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         u32 head;
 103         u32 tail;
 104         int space;
 105         int size;
 106         int effective_size;
 107
 108         /** We track the position of the requests in the ring buffer, and
 109          * when each is retired we increment last_retired_head as the GPU
 110          * must have finished processing the request and so we know we
 111          * can advance the ringbuffer up to that position.
 112          *
 113          * last_retired_head is set to -1 after the value is consumed so
 114          * we can detect new retirements.
 115          */
 116         u32 last_retired_head;
 117 };
 118
 119 struct  intel_context;
 120
 121 struct  intel_engine_cs {
 122         const char      *name;
 123         enum intel_ring_id {
 124                 RCS = 0x0,
 125                 VCS,
 126                 BCS,
 127                 VECS,
 128                 VCS2
 129         } id;
 130 #define I915_NUM_RINGS 5
 131 #define LAST_USER_RING (VECS + 1)
 132         u32             mmio_base;
 133         struct          drm_device *dev;
 134         struct intel_ringbuffer *buffer;
 135
 136         struct intel_hw_status_page status_page;
 137
 138         unsigned irq_refcount; /* protected by dev_priv->irq_lock */
 139         u32             irq_enable_mask;        /* bitmask to enable ring interrupt */
 140         struct drm_i915_gem_request *trace_irq_req;
 141         bool __must_check (*irq_get)(struct intel_engine_cs *ring);
 142         void            (*irq_put)(struct intel_engine_cs *ring);
 143
 144         int             (*init_hw)(struct intel_engine_cs *ring);
 145
 146         int             (*init_context)(struct intel_engine_cs *ring,
 147                                         struct intel_context *ctx);
 148
 149         void            (*write_tail)(struct intel_engine_cs *ring,
 150                                       u32 value);
 151         int __must_check (*flush)(struct intel_engine_cs *ring,
 152                                   u32   invalidate_domains,
 153                                   u32   flush_domains);
 154         int             (*add_request)(struct intel_engine_cs *ring);
 155         /* Some chipsets are not quite as coherent as advertised and need
 156          * an expensive kick to force a true read of the up-to-date seqno.
 157          * However, the up-to-date seqno is not always required and the last
 158          * seen value is good enough. Note that the seqno will always be
 159          * monotonic, even if not coherent.
 160          */
 161         u32             (*get_seqno)(struct intel_engine_cs *ring,
 162                                      bool lazy_coherency);
 163         void            (*set_seqno)(struct intel_engine_cs *ring,
 164                                      u32 seqno);
 165         int             (*dispatch_execbuffer)(struct intel_engine_cs *ring,
 166                                                u64 offset, u32 length,
 167                                                unsigned dispatch_flags);
 168 #define I915_DISPATCH_SECURE 0x1
 169 #define I915_DISPATCH_PINNED 0x2
 170         void            (*cleanup)(struct intel_engine_cs *ring);
 171
 172         /* GEN8 signal/wait table - never trust comments!
 173          *        signal to     signal to    signal to   signal to      signal to
 174          *          RCS            VCS          BCS        VECS          VCS2
 175          *      --------------------------------------------------------------------
 176          *  RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
 177          *      |-------------------------------------------------------------------
 178          *  VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
 179          *      |-------------------------------------------------------------------
 180          *  BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
 181          *      |-------------------------------------------------------------------
 182          * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) |  NOP (0x90) | VCS2 (0x98) |
 183          *      |-------------------------------------------------------------------
 184          * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP  (0xc0) |
 185          *      |-------------------------------------------------------------------
 186          *
 187          * Generalization:
 188          *  f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
 189          *  ie. transpose of g(x, y)
 190          *
 191          *       sync from      sync from    sync from    sync from     sync from
 192          *          RCS            VCS          BCS        VECS          VCS2
 193          *      --------------------------------------------------------------------
 194          *  RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
 195          *      |-------------------------------------------------------------------
 196          *  VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
 197          *      |-------------------------------------------------------------------
 198          *  BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
 199          *      |-------------------------------------------------------------------
 200          * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) |  NOP (0x90) | VCS2 (0xb8) |
 201          *      |-------------------------------------------------------------------
 202          * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) |  NOP (0xc0) |
 203          *      |-------------------------------------------------------------------
 204          *
 205          * Generalization:
 206          *  g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
 207          *  ie. transpose of f(x, y)
 208          */
 209         struct {
 210                 u32     sync_seqno[I915_NUM_RINGS-1];
 211
 212                 union {
 213                         struct {
