Merge tag 'keystone-dts-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/ssant...
[linux-drm-fsl-dcu.git] / drivers / gpu / drm / i915 / intel_ringbuffer.h
1 #ifndef _INTEL_RINGBUFFER_H_
2 #define _INTEL_RINGBUFFER_H_
3
4 #include <linux/hashtable.h>
5 #include "i915_gem_batch_pool.h"
6
7 #define I915_CMD_HASH_ORDER 9
8
9 /* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill,
10  * but keeps the logic simple. Indeed, the whole purpose of this macro is just
11  * to give some inclination as to some of the magic values used in the various
12  * workarounds!
13  */
14 #define CACHELINE_BYTES 64
15
16 /*
17  * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
18  * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
19  * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
20  *
21  * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
22  * cacheline, the Head Pointer must not be greater than the Tail
23  * Pointer."
24  */
25 #define I915_RING_FREE_SPACE 64
26
27 struct  intel_hw_status_page {
28         u32             *page_addr;
29         unsigned int    gfx_addr;
30         struct          drm_i915_gem_object *obj;
31 };
32
33 #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base))
34 #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val)
35
36 #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base))
37 #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val)
38
39 #define I915_READ_HEAD(ring)  I915_READ(RING_HEAD((ring)->mmio_base))
40 #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val)
41
42 #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base))
43 #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val)
44
45 #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base))
46 #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val)
47
48 #define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base))
49 #define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val)
50
51 /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to
52  * do the writes, and that must have qw aligned offsets, simply pretend it's 8b.
53  */
54 #define i915_semaphore_seqno_size sizeof(uint64_t)
55 #define GEN8_SIGNAL_OFFSET(__ring, to)                       \
56         (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
57         ((__ring)->id * I915_NUM_RINGS * i915_semaphore_seqno_size) +   \
58         (i915_semaphore_seqno_size * (to)))
59
60 #define GEN8_WAIT_OFFSET(__ring, from)                       \
61         (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \
62         ((from) * I915_NUM_RINGS * i915_semaphore_seqno_size) + \
63         (i915_semaphore_seqno_size * (__ring)->id))
64
65 #define GEN8_RING_SEMAPHORE_INIT do { \
66         if (!dev_priv->semaphore_obj) { \
67                 break; \
68         } \
69         ring->semaphore.signal_ggtt[RCS] = GEN8_SIGNAL_OFFSET(ring, RCS); \
70         ring->semaphore.signal_ggtt[VCS] = GEN8_SIGNAL_OFFSET(ring, VCS); \
71         ring->semaphore.signal_ggtt[BCS] = GEN8_SIGNAL_OFFSET(ring, BCS); \
72         ring->semaphore.signal_ggtt[VECS] = GEN8_SIGNAL_OFFSET(ring, VECS); \
73         ring->semaphore.signal_ggtt[VCS2] = GEN8_SIGNAL_OFFSET(ring, VCS2); \
74         ring->semaphore.signal_ggtt[ring->id] = MI_SEMAPHORE_SYNC_INVALID; \
75         } while(0)
76
77 enum intel_ring_hangcheck_action {
78         HANGCHECK_IDLE = 0,
79         HANGCHECK_WAIT,
80         HANGCHECK_ACTIVE,
81         HANGCHECK_ACTIVE_LOOP,
82         HANGCHECK_KICK,
83         HANGCHECK_HUNG,
84 };
85
86 #define HANGCHECK_SCORE_RING_HUNG 31
87
88 struct intel_ring_hangcheck {
89         u64 acthd;
90         u64 max_acthd;
91         u32 seqno;
92         int score;
93         enum intel_ring_hangcheck_action action;
94         int deadlock;
95 };
96
97 struct intel_ringbuffer {
98         struct drm_i915_gem_object *obj;
99         void __iomem *virtual_start;
100
101         struct intel_engine_cs *ring;
102
103         u32 head;
104         u32 tail;
105         int space;
106         int size;
107         int effective_size;
108
109         /** We track the position of the requests in the ring buffer, and
110          * when each is retired we increment last_retired_head as the GPU
111          * must have finished processing the request and so we know we
112          * can advance the ringbuffer up to that position.
