| 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 | /* |
| 9 | * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use" |
| 10 | * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use" |
| 11 | * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use" |
| 12 | * |
| 13 | * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same |
| 14 | * cacheline, the Head Pointer must not be greater than the Tail |
| 15 | * Pointer." |
| 16 | */ |
| 17 | #define I915_RING_FREE_SPACE 64 |
| 18 | |
| 19 | struct intel_hw_status_page { |
| 20 | u32 *page_addr; |
| 21 | unsigned int gfx_addr; |
| 22 | struct drm_i915_gem_object *obj; |
| 23 | }; |
| 24 | |
| 25 | #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base)) |
| 26 | #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val) |
| 27 | |
| 28 | #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base)) |
| 29 | #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val) |
| 30 | |
| 31 | #define I915_READ_HEAD(ring) I915_READ(RING_HEAD((ring)->mmio_base)) |
| 32 | #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val) |
| 33 | |
| 34 | #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base)) |
| 35 | #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val) |
| 36 | |
| 37 | #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base)) |
| 38 | #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val) |
| 39 | |
| 40 | #define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base)) |
| 41 | #define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val) |
| 42 | |
| 43 | /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to |
| 44 | * do the writes, and that must have qw aligned offsets, simply pretend it's 8b. |
| 45 | */ |
| 46 | #define i915_semaphore_seqno_size sizeof(uint64_t) |
| 47 | #define GEN8_SIGNAL_OFFSET(__ring, to) \ |
| 48 | (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \ |
| 49 | ((__ring)->id * I915_NUM_RINGS * i915_semaphore_seqno_size) + \ |
| 50 | (i915_semaphore_seqno_size * (to))) |
| 51 | |
| 52 | #define GEN8_WAIT_OFFSET(__ring, from) \ |
| 53 | (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \ |
| 54 | ((from) * I915_NUM_RINGS * i915_semaphore_seqno_size) + \ |
| 55 | (i915_semaphore_seqno_size * (__ring)->id)) |
| 56 | |
| 57 | #define GEN8_RING_SEMAPHORE_INIT do { \ |
| 58 | if (!dev_priv->semaphore_obj) { \ |
| 59 | break; \ |
| 60 | } \ |
| 61 | ring->semaphore.signal_ggtt[RCS] = GEN8_SIGNAL_OFFSET(ring, RCS); \ |
| 62 | ring->semaphore.signal_ggtt[VCS] = GEN8_SIGNAL_OFFSET(ring, VCS); \ |
| 63 | ring->semaphore.signal_ggtt[BCS] = GEN8_SIGNAL_OFFSET(ring, BCS); \ |
| 64 | ring->semaphore.signal_ggtt[VECS] = GEN8_SIGNAL_OFFSET(ring, VECS); \ |
| 65 | ring->semaphore.signal_ggtt[VCS2] = GEN8_SIGNAL_OFFSET(ring, VCS2); \ |
| 66 | ring->semaphore.signal_ggtt[ring->id] = MI_SEMAPHORE_SYNC_INVALID; \ |
| 67 | } while(0) |
| 68 | |
| 69 | enum intel_ring_hangcheck_action { |
| 70 | HANGCHECK_IDLE = 0, |
| 71 | HANGCHECK_WAIT, |
| 72 | HANGCHECK_ACTIVE, |
| 73 | HANGCHECK_ACTIVE_LOOP, |
| 74 | HANGCHECK_KICK, |
| 75 | HANGCHECK_HUNG, |
| 76 | }; |
| 77 | |
| 78 | #define HANGCHECK_SCORE_RING_HUNG 31 |
| 79 | |
| 80 | struct intel_ring_hangcheck { |
| 81 | u64 acthd; |
| 82 | u64 max_acthd; |
| 83 | u32 seqno; |
| 84 | int score; |
| 85 | enum intel_ring_hangcheck_action action; |
| 86 | int deadlock; |
| 87 | }; |
| 88 | |
| 89 | struct intel_ringbuffer { |
| 90 | struct drm_i915_gem_object *obj; |
| 91 | void __iomem *virtual_start; |
| 92 | |
| 93 | u32 head; |
| 94 | u32 tail; |
| 95 | int space; |
| 96 | int size; |
| 97 | int effective_size; |
| 98 | |
| 99 | /** We track the position of the requests in the ring buffer, and |
| 100 | * when each is retired we increment last_retired_head as the GPU |
| 101 | * must have finished processing the request and so we know we |
| 102 | * can advance the ringbuffer up to that position. |
| 103 | * |
| 104 | * last_retired_head is set to -1 after the value is consumed so |
| 105 | * we can detect new retirements. |
| 106 | */ |
| 107 | u32 last_retired_head; |
