Rev 5354 | Rev 6937 | Go to most recent revision | Only display areas with differences | Regard whitespace | Details | Blame | Last modification | View Log | RSS feed
Rev 5354 | Rev 6084 | ||
---|---|---|---|
1 | #ifndef _INTEL_RINGBUFFER_H_ |
1 | #ifndef _INTEL_RINGBUFFER_H_ |
2 | #define _INTEL_RINGBUFFER_H_ |
2 | #define _INTEL_RINGBUFFER_H_ |
3 | 3 | ||
4 | #include |
4 | #include |
- | 5 | #include "i915_gem_batch_pool.h" |
|
5 | 6 | ||
6 | #define I915_CMD_HASH_ORDER 9 |
7 | #define I915_CMD_HASH_ORDER 9 |
7 | 8 | ||
8 | /* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill, |
9 | /* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill, |
9 | * but keeps the logic simple. Indeed, the whole purpose of this macro is just |
10 | * but keeps the logic simple. Indeed, the whole purpose of this macro is just |
10 | * to give some inclination as to some of the magic values used in the various |
11 | * to give some inclination as to some of the magic values used in the various |
11 | * workarounds! |
12 | * workarounds! |
12 | */ |
13 | */ |
13 | #define CACHELINE_BYTES 64 |
14 | #define CACHELINE_BYTES 64 |
- | 15 | #define CACHELINE_DWORDS (CACHELINE_BYTES / sizeof(uint32_t)) |
|
14 | 16 | ||
15 | /* |
17 | /* |
16 | * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use" |
18 | * Gen2 BSpec "1. Programming Environment" / 1.4.4.6 "Ring Buffer Use" |
17 | * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use" |
19 | * Gen3 BSpec "vol1c Memory Interface Functions" / 2.3.4.5 "Ring Buffer Use" |
18 | * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use" |
20 | * Gen4+ BSpec "vol1c Memory Interface and Command Stream" / 5.3.4.5 "Ring Buffer Use" |
19 | * |
21 | * |
20 | * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same |
22 | * "If the Ring Buffer Head Pointer and the Tail Pointer are on the same |
21 | * cacheline, the Head Pointer must not be greater than the Tail |
23 | * cacheline, the Head Pointer must not be greater than the Tail |
22 | * Pointer." |
24 | * Pointer." |
23 | */ |
25 | */ |
24 | #define I915_RING_FREE_SPACE 64 |
26 | #define I915_RING_FREE_SPACE 64 |
25 | 27 | ||
26 | struct intel_hw_status_page { |
28 | struct intel_hw_status_page { |
27 | u32 *page_addr; |
29 | u32 *page_addr; |
28 | unsigned int gfx_addr; |
30 | unsigned int gfx_addr; |
29 | struct drm_i915_gem_object *obj; |
31 | struct drm_i915_gem_object *obj; |
30 | }; |
32 | }; |
31 | 33 | ||
32 | #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base)) |
34 | #define I915_READ_TAIL(ring) I915_READ(RING_TAIL((ring)->mmio_base)) |
33 | #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val) |
35 | #define I915_WRITE_TAIL(ring, val) I915_WRITE(RING_TAIL((ring)->mmio_base), val) |
34 | 36 | ||
35 | #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base)) |
37 | #define I915_READ_START(ring) I915_READ(RING_START((ring)->mmio_base)) |
36 | #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val) |
38 | #define I915_WRITE_START(ring, val) I915_WRITE(RING_START((ring)->mmio_base), val) |
37 | 39 | ||
38 | #define I915_READ_HEAD(ring) I915_READ(RING_HEAD((ring)->mmio_base)) |
40 | #define I915_READ_HEAD(ring) I915_READ(RING_HEAD((ring)->mmio_base)) |
39 | #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val) |
41 | #define I915_WRITE_HEAD(ring, val) I915_WRITE(RING_HEAD((ring)->mmio_base), val) |
40 | 42 | ||
41 | #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base)) |
43 | #define I915_READ_CTL(ring) I915_READ(RING_CTL((ring)->mmio_base)) |
42 | #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val) |
44 | #define I915_WRITE_CTL(ring, val) I915_WRITE(RING_CTL((ring)->mmio_base), val) |
43 | 45 | ||
44 | #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base)) |
46 | #define I915_READ_IMR(ring) I915_READ(RING_IMR((ring)->mmio_base)) |
45 | #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val) |
47 | #define I915_WRITE_IMR(ring, val) I915_WRITE(RING_IMR((ring)->mmio_base), val) |
46 | 48 | ||
47 | #define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base)) |
49 | #define I915_READ_MODE(ring) I915_READ(RING_MI_MODE((ring)->mmio_base)) |
48 | #define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val) |
50 | #define I915_WRITE_MODE(ring, val) I915_WRITE(RING_MI_MODE((ring)->mmio_base), val) |
49 | 51 | ||
50 | /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to |
52 | /* seqno size is actually only a uint32, but since we plan to use MI_FLUSH_DW to |
51 | * do the writes, and that must have qw aligned offsets, simply pretend it's 8b. |
53 | * do the writes, and that must have qw aligned offsets, simply pretend it's 8b. |
52 | */ |
54 | */ |
53 | #define i915_semaphore_seqno_size sizeof(uint64_t) |
55 | #define i915_semaphore_seqno_size sizeof(uint64_t) |
