drivers/gpu/drm/i915/intel_ringbuffer.h

   1 #ifndef _INTEL_RINGBUFFER_H_
   2 #define _INTEL_RINGBUFFER_H_
   3
   4 #include <linux/hashtable.h>
   5
   6 #define I915_CMD_HASH_ORDER 9
   7
   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
  11  * workarounds!
  12  */
  13 #define CACHELINE_BYTES 64
  14
  15 /*
  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"
  19  *
  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
  22  * Pointer."
  23  */
  24 #define I915_RING_FREE_SPACE 64
  25
  26 struct  intel_hw_status_page {
  27         u32             *page_addr;
  28         unsigned int    gfx_addr;
  29         struct          drm_i915_gem_object *obj;
  30 };
  31
  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)
  34
  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)
  37
  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)
  40
  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)
  43
  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)
  46
  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)
  49
  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.
  52  */
  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)))
  58
  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))
  63
  64 #define GEN8_RING_SEMAPHORE_INIT do { \
  65         if (!dev_priv->semaphore_obj) { \
  66                 break; \
  67         } \
  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; \
  74         } while(0)
  75
  76 enum intel_ring_hangcheck_action {
  77         HANGCHECK_IDLE = 0,
  78         HANGCHECK_WAIT,
  79         HANGCHECK_ACTIVE,
  80         HANGCHECK_ACTIVE_LOOP,
  81         HANGCHECK_KICK,
  82         HANGCHECK_HUNG,
  83 };
  84
  85 #define HANGCHECK_SCORE_RING_HUNG 31
  86
  87 struct intel_ring_hangcheck {
  88         u64 acthd;
  89         u64 max_acthd;
  90         u32 seqno;
  91         int score;
  92         enum intel_ring_hangcheck_action action;
  93         int deadlock;
  94 };
  95
  96 struct intel_ringbuffer {
  97         struct drm_i915_gem_object *obj;
  98         void __iomem *virtual_start;
  99
 100         struct intel_engine_cs *ring;
 101
 102         /*
 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
 105          * code has landed.
 106          */
 107         struct intel_context *FIXME_lrc_ctx;
 108
 109         u32 head;
 110         u32 tail;
 111         int space;
 112         int size;
 113         int effective_size;
 114
 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.
 119          *
 120          * last_retired_head is set to -1 after the value is consumed so
 121          * we can detect new retirements.
 122          */
 123         u32 last_retired_head;
 124 };
 125
 126 struct  intel_engine_cs {
 127         const char      *name;
 128         enum intel_ring_id {
 129                 RCS = 0x0,
 130                 VCS,
 131                 BCS,
 132                 VECS,
 133                 VCS2
 134         } id;
 135 #define I915_NUM_RINGS 5
 136 #define LAST_USER_RING (VECS + 1)
 137         u32             mmio_base;
 138         struct          drm_device *dev;
 139         struct intel_ringbuffer *buffer;
 140
 141         struct intel_hw_status_page status_page;
 142
 143         unsigned irq_refcount; /* protected by dev_priv->irq_lock */
 144         u32             irq_enable_mask;        /* bitmask to enable ring interrupt */
 145         u32             trace_irq_seqno;
 146         bool __must_check (*irq_get)(struct intel_engine_cs *ring);
 147         void            (*irq_put)(struct intel_engine_cs *ring);
 148
 149         int             (*init)(struct intel_engine_cs *ring);
 150
 151         void            (*write_tail)(struct intel_engine_cs *ring,
 152                                       u32 value);
 153         int __must_check (*flush)(struct intel_engine_cs *ring,
 154                                   u32   invalidate_domains,
 155                                   u32   flush_domains);
 156         int             (*add_request)(struct intel_engine_cs *ring);
 157         /* Some chipsets are not quite as coherent as advertised and need
 158          * an expensive kick to force a true read of the up-to-date seqno.
 159          * However, the up-to-date seqno is not always required and the last
 160          * seen value is good enough. Note that the seqno will always be
 161          * monotonic, even if not coherent.
 162          */
 163         u32             (*get_seqno)(struct intel_engine_cs *ring,
 164                                      bool lazy_coherency);
 165         void            (*set_seqno)(struct intel_engine_cs *ring,
 166                                      u32 seqno);
 167         int             (*dispatch_execbuffer)(struct intel_engine_cs *ring,
 168                                                u64 offset, u32 length,
 169                                                unsigned flags);
 170 #define I915_DISPATCH_SECURE 0x1
 171 #define I915_DISPATCH_PINNED 0x2
 172         void            (*cleanup)(struct intel_engine_cs *ring);
 173
 174         /* GEN8 signal/wait table - never trust comments!
