xref: /dragonfly/sys/dev/drm/i915/intel_ringbuffer.h (revision 896f2e3a)
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 enum intel_ring_hangcheck_action {
37 	HANGCHECK_IDLE = 0,
38 	HANGCHECK_WAIT,
39 	HANGCHECK_ACTIVE,
40 	HANGCHECK_KICK,
41 	HANGCHECK_HUNG,
42 };
43 
44 struct intel_ring_hangcheck {
45 	bool deadlock;
46 	u32 seqno;
47 	u32 acthd;
48 	int score;
49 	enum intel_ring_hangcheck_action action;
50 };
51 
52 struct  intel_ring_buffer {
53 	const char	*name;
54 	enum intel_ring_id {
55 		RCS = 0x0,
56 		VCS,
57 		BCS,
58 		VECS,
59 	} id;
60 #define I915_NUM_RINGS 4
61 	u32		mmio_base;
62 	void		__iomem *virtual_start;
63 	struct		drm_device *dev;
64 	struct		drm_i915_gem_object *obj;
65 
66 	u32		head;
67 	u32		tail;
68 	int		space;
69 	int		size;
70 	int		effective_size;
71 	struct intel_hw_status_page status_page;
72 
73 	/** We track the position of the requests in the ring buffer, and
74 	 * when each is retired we increment last_retired_head as the GPU
75 	 * must have finished processing the request and so we know we
76 	 * can advance the ringbuffer up to that position.
77 	 *
78 	 * last_retired_head is set to -1 after the value is consumed so
79 	 * we can detect new retirements.
80 	 */
81 	u32		last_retired_head;
82 
83 	unsigned irq_refcount; /* protected by dev_priv->irq_lock */
84 	u32		irq_enable_mask;	/* bitmask to enable ring interrupt */
85 	u32		trace_irq_seqno;
86 	u32		sync_seqno[I915_NUM_RINGS-1];
87 	bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
88 	void		(*irq_put)(struct intel_ring_buffer *ring);
89 
90 	int		(*init)(struct intel_ring_buffer *ring);
91 
92 	void		(*write_tail)(struct intel_ring_buffer *ring,
93 				      u32 value);
94 	int __must_check (*flush)(struct intel_ring_buffer *ring,
95 				  u32	invalidate_domains,
96 				  u32	flush_domains);
97 	int		(*add_request)(struct intel_ring_buffer *ring);
98 	/* Some chipsets are not quite as coherent as advertised and need
99 	 * an expensive kick to force a true read of the up-to-date seqno.
100 	 * However, the up-to-date seqno is not always required and the last
101 	 * seen value is good enough. Note that the seqno will always be
102 	 * monotonic, even if not coherent.
103 	 */
104 	u32		(*get_seqno)(struct intel_ring_buffer *ring,
105 				     bool lazy_coherency);
106 	void		(*set_seqno)(struct intel_ring_buffer *ring,
107 				     u32 seqno);
108 	int		(*dispatch_execbuffer)(struct intel_ring_buffer *ring,
109 					       u32 offset, u32 length,
110 					       unsigned flags);
111 #define I915_DISPATCH_SECURE 0x1
112 #define I915_DISPATCH_PINNED 0x2
113 	void		(*cleanup)(struct intel_ring_buffer *ring);
114 	int		(*sync_to)(struct intel_ring_buffer *ring,
115 				   struct intel_ring_buffer *to,
116 				   u32 seqno);
117 
118 	/* our mbox written by others */
119 	u32		semaphore_register[I915_NUM_RINGS];
120 	/* mboxes this ring signals to */
121 	u32		signal_mbox[I915_NUM_RINGS];
122 
123 	/**
124 	 * List of objects currently involved in rendering from the
125 	 * ringbuffer.
126 	 *
127 	 * Includes buffers having the contents of their GPU caches
128 	 * flushed, not necessarily primitives.  last_rendering_seqno
129 	 * represents when the rendering involved will be completed.
130 	 *
131 	 * A reference is held on the buffer while on this list.
132 	 */
133 	struct list_head active_list;
134 
135 	/**
136 	 * List of breadcrumbs associated with GPU requests currently
137 	 * outstanding.
138 	 */
139 	struct list_head request_list;
140 
141 	/**
142 	 * Do we have some not yet emitted requests outstanding?