 214                                 /* our mbox written by others */
 215                                 u32             wait[I915_NUM_RINGS];
 216                                 /* mboxes this ring signals to */
 217                                 u32             signal[I915_NUM_RINGS];
 218                         } mbox;
 219                         u64             signal_ggtt[I915_NUM_RINGS];
 220                 };
 221
 222                 /* AKA wait() */
 223                 int     (*sync_to)(struct intel_engine_cs *ring,
 224                                    struct intel_engine_cs *to,
 225                                    u32 seqno);
 226                 int     (*signal)(struct intel_engine_cs *signaller,
 227                                   /* num_dwords needed by caller */
 228                                   unsigned int num_dwords);
 229         } semaphore;
 230
 231         /* Execlists */
 232         spinlock_t execlist_lock;
 233         struct list_head execlist_queue;
 234         struct list_head execlist_retired_req_list;
 235         u8 next_context_status_buffer;
 236         u32             irq_keep_mask; /* bitmask for interrupts that should not be masked */
 237         int             (*emit_request)(struct intel_ringbuffer *ringbuf,
 238                                         struct drm_i915_gem_request *request);
 239         int             (*emit_flush)(struct intel_ringbuffer *ringbuf,
 240                                       struct intel_context *ctx,
 241                                       u32 invalidate_domains,
 242                                       u32 flush_domains);
 243         int             (*emit_bb_start)(struct intel_ringbuffer *ringbuf,
 244                                          struct intel_context *ctx,
 245                                          u64 offset, unsigned dispatch_flags);
 246
 247         /**
 248          * List of objects currently involved in rendering from the
 249          * ringbuffer.
 250          *
 251          * Includes buffers having the contents of their GPU caches
 252          * flushed, not necessarily primitives.  last_read_req
 253          * represents when the rendering involved will be completed.
 254          *
 255          * A reference is held on the buffer while on this list.
 256          */
 257         struct list_head active_list;
 258
 259         /**
 260          * List of breadcrumbs associated with GPU requests currently
 261          * outstanding.
 262          */
 263         struct list_head request_list;
 264
 265         /**
 266          * Do we have some not yet emitted requests outstanding?
 267          */
 268         struct drm_i915_gem_request *outstanding_lazy_request;
 269         bool gpu_caches_dirty;
 270
 271         wait_queue_head_t irq_queue;
 272
 273         struct intel_context *default_context;
 274         struct intel_context *last_context;
 275
 276         struct intel_ring_hangcheck hangcheck;
 277
 278         struct {
 279                 struct drm_i915_gem_object *obj;
 280                 u32 gtt_offset;
 281                 volatile u32 *cpu_page;
 282         } scratch;
 283
 284         bool needs_cmd_parser;
 285
 286         /*
 287          * Table of commands the command parser needs to know about
 288          * for this ring.
 289          */
 290         DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
 291
 292         /*
 293          * Table of registers allowed in commands that read/write registers.
 294          */
 295         const u32 *reg_table;
 296         int reg_count;
 297
 298         /*
 299          * Table of registers allowed in commands that read/write registers, but
 300          * only from the DRM master.
 301          */
 302         const u32 *master_reg_table;
 303         int master_reg_count;
 304
 305         /*
 306          * Returns the bitmask for the length field of the specified command.
 307          * Return 0 for an unrecognized/invalid command.
 308          *
 309          * If the command parser finds an entry for a command in the ring's
 310          * cmd_tables, it gets the command's length based on the table entry.
 311          * If not, it calls this function to determine the per-ring length field
 312          * encoding for the command (i.e. certain opcode ranges use certain bits
 313          * to encode the command length in the header).
 314          */
 315         u32 (*get_cmd_length_mask)(u32 cmd_header);
 316 };
 317
 318 bool intel_ring_initialized(struct intel_engine_cs *ring);
 319
 320 static inline unsigned
 321 intel_ring_flag(struct intel_engine_cs *ring)
 322 {
 323         return 1 << ring->id;
 324 }
 325
 326 static inline u32
 327 intel_ring_sync_index(struct intel_engine_cs *ring,
 328                       struct intel_engine_cs *other)
 329 {
 330         int idx;
 331
 332         /*
 333          * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
 334          * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
 335          * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
 336          * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
 337          * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
 338          */
 339
 340         idx = (other - ring) - 1;
 341         if (idx < 0)
 342                 idx += I915_NUM_RINGS;
 343
 344         return idx;
 345 }
 346
 347 static inline u32
 348 intel_read_status_page(struct intel_engine_cs *ring,
 349                        int reg)
 350 {
 351         /* Ensure that the compiler doesn't optimize away the load. */
 352         barrier();
 353         return ring->status_page.page_addr[reg];
 354 }
 355
 356 static inline void
 357 intel_write_status_page(struct intel_engine_cs *ring,
 358                         int reg, u32 value)
 359 {
 360         ring->status_page.page_addr[reg] = value;
 361 }
 362
 363 /**
 364  * Reads a dword out of the status page, which is written to from the command
 365  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
 366  * MI_STORE_DATA_IMM.
