32 last_retired_head
;
126 struct intel_engine_cs
{
135 #define I915_NUM_RINGS 5
136 #define LAST_USER_RING (VECS + 1)
138 struct drm_device
*dev
;
139 struct intel_ringbuffer
*buffer
;
141 struct intel_hw_status_page status_page
;
143 unsigned irq_refcount
; /* protected by dev_priv->irq_lock */
144 u32 irq_enable_mask
; /* bitmask to enable ring interrupt */
146 bool __must_check (*irq_get
)(struct intel_engine_cs
*ring
);
147 void (*irq_put
)(struct intel_engine_cs
*ring
);
149 int (*init
)(struct intel_engine_cs
*ring
);
151 int (*init_context
)(struct intel_engine_cs
*ring
,
152 struct intel_context
*ctx
);
154 void (*write_tail
)(struct intel_engine_cs
*ring
,
156 int __must_check (*flush
)(struct intel_engine_cs
*ring
,
157 u32 invalidate_domains
,
159 int (*add_request
)(struct intel_engine_cs
*ring
);
160 /* Some chipsets are not quite as coherent as advertised and need
161 * an expensive kick to force a true read of the up-to-date seqno.
162 * However, the up-to-date seqno is not always required and the last
163 * seen value is good enough. Note that the seqno will always be
164 * monotonic, even if not coherent.
166 u32 (*get_seqno
)(struct intel_engine_cs
*ring
,
167 bool lazy_coherency
);
168 void (*set_seqno
)(struct intel_engine_cs
*ring
,
170 int (*dispatch_execbuffer
)(struct intel_engine_cs
*ring
,
171 u64 offset
, u32 length
,
173 #define I915_DISPATCH_SECURE 0x1
174 #define I915_DISPATCH_PINNED 0x2
175 void (*cleanup
)(struct intel_engine_cs
*ring
);
177 /* GEN8 signal/wait table - never trust comments!
178 * signal to signal to signal to signal to signal to
179 * RCS VCS BCS VECS VCS2
180 * --------------------------------------------------------------------
181 * RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
182 * |-------------------------------------------------------------------
183 * VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
184 * |-------------------------------------------------------------------
185 * BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
186 * |-------------------------------------------------------------------
187 * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) | NOP (0x90) | VCS2 (0x98) |
188 * |-------------------------------------------------------------------
189 * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP (0xc0) |
190 * |-------------------------------------------------------------------
193 * f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
194 * ie. transpose of g(x, y)
196 * sync from sync from sync from sync from sync from
197 * RCS VCS BCS VECS VCS2
198 * --------------------------------------------------------------------
199 * RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
200 * |-------------------------------------------------------------------
201 * VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
202 * |-------------------------------------------------------------------
203 * BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
204 * |-------------------------------------------------------------------
205 * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) | NOP (0x90) | VCS2 (0xb8) |
206 * |-------------------------------------------------------------------
207 * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) | NOP (0xc0) |
208 * |-------------------------------------------------------------------
211 * g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
212 * ie. transpose of f(x, y)
215 u32 sync_seqno
[I915_NUM_RINGS
-1];
219 /* our mbox written by others */
220 u32 wait
[I915_NUM_RINGS
];
221 /* mboxes this ring signals to */
222 u32 signal
[I915_NUM_RINGS
];
224 u64 signal_ggtt
[I915_NUM_RINGS
];
228 int (*sync_to
)(struct intel_engine_cs
*ring
,
229 struct intel_engine_cs
*to
,
231 int (*signal
)(struct intel_engine_cs
*signaller
,
232 /* num_dwords needed by caller */
233 unsigned int num_dwords
);
237 spinlock_t execlist_lock
;
238 struct list_head execlist_queue
;
239 struct list_head execlist_retired_req_list
;
240 u8 next_context_status_buffer
;
241 u32 irq_keep_mask
; /* bitmask for interrupts that should not be masked */
242 int (*emit_request
)(struct intel_ringbuffer
*ringbuf
);
243 int (*emit_flush
)(struct intel_ringbuffer
*ringbuf
,
244 u32 invalidate_domains
,
246 int (*emit_bb_start
)(struct intel_ringbuffer
*ringbuf
,
247 u64 offset
, unsigned flags
);
250 * List of objects currently involved in rendering from the
253 * Includes buffers having the contents of their GPU caches
254 * flushed, not necessarily primitives. last_rendering_seqno
255 * represents when the rendering involved will be completed.
257 * A reference is held on the buffer while on this list.
259 struct list_head active_list
;
262 * List of breadcrumbs associated with GPU requests currently
265 struct list_head request_list
;
268 * Do we have some not yet emitted requests outstanding?
270 struct drm_i915_gem_request
*preallocated_lazy_request
;
271 u32 outstanding_lazy_seqno
;
272 bool gpu_caches_dirty
;
275 wait_queue_head_t irq_queue
;
277 struct intel_context
*default_context
;
278 struct intel_context
*last_context
;
280 struct intel_ring_hangcheck hangcheck
;
283 struct drm_i915_gem_object
*obj
;
285 volatile u32
*cpu_page
;
288 bool needs_cmd_parser
;
291 * Table of commands the command parser needs to know about
294 DECLARE_HASHTABLE(cmd_hash
, I915_CMD_HASH_ORDER
);
297 * Table of registers allowed in commands that read/write registers.
299 const u32
*reg_table
;
303 * Table of registers allowed in commands that read/write registers, but
304 * only from the DRM master.