54 | #define GEN8_SIGNAL_OFFSET(__ring, to) \ |
56 | #define GEN8_SIGNAL_OFFSET(__ring, to) \ |
55 | (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \ |
57 | (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \ |
56 | ((__ring)->id * I915_NUM_RINGS * i915_semaphore_seqno_size) + \ |
58 | ((__ring)->id * I915_NUM_RINGS * i915_semaphore_seqno_size) + \ |
57 | (i915_semaphore_seqno_size * (to))) |
59 | (i915_semaphore_seqno_size * (to))) |
58 | 60 | ||
59 | #define GEN8_WAIT_OFFSET(__ring, from) \ |
61 | #define GEN8_WAIT_OFFSET(__ring, from) \ |
60 | (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \ |
62 | (i915_gem_obj_ggtt_offset(dev_priv->semaphore_obj) + \ |
61 | ((from) * I915_NUM_RINGS * i915_semaphore_seqno_size) + \ |
63 | ((from) * I915_NUM_RINGS * i915_semaphore_seqno_size) + \ |
62 | (i915_semaphore_seqno_size * (__ring)->id)) |
64 | (i915_semaphore_seqno_size * (__ring)->id)) |
63 | 65 | ||
64 | #define GEN8_RING_SEMAPHORE_INIT do { \ |
66 | #define GEN8_RING_SEMAPHORE_INIT do { \ |
65 | if (!dev_priv->semaphore_obj) { \ |
67 | if (!dev_priv->semaphore_obj) { \ |
66 | break; \ |
68 | break; \ |
67 | } \ |
69 | } \ |
68 | ring->semaphore.signal_ggtt[RCS] = GEN8_SIGNAL_OFFSET(ring, RCS); \ |
70 | ring->semaphore.signal_ggtt[RCS] = GEN8_SIGNAL_OFFSET(ring, RCS); \ |
69 | ring->semaphore.signal_ggtt[VCS] = GEN8_SIGNAL_OFFSET(ring, VCS); \ |
71 | ring->semaphore.signal_ggtt[VCS] = GEN8_SIGNAL_OFFSET(ring, VCS); \ |
70 | ring->semaphore.signal_ggtt[BCS] = GEN8_SIGNAL_OFFSET(ring, BCS); \ |
72 | ring->semaphore.signal_ggtt[BCS] = GEN8_SIGNAL_OFFSET(ring, BCS); \ |
71 | ring->semaphore.signal_ggtt[VECS] = GEN8_SIGNAL_OFFSET(ring, VECS); \ |
73 | ring->semaphore.signal_ggtt[VECS] = GEN8_SIGNAL_OFFSET(ring, VECS); \ |
72 | ring->semaphore.signal_ggtt[VCS2] = GEN8_SIGNAL_OFFSET(ring, VCS2); \ |
74 | ring->semaphore.signal_ggtt[VCS2] = GEN8_SIGNAL_OFFSET(ring, VCS2); \ |
73 | ring->semaphore.signal_ggtt[ring->id] = MI_SEMAPHORE_SYNC_INVALID; \ |
75 | ring->semaphore.signal_ggtt[ring->id] = MI_SEMAPHORE_SYNC_INVALID; \ |
74 | } while(0) |
76 | } while(0) |
75 | 77 | ||
76 | enum intel_ring_hangcheck_action { |
78 | enum intel_ring_hangcheck_action { |
77 | HANGCHECK_IDLE = 0, |
79 | HANGCHECK_IDLE = 0, |
78 | HANGCHECK_WAIT, |
80 | HANGCHECK_WAIT, |
79 | HANGCHECK_ACTIVE, |
81 | HANGCHECK_ACTIVE, |
80 | HANGCHECK_ACTIVE_LOOP, |
82 | HANGCHECK_ACTIVE_LOOP, |
81 | HANGCHECK_KICK, |
83 | HANGCHECK_KICK, |
82 | HANGCHECK_HUNG, |
84 | HANGCHECK_HUNG, |
83 | }; |
85 | }; |
84 | 86 | ||
85 | #define HANGCHECK_SCORE_RING_HUNG 31 |
87 | #define HANGCHECK_SCORE_RING_HUNG 31 |
86 | 88 | ||
87 | struct intel_ring_hangcheck { |
89 | struct intel_ring_hangcheck { |
88 | u64 acthd; |
90 | u64 acthd; |
89 | u64 max_acthd; |
91 | u64 max_acthd; |
90 | u32 seqno; |
92 | u32 seqno; |
91 | int score; |
93 | int score; |
92 | enum intel_ring_hangcheck_action action; |
94 | enum intel_ring_hangcheck_action action; |
93 | int deadlock; |
95 | int deadlock; |
94 | }; |
96 | }; |
95 | 97 | ||
96 | struct intel_ringbuffer { |
98 | struct intel_ringbuffer { |
97 | struct drm_i915_gem_object *obj; |
99 | struct drm_i915_gem_object *obj; |
98 | void __iomem *virtual_start; |
100 | void __iomem *virtual_start; |
99 | 101 | ||
100 | struct intel_engine_cs *ring; |
102 | struct intel_engine_cs *ring; |
101 | - | ||
102 | /* |
- | |
103 | * FIXME: This backpointer is an artifact of the history of how the |
- | |
104 | * execlist patches came into being. It will get removed once the basic |
- | |
105 | * code has landed. |
- | |
106 | */ |
- | |
107 | struct intel_context *FIXME_lrc_ctx; |
- | |
108 | 103 | ||
109 | u32 head; |
104 | u32 head; |
110 | u32 tail; |
105 | u32 tail; |
111 | int space; |
106 | int space; |
112 | int size; |
107 | int size; |
113 | int effective_size; |
108 | int effective_size; |
- | 109 | int reserved_size; |
|
- | 110 | int reserved_tail; |
|
- | 111 | bool reserved_in_use; |
|
114 | 112 | ||
115 | /** We track the position of the requests in the ring buffer, and |
113 | /** We track the position of the requests in the ring buffer, and |
116 | * when each is retired we increment last_retired_head as the GPU |
114 | * when each is retired we increment last_retired_head as the GPU |
117 | * must have finished processing the request and so we know we |
115 | * must have finished processing the request and so we know we |
118 | * can advance the ringbuffer up to that position. |
116 | * can advance the ringbuffer up to that position. |
119 | * |
117 | * |
120 | * last_retired_head is set to -1 after the value is consumed so |
118 | * last_retired_head is set to -1 after the value is consumed so |
121 | * we can detect new retirements. |
119 | * we can detect new retirements. |
122 | */ |
120 | */ |
123 | u32 last_retired_head; |