143 	 */
144 	struct drm_i915_gem_request *preallocated_lazy_request;
145 	u32 outstanding_lazy_seqno;
146 	bool gpu_caches_dirty;
147 	bool fbc_dirty;
148 
149 	wait_queue_head_t irq_queue;
150 
151 	/**
152 	 * Do an explicit TLB flush before MI_SET_CONTEXT
153 	 */
154 	bool itlb_before_ctx_switch;
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 static inline bool
168 intel_ring_initialized(struct intel_ring_buffer *ring)
169 {
170 	return ring->obj != NULL;
171 }
172 
173 static inline unsigned
174 intel_ring_flag(struct intel_ring_buffer *ring)
175 {
176 	return 1 << ring->id;
177 }
178 
179 static inline u32
180 intel_ring_sync_index(struct intel_ring_buffer *ring,
181 		      struct intel_ring_buffer *other)
182 {
183 	int idx;
184 
185 	/*
186 	 * cs -> 0 = vcs, 1 = bcs
187 	 * vcs -> 0 = bcs, 1 = cs,
188 	 * bcs -> 0 = cs, 1 = vcs.
189 	 */
190 
191 	idx = (other - ring) - 1;
192 	if (idx < 0)
193 		idx += I915_NUM_RINGS;
194 
195 	return idx;
196 }
197 
198 static inline u32
199 intel_read_status_page(struct intel_ring_buffer *ring,
200 		       int reg)
201 {
202 	/* Ensure that the compiler doesn't optimize away the load. */
203 	barrier();
204 	return ring->status_page.page_addr[reg];
205 }
206 
207 static inline void
208 intel_write_status_page(struct intel_ring_buffer *ring,
209 			int reg, u32 value)
210 {
211 	ring->status_page.page_addr[reg] = value;
212 }
213 
214 /**
215  * Reads a dword out of the status page, which is written to from the command
216  * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or
217  * MI_STORE_DATA_IMM.
218  *
219  * The following dwords have a reserved meaning:
220  * 0x00: ISR copy, updated when an ISR bit not set in the HWSTAM changes.
221  * 0x04: ring 0 head pointer
222  * 0x05: ring 1 head pointer (915-class)
223  * 0x06: ring 2 head pointer (915-class)
224  * 0x10-0x1b: Context status DWords (GM45)
225  * 0x1f: Last written status offset. (GM45)
226  *
227  * The area from dword 0x20 to 0x3ff is available for driver usage.
228  */
229 #define I915_GEM_HWS_INDEX		0x20
230 #define I915_GEM_HWS_SCRATCH_INDEX	0x30
231 #define I915_GEM_HWS_SCRATCH_ADDR (I915_GEM_HWS_SCRATCH_INDEX << MI_STORE_DWORD_INDEX_SHIFT)
232 
233 void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring);
234 
235 int __must_check intel_ring_begin(struct intel_ring_buffer *ring, int n);
236 int __must_check intel_ring_cacheline_align(struct intel_ring_buffer *ring);
237 static inline void intel_ring_emit(struct intel_ring_buffer *ring,
238 				   u32 data)
239 {
240 	iowrite32(data, ring->virtual_start + ring->tail);
241 	ring->tail += 4;
242 }
243 static inline void intel_ring_advance(struct intel_ring_buffer *ring)
244 {
245 	ring->tail &= ring->size - 1;
246 }
247 void __intel_ring_advance(struct intel_ring_buffer *ring);
248 
249 int __must_check intel_ring_idle(struct intel_ring_buffer *ring);
250 void intel_ring_init_seqno(struct intel_ring_buffer *ring, u32 seqno);
251 int intel_ring_flush_all_caches(struct intel_ring_buffer *ring);
252 int intel_ring_invalidate_all_caches(struct intel_ring_buffer *ring);
253 
254 int intel_init_render_ring_buffer(struct drm_device *dev);
255 int intel_init_bsd_ring_buffer(struct drm_device *dev);
256 int intel_init_blt_ring_buffer(struct drm_device *dev);
257 int intel_init_vebox_ring_buffer(struct drm_device *dev);
258 
259 u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
260 void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
261 
262 static inline u32 intel_ring_get_tail(struct intel_ring_buffer *ring)
263 {
264 	return ring->tail;
265 }
266 
267 static inline u32 intel_ring_get_seqno(struct intel_ring_buffer *ring)
268 {
269 	BUG_ON(ring->outstanding_lazy_seqno == 0);
270 	return ring->outstanding_lazy_seqno;
271 }
272 
273 static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
274 {
275 	if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
276 		ring->trace_irq_seqno = seqno;
277 }
278 
279 /* DRI warts */
280 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);
281 
282 #endif /* _INTEL_RINGBUFFER_H_ */
283