113          *
114          * last_retired_head is set to -1 after the value is consumed so
115          * we can detect new retirements.
116          */
117         u32 last_retired_head;
118 };
119
120 struct  intel_context;
121 struct drm_i915_reg_descriptor;
122
123 struct  intel_engine_cs {
124         const char      *name;
125         enum intel_ring_id {
126                 RCS = 0x0,
127                 VCS,
128                 BCS,
129                 VECS,
130                 VCS2
131         } id;
132 #define I915_NUM_RINGS 5
133 #define LAST_USER_RING (VECS + 1)
134         u32             mmio_base;
135         struct          drm_device *dev;
136         struct intel_ringbuffer *buffer;
137
138         /*
139          * A pool of objects to use as shadow copies of client batch buffers
140          * when the command parser is enabled. Prevents the client from
141          * modifying the batch contents after software parsing.
142          */
143         struct i915_gem_batch_pool batch_pool;
144
145         struct intel_hw_status_page status_page;
146
147         unsigned irq_refcount; /* protected by dev_priv->irq_lock */
148         u32             irq_enable_mask;        /* bitmask to enable ring interrupt */
149         struct drm_i915_gem_request *trace_irq_req;
150         bool __must_check (*irq_get)(struct intel_engine_cs *ring);
151         void            (*irq_put)(struct intel_engine_cs *ring);
152
153         int             (*init_hw)(struct intel_engine_cs *ring);
154
155         int             (*init_context)(struct intel_engine_cs *ring,
156                                         struct intel_context *ctx);
157
158         void            (*write_tail)(struct intel_engine_cs *ring,
159                                       u32 value);
160         int __must_check (*flush)(struct intel_engine_cs *ring,
161                                   u32   invalidate_domains,
162                                   u32   flush_domains);
163         int             (*add_request)(struct intel_engine_cs *ring);
164         /* Some chipsets are not quite as coherent as advertised and need
165          * an expensive kick to force a true read of the up-to-date seqno.
166          * However, the up-to-date seqno is not always required and the last
167          * seen value is good enough. Note that the seqno will always be
168          * monotonic, even if not coherent.
169          */
170         u32             (*get_seqno)(struct intel_engine_cs *ring,
171                                      bool lazy_coherency);
172         void            (*set_seqno)(struct intel_engine_cs *ring,
173                                      u32 seqno);
174         int             (*dispatch_execbuffer)(struct intel_engine_cs *ring,
175                                                u64 offset, u32 length,
176                                                unsigned dispatch_flags);
177 #define I915_DISPATCH_SECURE 0x1
178 #define I915_DISPATCH_PINNED 0x2
179         void            (*cleanup)(struct intel_engine_cs *ring);
180
181         /* GEN8 signal/wait table - never trust comments!
182          *        signal to     signal to    signal to   signal to      signal to
183          *          RCS            VCS          BCS        VECS          VCS2
184          *      --------------------------------------------------------------------
185          *  RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) |
186          *      |-------------------------------------------------------------------
187          *  VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) |
188          *      |-------------------------------------------------------------------
189          *  BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) |
190          *      |-------------------------------------------------------------------
191          * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) |  NOP (0x90) | VCS2 (0x98) |
192          *      |-------------------------------------------------------------------
193          * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP  (0xc0) |
194          *      |-------------------------------------------------------------------
195          *
196          * Generalization:
197          *  f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id)
198          *  ie. transpose of g(x, y)
199          *
200          *       sync from      sync from    sync from    sync from     sync from
201          *          RCS            VCS          BCS        VECS          VCS2
202          *      --------------------------------------------------------------------
203          *  RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) |
204          *      |-------------------------------------------------------------------
205          *  VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) |
206          *      |-------------------------------------------------------------------
207          *  BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) |
208          *      |-------------------------------------------------------------------
209          * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) |  NOP (0x90) | VCS2 (0xb8) |
210          *      |-------------------------------------------------------------------
211          * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) |  NOP (0xc0) |
212          *      |-------------------------------------------------------------------
213          *
214          * Generalization:
215          *  g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id)
216          *  ie. transpose of f(x, y)
217          */
218         struct {
219                 u32     sync_seqno[I915_NUM_RINGS-1];
220
221                 union {
222                         struct {
223                                 /* our mbox written by others */