121 | u32 last_retired_head; |
124 | }; |
122 | }; |
- | 123 | ||
- | 124 | struct intel_context; |
|
- | 125 | struct drm_i915_reg_descriptor; |
|
- | 126 | ||
- | 127 | /* |
|
- | 128 | * we use a single page to load ctx workarounds so all of these |
|
- | 129 | * values are referred in terms of dwords |
|
- | 130 | * |
|
- | 131 | * struct i915_wa_ctx_bb: |
|
- | 132 | * offset: specifies batch starting position, also helpful in case |
|
- | 133 | * if we want to have multiple batches at different offsets based on |
|
- | 134 | * some criteria. It is not a requirement at the moment but provides |
|
- | 135 | * an option for future use. |
|
- | 136 | * size: size of the batch in DWORDS |
|
- | 137 | */ |
|
- | 138 | struct i915_ctx_workarounds { |
|
- | 139 | struct i915_wa_ctx_bb { |
|
- | 140 | u32 offset; |
|
- | 141 | u32 size; |
|
- | 142 | } indirect_ctx, per_ctx; |
|
- | 143 | struct drm_i915_gem_object *obj; |
|
- | 144 | }; |
|
125 | 145 | ||
126 | struct intel_engine_cs { |
146 | struct intel_engine_cs { |
127 | const char *name; |
147 | const char *name; |
128 | enum intel_ring_id { |
148 | enum intel_ring_id { |
129 | RCS = 0x0, |
149 | RCS = 0x0, |
130 | VCS, |
150 | VCS, |
131 | BCS, |
151 | BCS, |
132 | VECS, |
152 | VECS, |
133 | VCS2 |
153 | VCS2 |
134 | } id; |
154 | } id; |
135 | #define I915_NUM_RINGS 5 |
155 | #define I915_NUM_RINGS 5 |
136 | #define LAST_USER_RING (VECS + 1) |
156 | #define LAST_USER_RING (VECS + 1) |
137 | u32 mmio_base; |
157 | u32 mmio_base; |
138 | struct drm_device *dev; |
158 | struct drm_device *dev; |
139 | struct intel_ringbuffer *buffer; |
159 | struct intel_ringbuffer *buffer; |
- | 160 | ||
- | 161 | /* |
|
- | 162 | * A pool of objects to use as shadow copies of client batch buffers |
|
- | 163 | * when the command parser is enabled. Prevents the client from |
|
- | 164 | * modifying the batch contents after software parsing. |
|
- | 165 | */ |
|
- | 166 | struct i915_gem_batch_pool batch_pool; |
|
140 | 167 | ||
- | 168 | struct intel_hw_status_page status_page; |
|
141 | struct intel_hw_status_page status_page; |
169 | struct i915_ctx_workarounds wa_ctx; |
142 | 170 | ||
143 | unsigned irq_refcount; /* protected by dev_priv->irq_lock */ |
171 | unsigned irq_refcount; /* protected by dev_priv->irq_lock */ |
144 | u32 irq_enable_mask; /* bitmask to enable ring interrupt */ |
172 | u32 irq_enable_mask; /* bitmask to enable ring interrupt */ |
145 | u32 trace_irq_seqno; |
173 | struct drm_i915_gem_request *trace_irq_req; |
146 | bool __must_check (*irq_get)(struct intel_engine_cs *ring); |
174 | bool __must_check (*irq_get)(struct intel_engine_cs *ring); |
147 | void (*irq_put)(struct intel_engine_cs *ring); |
175 | void (*irq_put)(struct intel_engine_cs *ring); |
148 | 176 | ||
149 | int (*init)(struct intel_engine_cs *ring); |
177 | int (*init_hw)(struct intel_engine_cs *ring); |
150 | - | ||
151 | int (*init_context)(struct intel_engine_cs *ring, |
178 | |
152 | struct intel_context *ctx); |
179 | int (*init_context)(struct drm_i915_gem_request *req); |
153 | 180 | ||
154 | void (*write_tail)(struct intel_engine_cs *ring, |
181 | void (*write_tail)(struct intel_engine_cs *ring, |
155 | u32 value); |
182 | u32 value); |
156 | int __must_check (*flush)(struct intel_engine_cs *ring, |
183 | int __must_check (*flush)(struct drm_i915_gem_request *req, |
157 | u32 invalidate_domains, |
184 | u32 invalidate_domains, |
158 | u32 flush_domains); |
185 | u32 flush_domains); |
159 | int (*add_request)(struct intel_engine_cs *ring); |
186 | int (*add_request)(struct drm_i915_gem_request *req); |
160 | /* Some chipsets are not quite as coherent as advertised and need |
187 | /* Some chipsets are not quite as coherent as advertised and need |
161 | * an expensive kick to force a true read of the up-to-date seqno. |
188 | * an expensive kick to force a true read of the up-to-date seqno. |
162 | * However, the up-to-date seqno is not always required and the last |
189 | * However, the up-to-date seqno is not always required and the last |
163 | * seen value is good enough. Note that the seqno will always be |
190 | * seen value is good enough. Note that the seqno will always be |
164 | * monotonic, even if not coherent. |
191 | * monotonic, even if not coherent. |
165 | */ |
192 | */ |
166 | u32 (*get_seqno)(struct intel_engine_cs *ring, |
193 | u32 (*get_seqno)(struct intel_engine_cs *ring, |
167 | bool lazy_coherency); |
194 | bool lazy_coherency); |
168 | void (*set_seqno)(struct intel_engine_cs *ring, |
195 | void (*set_seqno)(struct intel_engine_cs *ring, |
169 | u32 seqno); |
196 | u32 seqno); |
170 | int (*dispatch_execbuffer)(struct intel_engine_cs *ring, |
197 | int (*dispatch_execbuffer)(struct drm_i915_gem_request *req, |
171 | u64 offset, u32 length, |
