]> git.karo-electronics.de Git - karo-tx-linux.git/blob - drivers/gpu/drm/i915/intel_ringbuffer.h
drm/i915: Move ring_begin to signal()
[karo-tx-linux.git] / drivers / gpu / drm / i915 / intel_ringbuffer.h
1 #ifndef _INTEL_RINGBUFFER_H_
2 #define _INTEL_RINGBUFFER_H_
3
4 /*
5  * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use"
6  * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use"
7  * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use"
8  *
9  * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same
10  * cacheline, the Head Pointer must not be greater than the Tail
11  * Pointer."
12  */
13 #define I915_RING_FREE_SPACE 64
14
15 struct  intel_hw_status_page {
16         u32             *page_addr;
17         unsigned int    gfx_addr;
18         struct          drm_i915_gem_object *obj;
19 };
20
21 #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base))
22 #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val)
23
24 #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base))
25 #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val)
26
27 #define I915_READ_HEAD(ring)  I915_READ(RING_HEAD((ring)->mmio_base))
28 #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val)
29
30 #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base))
31 #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val)
32
33 #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base))
34 #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val)
35
36 #define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base))
37 #define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val)
38
39 enum intel_ring_hangcheck_action {
40         HANGCHECK_IDLE = 0,
41         HANGCHECK_WAIT,
42         HANGCHECK_ACTIVE,
43         HANGCHECK_KICK,
44         HANGCHECK_HUNG,
45 };
46
47 #define HANGCHECK_SCORE_RING_HUNG 31
48
49 struct intel_ring_hangcheck {
50         u64 acthd;
51         u32 seqno;
52         int score;
53         enum intel_ring_hangcheck_action action;
54         bool deadlock;
55 };
56
57 struct  intel_ring_buffer {
58         const char      *name;
59         enum intel_ring_id {
60                 RCS = 0x0,
61                 VCS,
62                 BCS,
63                 VECS,
64                 VCS2
65         } id;
66 #define I915_NUM_RINGS 5
67 #define LAST_USER_RING (VECS + 1)
68         u32             mmio_base;
69         void            __iomem *virtual_start;
70         struct          drm_device *dev;
71         struct          drm_i915_gem_object *obj;
72
73         u32             head;
74         u32             tail;
75         int             space;
76         int             size;
77         int             effective_size;
78         struct intel_hw_status_page status_page;
79
80         /** We track the position of the requests in the ring buffer, and
81          * when each is retired we increment last_retired_head as the GPU
82          * must have finished processing the request and so we know we
83          * can advance the ringbuffer up to that position.
84          *
85          * last_retired_head is set to -1 after the value is consumed so
86          * we can detect new retirements.
87          */
88         u32             last_retired_head;
89
90         unsigned irq_refcount; /* protected by dev_priv->irq_lock */
91         u32             irq_enable_mask;        /* bitmask to enable ring interrupt */
92         u32             trace_irq_seqno;
93         bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
94         void            (*irq_put)(struct intel_ring_buffer *ring);
95
96         int             (*init)(struct intel_ring_buffer *ring);
97
98         void            (*write_tail)(struct intel_ring_buffer *ring,
99                                       u32 value);
100         int __must_check (*flush)(struct intel_ring_buffer *ring,
101                                   u32   invalidate_domains,
102                                   u32   flush_domains);
103         int             (*add_request)(struct intel_ring_buffer *ring);
104         /* Some chipsets are not quite as coherent as advertised and need
105          * an expensive kick to force a true read of the up-to-date seqno.
106          * However, the up-to-date seqno is not always required and the last
107          * seen value is good enough. Note that the seqno will always be
108          * monotonic, even if not coherent.
109          */
110         u32             (*get_seqno)(struct intel_ring_buffer *ring,
111                                      bool lazy_coherency);
112         void            (*set_seqno)(struct intel_ring_buffer *ring,
113                                      u32 seqno);
114         int             (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
115                                                u32 offset, u32 length,
116                                                unsigned flags);
117 #define I915_DISPATCH_SECURE 0x1
118 #define I915_DISPATCH_PINNED 0x2
119         void            (*cleanup)(struct intel_ring_buffer *ring);
120
121         struct {
122                 u32     sync_seqno[I915_NUM_RINGS-1];
123
124                 struct {
125                         /* our mbox written by others */
126                         u32             wait[I915_NUM_RINGS];
127                         /* mboxes this ring signals to */
128                         u32             signal[I915_NUM_RINGS];
129                 } mbox;
130
131                 /* AKA wait() */
132                 int     (*sync_to)(struct intel_ring_buffer *ring,
133                                    struct intel_ring_buffer *to,
134                                    u32 seqno);
135                 int     (*signal)(struct intel_ring_buffer *signaller,
136                                   /* num_dwords needed by caller */
137                                   unsigned int num_dwords);
138         } semaphore;
139
140         /**
141          * List of objects currently involved in rendering from the
142          * ringbuffer.
143          *
144          * Includes buffers having the contents of their GPU caches
145          * flushed, not necessarily primitives.  last_rendering_seqno
146          * represents when the rendering involved will be completed.
147          *
148          * A reference is held on the buffer while on this list.
149          */
150         struct list_head active_list;
151
152         /**
153          * List of breadcrumbs associated with GPU requests currently
154          * outstanding.
