intel_ringbuffer.h 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274
  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. enum intel_ring_hangcheck_action {
  29. HANGCHECK_WAIT,
  30. HANGCHECK_ACTIVE,
  31. HANGCHECK_KICK,
  32. HANGCHECK_HUNG,
  33. };
  34. struct intel_ring_hangcheck {
  35. bool deadlock;
  36. u32 seqno;
  37. u32 acthd;
  38. int score;
  39. enum intel_ring_hangcheck_action action;
  40. };
  41. struct intel_ring_buffer {
  42. const char *name;
  43. enum intel_ring_id {
  44. RCS = 0x0,
  45. VCS,
  46. BCS,
  47. VECS,
  48. } id;
  49. #define I915_NUM_RINGS 4
  50. u32 mmio_base;
  51. void __iomem *virtual_start;
  52. struct drm_device *dev;
  53. struct drm_i915_gem_object *obj;
  54. u32 head;
  55. u32 tail;
  56. int space;
  57. int size;
  58. int effective_size;
  59. 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.
  64. *
  65. * last_retired_head is set to -1 after the value is consumed so
  66. * we can detect new retirements.
  67. */
  68. u32 last_retired_head;
  69. unsigned irq_refcount; /* protected by dev_priv->irq_lock */
  70. u32 irq_enable_mask; /* bitmask to enable ring interrupt */
  71. u32 trace_irq_seqno;
  72. u32 sync_seqno[I915_NUM_RINGS-1];
  73. bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
  74. void (*irq_put)(struct intel_ring_buffer *ring);
  75. int (*init)(struct intel_ring_buffer *ring);
  76. void (*write_tail)(struct intel_ring_buffer *ring,
  77. u32 value);
  78. int __must_check (*flush)(struct intel_ring_buffer *ring,
  79. u32 invalidate_domains,
  80. u32 flush_domains);
  81. int (*add_request)(struct intel_ring_buffer *ring);
  82. /* Some chipsets are not quite as coherent as advertised and need
  83. * an expensive kick to force a true read of the up-to-date seqno.
  84. * However, the up-to-date seqno is not always required and the last
  85. * seen value is good enough. Note that the seqno will always be
  86. * monotonic, even if not coherent.
  87. */
  88. u32 (*get_seqno)(struct intel_ring_buffer *ring,
  89. bool lazy_coherency);
  90. void (*set_seqno)(struct intel_ring_buffer *ring,
  91. u32 seqno);
  92. int (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
  93. u32 offset, u32 length,
  94. unsigned flags);
  95. #define I915_DISPATCH_SECURE 0x1
  96. #define I915_DISPATCH_PINNED 0x2
  97. void (*cleanup)(struct intel_ring_buffer *ring);
  98. int (*sync_to)(struct intel_ring_buffer *ring,
  99. struct intel_ring_buffer *to,
  100. u32 seqno);
  101. /* our mbox written by others */
  102. u32 semaphore_register[I915_NUM_RINGS];
  103. /* mboxes this ring signals to */
  104. u32 signal_mbox[I915_NUM_RINGS];
  105. /**
  106. * List of objects currently involved in rendering from the
  107. * ringbuffer.
  108. *
  109. * Includes buffers having the contents of their GPU caches
  110. * flushed, not necessarily primitives. last_rendering_seqno
  111. * represents when the rendering involved will be completed.
  112. *
  113. * A reference is held on the buffer while on this list.
  114. */
  115. struct list_head active_list;
  116. /**
  117. * List of breadcrumbs associated with GPU requests currently
  118. * outstanding.
  119. */
  120. struct list_head request_list;
  121. /**
  122. * Do we have some not yet emitted requests outstanding?
  123. */
  124. u32 outstanding_lazy_request;
  125. bool gpu_caches_dirty;
  126. bool fbc_dirty;
  127. wait_queue_head_t irq_queue;
  128. /**
  129. * Do an explicit TLB flush before MI_SET_CONTEXT
  130. */
  131. bool itlb_before_ctx_switch;
  132. struct i915_hw_context *default_context;
  133. struct i915_hw_context *last_context;
  134. struct intel_ring_hangcheck hangcheck;
  135. struct {
  136. struct drm_i915_gem_object *obj;
  137. u32 gtt_offset;
  138. volatile u32 *cpu_page;
  139. } scratch;
  140. };
  141. static inline bool
  142. intel_ring_initialized(struct intel_ring_buffer *ring)
  143. {
  144. return ring->obj != NULL;
  145. }
  146. static inline unsigned
  147. intel_ring_flag(struct intel_ring_buffer *ring)
  148. {
  149. return 1 << ring->id;
  150. }
  151. static inline u32
  152. intel_ring_sync_index(struct intel_ring_buffer *ring,
  153. struct intel_ring_buffer *other)
  154. {
  155. int idx;
  156. /*
  157. * cs -> 0 = vcs, 1 = bcs
  158. * vcs -> 0 = bcs, 1 = cs,
  159. * bcs -> 0 = cs, 1 = vcs.
  160. */
  161. idx = (other - ring) - 1;
  162. if (idx < 0)
  163. idx += I915_NUM_RINGS;
  164. return idx;
  165. }
  166. static inline u32
  167. intel_read_status_page(struct intel_ring_buffer *ring,
  168. int reg)
  169. {
  170. /* Ensure that the compiler doesn't optimize away the load. */
  171. barrier();
  172. return ring->status_page.page_addr[reg];
  173. }
  174. static inline void
  175. intel_write_status_page(struct intel_ring_buffer *ring,
  176. int reg, u32 value)
  177. {
  178. ring->status_page.page_addr[reg] = value;
  179. }
  180. /**
  181. * Reads a dword out of the status page, which is written to from the command
  182. * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
  183. * MI_STORE_DATA_IMM.
  184. *
  185. * The following dwords have a reserved meaning:
  186. * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
  187. * 0x04: ring 0 head pointer
  188. * 0x05: ring 1 head pointer (915-class)
  189. * 0x06: ring 2 head pointer (915-class)
  190. * 0x10-0x1b: Context status DWords (GM45)
  191. * 0x1f: Last written status offset. (GM45)
  192. *
  193. * The area from dword 0x20 to 0x3ff is available for driver usage.
  194. */
  195. #define I915_GEM_HWS_INDEX 0x20
  196. #define I915_GEM_HWS_SCRATCH_INDEX 0x30
  197. #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
  198. void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
  199. int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
  200. static inline void intel_ring_emit(struct intel_ring_buffer *ring,
  201. u32 data)
  202. {
  203. iowrite32(data, ring->virtual_start + ring->tail);
  204. ring->tail += 4;
  205. }
  206. void intel_ring_advance(struct intel_ring_buffer *ring);
  207. int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
  208. void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
  209. int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
  210. int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
  211. int intel_init_render_ring_buffer(struct drm_device *dev);
  212. int intel_init_bsd_ring_buffer(struct drm_device *dev);
  213. int intel_init_blt_ring_buffer(struct drm_device *dev);
  214. int intel_init_vebox_ring_buffer(struct drm_device *dev);
  215. u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
  216. void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
  217. static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
  218. {
  219. return ring->tail;
  220. }
  221. static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
  222. {
  223. BUG_ON(ring->outstanding_lazy_request == 0);
  224. return ring->outstanding_lazy_request;
  225. }
  226. static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
  227. {
  228. if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
  229. ring->trace_irq_seqno = seqno;
  230. }
  231. /* DRI warts */
  232. int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
  233. #endif /* _INTEL_RINGBUFFER_H_ */