198 | u64 offset, u32 length, |
172 | unsigned flags); |
199 | unsigned dispatch_flags); |
173 | #define I915_DISPATCH_SECURE 0x1 |
200 | #define I915_DISPATCH_SECURE 0x1 |
174 | #define I915_DISPATCH_PINNED 0x2 |
201 | #define I915_DISPATCH_PINNED 0x2 |
- | 202 | #define I915_DISPATCH_RS 0x4 |
|
175 | void (*cleanup)(struct intel_engine_cs *ring); |
203 | void (*cleanup)(struct intel_engine_cs *ring); |
176 | 204 | ||
177 | /* GEN8 signal/wait table - never trust comments! |
205 | /* GEN8 signal/wait table - never trust comments! |
178 | * signal to signal to signal to signal to signal to |
206 | * signal to signal to signal to signal to signal to |
179 | * RCS VCS BCS VECS VCS2 |
207 | * RCS VCS BCS VECS VCS2 |
180 | * -------------------------------------------------------------------- |
208 | * -------------------------------------------------------------------- |
181 | * RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) | |
209 | * RCS | NOP (0x00) | VCS (0x08) | BCS (0x10) | VECS (0x18) | VCS2 (0x20) | |
182 | * |------------------------------------------------------------------- |
210 | * |------------------------------------------------------------------- |
183 | * VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) | |
211 | * VCS | RCS (0x28) | NOP (0x30) | BCS (0x38) | VECS (0x40) | VCS2 (0x48) | |
184 | * |------------------------------------------------------------------- |
212 | * |------------------------------------------------------------------- |
185 | * BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) | |
213 | * BCS | RCS (0x50) | VCS (0x58) | NOP (0x60) | VECS (0x68) | VCS2 (0x70) | |
186 | * |------------------------------------------------------------------- |
214 | * |------------------------------------------------------------------- |
187 | * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) | NOP (0x90) | VCS2 (0x98) | |
215 | * VECS | RCS (0x78) | VCS (0x80) | BCS (0x88) | NOP (0x90) | VCS2 (0x98) | |
188 | * |------------------------------------------------------------------- |
216 | * |------------------------------------------------------------------- |
189 | * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP (0xc0) | |
217 | * VCS2 | RCS (0xa0) | VCS (0xa8) | BCS (0xb0) | VECS (0xb8) | NOP (0xc0) | |
190 | * |------------------------------------------------------------------- |
218 | * |------------------------------------------------------------------- |
191 | * |
219 | * |
192 | * Generalization: |
220 | * Generalization: |
193 | * f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id) |
221 | * f(x, y) := (x->id * NUM_RINGS * seqno_size) + (seqno_size * y->id) |
194 | * ie. transpose of g(x, y) |
222 | * ie. transpose of g(x, y) |
195 | * |
223 | * |
196 | * sync from sync from sync from sync from sync from |
224 | * sync from sync from sync from sync from sync from |
197 | * RCS VCS BCS VECS VCS2 |
225 | * RCS VCS BCS VECS VCS2 |
198 | * -------------------------------------------------------------------- |
226 | * -------------------------------------------------------------------- |
199 | * RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) | |
227 | * RCS | NOP (0x00) | VCS (0x28) | BCS (0x50) | VECS (0x78) | VCS2 (0xa0) | |
200 | * |------------------------------------------------------------------- |
228 | * |------------------------------------------------------------------- |
201 | * VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) | |
229 | * VCS | RCS (0x08) | NOP (0x30) | BCS (0x58) | VECS (0x80) | VCS2 (0xa8) | |
202 | * |------------------------------------------------------------------- |
230 | * |------------------------------------------------------------------- |
203 | * BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) | |
231 | * BCS | RCS (0x10) | VCS (0x38) | NOP (0x60) | VECS (0x88) | VCS2 (0xb0) | |
204 | * |------------------------------------------------------------------- |
232 | * |------------------------------------------------------------------- |
205 | * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) | NOP (0x90) | VCS2 (0xb8) | |
233 | * VECS | RCS (0x18) | VCS (0x40) | BCS (0x68) | NOP (0x90) | VCS2 (0xb8) | |
206 | * |------------------------------------------------------------------- |
234 | * |------------------------------------------------------------------- |
207 | * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) | NOP (0xc0) | |
235 | * VCS2 | RCS (0x20) | VCS (0x48) | BCS (0x70) | VECS (0x98) | NOP (0xc0) | |
208 | * |------------------------------------------------------------------- |
236 | * |------------------------------------------------------------------- |
209 | * |
237 | * |
210 | * Generalization: |
238 | * Generalization: |
211 | * g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id) |