155          */
156         struct list_head request_list;
157
158         /**
159          * Do we have some not yet emitted requests outstanding?
160          */
161         struct drm_i915_gem_request *preallocated_lazy_request;
162         u32 outstanding_lazy_seqno;
163         bool gpu_caches_dirty;
164         bool fbc_dirty;
165
166         wait_queue_head_t irq_queue;
167
168         struct i915_hw_context *default_context;
169         struct i915_hw_context *last_context;
170
171         struct intel_ring_hangcheck hangcheck;
172
173         struct {
174                 struct drm_i915_gem_object *obj;
175                 u32 gtt_offset;
176                 volatile u32 *cpu_page;
177         } scratch;
178
179         /*
180          * Tables of commands the command parser needs to know about
181          * for this ring.
182          */
183         const struct drm_i915_cmd_table *cmd_tables;
184         int cmd_table_count;
185
186         /*
187          * Table of registers allowed in commands that read/write registers.
188          */
189         const u32 *reg_table;
190         int reg_count;
191
192         /*
193          * Table of registers allowed in commands that read/write registers, but
194          * only from the DRM master.
195          */
196         const u32 *master_reg_table;
197         int master_reg_count;
198
199         /*
200          * Returns the bitmask for the length field of the specified command.
201          * Return 0 for an unrecognized/invalid command.
202          *
203          * If the command parser finds an entry for a command in the ring's
204          * cmd_tables, it gets the command's length based on the table entry.
205          * If not, it calls this function to determine the per-ring length field
206          * encoding for the command (i.e. certain opcode ranges use certain bits
207          * to encode the command length in the header).
208          */
209         u32 (*get_cmd_length_mask)(u32 cmd_header);
210 };
211
212 static inline bool
213 intel_ring_initialized(struct intel_ring_buffer *ring)
214 {
215         return ring->obj != NULL;
216 }
217
218 static inline unsigned
219 intel_ring_flag(struct intel_ring_buffer *ring)
220 {
221         return 1 << ring->id;
222 }
223
224 static inline u32
225 intel_ring_sync_index(struct intel_ring_buffer *ring,
226                       struct intel_ring_buffer *other)
227 {
228         int idx;
229
230         /*
231          * cs -> 0 = vcs, 1 = bcs
232          * vcs -> 0 = bcs, 1 = cs,
233          * bcs -> 0 = cs, 1 = vcs.
234          */
235
236         idx = (other - ring) - 1;
237         if (idx < 0)
238                 idx += I915_NUM_RINGS;
239
240         return idx;
241 }
242
243 static inline u32
244 intel_read_status_page(struct intel_ring_buffer *ring,
245                        int reg)
246 {
247         /* Ensure that the compiler doesn't optimize away the load. */
248         barrier();
249         return ring->status_page.page_addr[reg];
250 }
251
252 static inline void
253 intel_write_status_page(struct intel_ring_buffer *ring,
254                         int reg, u32 value)
255 {
256         ring->status_page.page_addr[reg] = value;
257 }
258
259 /**
260  * Reads a dword out of the status page, which is written to from the command
261  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
262  * MI_STORE_DATA_IMM.
263  *
264  * The following dwords have a reserved meaning:
265  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
266  * 0x04: ring 0 head pointer
267  * 0x05: ring 1 head pointer (915-class)
268  * 0x06: ring 2 head pointer (915-class)
269  * 0x10-0x1b: Context status DWords (GM45)
270  * 0x1f: Last written status offset. (GM45)
271  *
272  * The area from dword 0x20 to 0x3ff is available for driver usage.
273  */
274 #define I915_GEM_HWS_INDEX              0x20
275 #define I915_GEM_HWS_SCRATCH_INDEX      0x30
276 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
277
278 void intel_stop_ring_buffer(struct intel_ring_buffer *ring);
279 void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
280
281 int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
282 int __must_check intel_ring_cacheline_align(struct intel_ring_buffer *ring);
283 static inline void intel_ring_emit(struct intel_ring_buffer *ring,
284                                    u32 data)
285 {
286         iowrite32(data, ring->virtual_start + ring->tail);
287         ring->tail += 4;
288 }
289 static inline void intel_ring_advance(struct intel_ring_buffer *ring)
290 {
291         ring->tail &= ring->size - 1;
292 }
293 void __intel_ring_advance(struct intel_ring_buffer *ring);
294
295 int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
296 void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
297 int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
298 int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
299
300 int intel_init_render_ring_buffer(struct drm_device *dev);
301 int intel_init_bsd_ring_buffer(struct drm_device *dev);
302 int intel_init_bsd2_ring_buffer(struct drm_device *dev);
303 int intel_init_blt_ring_buffer(struct drm_device *dev);
304 int intel_init_vebox_ring_buffer(struct drm_device *dev);
305
306 u64 intel_ring_get_active_head(struct intel_ring_buffer *ring);
307 void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
308
309 static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
310 {
311         return ring->tail;
312 }
313
314 static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
315 {
316         BUG_ON(ring->outstanding_lazy_seqno == 0);
317         return ring->outstanding_lazy_seqno;
318 }
319
320 static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
321 {
322         if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
323                 ring->trace_irq_seqno = seqno;
324 }
325
326 /* DRI warts */
327 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
328
329 #endif /* _INTEL_RINGBUFFER_H_ */