drm/i915: Invariably invalidate before ctx switch
[deliverable/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
38 enum intel_ring_hangcheck_action {
39 HANGCHECK_IDLE = 0,
40 HANGCHECK_WAIT,
41 HANGCHECK_ACTIVE,
42 HANGCHECK_KICK,
43 HANGCHECK_HUNG,
44 };
45
46 #define HANGCHECK_SCORE_RING_HUNG 31
47
48 struct intel_ring_hangcheck {
49 u64 acthd;
50 u32 seqno;
51 int score;
52 enum intel_ring_hangcheck_action action;
53 bool deadlock;
54 };
55
56 struct intel_ring_buffer {
57 const char *name;
58 enum intel_ring_id {
59 RCS = 0x0,
60 VCS,
61 BCS,
62 VECS,
63 } id;
64 #define I915_NUM_RINGS 4
65 u32 mmio_base;
66 void __iomem *virtual_start;
67 struct drm_device *dev;
68 struct drm_i915_gem_object *obj;
69
70 u32 head;
71 u32 tail;
72 int space;
73 int size;
74 int effective_size;
75 struct intel_hw_status_page status_page;
76
77 /** We track the position of the requests in the ring buffer, and
78 * when each is retired we increment last_retired_head as the GPU
79 * must have finished processing the request and so we know we
80 * can advance the ringbuffer up to that position.
81 *
82 * last_retired_head is set to -1 after the value is consumed so
83 * we can detect new retirements.
84 */
85 u32 last_retired_head;
86
87 unsigned irq_refcount; /* protected by dev_priv->irq_lock */
88 u32 irq_enable_mask; /* bitmask to enable ring interrupt */
89 u32 trace_irq_seqno;
90 u32 sync_seqno[I915_NUM_RINGS-1];
91 bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
92 void (*irq_put)(struct intel_ring_buffer *ring);
93
94 int (*init)(struct intel_ring_buffer *ring);
95
96 void (*write_tail)(struct intel_ring_buffer *ring,
97 u32 value);
98 int __must_check (*flush)(struct intel_ring_buffer *ring,
99 u32 invalidate_domains,
100 u32 flush_domains);
101 int (*add_request)(struct intel_ring_buffer *ring);
102 /* Some chipsets are not quite as coherent as advertised and need
103 * an expensive kick to force a true read of the up-to-date seqno.
104 * However, the up-to-date seqno is not always required and the last
105 * seen value is good enough. Note that the seqno will always be
106 * monotonic, even if not coherent.
107 */
108 u32 (*get_seqno)(struct intel_ring_buffer *ring,
109 bool lazy_coherency);
110 void (*set_seqno)(struct intel_ring_buffer *ring,
111 u32 seqno);
112 int (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
113 u32 offset, u32 length,
114 unsigned flags);
115 #define I915_DISPATCH_SECURE 0x1
116 #define I915_DISPATCH_PINNED 0x2
117 void (*cleanup)(struct intel_ring_buffer *ring);
118 int (*sync_to)(struct intel_ring_buffer *ring,
119 struct intel_ring_buffer *to,
120 u32 seqno);
121
122 /* our mbox written by others */
123 u32 semaphore_register[I915_NUM_RINGS];
124 /* mboxes this ring signals to */
125 u32 signal_mbox[I915_NUM_RINGS];
126
127 /**
128 * List of objects currently involved in rendering from the
129 * ringbuffer.
130 *
131 * Includes buffers having the contents of their GPU caches
132 * flushed, not necessarily primitives. last_rendering_seqno
133 * represents when the rendering involved will be completed.
134 *
135 * A reference is held on the buffer while on this list.
136 */
137 struct list_head active_list;
138
139 /**
140 * List of breadcrumbs associated with GPU requests currently
141 * outstanding.
142 */
143 struct list_head request_list;
144
145 /**
146 * Do we have some not yet emitted requests outstanding?
147 */
148 struct drm_i915_gem_request *preallocated_lazy_request;
149 u32 outstanding_lazy_seqno;
150 bool gpu_caches_dirty;
151 bool fbc_dirty;
152
153 wait_queue_head_t irq_queue;
154
155 struct i915_hw_context *default_context;
156 struct i915_hw_context *last_context;
157
158 struct intel_ring_hangcheck hangcheck;
159
160 struct {
161 struct drm_i915_gem_object *obj;
162 u32 gtt_offset;
163 volatile u32 *cpu_page;
164 } scratch;
165
166 /*
167 * Tables of commands the command parser needs to know about
168 * for this ring.