239 | * g(x, y) := (y->id * NUM_RINGS * seqno_size) + (seqno_size * x->id) |
212 | * ie. transpose of f(x, y) |
240 | * ie. transpose of f(x, y) |
213 | */ |
241 | */ |
214 | struct { |
242 | struct { |
215 | u32 sync_seqno[I915_NUM_RINGS-1]; |
243 | u32 sync_seqno[I915_NUM_RINGS-1]; |
216 | 244 | ||
217 | union { |
245 | union { |
218 | struct { |
246 | struct { |
219 | /* our mbox written by others */ |
247 | /* our mbox written by others */ |
220 | u32 wait[I915_NUM_RINGS]; |
248 | u32 wait[I915_NUM_RINGS]; |
221 | /* mboxes this ring signals to */ |
249 | /* mboxes this ring signals to */ |
222 | u32 signal[I915_NUM_RINGS]; |
250 | u32 signal[I915_NUM_RINGS]; |
223 | } mbox; |
251 | } mbox; |
224 | u64 signal_ggtt[I915_NUM_RINGS]; |
252 | u64 signal_ggtt[I915_NUM_RINGS]; |
225 | }; |
253 | }; |
226 | 254 | ||
227 | /* AKA wait() */ |
255 | /* AKA wait() */ |
228 | int (*sync_to)(struct intel_engine_cs *ring, |
256 | int (*sync_to)(struct drm_i915_gem_request *to_req, |
229 | struct intel_engine_cs *to, |
257 | struct intel_engine_cs *from, |
230 | u32 seqno); |
258 | u32 seqno); |
231 | int (*signal)(struct intel_engine_cs *signaller, |
259 | int (*signal)(struct drm_i915_gem_request *signaller_req, |
232 | /* num_dwords needed by caller */ |
260 | /* num_dwords needed by caller */ |
233 | unsigned int num_dwords); |
261 | unsigned int num_dwords); |
234 | } semaphore; |
262 | } semaphore; |
235 | 263 | ||
236 | /* Execlists */ |
264 | /* Execlists */ |
237 | spinlock_t execlist_lock; |
265 | spinlock_t execlist_lock; |
238 | struct list_head execlist_queue; |
266 | struct list_head execlist_queue; |
239 | struct list_head execlist_retired_req_list; |
267 | struct list_head execlist_retired_req_list; |
240 | u8 next_context_status_buffer; |
268 | u8 next_context_status_buffer; |
241 | u32 irq_keep_mask; /* bitmask for interrupts that should not be masked */ |
269 | u32 irq_keep_mask; /* bitmask for interrupts that should not be masked */ |
242 | int (*emit_request)(struct intel_ringbuffer *ringbuf); |
270 | int (*emit_request)(struct drm_i915_gem_request *request); |
243 | int (*emit_flush)(struct intel_ringbuffer *ringbuf, |
271 | int (*emit_flush)(struct drm_i915_gem_request *request, |
244 | u32 invalidate_domains, |
272 | u32 invalidate_domains, |
245 | u32 flush_domains); |
273 | u32 flush_domains); |
246 | int (*emit_bb_start)(struct intel_ringbuffer *ringbuf, |
274 | int (*emit_bb_start)(struct drm_i915_gem_request *req, |
247 | u64 offset, unsigned flags); |
275 | u64 offset, unsigned dispatch_flags); |
248 | 276 | ||
249 | /** |
277 | /** |
250 | * List of objects currently involved in rendering from the |
278 | * List of objects currently involved in rendering from the |
251 | * ringbuffer. |
279 | * ringbuffer. |
252 | * |
280 | * |
253 | * Includes buffers having the contents of their GPU caches |
281 | * Includes buffers having the contents of their GPU caches |
254 | * flushed, not necessarily primitives. last_rendering_seqno |
282 | * flushed, not necessarily primitives. last_read_req |
255 | * represents when the rendering involved will be completed. |
283 | * represents when the rendering involved will be completed. |
256 | * |
284 | * |
257 | * A reference is held on the buffer while on this list. |
285 | * A reference is held on the buffer while on this list. |
258 | */ |
286 | */ |
259 | struct list_head active_list; |
287 | struct list_head active_list; |
260 | 288 | ||
261 | /** |
289 | /** |
262 | * List of breadcrumbs associated with GPU requests currently |
290 | * List of breadcrumbs associated with GPU requests currently |
263 | * outstanding. |
291 | * outstanding. |
264 | */ |
292 | */ |
265 | struct list_head request_list; |
293 | struct list_head request_list; |
266 | 294 | ||
267 | /** |
295 | /** |
268 | * Do we have some not yet emitted requests outstanding? |
296 | * Seqno of request most recently submitted to request_list. |
- | 297 | * Used exclusively by hang checker to avoid grabbing lock while |
|
- | 298 | * inspecting request list. |
|
269 | */ |
299 | */ |
270 | struct drm_i915_gem_request *preallocated_lazy_request; |
- | |
271 | u32 outstanding_lazy_seqno; |
300 | u32 last_submitted_seqno; |
- | 301 | ||
272 | bool gpu_caches_dirty; |
302 | bool gpu_caches_dirty; |
273 | bool fbc_dirty; |
- | |
274 | 303 | ||
275 | wait_queue_head_t irq_queue; |
304 | wait_queue_head_t irq_queue; |
276 | 305 | ||
277 | struct intel_context *default_context; |
306 | struct intel_context *default_context; |
278 | struct intel_context *last_context; |
307 | struct intel_context *last_context; |
279 | 308 | ||
280 | struct intel_ring_hangcheck hangcheck; |
309 | struct intel_ring_hangcheck hangcheck; |
281 | 310 | ||
282 | struct { |
311 | struct { |