 175          *        signal to     signal to    signal to   signal to      signal to
 176          *          RCS            VCS          BCS        VECS          VCS2
 177          *      --------------------------------------------------------------------
 178          *  RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
 179          *      |-------------------------------------------------------------------
 180          *  VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
 181          *      |-------------------------------------------------------------------
 182          *  BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
 183          *      |-------------------------------------------------------------------
 184          * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) |  NOP (0x90) | VCS2 (0x98) |
 185          *      |-------------------------------------------------------------------
 186          * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP  (0xc0) |
 187          *      |-------------------------------------------------------------------
 188          *
 189          * Generalization:
 190          *  f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
 191          *  ie. transpose of g(x, y)
 192          *
 193          *       sync from      sync from    sync from    sync from     sync from
 194          *          RCS            VCS          BCS        VECS          VCS2
 195          *      --------------------------------------------------------------------
 196          *  RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
 197          *      |-------------------------------------------------------------------
 198          *  VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
 199          *      |-------------------------------------------------------------------
 200          *  BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
 201          *      |-------------------------------------------------------------------
 202          * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) |  NOP (0x90) | VCS2 (0xb8) |
 203          *      |-------------------------------------------------------------------
 204          * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) |  NOP (0xc0) |
 205          *      |-------------------------------------------------------------------
 206          *
 207          * Generalization:
 208          *  g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
 209          *  ie. transpose of f(x, y)
 210          */
 211         struct {
 212                 u32     sync_seqno[I915_NUM_RINGS-1];
 213
 214                 union {
 215                         struct {
 216                                 /* our mbox written by others */
 217                                 u32             wait[I915_NUM_RINGS];
 218                                 /* mboxes this ring signals to */
 219                                 u32             signal[I915_NUM_RINGS];
 220                         } mbox;
 221                         u64             signal_ggtt[I915_NUM_RINGS];
 222                 };
 223
 224                 /* AKA wait() */
 225                 int     (*sync_to)(struct intel_engine_cs *ring,
 226                                    struct intel_engine_cs *to,
 227                                    u32 seqno);
 228                 int     (*signal)(struct intel_engine_cs *signaller,
 229                                   /* num_dwords needed by caller */
 230                                   unsigned int num_dwords);
 231         } semaphore;
 232
 233         /* Execlists */
 234         u32             irq_keep_mask; /* bitmask for interrupts that should not be masked */
 235         int             (*emit_request)(struct intel_ringbuffer *ringbuf);
 236         int             (*emit_flush)(struct intel_ringbuffer *ringbuf,
 237                                       u32 invalidate_domains,
 238                                       u32 flush_domains);
 239         int             (*emit_bb_start)(struct intel_ringbuffer *ringbuf,
 240                                          u64 offset, unsigned flags);
 241
 242         /**
 243          * List of objects currently involved in rendering from the
 244          * ringbuffer.
 245          *
 246          * Includes buffers having the contents of their GPU caches
 247          * flushed, not necessarily primitives.  last_rendering_seqno
 248          * represents when the rendering involved will be completed.
 249          *
 250          * A reference is held on the buffer while on this list.
 251          */
 252         struct list_head active_list;
 253
 254         /**
 255          * List of breadcrumbs associated with GPU requests currently
 256          * outstanding.
 257          */
 258         struct list_head request_list;
 259
 260         /**
 261          * Do we have some not yet emitted requests outstanding?
 262          */
 263         struct drm_i915_gem_request *preallocated_lazy_request;
 264         u32 outstanding_lazy_seqno;
 265         bool gpu_caches_dirty;
 266         bool fbc_dirty;
 267
 268         wait_queue_head_t irq_queue;
 269
 270         struct intel_context *default_context;
 271         struct intel_context *last_context;
 272
 273         struct intel_ring_hangcheck hangcheck;
 274
 275         struct {
 276                 struct drm_i915_gem_object *obj;
 277                 u32 gtt_offset;
 278                 volatile u32 *cpu_page;
 279         } scratch;
 280
 281         bool needs_cmd_parser;
 282
 283         /*
 284          * Table of commands the command parser needs to know about
 285          * for this ring.
 286          */
 287         DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
 288
 289         /*
 290          * Table of registers allowed in commands that read/write registers.
 291          */
 292         const u32 *reg_table;
 293         int reg_count;
 294
 295         /*
 296          * Table of registers allowed in commands that read/write registers, but
 297          * only from the DRM master.
 298          */
 299         const u32 *master_reg_table;
 300         int master_reg_count;
 301
 302         /*
 303          * Returns the bitmask for the length field of the specified command.
 304          * Return 0 for an unrecognized/invalid command.
 305          *
 306          * If the command parser finds an entry for a command in the ring's
 307          * cmd_tables, it gets the command's length based on the table entry.
 308          * If not, it calls this function to determine the per-ring length field
 309          * encoding for the command (i.e. certain opcode ranges use certain bits
 310          * to encode the command length in the header).