169 */
170 const struct drm_i915_cmd_table *cmd_tables;
171 int cmd_table_count;
172
173 /*
174 * Table of registers allowed in commands that read/write registers.
175 */
176 const u32 *reg_table;
177 int reg_count;
178
179 /*
180 * Table of registers allowed in commands that read/write registers, but
181 * only from the DRM master.
182 */
183 const u32 *master_reg_table;
184 int master_reg_count;
185
186 /*
187 * Returns the bitmask for the length field of the specified command.
188 * Return 0 for an unrecognized/invalid command.
189 *
190 * If the command parser finds an entry for a command in the ring's
191 * cmd_tables, it gets the command's length based on the table entry.
192 * If not, it calls this function to determine the per-ring length field
193 * encoding for the command (i.e. certain opcode ranges use certain bits
194 * to encode the command length in the header).
195 */
196 u32 (*get_cmd_length_mask)(u32 cmd_header);
197 };
198
199 static inline bool
200 intel_ring_initialized(struct intel_ring_buffer *ring)
201 {
202 return ring->obj != NULL;
203 }
204
205 static inline unsigned
206 intel_ring_flag(struct intel_ring_buffer *ring)
207 {
208 return 1 << ring->id;
209 }
210
211 static inline u32
212 intel_ring_sync_index(struct intel_ring_buffer *ring,
213 struct intel_ring_buffer *other)
214 {
215 int idx;
216
217 /*
218 * cs -> 0 = vcs, 1 = bcs
219 * vcs -> 0 = bcs, 1 = cs,
220 * bcs -> 0 = cs, 1 = vcs.
221 */
222
223 idx = (other - ring) - 1;
224 if (idx < 0)
225 idx += I915_NUM_RINGS;
226
227 return idx;
228 }
229
230 static inline u32
231 intel_read_status_page(struct intel_ring_buffer *ring,
232 int reg)
233 {
234 /* Ensure that the compiler doesn't optimize away the load. */
235 barrier();
236 return ring->status_page.page_addr[reg];
237 }
238
239 static inline void
240 intel_write_status_page(struct intel_ring_buffer *ring,
241 int reg, u32 value)
242 {
243 ring->status_page.page_addr[reg] = value;
244 }
245
246 /**
247 * Reads a dword out of the status page, which is written to from the command
248 * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
249 * MI_STORE_DATA_IMM.
250 *
251 * The following dwords have a reserved meaning:
252 * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
253 * 0x04: ring 0 head pointer
254 * 0x05: ring 1 head pointer (915-class)
255 * 0x06: ring 2 head pointer (915-class)
256 * 0x10-0x1b: Context status DWords (GM45)
257 * 0x1f: Last written status offset. (GM45)
258 *
259 * The area from dword 0x20 to 0x3ff is available for driver usage.
260 */
261 #define I915_GEM_HWS_INDEX 0x20
262 #define I915_GEM_HWS_SCRATCH_INDEX 0x30
263 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
264
265 void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
266
267 int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
268 int __must_check intel_ring_cacheline_align(struct intel_ring_buffer *ring);
269 static inline void intel_ring_emit(struct intel_ring_buffer *ring,
270 u32 data)
271 {
272 iowrite32(data, ring->virtual_start + ring->tail);
273 ring->tail += 4;
274 }
275 static inline void intel_ring_advance(struct intel_ring_buffer *ring)
276 {
277 ring->tail &= ring->size - 1;
278 }
279 void __intel_ring_advance(struct intel_ring_buffer *ring);
280
281 int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
282 void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
283 int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
284 int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
285
286 int intel_init_render_ring_buffer(struct drm_device *dev);
287 int intel_init_bsd_ring_buffer(struct drm_device *dev);
288 int intel_init_blt_ring_buffer(struct drm_device *dev);
289 int intel_init_vebox_ring_buffer(struct drm_device *dev);
290
291 u64 intel_ring_get_active_head(struct intel_ring_buffer *ring);
292 void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
293
294 static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
295 {
296 return ring->tail;
297 }
298
299 static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
300 {
301 BUG_ON(ring->outstanding_lazy_seqno == 0);
302 return ring->outstanding_lazy_seqno;
303 }
304
305 static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
306 {
307 if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
308 ring->trace_irq_seqno = seqno;
309 }
310
311 /* DRI warts */
312 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
313
314 #endif /* _INTEL_RINGBUFFER_H_ */
This page took 0.093601 seconds and 6 git commands to generate.