283 | struct drm_i915_gem_object *obj; |
312 | struct drm_i915_gem_object *obj; |
284 | u32 gtt_offset; |
313 | u32 gtt_offset; |
285 | volatile u32 *cpu_page; |
314 | volatile u32 *cpu_page; |
286 | } scratch; |
315 | } scratch; |
287 | 316 | ||
288 | bool needs_cmd_parser; |
317 | bool needs_cmd_parser; |
289 | 318 | ||
290 | /* |
319 | /* |
291 | * Table of commands the command parser needs to know about |
320 | * Table of commands the command parser needs to know about |
292 | * for this ring. |
321 | * for this ring. |
293 | */ |
322 | */ |
294 | DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER); |
323 | DECLARE_HASHTABLE(cmd_hash, I915_CMD_HASH_ORDER); |
295 | 324 | ||
296 | /* |
325 | /* |
297 | * Table of registers allowed in commands that read/write registers. |
326 | * Table of registers allowed in commands that read/write registers. |
298 | */ |
327 | */ |
299 | const u32 *reg_table; |
328 | const struct drm_i915_reg_descriptor *reg_table; |
300 | int reg_count; |
329 | int reg_count; |
301 | 330 | ||
302 | /* |
331 | /* |
303 | * Table of registers allowed in commands that read/write registers, but |
332 | * Table of registers allowed in commands that read/write registers, but |
304 | * only from the DRM master. |
333 | * only from the DRM master. |
305 | */ |
334 | */ |
306 | const u32 *master_reg_table; |
335 | const struct drm_i915_reg_descriptor *master_reg_table; |
307 | int master_reg_count; |
336 | int master_reg_count; |
308 | 337 | ||
309 | /* |
338 | /* |
310 | * Returns the bitmask for the length field of the specified command. |
339 | * Returns the bitmask for the length field of the specified command. |
311 | * Return 0 for an unrecognized/invalid command. |
340 | * Return 0 for an unrecognized/invalid command. |
312 | * |
341 | * |
313 | * If the command parser finds an entry for a command in the ring's |
342 | * If the command parser finds an entry for a command in the ring's |
314 | * cmd_tables, it gets the command's length based on the table entry. |
343 | * cmd_tables, it gets the command's length based on the table entry. |
315 | * If not, it calls this function to determine the per-ring length field |
344 | * If not, it calls this function to determine the per-ring length field |
316 | * encoding for the command (i.e. certain opcode ranges use certain bits |
345 | * encoding for the command (i.e. certain opcode ranges use certain bits |
317 | * to encode the command length in the header). |
346 | * to encode the command length in the header). |
318 | */ |
347 | */ |
319 | u32 (*get_cmd_length_mask)(u32 cmd_header); |
348 | u32 (*get_cmd_length_mask)(u32 cmd_header); |
320 | }; |
349 | }; |
321 | 350 | ||
322 | bool intel_ring_initialized(struct intel_engine_cs *ring); |
351 | bool intel_ring_initialized(struct intel_engine_cs *ring); |
323 | 352 | ||
324 | static inline unsigned |
353 | static inline unsigned |
325 | intel_ring_flag(struct intel_engine_cs *ring) |
354 | intel_ring_flag(struct intel_engine_cs *ring) |
326 | { |
355 | { |
327 | return 1 << ring->id; |
356 | return 1 << ring->id; |
328 | } |
357 | } |
329 | 358 | ||
330 | static inline u32 |
359 | static inline u32 |
331 | intel_ring_sync_index(struct intel_engine_cs *ring, |
360 | intel_ring_sync_index(struct intel_engine_cs *ring, |
332 | struct intel_engine_cs *other) |
361 | struct intel_engine_cs *other) |
333 | { |
362 | { |
334 | int idx; |
363 | int idx; |
335 | 364 | ||
336 | /* |
365 | /* |
337 | * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2; |
366 | * rcs -> 0 = vcs, 1 = bcs, 2 = vecs, 3 = vcs2; |
338 | * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs; |
367 | * vcs -> 0 = bcs, 1 = vecs, 2 = vcs2, 3 = rcs; |
339 | * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs; |
368 | * bcs -> 0 = vecs, 1 = vcs2. 2 = rcs, 3 = vcs; |
340 | * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs; |
369 | * vecs -> 0 = vcs2, 1 = rcs, 2 = vcs, 3 = bcs; |
341 | * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs; |
370 | * vcs2 -> 0 = rcs, 1 = vcs, 2 = bcs, 3 = vecs; |
342 | */ |
371 | */ |
343 | 372 | ||
344 | idx = (other - ring) - 1; |
373 | idx = (other - ring) - 1; |
345 | if (idx < 0) |
374 | if (idx < 0) |
346 | idx += I915_NUM_RINGS; |
375 | idx += I915_NUM_RINGS; |
347 | 376 | ||
348 | return idx; |
377 | return idx; |
349 | } |
378 | } |
- | 379 | ||
- | 380 | static inline void |
|
- | 381 | intel_flush_status_page(struct intel_engine_cs *ring, int reg) |
|
- | 382 | { |
|
- | 383 | drm_clflush_virt_range(&ring->status_page.page_addr[reg], |
|
- | 384 | sizeof(uint32_t)); |
|
- | 385 | } |
|
350 | 386 | ||
351 | static inline u32 |
387 | static inline u32 |
352 | intel_read_status_page(struct intel_engine_cs *ring, |
388 | intel_read_status_page(struct intel_engine_cs *ring, |
