]> Pileus Git - ~andy/linux/blob - drivers/gpu/drm/i915/intel_ringbuffer.h
drm/i915: Track when we dirty the scanout with render commands
[~andy/linux] / 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_NOPID(ring) I915_READ(RING_NOPID((ring)->mmio_base))
37 #define I915_READ_SYNC_0(ring) I915_READ(RING_SYNC_0((ring)->mmio_base))
38 #define I915_READ_SYNC_1(ring) I915_READ(RING_SYNC_1((ring)->mmio_base))
39
40 struct intel_ring_hangcheck {
41         u32 seqno;
42         u32 acthd;
43         int score;
44 };
45
46 struct  intel_ring_buffer {
47         const char      *name;
48         enum intel_ring_id {
49                 RCS = 0x0,
50                 VCS,
51                 BCS,
52                 VECS,
53         } id;
54 #define I915_NUM_RINGS 4
55         u32             mmio_base;
56         void            __iomem *virtual_start;
57         struct          drm_device *dev;
58         struct          drm_i915_gem_object *obj;
59
60         u32             head;
61         u32             tail;
62         int             space;
63         int             size;
64         int             effective_size;
65         struct intel_hw_status_page status_page;
66
67         /** We track the position of the requests in the ring buffer, and
68          * when each is retired we increment last_retired_head as the GPU
69          * must have finished processing the request and so we know we
70          * can advance the ringbuffer up to that position.
71          *
72          * last_retired_head is set to -1 after the value is consumed so
73          * we can detect new retirements.
74          */
75         u32             last_retired_head;
76
77         struct {
78                 u32     gt; /*  protected by dev_priv->irq_lock */
79                 u32     pm; /*  protected by dev_priv->rps.lock (sucks) */
80         } irq_refcount;
81         u32             irq_enable_mask;        /* bitmask to enable ring interrupt */
82         u32             trace_irq_seqno;
83         u32             sync_seqno[I915_NUM_RINGS-1];
84         bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
85         void            (*irq_put)(struct intel_ring_buffer *ring);
86
87         int             (*init)(struct intel_ring_buffer *ring);
88
89         void            (*write_tail)(struct intel_ring_buffer *ring,
90                                       u32 value);
91         int __must_check (*flush)(struct intel_ring_buffer *ring,
92                                   u32   invalidate_domains,
93                                   u32   flush_domains);
94         int             (*add_request)(struct intel_ring_buffer *ring);
95         /* Some chipsets are not quite as coherent as advertised and need
96          * an expensive kick to force a true read of the up-to-date seqno.
97          * However, the up-to-date seqno is not always required and the last
98          * seen value is good enough. Note that the seqno will always be
99          * monotonic, even if not coherent.
100          */
101         u32             (*get_seqno)(struct intel_ring_buffer *ring,
102                                      bool lazy_coherency);
103         void            (*set_seqno)(struct intel_ring_buffer *ring,
104                                      u32 seqno);
105         int             (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
106                                                u32 offset, u32 length,
107                                                unsigned flags);
108 #define I915_DISPATCH_SECURE 0x1
109 #define I915_DISPATCH_PINNED 0x2
110         void            (*cleanup)(struct intel_ring_buffer *ring);
111         int             (*sync_to)(struct intel_ring_buffer *ring,
112                                    struct intel_ring_buffer *to,
113                                    u32 seqno);
114
115         /* our mbox written by others */
116         u32             semaphore_register[I915_NUM_RINGS];
117         /* mboxes this ring signals to */
118         u32             signal_mbox[I915_NUM_RINGS];
119
120         /**
121          * List of objects currently involved in rendering from the
122          * ringbuffer.
123          *
124          * Includes buffers having the contents of their GPU caches
125          * flushed, not necessarily primitives.  last_rendering_seqno
126          * represents when the rendering involved will be completed.
127          *
128          * A reference is held on the buffer while on this list.
