intel_ringbuffer.h 7.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257
  1. #ifndef _INTEL_RINGBUFFER_H_
  2. #define _INTEL_RINGBUFFER_H_
  3. /*
  4. * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
  5. * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
  6. * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
  7. *
  8. * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
  9. * cacheline, the Head Pointer must not be greater than the Tail
  10. * Pointer."
  11. */
  12. #define I915_RING_FREE_SPACE 64
  13. struct intel_hw_status_page {
  14. u32 *page_addr;
  15. unsigned int gfx_addr;
  16. struct drm_i915_gem_object *obj;
  17. };
  18. #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base))
  19. #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val)
  20. #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base))
  21. #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val)
  22. #define I915_READ_HEAD(ring) I915_READ(RING_HEAD((ring)->mmio_base))
  23. #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val)
  24. #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base))
  25. #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val)
  26. #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base))
  27. #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val)
  28. #define I915_READ_NOPID(ring) I915_READ(RING_NOPID((ring)->mmio_base))
  29. #define I915_READ_SYNC_0(ring) I915_READ(RING_SYNC_0((ring)->mmio_base))
  30. #define I915_READ_SYNC_1(ring) I915_READ(RING_SYNC_1((ring)->mmio_base))
  31. struct intel_ring_hangcheck {
  32. u32 seqno;
  33. };
  34. struct intel_ring_buffer {
  35. const char *name;
  36. enum intel_ring_id {
  37. RCS = 0x0,
  38. VCS,
  39. BCS,
  40. } id;
  41. #define I915_NUM_RINGS 3
  42. u32 mmio_base;
  43. void __iomem *virtual_start;
  44. struct drm_device *dev;
  45. struct drm_i915_gem_object *obj;
  46. u32 head;
  47. u32 tail;
  48. int space;
  49. int size;
  50. int effective_size;
  51. struct intel_hw_status_page status_page;
  52. /** We track the position of the requests in the ring buffer, and
  53. * when each is retired we increment last_retired_head as the GPU
  54. * must have finished processing the request and so we know we
  55. * can advance the ringbuffer up to that position.
  56. *
  57. * last_retired_head is set to -1 after the value is consumed so
  58. * we can detect new retirements.
  59. */
  60. u32 last_retired_head;
  61. u32 irq_refcount; /* protected by dev_priv->irq_lock */
  62. u32 irq_enable_mask; /* bitmask to enable ring interrupt */
  63. u32 trace_irq_seqno;
  64. u32 sync_seqno[I915_NUM_RINGS-1];
  65. bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
  66. void (*irq_put)(struct intel_ring_buffer *ring);
  67. int (*init)(struct intel_ring_buffer *ring);
  68. void (*write_tail)(struct intel_ring_buffer *ring,
  69. u32 value);
  70. int __must_check (*flush)(struct intel_ring_buffer *ring,
  71. u32 invalidate_domains,
  72. u32 flush_domains);
  73. int (*add_request)(struct intel_ring_buffer *ring);
  74. /* Some chipsets are not quite as coherent as advertised and need
  75. * an expensive kick to force a true read of the up-to-date seqno.
  76. * However, the up-to-date seqno is not always required and the last
  77. * seen value is good enough. Note that the seqno will always be
  78. * monotonic, even if not coherent.
  79. */
  80. u32 (*get_seqno)(struct intel_ring_buffer *ring,
  81. bool lazy_coherency);
  82. void (*set_seqno)(struct intel_ring_buffer *ring,
  83. u32 seqno);
  84. int (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
  85. u32 offset, u32 length,
  86. unsigned flags);
  87. #define I915_DISPATCH_SECURE 0x1
  88. #define I915_DISPATCH_PINNED 0x2
  89. void (*cleanup)(struct intel_ring_buffer *ring);
  90. int (*sync_to)(struct intel_ring_buffer *ring,
  91. struct intel_ring_buffer *to,
  92. u32 seqno);
  93. u32 semaphore_register[3]; /*our mbox written by others */
  94. u32 signal_mbox[2]; /* mboxes this ring signals to */
  95. /**
  96. * List of objects currently involved in rendering from the
  97. * ringbuffer.