353 | int reg) |
389 | int reg) |
354 | { |
390 | { |
355 | /* Ensure that the compiler doesn't optimize away the load. */ |
391 | /* Ensure that the compiler doesn't optimize away the load. */ |
356 | barrier(); |
392 | barrier(); |
357 | return ring->status_page.page_addr[reg]; |
393 | return ring->status_page.page_addr[reg]; |
358 | } |
394 | } |
359 | 395 | ||
360 | static inline void |
396 | static inline void |
361 | intel_write_status_page(struct intel_engine_cs *ring, |
397 | intel_write_status_page(struct intel_engine_cs *ring, |
362 | int reg, u32 value) |
398 | int reg, u32 value) |
363 | { |
399 | { |
364 | ring->status_page.page_addr[reg] = value; |
400 | ring->status_page.page_addr[reg] = value; |
365 | } |
401 | } |
366 | 402 | ||
367 | /** |
403 | /** |
368 | * Reads a dword out of the status page, which is written to from the command |
404 | * Reads a dword out of the status page, which is written to from the command |
369 | * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or |
405 | * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or |
370 | * MI_STORE_DATA_IMM. |
406 | * MI_STORE_DATA_IMM. |
371 | * |
407 | * |
372 | * The following dwords have a reserved meaning: |
408 | * The following dwords have a reserved meaning: |
373 | * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes. |
409 | * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes. |
374 | * 0x04: ring 0 head pointer |
410 | * 0x04: ring 0 head pointer |
375 | * 0x05: ring 1 head pointer (915-class) |
411 | * 0x05: ring 1 head pointer (915-class) |
376 | * 0x06: ring 2 head pointer (915-class) |
412 | * 0x06: ring 2 head pointer (915-class) |
377 | * 0x10-0x1b: Context status DWords (GM45) |
413 | * 0x10-0x1b: Context status DWords (GM45) |
378 | * 0x1f: Last written status offset. (GM45) |
414 | * 0x1f: Last written status offset. (GM45) |
- | 415 | * 0x20-0x2f: Reserved (Gen6+) |
|
379 | * |
416 | * |
380 | * The area from dword 0x20 to 0x3ff is available for driver usage. |
417 | * The area from dword 0x30 to 0x3ff is available for driver usage. |
381 | */ |
418 | */ |
382 | #define I915_GEM_HWS_INDEX 0x20 |
419 | #define I915_GEM_HWS_INDEX 0x30 |
383 | #define I915_GEM_HWS_SCRATCH_INDEX 0x30 |
420 | #define I915_GEM_HWS_SCRATCH_INDEX 0x40 |
384 | #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT) |
421 | #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT) |
- | 422 | ||
385 | 423 | struct intel_ringbuffer * |
|
386 | void intel_unpin_ringbuffer_obj(struct intel_ringbuffer *ringbuf); |
424 | intel_engine_create_ringbuffer(struct intel_engine_cs *engine, int size); |
387 | int intel_pin_and_map_ringbuffer_obj(struct drm_device *dev, |
425 | int intel_pin_and_map_ringbuffer_obj(struct drm_device *dev, |
388 | struct intel_ringbuffer *ringbuf); |
426 | struct intel_ringbuffer *ringbuf); |
389 | void intel_destroy_ringbuffer_obj(struct intel_ringbuffer *ringbuf); |
- | |
390 | int intel_alloc_ringbuffer_obj(struct drm_device *dev, |
427 | void intel_unpin_ringbuffer_obj(struct intel_ringbuffer *ringbuf); |
391 | struct intel_ringbuffer *ringbuf); |
428 | void intel_ringbuffer_free(struct intel_ringbuffer *ring); |
392 | 429 | ||
393 | void intel_stop_ring_buffer(struct intel_engine_cs *ring); |
430 | void intel_stop_ring_buffer(struct intel_engine_cs *ring); |
394 | void intel_cleanup_ring_buffer(struct intel_engine_cs *ring); |
431 | void intel_cleanup_ring_buffer(struct intel_engine_cs *ring); |
- | 432 | ||
- | 433 | int intel_ring_alloc_request_extras(struct drm_i915_gem_request *request); |
|
395 | 434 | ||
396 | int __must_check intel_ring_begin(struct intel_engine_cs *ring, int n); |
435 | int __must_check intel_ring_begin(struct drm_i915_gem_request *req, int n); |
397 | int __must_check intel_ring_cacheline_align(struct intel_engine_cs *ring); |
436 | int __must_check intel_ring_cacheline_align(struct drm_i915_gem_request *req); |
398 | static inline void intel_ring_emit(struct intel_engine_cs *ring, |
437 | static inline void intel_ring_emit(struct intel_engine_cs *ring, |
399 | u32 data) |
438 | u32 data) |
400 | { |
439 | { |
401 | struct intel_ringbuffer *ringbuf = ring->buffer; |
440 | struct intel_ringbuffer *ringbuf = ring->buffer; |
402 | iowrite32(data, ringbuf->virtual_start + ringbuf->tail); |
441 | iowrite32(data, ringbuf->virtual_start + ringbuf->tail); |
403 | ringbuf->tail += 4; |
442 | ringbuf->tail += 4; |
404 | } |
443 | } |
405 | static inline void intel_ring_advance(struct intel_engine_cs *ring) |
444 | static inline void intel_ring_advance(struct intel_engine_cs *ring) |
406 | { |
445 | { |
407 | struct intel_ringbuffer *ringbuf = ring->buffer; |
446 | struct intel_ringbuffer *ringbuf = ring->buffer; |
408 | ringbuf->tail &= ringbuf->size - 1; |
447 | ringbuf->tail &= ringbuf->size - 1; |
