Merge tag 'mvebu-dt-3.17-2' of git://git.infradead.org/linux-mvebu into next/dt
[cascardo/linux.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
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 enum intel_ring_hangcheck_action {
44         HANGCHECK_IDLE = 0,
45         HANGCHECK_WAIT,
46         HANGCHECK_ACTIVE,
47         HANGCHECK_KICK,
48         HANGCHECK_HUNG,
49 };
50
51 #define HANGCHECK_SCORE_RING_HUNG 31
52
53 struct intel_ring_hangcheck {
54         u64 acthd;
55         u32 seqno;
56         int score;
57         enum intel_ring_hangcheck_action action;
58         int deadlock;
59 };
60
61 struct intel_ringbuffer {
62         struct drm_i915_gem_object *obj;
63         void __iomem *virtual_start;
64
65         u32 head;
66         u32 tail;
67         int space;
68         int size;
69         int effective_size;
70
71         /** We track the position of the requests in the ring buffer, and
72          * when each is retired we increment last_retired_head as the GPU
73          * must have finished processing the request and so we know we
74          * can advance the ringbuffer up to that position.
75          *
76          * last_retired_head is set to -1 after the value is consumed so
77          * we can detect new retirements.
78          */
79         u32 last_retired_head;
80 };
81
82 struct  intel_engine_cs {
83         const char      *name;
84         enum intel_ring_id {
85                 RCS = 0x0,
86                 VCS,
87                 BCS,
88                 VECS,
89                 VCS2
90         } id;
91 #define I915_NUM_RINGS 5
92 #define LAST_USER_RING (VECS + 1)
93         u32             mmio_base;
94         struct          drm_device *dev;
95         struct intel_ringbuffer *buffer;
96
97         struct intel_hw_status_page status_page;
98
99         unsigned irq_refcount; /* protected by dev_priv->irq_lock */
100         u32             irq_enable_mask;        /* bitmask to enable ring interrupt */
101         u32             trace_irq_seqno;
102         bool __must_check (*irq_get)(struct intel_engine_cs *ring);
103         void            (*irq_put)(struct intel_engine_cs *ring);
104
105         int             (*init)(struct intel_engine_cs *ring);
106
107         void            (*write_tail)(struct intel_engine_cs *ring,
108                                       u32 value);
109         int __must_check (*flush)(struct intel_engine_cs *ring,
110                                   u32   invalidate_domains,
111                                   u32   flush_domains);
112         int             (*add_request)(struct intel_engine_cs *ring);
113         /* Some chipsets are not quite as coherent as advertised and need
114          * an expensive kick to force a true read of the up-to-date seqno.
115          * However, the up-to-date seqno is not always required and the last
116          * seen value is good enough. Note that the seqno will always be
117          * monotonic, even if not coherent.
118          */
119         u32             (*get_seqno)(struct intel_engine_cs *ring,
120                                      bool lazy_coherency);
121         void            (*set_seqno)(struct intel_engine_cs *ring,
122                                      u32 seqno);
123         int             (*dispatch_execbuffer)(struct intel_engine_cs *ring,
124                                                u64 offset, u32 length,
125                                                unsigned flags);
126 #define I915_DISPATCH_SECURE 0x1
127 #define I915_DISPATCH_PINNED 0x2
128         void            (*cleanup)(struct intel_engine_cs *ring);
129
130         struct {
131                 u32     sync_seqno[I915_NUM_RINGS-1];
132
133                 struct {
134                         /* our mbox written by others */
135                         u32             wait[I915_NUM_RINGS];
136                         /* mboxes this ring signals to */
137                         u32             signal[I915_NUM_RINGS];
138                 } mbox;
139
140                 /* AKA wait() */
141                 int     (*sync_to)(struct intel_engine_cs *ring,
142                                    struct intel_engine_cs *to,
143                                    u32 seqno);
144                 int     (*signal)(struct intel_engine_cs *signaller,
145                                   /* num_dwords needed by caller */
146                                   unsigned int num_dwords);
147         } semaphore;
148
149         /**
150          * List of objects currently involved in rendering from the
151          * ringbuffer.
152          *
153          * Includes buffers having the contents of their GPU caches
154          * flushed, not necessarily primitives.  last_rendering_seqno
155          * represents when the rendering involved will be completed.
156          *
157          * A reference is held on the buffer while on this list.
158          */
159         struct list_head active_list;
160
161         /**
162          * List of breadcrumbs associated with GPU requests currently
163          * outstanding.
164          */
165         struct list_head request_list;
166
167         /**
168          * Do we have some not yet emitted requests outstanding?
169          */
170         struct drm_i915_gem_request *preallocated_lazy_request;
171         u32 outstanding_lazy_seqno;
172         bool gpu_caches_dirty;
173         bool fbc_dirty;
174
175         wait_queue_head_t irq_queue;
176
177         struct intel_context *default_context;
178         struct intel_context *last_context;
179
180         struct intel_ring_hangcheck hangcheck;
181
182         struct {
183                 struct drm_i915_gem_object *obj;
184                 u32 gtt_offset;
185                 volatile u32 *cpu_page;
186         } scratch;
187
188         bool needs_cmd_parser;
189
190         /*
191          * Table of commands the command parser needs to know about
192          * for this ring.