| 108 | }; |
| 109 | |
| 110 | struct intel_engine_cs { |
| 111 | const char *name; |
| 112 | enum intel_ring_id { |
| 113 | RCS = 0x0, |
| 114 | VCS, |
| 115 | BCS, |
| 116 | VECS, |
| 117 | VCS2 |
| 118 | } id; |
| 119 | #define I915_NUM_RINGS 5 |
| 120 | #define LAST_USER_RING (VECS + 1) |
| 121 | u32 mmio_base; |
| 122 | struct drm_device *dev; |
| 123 | struct intel_ringbuffer *buffer; |
| 124 | |
| 125 | struct intel_hw_status_page status_page; |
| 126 | |
| 127 | unsigned irq_refcount; /* protected by dev_priv->irq_lock */ |
| 128 | u32 irq_enable_mask; /* bitmask to enable ring interrupt */ |
| 129 | u32 trace_irq_seqno; |
| 130 | bool __must_check (*irq_get)(struct intel_engine_cs *ring); |
| 131 | void (*irq_put)(struct intel_engine_cs *ring); |
| 132 | |
| 133 | int (*init)(struct intel_engine_cs *ring); |
| 134 | |
| 135 | void (*write_tail)(struct intel_engine_cs *ring, |
| 136 | u32 value); |
| 137 | int __must_check (*flush)(struct intel_engine_cs *ring, |
| 138 | u32 invalidate_domains, |
| 139 | u32 flush_domains); |
| 140 | int (*add_request)(struct intel_engine_cs *ring); |
| 141 | /* Some chipsets are not quite as coherent as advertised and need |
| 142 | * an expensive kick to force a true read of the up-to-date seqno. |
| 143 | * However, the up-to-date seqno is not always required and the last |
| 144 | * seen value is good enough. Note that the seqno will always be |
| 145 | * monotonic, even if not coherent. |
| 146 | */ |
| 147 | u32 (*get_seqno)(struct intel_engine_cs *ring, |
| 148 | bool lazy_coherency); |
| 149 | void (*set_seqno)(struct intel_engine_cs *ring, |
| 150 | u32 seqno); |
| 151 | int (*dispatch_execbuffer)(struct intel_engine_cs *ring, |
| 152 | u64 offset, u32 length, |
| 153 | unsigned flags); |
| 154 | #define I915_DISPATCH_SECURE 0x1 |
| 155 | #define I915_DISPATCH_PINNED 0x2 |
| 156 | void (*cleanup)(struct intel_engine_cs *ring); |
| 157 | |
| 158 | /* GEN8 signal/wait table - never trust comments! |
| 159 | * signal to signal to signal to signal to signal to |
| 160 | * RCS VCS BCS VECS VCS2 |
| 161 | * -------------------------------------------------------------------- |
| 162 | * RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) | |
| 163 | * |------------------------------------------------------------------- |
| 164 | * VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) | |
| 165 | * |------------------------------------------------------------------- |
| 166 | * BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) | |
| 167 | * |------------------------------------------------------------------- |
| 168 | * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) | NOP (0x90) | VCS2 (0x98) | |
| 169 | * |------------------------------------------------------------------- |
| 170 | * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP (0xc0) | |
| 171 | * |------------------------------------------------------------------- |
| 172 | * |
| 173 | * Generalization: |
| 174 | * f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id) |
| 175 | * ie. transpose of g(x, y) |
| 176 | * |
| 177 | * sync from sync from sync from sync from sync from |
| 178 | * RCS VCS BCS VECS VCS2 |
| 179 | * -------------------------------------------------------------------- |
| 180 | * RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) | |
| 181 | * |------------------------------------------------------------------- |
| 182 | * VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) | |
| 183 | * |------------------------------------------------------------------- |
| 184 | * BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) | |
| 185 | * |------------------------------------------------------------------- |
| 186 | * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) | NOP (0x90) | VCS2 (0xb8) | |
| 187 | * |------------------------------------------------------------------- |
| 188 | * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) | NOP (0xc0) | |
| 189 | * |------------------------------------------------------------------- |
