drm/i915/bdw: Implement Wa4x4STCOptimizationDisable:bdw
[deliverable/linux.git] / drivers / gpu / drm / i915 / intel_ringbuffer.h
CommitLineData
8187a2b7
ZN
1#ifndef _INTEL_RINGBUFFER_H_
2#define _INTEL_RINGBUFFER_H_
3
633cf8f5
VS
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
8187a2b7 15struct intel_hw_status_page {
4225d0f2 16 u32 *page_addr;
8187a2b7 17 unsigned int gfx_addr;
05394f39 18 struct drm_i915_gem_object *obj;
8187a2b7
ZN
19};
20
b7287d80
BW
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)
cae5852d 23
b7287d80
BW
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)
cae5852d 26
b7287d80
BW
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)
cae5852d 29
b7287d80
BW
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)
cae5852d 32
b7287d80
BW
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)
870e86dd 35
e9fea574
NKK
36#define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base))
37
f2f4d82f 38enum intel_ring_hangcheck_action {
da661464 39 HANGCHECK_IDLE = 0,
f2f4d82f
JN
40 HANGCHECK_WAIT,
41 HANGCHECK_ACTIVE,
42 HANGCHECK_KICK,
43 HANGCHECK_HUNG,
44};
ad8beaea 45
b6b0fac0
MK
46#define HANGCHECK_SCORE_RING_HUNG 31
47
92cab734 48struct intel_ring_hangcheck {
6274f212 49 bool deadlock;
92cab734 50 u32 seqno;
05407ff8
MK
51 u32 acthd;
52 int score;
ad8beaea 53 enum intel_ring_hangcheck_action action;
92cab734
MK
54};
55
8187a2b7
ZN
56struct intel_ring_buffer {
57 const char *name;
9220434a 58 enum intel_ring_id {
96154f2f
DV
59 RCS = 0x0,
60 VCS,
61 BCS,
4a3dd19d 62 VECS,
9220434a 63 } id;
4a3dd19d 64#define I915_NUM_RINGS 4
333e9fe9 65 u32 mmio_base;
311bd68e 66 void __iomem *virtual_start;
8187a2b7 67 struct drm_device *dev;
05394f39 68 struct drm_i915_gem_object *obj;
8187a2b7 69
8c0a6bfe
CW
70 u32 head;
71 u32 tail;
780f0ca3 72 int space;
c2c347a9 73 int size;
55249baa 74 int effective_size;
8187a2b7
ZN
75 struct intel_hw_status_page status_page;
76
a71d8d94
CW
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
c7113cc3 87 unsigned irq_refcount; /* protected by dev_priv->irq_lock */
6a848ccb 88 u32 irq_enable_mask; /* bitmask to enable ring interrupt */
db53a302 89 u32 trace_irq_seqno;
1ec14ad3 90 u32 sync_seqno[I915_NUM_RINGS-1];
b13c2b96 91 bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
1ec14ad3 92 void (*irq_put)(struct intel_ring_buffer *ring);
8187a2b7 93
78501eac 94 int (*init)(struct intel_ring_buffer *ring);
8187a2b7 95
78501eac 96 void (*write_tail)(struct intel_ring_buffer *ring,
297b0c5b 97 u32 value);
b72f3acb
CW
98 int __must_check (*flush)(struct intel_ring_buffer *ring,
99 u32 invalidate_domains,
100 u32 flush_domains);
9d773091 101 int (*add_request)(struct intel_ring_buffer *ring);
b2eadbc8
CW
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);
b70ec5bf
MK
110 void (*set_seqno)(struct intel_ring_buffer *ring,
111 u32 seqno);
78501eac 112 int (*dispatch_execbuffer)(struct intel_ring_buffer *ring,
d7d4eedd
CW
113 u32 offset, u32 length,
114 unsigned flags);
115#define I915_DISPATCH_SECURE 0x1
b45305fc 116#define I915_DISPATCH_PINNED 0x2
8d19215b 117 void (*cleanup)(struct intel_ring_buffer *ring);
c8c99b0f
BW
118 int (*sync_to)(struct intel_ring_buffer *ring,
119 struct intel_ring_buffer *to,
120 u32 seqno);
ad776f8b 121
5586181f
BW
122 /* our mbox written by others */
123 u32 semaphore_register[I915_NUM_RINGS];
ad776f8b
BW
124 /* mboxes this ring signals to */
125 u32 signal_mbox[I915_NUM_RINGS];
126
8187a2b7
ZN
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
a56ba56c
CW
145 /**
146 * Do we have some not yet emitted requests outstanding?
147 */
3c0e234c 148 struct drm_i915_gem_request *preallocated_lazy_request;
1823521d 149 u32 outstanding_lazy_seqno;
cc889e0f 150 bool gpu_caches_dirty;
c65355bb 151 bool fbc_dirty;
a56ba56c 152
8187a2b7 153 wait_queue_head_t irq_queue;
8d19215b 154
12b0286f
BW
155 /**
156 * Do an explicit TLB flush before MI_SET_CONTEXT
157 */
158 bool itlb_before_ctx_switch;
40521054 159 struct i915_hw_context *default_context;
112522f6 160 struct i915_hw_context *last_context;
40521054 161
92cab734
MK
162 struct intel_ring_hangcheck hangcheck;
163
0d1aacac
CW
164 struct {
165 struct drm_i915_gem_object *obj;
166 u32 gtt_offset;
167 volatile u32 *cpu_page;
168 } scratch;
351e3db2
BV
169
170 /*
171 * Tables of commands the command parser needs to know about
172 * for this ring.