129          */
130         struct list_head active_list;
131
132         /**
133          * List of breadcrumbs associated with GPU requests currently
134          * outstanding.
135          */
136         struct list_head request_list;
137
138         /**
139          * Do we have some not yet emitted requests outstanding?
140          */
141         u32 outstanding_lazy_request;
142         bool gpu_caches_dirty;
143         bool fbc_dirty;
144
145         wait_queue_head_t irq_queue;
146
147         /**
148          * Do an explicit TLB flush before MI_SET_CONTEXT
149          */
150         bool itlb_before_ctx_switch;
151         struct i915_hw_context *default_context;
152         struct i915_hw_context *last_context;
153
154         struct intel_ring_hangcheck hangcheck;
155
156         void *private;
157 };
158
159 static inline bool
160 intel_ring_initialized(struct intel_ring_buffer *ring)
161 {
162         return ring->obj != NULL;
163 }
164
165 static inline unsigned
166 intel_ring_flag(struct intel_ring_buffer *ring)
167 {
168         return 1 << ring->id;
169 }
170
171 static inline u32
172 intel_ring_sync_index(struct intel_ring_buffer *ring,
173                       struct intel_ring_buffer *other)
174 {
175         int idx;
176
177         /*
178          * cs -> 0 = vcs, 1 = bcs
179          * vcs -> 0 = bcs, 1 = cs,
180          * bcs -> 0 = cs, 1 = vcs.
181          */
182
183         idx = (other - ring) - 1;
184         if (idx < 0)
185                 idx += I915_NUM_RINGS;
186
187         return idx;
188 }
189
190 static inline u32
191 intel_read_status_page(struct intel_ring_buffer *ring,
192                        int reg)
193 {
194         /* Ensure that the compiler doesn't optimize away the load. */
195         barrier();
196         return ring->status_page.page_addr[reg];
197 }
198
199 static inline void
200 intel_write_status_page(struct intel_ring_buffer *ring,
201                         int reg, u32 value)
202 {
203         ring->status_page.page_addr[reg] = value;
204 }
205
206 /**
207  * Reads a dword out of the status page, which is written to from the command
208  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
209  * MI_STORE_DATA_IMM.
210  *
211  * The following dwords have a reserved meaning:
212  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
213  * 0x04: ring 0 head pointer
214  * 0x05: ring 1 head pointer (915-class)
215  * 0x06: ring 2 head pointer (915-class)
216  * 0x10-0x1b: Context status DWords (GM45)
217  * 0x1f: Last written status offset. (GM45)
218  *
219  * The area from dword 0x20 to 0x3ff is available for driver usage.
220  */
221 #define I915_GEM_HWS_INDEX              0x20
222 #define I915_GEM_HWS_SCRATCH_INDEX      0x30
223 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
224
225 void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
226
227 int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
228 static inline void intel_ring_emit(struct intel_ring_buffer *ring,
229                                    u32 data)
230 {
231         iowrite32(data, ring->virtual_start + ring->tail);
232         ring->tail += 4;
233 }
234 void intel_ring_advance(struct intel_ring_buffer *ring);
235 int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
236 void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
237 int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
238 int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
239
240 int intel_init_render_ring_buffer(struct drm_device *dev);
241 int intel_init_bsd_ring_buffer(struct drm_device *dev);
242 int intel_init_blt_ring_buffer(struct drm_device *dev);
243 int intel_init_vebox_ring_buffer(struct drm_device *dev);
244
245 u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
246 void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
247
248 static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
249 {
250         return ring->tail;
251 }
252
253 static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
254 {
255         BUG_ON(ring->outstanding_lazy_request == 0);
256         return ring->outstanding_lazy_request;
257 }
258
259 static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
260 {
261         if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
262                 ring->trace_irq_seqno = seqno;
263 }
264
265 /* DRI warts */
266 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
267
268 #endif /* _INTEL_RINGBUFFER_H_ */