224                                 u32             wait[I915_NUM_RINGS];
225                                 /* mboxes this ring signals to */
226                                 u32             signal[I915_NUM_RINGS];
227                         } mbox;
228                         u64             signal_ggtt[I915_NUM_RINGS];
229                 };
230
231                 /* AKA wait() */
232                 int     (*sync_to)(struct intel_engine_cs *ring,
233                                    struct intel_engine_cs *to,
234                                    u32 seqno);
235                 int     (*signal)(struct intel_engine_cs *signaller,
236                                   /* num_dwords needed by caller */
237                                   unsigned int num_dwords);
238         } semaphore;
239
240         /* Execlists */
241         spinlock_t execlist_lock;
242         struct list_head execlist_queue;
243         struct list_head execlist_retired_req_list;
244         u8 next_context_status_buffer;
245         u32             irq_keep_mask; /* bitmask for interrupts that should not be masked */
246         int             (*emit_request)(struct intel_ringbuffer *ringbuf,
247                                         struct drm_i915_gem_request *request);
248         int             (*emit_flush)(struct intel_ringbuffer *ringbuf,
249                                       struct intel_context *ctx,
250                                       u32 invalidate_domains,
251                                       u32 flush_domains);
252         int             (*emit_bb_start)(struct intel_ringbuffer *ringbuf,
253                                          struct intel_context *ctx,
254                                          u64 offset, unsigned dispatch_flags);
255
256         /**
257          * List of objects currently involved in rendering from the
258          * ringbuffer.
259          *
260          * Includes buffers having the contents of their GPU caches
261          * flushed, not necessarily primitives.  last_read_req
262          * represents when the rendering involved will be completed.
263          *
264          * A reference is held on the buffer while on this list.
265          */
266         struct list_head active_list;
267
268         /**
269          * List of breadcrumbs associated with GPU requests currently
270          * outstanding.
271          */
272         struct list_head request_list;
273
274         /**
275          * Do we have some not yet emitted requests outstanding?
276          */
277         struct drm_i915_gem_request *outstanding_lazy_request;
278         bool gpu_caches_dirty;
279
280         wait_queue_head_t irq_queue;
281
282         struct intel_context *default_context;
283         struct intel_context *last_context;
284
285         struct intel_ring_hangcheck hangcheck;
286
287         struct {
288                 struct drm_i915_gem_object *obj;
289                 u32 gtt_offset;
290                 volatile u32 *cpu_page;
291         } scratch;
292
293         bool needs_cmd_parser;
294
295         /*
296          * Table of commands the command parser needs to know about
297          * for this ring.
298          */
299         DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
300
301         /*
302          * Table of registers allowed in commands that read/write registers.
303          */
304         const struct drm_i915_reg_descriptor *reg_table;
305         int reg_count;
306
307         /*
308          * Table of registers allowed in commands that read/write registers, but
309          * only from the DRM master.
310          */
311         const struct drm_i915_reg_descriptor *master_reg_table;
312         int master_reg_count;
313
314         /*
315          * Returns the bitmask for the length field of the specified command.
316          * Return 0 for an unrecognized/invalid command.
317          *
318          * If the command parser finds an entry for a command in the ring's
319          * cmd_tables, it gets the command's length based on the table entry.
320          * If not, it calls this function to determine the per-ring length field
321          * encoding for the command (i.e. certain opcode ranges use certain bits
322          * to encode the command length in the header).
323          */
324         u32 (*get_cmd_length_mask)(u32 cmd_header);
325 };
326
327 bool intel_ring_initialized(struct intel_engine_cs *ring);
328
329 static inline unsigned
330 intel_ring_flag(struct intel_engine_cs *ring)
331 {
332         return 1 << ring->id;
333 }
334
335 static inline u32
336 intel_ring_sync_index(struct intel_engine_cs *ring,
337                       struct intel_engine_cs *other)
338 {
339         int idx;
340
341         /*
342          * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2;
343          * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs;
344          * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs;
345          * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs;
346          * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs;
347          */
348
349         idx = (other - ring) - 1;
350         if (idx < 0)
351                 idx += I915_NUM_RINGS;
352
353         return idx;
354 }
355
356 static inline u32
357 intel_read_status_page(struct intel_engine_cs *ring,
358                        int reg)
359 {
360         /* Ensure that the compiler doesn't optimize away the load. */
361         barrier();
362         return ring->status_page.page_addr[reg];
363 }
364
365 static inline void
366 intel_write_status_page(struct intel_engine_cs *ring,
367                         int reg, u32 value)
368 {
369         ring->status_page.page_addr[reg] = value;
370 }
371
372 /**
373  * Reads a dword out of the status page, which is written to from the command
374  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
375  * MI_STORE_DATA_IMM.