173 */
174 const struct drm_i915_cmd_table *cmd_tables;
175 int cmd_table_count;
176
177 /*
178 * Table of registers allowed in commands that read/write registers.
179 */
180 const u32 *reg_table;
181 int reg_count;
182
183 /*
184 * Table of registers allowed in commands that read/write registers, but
185 * only from the DRM master.
186 */
187 const u32 *master_reg_table;
188 int master_reg_count;
189
190 /*
191 * Returns the bitmask for the length field of the specified command.
192 * Return 0 for an unrecognized/invalid command.
193 *
194 * If the command parser finds an entry for a command in the ring's
195 * cmd_tables, it gets the command's length based on the table entry.
196 * If not, it calls this function to determine the per-ring length field
197 * encoding for the command (i.e. certain opcode ranges use certain bits
198 * to encode the command length in the header).
199 */
200 u32 (*get_cmd_length_mask)(u32 cmd_header);
8187a2b7
ZN
201};
202
b4519513
CW
203static inline bool
204intel_ring_initialized(struct intel_ring_buffer *ring)
205{
206 return ring->obj != NULL;
207}
208
96154f2f
DV
209static inline unsigned
210intel_ring_flag(struct intel_ring_buffer *ring)
211{
212 return 1 << ring->id;
213}
214
1ec14ad3
CW
215static inline u32
216intel_ring_sync_index(struct intel_ring_buffer *ring,
217 struct intel_ring_buffer *other)
218{
219 int idx;
220
221 /*
222 * cs -> 0 = vcs, 1 = bcs
223 * vcs -> 0 = bcs, 1 = cs,
224 * bcs -> 0 = cs, 1 = vcs.
225 */
226
227 idx = (other - ring) - 1;
228 if (idx < 0)
229 idx += I915_NUM_RINGS;
230
231 return idx;
232}
233
8187a2b7
ZN
234static inline u32
235intel_read_status_page(struct intel_ring_buffer *ring,
78501eac 236 int reg)
8187a2b7 237{
4225d0f2
DV
238 /* Ensure that the compiler doesn't optimize away the load. */
239 barrier();
240 return ring->status_page.page_addr[reg];
8187a2b7
ZN
241}
242
b70ec5bf
MK
243static inline void
244intel_write_status_page(struct intel_ring_buffer *ring,
245 int reg, u32 value)
246{
247 ring->status_page.page_addr[reg] = value;
248}
249
311bd68e
CW
250/**
251 * Reads a dword out of the status page, which is written to from the command
252 * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
253 * MI_STORE_DATA_IMM.
254 *
255 * The following dwords have a reserved meaning:
256 * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
257 * 0x04: ring 0 head pointer
258 * 0x05: ring 1 head pointer (915-class)
259 * 0x06: ring 2 head pointer (915-class)
260 * 0x10-0x1b: Context status DWords (GM45)
261 * 0x1f: Last written status offset. (GM45)
262 *
263 * The area from dword 0x20 to 0x3ff is available for driver usage.
264 */
311bd68e 265#define I915_GEM_HWS_INDEX 0x20
9a289771
JB
266#define I915_GEM_HWS_SCRATCH_INDEX 0x30
267#define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
311bd68e 268
78501eac 269void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
96f298aa 270
e1f99ce6 271int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
753b1ad4 272int __must_check intel_ring_cacheline_align(struct intel_ring_buffer *ring);
78501eac
CW
273static inline void intel_ring_emit(struct intel_ring_buffer *ring,
274 u32 data)
e898cd22 275{
78501eac 276 iowrite32(data, ring->virtual_start + ring->tail);
e898cd22
CW
277 ring->tail += 4;
278}
09246732
CW
279static inline void intel_ring_advance(struct intel_ring_buffer *ring)
280{
281 ring->tail &= ring->size - 1;
282}
283void __intel_ring_advance(struct intel_ring_buffer *ring);
284
3e960501 285int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
f7e98ad4 286void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
a7b9761d
CW
287int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
288int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
8187a2b7 289
5c1143bb
XH
290int intel_init_render_ring_buffer(struct drm_device *dev);
291int intel_init_bsd_ring_buffer(struct drm_device *dev);
549f7365 292int intel_init_blt_ring_buffer(struct drm_device *dev);
9a8a2213 293int intel_init_vebox_ring_buffer(struct drm_device *dev);
8187a2b7 294
78501eac
CW
295u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
296void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
79f321b7 297
a71d8d94
CW
298static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
299{
300 return ring->tail;
301}
302
9d773091
CW
303static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
304{
1823521d
CW
305 BUG_ON(ring->outstanding_lazy_seqno == 0);
306 return ring->outstanding_lazy_seqno;
9d773091
CW
307}
308
db53a302
CW
309static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
310{
311 if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
312 ring->trace_irq_seqno = seqno;
313}
314
e8616b6c
CW
315/* DRI warts */
316int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
317
8187a2b7 318#endif /* _INTEL_RINGBUFFER_H_ */
This page took 0.224922 seconds and 5 git commands to generate.