 367  *
 368  * The following dwords have a reserved meaning:
 369  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
 370  * 0x04: ring 0 head pointer
 371  * 0x05: ring 1 head pointer (915-class)
 372  * 0x06: ring 2 head pointer (915-class)
 373  * 0x10-0x1b: Context status DWords (GM45)
 374  * 0x1f: Last written status offset. (GM45)
 375  * 0x20-0x2f: Reserved (Gen6+)
 376  *
 377  * The area from dword 0x30 to 0x3ff is available for driver usage.
 378  */
 379 #define I915_GEM_HWS_INDEX              0x30
 380 #define I915_GEM_HWS_SCRATCH_INDEX      0x40
 381 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
 382
 383 void intel_unpin_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
 384 int intel_pin_and_map_ringbuffer_obj(struct drm_device *dev,
 385                                      struct intel_ringbuffer *ringbuf);
 386 void intel_destroy_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
 387 int intel_alloc_ringbuffer_obj(struct drm_device *dev,
 388                                struct intel_ringbuffer *ringbuf);
 389
 390 void intel_stop_ring_buffer(struct intel_engine_cs *ring);
 391 void intel_cleanup_ring_buffer(struct intel_engine_cs *ring);
 392
 393 int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
 394
 395 int __must_check intel_ring_begin(struct intel_engine_cs *ring, int n);
 396 int __must_check intel_ring_cacheline_align(struct intel_engine_cs *ring);
 397 static inline void intel_ring_emit(struct intel_engine_cs *ring,
 398                                    u32 data)
 399 {
 400         struct intel_ringbuffer *ringbuf = ring->buffer;
 401         iowrite32(data, ringbuf->virtual_start + ringbuf->tail);
 402         ringbuf->tail += 4;
 403 }
 404 static inline void intel_ring_advance(struct intel_engine_cs *ring)
 405 {
 406         struct intel_ringbuffer *ringbuf = ring->buffer;
 407         ringbuf->tail &= ringbuf->size - 1;
 408 }
 409 int __intel_ring_space(int head, int tail, int size);
 410 void intel_ring_update_space(struct intel_ringbuffer *ringbuf);
 411 int intel_ring_space(struct intel_ringbuffer *ringbuf);
 412 bool intel_ring_stopped(struct intel_engine_cs *ring);
 413 void __intel_ring_advance(struct intel_engine_cs *ring);
 414
 415 int __must_check intel_ring_idle(struct intel_engine_cs *ring);
 416 void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno);
 417 int intel_ring_flush_all_caches(struct intel_engine_cs *ring);
 418 int intel_ring_invalidate_all_caches(struct intel_engine_cs *ring);
 419
 420 void intel_fini_pipe_control(struct intel_engine_cs *ring);
 421 int intel_init_pipe_control(struct intel_engine_cs *ring);
 422
 423 int intel_init_render_ring_buffer(struct drm_device *dev);
 424 int intel_init_bsd_ring_buffer(struct drm_device *dev);
 425 int intel_init_bsd2_ring_buffer(struct drm_device *dev);
 426 int intel_init_blt_ring_buffer(struct drm_device *dev);
 427 int intel_init_vebox_ring_buffer(struct drm_device *dev);
 428
 429 u64 intel_ring_get_active_head(struct intel_engine_cs *ring);
 430
 431 int init_workarounds_ring(struct intel_engine_cs *ring);
 432
 433 static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf)
 434 {
 435         return ringbuf->tail;
 436 }
 437
 438 static inline struct drm_i915_gem_request *
 439 intel_ring_get_request(struct intel_engine_cs *ring)
 440 {
 441         BUG_ON(ring->outstanding_lazy_request == NULL);
 442         return ring->outstanding_lazy_request;
 443 }
 444
 445 #endif /* _INTEL_RINGBUFFER_H_ */