 311          */
 312         u32 (*get_cmd_length_mask)(u32 cmd_header);
 313 };
 314
 315 bool intel_ring_initialized(struct intel_engine_cs *ring);
 316
 317 static inline unsigned
 318 intel_ring_flag(struct intel_engine_cs *ring)
 319 {
 320         return 1 << ring->id;
 321 }
 322
 323 static inline u32
 324 intel_ring_sync_index(struct intel_engine_cs *ring,
 325                       struct intel_engine_cs *other)
 326 {
 327         int idx;
 328
 329         /*
 330          * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
 331          * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
 332          * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
 333          * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
 334          * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
 335          */
 336
 337         idx = (other - ring) - 1;
 338         if (idx < 0)
 339                 idx += I915_NUM_RINGS;
 340
 341         return idx;
 342 }
 343
 344 static inline u32
 345 intel_read_status_page(struct intel_engine_cs *ring,
 346                        int reg)
 347 {
 348         /* Ensure that the compiler doesn't optimize away the load. */
 349         barrier();
 350         return ring->status_page.page_addr[reg];
 351 }
 352
 353 static inline void
 354 intel_write_status_page(struct intel_engine_cs *ring,
 355                         int reg, u32 value)
 356 {
 357         ring->status_page.page_addr[reg] = value;
 358 }
 359
 360 /**
 361  * Reads a dword out of the status page, which is written to from the command
 362  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
 363  * MI_STORE_DATA_IMM.
 364  *
 365  * The following dwords have a reserved meaning:
 366  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
 367  * 0x04: ring 0 head pointer
 368  * 0x05: ring 1 head pointer (915-class)
 369  * 0x06: ring 2 head pointer (915-class)
 370  * 0x10-0x1b: Context status DWords (GM45)
 371  * 0x1f: Last written status offset. (GM45)
 372  *
 373  * The area from dword 0x20 to 0x3ff is available for driver usage.
 374  */
 375 #define I915_GEM_HWS_INDEX              0x20
 376 #define I915_GEM_HWS_SCRATCH_INDEX      0x30
 377 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
 378
 379 void intel_destroy_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
 380 int intel_alloc_ringbuffer_obj(struct drm_device *dev,
 381                                struct intel_ringbuffer *ringbuf);
 382
 383 void intel_stop_ring_buffer(struct intel_engine_cs *ring);
 384 void intel_cleanup_ring_buffer(struct intel_engine_cs *ring);
 385
 386 int __must_check intel_ring_begin(struct intel_engine_cs *ring, int n);
 387 int __must_check intel_ring_cacheline_align(struct intel_engine_cs *ring);
 388 static inline void intel_ring_emit(struct intel_engine_cs *ring,
 389                                    u32 data)
 390 {
 391         struct intel_ringbuffer *ringbuf = ring->buffer;
 392         iowrite32(data, ringbuf->virtual_start + ringbuf->tail);
 393         ringbuf->tail += 4;
 394 }
 395 static inline void intel_ring_advance(struct intel_engine_cs *ring)
 396 {
 397         struct intel_ringbuffer *ringbuf = ring->buffer;
 398         ringbuf->tail &= ringbuf->size - 1;
 399 }
 400 int __intel_ring_space(int head, int tail, int size);
 401 int intel_ring_space(struct intel_ringbuffer *ringbuf);
 402 bool intel_ring_stopped(struct intel_engine_cs *ring);
 403 void __intel_ring_advance(struct intel_engine_cs *ring);
 404
 405 int __must_check intel_ring_idle(struct intel_engine_cs *ring);
 406 void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno);
 407 int intel_ring_flush_all_caches(struct intel_engine_cs *ring);
 408 int intel_ring_invalidate_all_caches(struct intel_engine_cs *ring);
 409
 410 void intel_fini_pipe_control(struct intel_engine_cs *ring);
 411 int intel_init_pipe_control(struct intel_engine_cs *ring);
 412
 413 int intel_init_render_ring_buffer(struct drm_device *dev);
 414 int intel_init_bsd_ring_buffer(struct drm_device *dev);
 415 int intel_init_bsd2_ring_buffer(struct drm_device *dev);
 416 int intel_init_blt_ring_buffer(struct drm_device *dev);
 417 int intel_init_vebox_ring_buffer(struct drm_device *dev);
 418
 419 u64 intel_ring_get_active_head(struct intel_engine_cs *ring);
 420 void intel_ring_setup_status_page(struct intel_engine_cs *ring);
 421
 422 static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf)
 423 {
 424         return ringbuf->tail;
 425 }
 426
 427 static inline u32 intel_ring_get_seqno(struct intel_engine_cs *ring)
 428 {
 429         BUG_ON(ring->outstanding_lazy_seqno == 0);
 430         return ring->outstanding_lazy_seqno;
 431 }
 432
 433 static inline void i915_trace_irq_get(struct intel_engine_cs *ring, u32 seqno)
 434 {
 435         if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
 436                 ring->trace_irq_seqno = seqno;
 437 }
 438
 439 /* DRI warts */
 440 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
 441
 442 #endif /* _INTEL_RINGBUFFER_H_ */