409 | } |
448 | } |
410 | int __intel_ring_space(int head, int tail, int size); |
449 | int __intel_ring_space(int head, int tail, int size); |
- | 450 | void intel_ring_update_space(struct intel_ringbuffer *ringbuf); |
|
411 | int intel_ring_space(struct intel_ringbuffer *ringbuf); |
451 | int intel_ring_space(struct intel_ringbuffer *ringbuf); |
412 | bool intel_ring_stopped(struct intel_engine_cs *ring); |
452 | bool intel_ring_stopped(struct intel_engine_cs *ring); |
413 | void __intel_ring_advance(struct intel_engine_cs *ring); |
- | |
414 | 453 | ||
415 | int __must_check intel_ring_idle(struct intel_engine_cs *ring); |
454 | int __must_check intel_ring_idle(struct intel_engine_cs *ring); |
416 | void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno); |
455 | void intel_ring_init_seqno(struct intel_engine_cs *ring, u32 seqno); |
417 | int intel_ring_flush_all_caches(struct intel_engine_cs *ring); |
456 | int intel_ring_flush_all_caches(struct drm_i915_gem_request *req); |
418 | int intel_ring_invalidate_all_caches(struct intel_engine_cs *ring); |
457 | int intel_ring_invalidate_all_caches(struct drm_i915_gem_request *req); |
419 | 458 | ||
420 | void intel_fini_pipe_control(struct intel_engine_cs *ring); |
459 | void intel_fini_pipe_control(struct intel_engine_cs *ring); |
421 | int intel_init_pipe_control(struct intel_engine_cs *ring); |
460 | int intel_init_pipe_control(struct intel_engine_cs *ring); |
422 | 461 | ||
423 | int intel_init_render_ring_buffer(struct drm_device *dev); |
462 | int intel_init_render_ring_buffer(struct drm_device *dev); |
424 | int intel_init_bsd_ring_buffer(struct drm_device *dev); |
463 | int intel_init_bsd_ring_buffer(struct drm_device *dev); |
425 | int intel_init_bsd2_ring_buffer(struct drm_device *dev); |
464 | int intel_init_bsd2_ring_buffer(struct drm_device *dev); |
426 | int intel_init_blt_ring_buffer(struct drm_device *dev); |
465 | int intel_init_blt_ring_buffer(struct drm_device *dev); |
427 | int intel_init_vebox_ring_buffer(struct drm_device *dev); |
466 | int intel_init_vebox_ring_buffer(struct drm_device *dev); |
428 | 467 | ||
429 | u64 intel_ring_get_active_head(struct intel_engine_cs *ring); |
468 | u64 intel_ring_get_active_head(struct intel_engine_cs *ring); |
430 | void intel_ring_setup_status_page(struct intel_engine_cs *ring); |
- | |
431 | 469 | ||
432 | int init_workarounds_ring(struct intel_engine_cs *ring); |
470 | int init_workarounds_ring(struct intel_engine_cs *ring); |
433 | 471 | ||
434 | static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf) |
472 | static inline u32 intel_ring_get_tail(struct intel_ringbuffer *ringbuf) |
435 | { |
473 | { |
436 | return ringbuf->tail; |
474 | return ringbuf->tail; |
437 | } |
475 | } |
438 | - | ||
439 | static inline u32 intel_ring_get_seqno(struct intel_engine_cs *ring) |
476 | |
440 | { |
477 | /* |
- | 478 | * Arbitrary size for largest possible 'add request' sequence. The code paths |
|
- | 479 | * are complex and variable. Empirical measurement shows that the worst case |
|
- | 480 | * is ILK at 136 words. Reserving too much is better than reserving too little |
|
441 | BUG_ON(ring->outstanding_lazy_seqno == 0); |
481 | * as that allows for corner cases that might have been missed. So the figure |
442 | return ring->outstanding_lazy_seqno; |
482 | * has been rounded up to 160 words. |
- | 483 | */ |
|
- | 484 | #define MIN_SPACE_FOR_ADD_REQUEST 160 |
|
443 | } |
485 | |
- | 486 | /* |
|
- | 487 | * Reserve space in the ring to guarantee that the i915_add_request() call |
|
444 | 488 | * will always have sufficient room to do its stuff. The request creation |
|
- | 489 | * code calls this automatically. |
|
- | 490 | */ |
|
- | 491 | void intel_ring_reserved_space_reserve(struct intel_ringbuffer *ringbuf, int size); |
|
- | 492 | /* Cancel the reservation, e.g. because the request is being discarded. */ |
|
445 | static inline void i915_trace_irq_get(struct intel_engine_cs *ring, u32 seqno) |
493 | void intel_ring_reserved_space_cancel(struct intel_ringbuffer *ringbuf); |
- | 494 | /* Use the reserved space - for use by i915_add_request() only. */ |
|
446 | { |
495 | void intel_ring_reserved_space_use(struct intel_ringbuffer *ringbuf); |
447 | if (ring->trace_irq_seqno == 0 && ring->irq_get(ring)) |
496 | /* Finish with the reserved space - for use by i915_add_request() only. */ |
- | 497 | void intel_ring_reserved_space_end(struct intel_ringbuffer *ringbuf); |
|
- | 498 | ||
448 | ring->trace_irq_seqno = seqno; |
499 | /* Legacy ringbuffer specific portion of reservation code: */ |
449 | } |
500 | int intel_ring_reserve_space(struct drm_i915_gem_request *request); |
450 | 501 | ||
451 | #endif /* _INTEL_RINGBUFFER_H_ */><>>><> |
502 | #endif /* _INTEL_RINGBUFFER_H_ */><>>><> |