306 const u32
*master_reg_table
;
307 int master_reg_count
;
310 * Returns the bitmask for the length field of the specified command.
311 * Return 0 for an unrecognized/invalid command.
313 * If the command parser finds an entry for a command in the ring's
314 * cmd_tables, it gets the command's length based on the table entry.
315 * If not, it calls this function to determine the per-ring length field
316 * encoding for the command (i.e. certain opcode ranges use certain bits
317 * to encode the command length in the header).
319 u32 (*get_cmd_length_mask
)(u32 cmd_header
);
322 bool intel_ring_initialized(struct intel_engine_cs
*ring
);
324 static inline unsigned
325 intel_ring_flag(struct intel_engine_cs
*ring
)
327 return 1 << ring
->id
;
331 intel_ring_sync_index(struct intel_engine_cs
*ring
,
332 struct intel_engine_cs
*other
)
337 * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
338 * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
339 * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
340 * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
341 * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
344 idx
= (other
- ring
) - 1;
346 idx
+= I915_NUM_RINGS
;
352 intel_read_status_page(struct intel_engine_cs
*ring
,
355 /* Ensure that the compiler doesn't optimize away the load. */
357 return ring
->status_page
.page_addr
[reg
];
361 intel_write_status_page(struct intel_engine_cs
*ring
,
364 ring
->status_page
.page_addr
[reg
] = value
;
368 * Reads a dword out of the status page, which is written to from the command
369 * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
372 * The following dwords have a reserved meaning:
373 * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
374 * 0x04: ring 0 head pointer
375 * 0x05: ring 1 head pointer (915-class)
376 * 0x06: ring 2 head pointer (915-class)
377 * 0x10-0x1b: Context status DWords (GM45)
378 * 0x1f: Last written status offset. (GM45)
380 * The area from dword 0x20 to 0x3ff is available for driver usage.
382 #define I915_GEM_HWS_INDEX 0x20
383 #define I915_GEM_HWS_SCRATCH_INDEX 0x30
384 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
386 void intel_unpin_ringbuffer_obj(struct intel_ringbuffer
*ringbuf
);
387 int intel_pin_and_map_ringbuffer_obj(struct drm_device
*dev
,
388 struct intel_ringbuffer
*ringbuf
);
389 void intel_destroy_ringbuffer_obj(struct intel_ringbuffer
*ringbuf
);
390 int intel_alloc_ringbuffer_obj(struct drm_device
*dev
,
391 struct intel_ringbuffer
*ringbuf
);
393 void intel_stop_ring_buffer(struct intel_engine_cs
*ring
);
394 void intel_cleanup_ring_buffer(struct intel_engine_cs
*ring
);
396 int __must_check
intel_ring_begin(struct intel_engine_cs
*ring
, int n
);
397 int __must_check
intel_ring_cacheline_align(struct intel_engine_cs
*ring
);
398 static inline void intel_ring_emit(struct intel_engine_cs
*ring
,
401 struct intel_ringbuffer
*ringbuf
= ring
->buffer
;
402 iowrite32(data
, ringbuf
->virtual_start
+ ringbuf
->tail
);
405 static inline void intel_ring_advance(struct intel_engine_cs
*ring
)
407 struct intel_ringbuffer
*ringbuf
= ring
->buffer
;
408 ringbuf
->tail
&= ringbuf
->size
- 1;
410 int __intel_ring_space(int head
, int tail
, int size
);
411 int intel_ring_space(struct intel_ringbuffer
*ringbuf
);
412 bool intel_ring_stopped(struct intel_engine_cs
*ring
);
413 void __intel_ring_advance(struct intel_engine_cs
*ring
);
415 int __must_check
intel_ring_idle(struct intel_engine_cs
*ring
);
416 void intel_ring_init_seqno(struct intel_engine_cs
*ring
, u32 seqno
);
417 int intel_ring_flush_all_caches(struct intel_engine_cs
*ring
);
418 int intel_ring_invalidate_all_caches(struct intel_engine_cs
*ring
);
420 void intel_fini_pipe_control(struct intel_engine_cs
*ring
);
421 int intel_init_pipe_control(struct intel_engine_cs
*ring
);
423 int intel_init_render_ring_buffer(struct drm_device
*dev
);
424 int intel_init_bsd_ring_buffer(struct drm_device
*dev
);
425 int intel_init_bsd2_ring_buffer(struct drm_device
*dev
);
426 int intel_init_blt_ring_buffer(struct drm_device
*dev
);
427 int intel_init_vebox_ring_buffer(struct drm_device
*dev
);
429 u64
intel_ring_get_active_head(struct intel_engine_cs
*ring
);
430 void intel_ring_setup_status_page(struct intel_engine_cs
*ring
);
432 int init_workarounds_ring(struct intel_engine_cs
*ring
);
434 static inline u32
intel_ring_get_tail(struct intel_ringbuffer
*ringbuf
)
436 return ringbuf
->tail
;
439 static inline u32
intel_ring_get_seqno(struct intel_engine_cs
*ring
)
441 BUG_ON(ring
->outstanding_lazy_seqno
== 0);
442 return ring
->outstanding_lazy_seqno
;
445 static inline void i915_trace_irq_get(struct intel_engine_cs
*ring
, u32 seqno
)
447 if (ring
->trace_irq_seqno
== 0 && ring
->irq_get(ring
))
448 ring
->trace_irq_seqno
= seqno
;
451 #endif /* _INTEL_RINGBUFFER_H_ */