| 190 | * |
| 191 | * Generalization: |
| 192 | * g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id) |
| 193 | * ie. transpose of f(x, y) |
| 194 | */ |
| 195 | struct { |
| 196 | u32 sync_seqno[I915_NUM_RINGS-1]; |
| 197 | |
| 198 | union { |
| 199 | struct { |
| 200 | /* our mbox written by others */ |
| 201 | u32 wait[I915_NUM_RINGS]; |
| 202 | /* mboxes this ring signals to */ |
| 203 | u32 signal[I915_NUM_RINGS]; |
| 204 | } mbox; |
| 205 | u64 signal_ggtt[I915_NUM_RINGS]; |
| 206 | }; |
| 207 | |
| 208 | /* AKA wait() */ |
| 209 | int (*sync_to)(struct intel_engine_cs *ring, |
| 210 | struct intel_engine_cs *to, |
| 211 | u32 seqno); |
| 212 | int (*signal)(struct intel_engine_cs *signaller, |
| 213 | /* num_dwords needed by caller */ |
| 214 | unsigned int num_dwords); |
| 215 | } semaphore; |
| 216 | |
| 217 | /** |
| 218 | * List of objects currently involved in rendering from the |
| 219 | * ringbuffer. |
| 220 | * |
| 221 | * Includes buffers having the contents of their GPU caches |
| 222 | * flushed, not necessarily primitives. last_rendering_seqno |
| 223 | * represents when the rendering involved will be completed. |
| 224 | * |
| 225 | * A reference is held on the buffer while on this list. |
| 226 | */ |
| 227 | struct list_head active_list; |
| 228 | |
| 229 | /** |
| 230 | * List of breadcrumbs associated with GPU requests currently |
| 231 | * outstanding. |
| 232 | */ |
| 233 | struct list_head request_list; |
| 234 | |
| 235 | /** |
| 236 | * Do we have some not yet emitted requests outstanding? |
| 237 | */ |
| 238 | struct drm_i915_gem_request *preallocated_lazy_request; |
| 239 | u32 outstanding_lazy_seqno; |
| 240 | bool gpu_caches_dirty; |
| 241 | bool fbc_dirty; |
| 242 | |
| 243 | wait_queue_head_t irq_queue; |
| 244 | |
| 245 | struct intel_context *default_context; |
| 246 | struct intel_context *last_context; |
| 247 | |
| 248 | struct intel_ring_hangcheck hangcheck; |
| 249 | |
| 250 | struct { |
| 251 | struct drm_i915_gem_object *obj; |
| 252 | u32 gtt_offset; |
| 253 | volatile u32 *cpu_page; |
| 254 | } scratch; |
| 255 | |
| 256 | bool needs_cmd_parser; |
| 257 | |
| 258 | /* |
| 259 | * Table of commands the command parser needs to know about |
| 260 | * for this ring. |
| 261 | */ |
| 262 | DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER); |
| 263 | |
| 264 | /* |
| 265 | * Table of registers allowed in commands that read/write registers. |
| 266 | */ |
| 267 | const u32 *reg_table; |
| 268 | int reg_count; |
| 269 | |
| 270 | /* |
| 271 | * Table of registers allowed in commands that read/write registers, but |
| 272 | * only from the DRM master. |
| 273 | */ |
| 274 | const u32 *master_reg_table; |
| 275 | int master_reg_count; |
| 276 | |
| 277 | /* |
| 278 | * Returns the bitmask for the length field of the specified command. |
| 279 | * Return 0 for an unrecognized/invalid command. |
| 280 | * |
| 281 | * If the command parser finds an entry for a command in the ring's |
| 282 | * cmd_tables, it gets the command's length based on the table entry. |
| 283 | * If not, it calls this function to determine the per-ring length field |
| 284 | * encoding for the command (i.e. certain opcode ranges use certain bits |
| 285 | * to encode the command length in the header). |
| 286 | */ |
| 287 | u32 (*get_cmd_length_mask)(u32 cmd_header); |
| 288 | }; |
| 289 | |
| 290 | static inline bool |
| 291 | intel_ring_initialized(struct intel_engine_cs *ring) |
| 292 | { |
| 293 | return ring->buffer && ring->buffer->obj; |
| 294 | } |
| 295 | |
| 296 | static inline unsigned |
| 297 | intel_ring_flag(struct intel_engine_cs *ring) |
| 298 | { |
| 299 | return 1 << ring->id; |
| 300 | } |
| 301 | |
| 302 | static inline u32 |
| 303 | intel_ring_sync_index(struct intel_engine_cs *ring, |
| 304 | struct intel_engine_cs *other) |
| 305 | { |
| 306 | int idx; |
| 307 | |
| 308 | /* |
| 309 | * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2; |
| 310 | * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs; |
| 311 | * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs; |
