1 #ifndef _INTEL_RINGBUFFER_H_
2 #define _INTEL_RINGBUFFER_H_
5 * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
6 * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
7 * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
9 * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
10 * cacheline, the Head Pointer must not be greater than the Tail
13 #define I915_RING_FREE_SPACE 64
15 struct intel_hw_status_page {
17 unsigned int gfx_addr;
18 struct drm_i915_gem_object *obj;
21 #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base))
22 #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val)
24 #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base))
25 #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val)
27 #define I915_READ_HEAD(ring) I915_READ(RING_HEAD((ring)->mmio_base))
28 #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val)
30 #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base))
31 #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val)
33 #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base))
34 #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val)
36 #define I915_READ_NOPID(ring) I915_READ(RING_NOPID((ring)->mmio_base))
37 #define I915_READ_SYNC_0(ring) I915_READ(RING_SYNC_0((ring)->mmio_base))
38 #define I915_READ_SYNC_1(ring) I915_READ(RING_SYNC_1((ring)->mmio_base))
40 struct intel_ring_buffer {
47 #define I915_NUM_RINGS 3
49 void __iomem *virtual_start;
50 struct drm_device *dev;
51 struct drm_i915_gem_object *obj;
58 struct intel_hw_status_page status_page;
60 /** We track the position of the requests in the ring buffer, and
61 * when each is retired we increment last_retired_head as the GPU
62 * must have finished processing the request and so we know we
63 * can advance the ringbuffer up to that position.
65 * last_retired_head is set to -1 after the value is consumed so
66 * we can detect new retirements.
68 u32 last_retired_head;
70 u32 irq_refcount; /* protected by dev_priv->irq_lock */
71 u32 irq_enable_mask; /* bitmask to enable ring interrupt */
73 u32 sync_seqno[I915_NUM_RINGS-1];
74 bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
75 void (*irq_put)(struct intel_ring_buffer *ring);
77 int (*init)(struct intel_ring_buffer *ring);
79 void (*write_tail)(struct intel_ring_buffer *ring,
81 int __must_check (*flush)(struct intel_ring_buffer *ring,
82 u32 invalidate_domains,
84 int (*add_request)(struct intel_ring_buffer *ring);
85 /* Some chipsets are not quite as coherent as advertised and need
86 * an expensive kick to force a true read of the up-to-date seqno.
87 * However, the up-to-date seqno is not always required and the last
88 * seen value is good enough. Note that the seqno will always be
89 * monotonic, even if not coherent.
91 u32 (*get_seqno)(struct intel_ring_buffer *ring,
93 void (*set_seqno)(struct intel_ring_buffer *ring,
95 int (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
96 u32 offset, u32 length,
98 #define I915_DISPATCH_SECURE 0x1
99 #define I915_DISPATCH_PINNED 0x2
100 void (*cleanup)(struct intel_ring_buffer *ring);
101 int (*sync_to)(struct intel_ring_buffer *ring,
102 struct intel_ring_buffer *to,
105 u32 semaphore_register[3]; /*our mbox written by others */
106 u32 signal_mbox[2]; /* mboxes this ring signals to */
108 * List of objects currently involved in rendering from the
111 * Includes buffers having the contents of their GPU caches
112 * flushed, not necessarily primitives. last_rendering_seqno
113 * represents when the rendering involved will be completed.
115 * A reference is held on the buffer while on this list.
117 struct list_head active_list;
120 * List of breadcrumbs associated with GPU requests currently
123 struct list_head request_list;
126 * Do we have some not yet emitted requests outstanding?
128 u32 outstanding_lazy_request;
129 bool gpu_caches_dirty;
131 wait_queue_head_t irq_queue;
134 * Do an explicit TLB flush before MI_SET_CONTEXT
136 bool itlb_before_ctx_switch;
137 struct i915_hw_context *default_context;
138 struct drm_i915_gem_object *last_context_obj;
144 intel_ring_initialized(struct intel_ring_buffer *ring)
146 return ring->obj != NULL;
149 static inline unsigned
150 intel_ring_flag(struct intel_ring_buffer *ring)
152 return 1 << ring->id;
156 intel_ring_sync_index(struct intel_ring_buffer *ring,
157 struct intel_ring_buffer *other)
162 * cs -> 0 = vcs, 1 = bcs
163 * vcs -> 0 = bcs, 1 = cs,
164 * bcs -> 0 = cs, 1 = vcs.
167 idx = (other - ring) - 1;
169 idx += I915_NUM_RINGS;
175 intel_read_status_page(struct intel_ring_buffer *ring,
178 /* Ensure that the compiler doesn't optimize away the load. */
180 return ring->status_page.page_addr[reg];
184 intel_write_status_page(struct intel_ring_buffer *ring,
187 ring->status_page.page_addr[reg] = value;
191 * Reads a dword out of the status page, which is written to from the command
192 * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
195 * The following dwords have a reserved meaning:
196 * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
197 * 0x04: ring 0 head pointer
198 * 0x05: ring 1 head pointer (915-class)
199 * 0x06: ring 2 head pointer (915-class)
200 * 0x10-0x1b: Context status DWords (GM45)
201 * 0x1f: Last written status offset. (GM45)
203 * The area from dword 0x20 to 0x3ff is available for driver usage.
205 #define I915_GEM_HWS_INDEX 0x20
206 #define I915_GEM_HWS_SCRATCH_INDEX 0x30
207 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
209 void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
211 int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
212 static inline void intel_ring_emit(struct intel_ring_buffer *ring,
215 iowrite32(data, ring->virtual_start + ring->tail);
218 void intel_ring_advance(struct intel_ring_buffer *ring);
219 int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
220 void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
221 int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
222 int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
224 int intel_init_render_ring_buffer(struct drm_device *dev);
225 int intel_init_bsd_ring_buffer(struct drm_device *dev);
226 int intel_init_blt_ring_buffer(struct drm_device *dev);
228 u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
229 void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
231 static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
236 static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
238 BUG_ON(ring->outstanding_lazy_request == 0);
239 return ring->outstanding_lazy_request;
242 static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
244 if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
245 ring->trace_irq_seqno = seqno;
249 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
251 #endif /* _INTEL_RINGBUFFER_H_ */