193          */
194         DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER);
195
196         /*
197          * Table of registers allowed in commands that read/write registers.
198          */
199         const u32 *reg_table;
200         int reg_count;
201
202         /*
203          * Table of registers allowed in commands that read/write registers, but
204          * only from the DRM master.
205          */
206         const u32 *master_reg_table;
207         int master_reg_count;
208
209         /*
210          * Returns the bitmask for the length field of the specified command.
211          * Return 0 for an unrecognized/invalid command.
212          *
213          * If the command parser finds an entry for a command in the ring's
214          * cmd_tables, it gets the command's length based on the table entry.
215          * If not, it calls this function to determine the per-ring length field
216          * encoding for the command (i.e. certain opcode ranges use certain bits
217          * to encode the command length in the header).
218          */
219         u32 (*get_cmd_length_mask)(u32 cmd_header);
220 };
221
222 static inline bool
223 intel_ring_initialized(struct intel_engine_cs *ring)
224 {
225         return ring->buffer && ring->buffer->obj;
226 }
227
228 static inline unsigned
229 intel_ring_flag(struct intel_engine_cs *ring)
230 {
231         return 1 << ring->id;
232 }
233
234 static inline u32
235 intel_ring_sync_index(struct intel_engine_cs *ring,
236                       struct intel_engine_cs *other)
237 {
238         int idx;
239
240         /*
241          * cs -> 0 = vcs, 1 = bcs
242          * vcs -> 0 = bcs, 1 = cs,
243          * bcs -> 0 = cs, 1 = vcs.
244          */
245
246         idx = (other - ring) - 1;
247         if (idx < 0)
248                 idx += I915_NUM_RINGS;
249
250         return idx;
251 }
252
253 static inline u32
254 intel_read_status_page(struct intel_engine_cs *ring,
255                        int reg)
256 {
257         /* Ensure that the compiler doesn't optimize away the load. */
258         barrier();
259         return ring->status_page.page_addr[reg];
260 }
261
262 static inline void
263 intel_write_status_page(struct intel_engine_cs *ring,
264                         int reg, u32 value)
265 {
266         ring->status_page.page_addr[reg] = value;
267 }
268
269 /**
270  * Reads a dword out of the status page, which is written to from the command
271  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
272  * MI_STORE_DATA_IMM.
273  *
274  * The following dwords have a reserved meaning:
275  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
276  * 0x04: ring 0 head pointer
277  * 0x05: ring 1 head pointer (915-class)
278  * 0x06: ring 2 head pointer (915-class)
279  * 0x10-0x1b: Context status DWords (GM45)
280  * 0x1f: Last written status offset. (GM45)
281  *
282  * The area from dword 0x20 to 0x3ff is available for driver usage.
283  */
284 #define I915_GEM_HWS_INDEX              0x20
285 #define I915_GEM_HWS_SCRATCH_INDEX      0x30
286 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
287
288 void intel_stop_ring_buffer(struct intel_engine_cs *ring);
289 void intel_cleanup_ring_buffer(struct intel_engine_cs *ring);
290
291 int __must_check intel_ring_begin(struct intel_engine_cs *ring, int n);
292 int __must_check intel_ring_cacheline_align(struct intel_engine_cs *ring);
293 static inline void intel_ring_emit(struct intel_engine_cs *ring,
294                                    u32 data)
295 {
296         struct intel_ringbuffer *ringbuf = ring->buffer;
297         iowrite32(data, ringbuf->virtual_start + ringbuf->tail);
298         ringbuf->tail += 4;
299 }
300 static inline void intel_ring_advance(struct intel_engine_cs *ring)
301 {
302         struct intel_ringbuffer *ringbuf = ring->buffer;
303         ringbuf->tail &= ringbuf->size - 1;
304 }
305 void __intel_ring_advance(struct intel_engine_cs *ring);
306
307 int __must_check intel_ring_idle(struct intel_engine_cs *ring);
308 void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno);
309 int intel_ring_flush_all_caches(struct intel_engine_cs *ring);
310 int intel_ring_invalidate_all_caches(struct intel_engine_cs *ring);
311
312 int intel_init_render_ring_buffer(struct drm_device *dev);
313 int intel_init_bsd_ring_buffer(struct drm_device *dev);
314 int intel_init_bsd2_ring_buffer(struct drm_device *dev);
315 int intel_init_blt_ring_buffer(struct drm_device *dev);
316 int intel_init_vebox_ring_buffer(struct drm_device *dev);
317
318 u64 intel_ring_get_active_head(struct intel_engine_cs *ring);
319 void intel_ring_setup_status_page(struct intel_engine_cs *ring);
320
321 static inline u32 intel_ring_get_tail(struct intel_engine_cs *ring)
322 {
323         return ring->buffer->tail;
324 }
325
326 static inline u32 intel_ring_get_seqno(struct intel_engine_cs *ring)
327 {
328         BUG_ON(ring->outstanding_lazy_seqno == 0);
329         return ring->outstanding_lazy_seqno;
330 }
331
332 static inline void i915_trace_irq_get(struct intel_engine_cs *ring, u32 seqno)
333 {
334         if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
335                 ring->trace_irq_seqno = seqno;
336 }
337
338 /* DRI warts */
339 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
340
341 #endif /* _INTEL_RINGBUFFER_H_ */