| 312 | * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs; |
| 313 | * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs; |
| 314 | */ |
| 315 | |
| 316 | idx = (other - ring) - 1; |
| 317 | if (idx < 0) |
| 318 | idx += I915_NUM_RINGS; |
| 319 | |
| 320 | return idx; |
| 321 | } |
| 322 | |
| 323 | static inline u32 |
| 324 | intel_read_status_page(struct intel_engine_cs *ring, |
| 325 | int reg) |
| 326 | { |
| 327 | /* Ensure that the compiler doesn't optimize away the load. */ |
| 328 | barrier(); |
| 329 | return ring->status_page.page_addr[reg]; |
| 330 | } |
| 331 | |
| 332 | static inline void |
| 333 | intel_write_status_page(struct intel_engine_cs *ring, |
| 334 | int reg, u32 value) |
| 335 | { |
| 336 | ring->status_page.page_addr[reg] = value; |
| 337 | } |
| 338 | |
| 339 | /** |
| 340 | * Reads a dword out of the status page, which is written to from the command |
| 341 | * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or |
| 342 | * MI_STORE_DATA_IMM. |
| 343 | * |
| 344 | * The following dwords have a reserved meaning: |
| 345 | * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes. |
| 346 | * 0x04: ring 0 head pointer |
| 347 | * 0x05: ring 1 head pointer (915-class) |
| 348 | * 0x06: ring 2 head pointer (915-class) |
| 349 | * 0x10-0x1b: Context status DWords (GM45) |
| 350 | * 0x1f: Last written status offset. (GM45) |
| 351 | * |
| 352 | * The area from dword 0x20 to 0x3ff is available for driver usage. |
| 353 | */ |
| 354 | #define I915_GEM_HWS_INDEX 0x20 |
| 355 | #define I915_GEM_HWS_SCRATCH_INDEX 0x30 |
| 356 | #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT) |
| 357 | |
| 358 | void intel_stop_ring_buffer(struct intel_engine_cs *ring); |
| 359 | void intel_cleanup_ring_buffer(struct intel_engine_cs *ring); |
| 360 | |
| 361 | int __must_check intel_ring_begin(struct intel_engine_cs *ring, int n); |
| 362 | int __must_check intel_ring_cacheline_align(struct intel_engine_cs *ring); |
| 363 | static inline void intel_ring_emit(struct intel_engine_cs *ring, |
| 364 | u32 data) |
| 365 | { |
| 366 | struct intel_ringbuffer *ringbuf = ring->buffer; |
| 367 | iowrite32(data, ringbuf->virtual_start + ringbuf->tail); |
| 368 | ringbuf->tail += 4; |
| 369 | } |
| 370 | static inline void intel_ring_advance(struct intel_engine_cs *ring) |
| 371 | { |
| 372 | struct intel_ringbuffer *ringbuf = ring->buffer; |
| 373 | ringbuf->tail &= ringbuf->size - 1; |
| 374 | } |
| 375 | void __intel_ring_advance(struct intel_engine_cs *ring); |
| 376 | |
| 377 | int __must_check intel_ring_idle(struct intel_engine_cs *ring); |
| 378 | void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno); |
| 379 | int intel_ring_flush_all_caches(struct intel_engine_cs *ring); |
| 380 | int intel_ring_invalidate_all_caches(struct intel_engine_cs *ring); |
| 381 | |
| 382 | int intel_init_render_ring_buffer(struct drm_device *dev); |
| 383 | int intel_init_bsd_ring_buffer(struct drm_device *dev); |
| 384 | int intel_init_bsd2_ring_buffer(struct drm_device *dev); |
| 385 | int intel_init_blt_ring_buffer(struct drm_device *dev); |
| 386 | int intel_init_vebox_ring_buffer(struct drm_device *dev); |
| 387 | |
| 388 | u64 intel_ring_get_active_head(struct intel_engine_cs *ring); |
| 389 | void intel_ring_setup_status_page(struct intel_engine_cs *ring); |
| 390 | |
| 391 | static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf) |
| 392 | { |
| 393 | return ringbuf->tail; |
| 394 | } |
| 395 | |
| 396 | static inline u32 intel_ring_get_seqno(struct intel_engine_cs *ring) |
| 397 | { |
| 398 | BUG_ON(ring->outstanding_lazy_seqno == 0); |
| 399 | return ring->outstanding_lazy_seqno; |
| 400 | } |
| 401 | |
| 402 | static inline void i915_trace_irq_get(struct intel_engine_cs *ring, u32 seqno) |
| 403 | { |
| 404 | if (ring->trace_irq_seqno == 0 && ring->irq_get(ring)) |
| 405 | ring->trace_irq_seqno = seqno; |
| 406 | } |
| 407 | |
| 408 | /* DRI warts */ |
| 409 | int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size); |
| 410 | |
| 411 | #endif /* _INTEL_RINGBUFFER_H_ */ |