376  *
377  * The following dwords have a reserved meaning:
378  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
379  * 0x04: ring 0 head pointer
380  * 0x05: ring 1 head pointer (915-class)
381  * 0x06: ring 2 head pointer (915-class)
382  * 0x10-0x1b: Context status DWords (GM45)
383  * 0x1f: Last written status offset. (GM45)
384  * 0x20-0x2f: Reserved (Gen6+)
385  *
386  * The area from dword 0x30 to 0x3ff is available for driver usage.
387  */
388 #define I915_GEM_HWS_INDEX              0x30
389 #define I915_GEM_HWS_SCRATCH_INDEX      0x40
390 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
391
392 void intel_unpin_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
393 int intel_pin_and_map_ringbuffer_obj(struct drm_device *dev,
394                                      struct intel_ringbuffer *ringbuf);
395 void intel_destroy_ringbuffer_obj(struct intel_ringbuffer *ringbuf);
396 int intel_alloc_ringbuffer_obj(struct drm_device *dev,
397                                struct intel_ringbuffer *ringbuf);
398
399 void intel_stop_ring_buffer(struct intel_engine_cs *ring);
400 void intel_cleanup_ring_buffer(struct intel_engine_cs *ring);
401
402 int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request);
403
404 int __must_check intel_ring_begin(struct intel_engine_cs *ring, int n);
405 int __must_check intel_ring_cacheline_align(struct intel_engine_cs *ring);
406 static inline void intel_ring_emit(struct intel_engine_cs *ring,
407                                    u32 data)
408 {
409         struct intel_ringbuffer *ringbuf = ring->buffer;
410         iowrite32(data, ringbuf->virtual_start + ringbuf->tail);
411         ringbuf->tail += 4;
412 }
413 static inline void intel_ring_advance(struct intel_engine_cs *ring)
414 {
415         struct intel_ringbuffer *ringbuf = ring->buffer;
416         ringbuf->tail &= ringbuf->size - 1;
417 }
418 int __intel_ring_space(int head, int tail, int size);
419 void intel_ring_update_space(struct intel_ringbuffer *ringbuf);
420 int intel_ring_space(struct intel_ringbuffer *ringbuf);
421 bool intel_ring_stopped(struct intel_engine_cs *ring);
422 void __intel_ring_advance(struct intel_engine_cs *ring);
423
424 int __must_check intel_ring_idle(struct intel_engine_cs *ring);
425 void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno);
426 int intel_ring_flush_all_caches(struct intel_engine_cs *ring);
427 int intel_ring_invalidate_all_caches(struct intel_engine_cs *ring);
428
429 void intel_fini_pipe_control(struct intel_engine_cs *ring);
430 int intel_init_pipe_control(struct intel_engine_cs *ring);
431
432 int intel_init_render_ring_buffer(struct drm_device *dev);
433 int intel_init_bsd_ring_buffer(struct drm_device *dev);
434 int intel_init_bsd2_ring_buffer(struct drm_device *dev);
435 int intel_init_blt_ring_buffer(struct drm_device *dev);
436 int intel_init_vebox_ring_buffer(struct drm_device *dev);
437
438 u64 intel_ring_get_active_head(struct intel_engine_cs *ring);
439
440 int init_workarounds_ring(struct intel_engine_cs *ring);
441
442 static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf)
443 {
444         return ringbuf->tail;
445 }
446
447 static inline struct drm_i915_gem_request *
448 intel_ring_get_request(struct intel_engine_cs *ring)
449 {
450         BUG_ON(ring->outstanding_lazy_request == NULL);
451         return ring->outstanding_lazy_request;
452 }
453
454 #endif /* _INTEL_RINGBUFFER_H_ */