  98. *
  99. * Includes buffers having the contents of their GPU caches
  100. * flushed, not necessarily primitives. last_rendering_seqno
  101. * represents when the rendering involved will be completed.
  102. *
  103. * A reference is held on the buffer while on this list.
  104. */
  105. struct list_head active_list;
  106. /**
  107. * List of breadcrumbs associated with GPU requests currently
  108. * outstanding.
  109. */
  110. struct list_head request_list;
  111. /**
  112. * Do we have some not yet emitted requests outstanding?
  113. */
  114. u32 outstanding_lazy_request;
  115. bool gpu_caches_dirty;
  116. wait_queue_head_t irq_queue;
  117. /**
  118. * Do an explicit TLB flush before MI_SET_CONTEXT
  119. */
  120. bool itlb_before_ctx_switch;
  121. struct i915_hw_context *default_context;
  122. struct i915_hw_context *last_context;
  123. struct intel_ring_hangcheck hangcheck;
  124. void *private;
  125. };
  126. static inline bool
  127. intel_ring_initialized(struct intel_ring_buffer *ring)
  128. {
  129. return ring->obj != NULL;
  130. }
  131. static inline unsigned
  132. intel_ring_flag(struct intel_ring_buffer *ring)
  133. {
  134. return 1 << ring->id;
  135. }
  136. static inline u32
  137. intel_ring_sync_index(struct intel_ring_buffer *ring,
  138. struct intel_ring_buffer *other)
  139. {
  140. int idx;
  141. /*
  142. * cs -> 0 = vcs, 1 = bcs
  143. * vcs -> 0 = bcs, 1 = cs,
  144. * bcs -> 0 = cs, 1 = vcs.
  145. */
  146. idx = (other - ring) - 1;
  147. if (idx < 0)
  148. idx += I915_NUM_RINGS;
  149. return idx;
  150. }
  151. static inline u32
  152. intel_read_status_page(struct intel_ring_buffer *ring,
  153. int reg)
  154. {
  155. /* Ensure that the compiler doesn't optimize away the load. */
  156. barrier();
  157. return ring->status_page.page_addr[reg];
  158. }
  159. static inline void
  160. intel_write_status_page(struct intel_ring_buffer *ring,
  161. int reg, u32 value)
  162. {
  163. ring->status_page.page_addr[reg] = value;
  164. }
  165. /**
  166. * Reads a dword out of the status page, which is written to from the command
  167. * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
  168. * MI_STORE_DATA_IMM.
  169. *
  170. * The following dwords have a reserved meaning:
  171. * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
  172. * 0x04: ring 0 head pointer
  173. * 0x05: ring 1 head pointer (915-class)
  174. * 0x06: ring 2 head pointer (915-class)
  175. * 0x10-0x1b: Context status DWords (GM45)
  176. * 0x1f: Last written status offset. (GM45)
  177. *
  178. * The area from dword 0x20 to 0x3ff is available for driver usage.
  179. */
  180. #define I915_GEM_HWS_INDEX 0x20
  181. #define I915_GEM_HWS_SCRATCH_INDEX 0x30
  182. #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
  183. void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
  184. int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
  185. static inline void intel_ring_emit(struct intel_ring_buffer *ring,
  186. u32 data)
  187. {
  188. iowrite32(data, ring->virtual_start + ring->tail);
  189. ring->tail += 4;
  190. }
  191. void intel_ring_advance(struct intel_ring_buffer *ring);
  192. int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
  193. void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
  194. int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
  195. int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
  196. int intel_init_render_ring_buffer(struct drm_device *dev);
  197. int intel_init_bsd_ring_buffer(struct drm_device *dev);
  198. int intel_init_blt_ring_buffer(struct drm_device *dev);
  199. u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
  200. void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
  201. static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
  202. {
  203. return ring->tail;
  204. }
  205. static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
  206. {
  207. BUG_ON(ring->outstanding_lazy_request == 0);
  208. return ring->outstanding_lazy_request;
  209. }
  210. static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
  211. {
  212. if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
  213. ring->trace_irq_seqno = seqno;
  214. }
  215. /* DRI warts */
  216. int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
  217. #endif /* _INTEL_RINGBUFFER_H_ */