intel_ringbuffer.h 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262
  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. VECS,
  41. } id;
  42. #define I915_NUM_RINGS 4
  43. u32 mmio_base;
  44. void __iomem *virtual_start;
  45. struct drm_device *dev;
  46. struct drm_i915_gem_object *obj;
  47. u32 head;
  48. u32 tail;
  49. int space;
  50. int size;
  51. int effective_size;
  52. struct intel_hw_status_page status_page;
  53. /** We track the position of the requests in the ring buffer, and
  54. * when each is retired we increment last_retired_head as the GPU
  55. * must have finished processing the request and so we know we
  56. * can advance the ringbuffer up to that position.
  57. *
  58. * last_retired_head is set to -1 after the value is consumed so
  59. * we can detect new retirements.
  60. */
  61. u32 last_retired_head;
  62. u32 irq_refcount; /* protected by dev_priv->irq_lock */
  63. u32 irq_enable_mask; /* bitmask to enable ring interrupt */
  64. u32 trace_irq_seqno;
  65. u32 sync_seqno[I915_NUM_RINGS-1];
  66. bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
  67. void (*irq_put)(struct intel_ring_buffer *ring);
  68. int (*init)(struct intel_ring_buffer *ring);
  69. void (*write_tail)(struct intel_ring_buffer *ring,
  70. u32 value);
  71. int __must_check (*flush)(struct intel_ring_buffer *ring,
  72. u32 invalidate_domains,
  73. u32 flush_domains);
  74. int (*add_request)(struct intel_ring_buffer *ring);
  75. /* Some chipsets are not quite as coherent as advertised and need
  76. * an expensive kick to force a true read of the up-to-date seqno.
  77. * However, the up-to-date seqno is not always required and the last
  78. * seen value is good enough. Note that the seqno will always be
  79. * monotonic, even if not coherent.
  80. */
  81. u32 (*get_seqno)(struct intel_ring_buffer *ring,
  82. bool lazy_coherency);
  83. void (*set_seqno)(struct intel_ring_buffer *ring,
  84. u32 seqno);
  85. int (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
  86. u32 offset, u32 length,
  87. unsigned flags);
  88. #define I915_DISPATCH_SECURE 0x1
  89. #define I915_DISPATCH_PINNED 0x2
  90. void (*cleanup)(struct intel_ring_buffer *ring);
  91. int (*sync_to)(struct intel_ring_buffer *ring,
  92. struct intel_ring_buffer *to,
  93. u32 seqno);
  94. /* our mbox written by others */
  95. u32 semaphore_register[I915_NUM_RINGS];
  96. /* mboxes this ring signals to */
  97. u32 signal_mbox[I915_NUM_RINGS];
  98. /**
  99. * List of objects currently involved in rendering from the
  100. * ringbuffer.
  101. *
  102. * Includes buffers having the contents of their GPU caches
  103. * flushed, not necessarily primitives. last_rendering_seqno
  104. * represents when the rendering involved will be completed.
  105. *
  106. * A reference is held on the buffer while on this list.
  107. */
  108. struct list_head active_list;
  109. /**
  110. * List of breadcrumbs associated with GPU requests currently
  111. * outstanding.
  112. */
  113. struct list_head request_list;
  114. /**
  115. * Do we have some not yet emitted requests outstanding?
  116. */
  117. u32 outstanding_lazy_request;
  118. bool gpu_caches_dirty;
  119. wait_queue_head_t irq_queue;
  120. /**
  121. * Do an explicit TLB flush before MI_SET_CONTEXT
  122. */
  123. bool itlb_before_ctx_switch;
  124. struct i915_hw_context *default_context;
  125. struct i915_hw_context *last_context;
  126. struct intel_ring_hangcheck hangcheck;
  127. void *private;
  128. };
  129. static inline bool
  130. intel_ring_initialized(struct intel_ring_buffer *ring)
  131. {
  132. return ring->obj != NULL;
  133. }
  134. static inline unsigned
  135. intel_ring_flag(struct intel_ring_buffer *ring)
  136. {
  137. return 1 << ring->id;
  138. }
  139. static inline u32
  140. intel_ring_sync_index(struct intel_ring_buffer *ring,
  141. struct intel_ring_buffer *other)
  142. {
  143. int idx;
  144. /*
  145. * cs -> 0 = vcs, 1 = bcs
  146. * vcs -> 0 = bcs, 1 = cs,
  147. * bcs -> 0 = cs, 1 = vcs.
  148. */
  149. idx = (other - ring) - 1;
  150. if (idx < 0)
  151. idx += I915_NUM_RINGS;
  152. return idx;
  153. }
  154. static inline u32
  155. intel_read_status_page(struct intel_ring_buffer *ring,
  156. int reg)
  157. {
  158. /* Ensure that the compiler doesn't optimize away the load. */
  159. barrier();
  160. return ring->status_page.page_addr[reg];
  161. }
  162. static inline void
  163. intel_write_status_page(struct intel_ring_buffer *ring,
  164. int reg, u32 value)
  165. {
  166. ring->status_page.page_addr[reg] = value;
  167. }
  168. /**
  169. * Reads a dword out of the status page, which is written to from the command
  170. * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
  171. * MI_STORE_DATA_IMM.
  172. *
  173. * The following dwords have a reserved meaning:
  174. * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
  175. * 0x04: ring 0 head pointer
  176. * 0x05: ring 1 head pointer (915-class)
  177. * 0x06: ring 2 head pointer (915-class)
  178. * 0x10-0x1b: Context status DWords (GM45)
  179. * 0x1f: Last written status offset. (GM45)
  180. *
  181. * The area from dword 0x20 to 0x3ff is available for driver usage.
  182. */
  183. #define I915_GEM_HWS_INDEX 0x20
  184. #define I915_GEM_HWS_SCRATCH_INDEX 0x30
  185. #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
  186. void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
  187. int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
  188. static inline void intel_ring_emit(struct intel_ring_buffer *ring,
  189. u32 data)
  190. {
  191. iowrite32(data, ring->virtual_start + ring->tail);
  192. ring->tail += 4;
  193. }
  194. void intel_ring_advance(struct intel_ring_buffer *ring);
  195. int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
  196. void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
  197. int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
  198. int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
  199. int intel_init_render_ring_buffer(struct drm_device *dev);
  200. int intel_init_bsd_ring_buffer(struct drm_device *dev);
  201. int intel_init_blt_ring_buffer(struct drm_device *dev);
  202. int intel_init_vebox_ring_buffer(struct drm_device *dev);
  203. u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
  204. void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
  205. static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
  206. {
  207. return ring->tail;
  208. }
  209. static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
  210. {
  211. BUG_ON(ring->outstanding_lazy_request == 0);
  212. return ring->outstanding_lazy_request;
  213. }
  214. static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
  215. {
  216. if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
  217. ring->trace_irq_seqno = seqno;
  218. }
  219. /* DRI warts */
  220. int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
  221. #endif /* _INTEL_RINGBUFFER_H_ */