1 /*
2  * Copyright 2016 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  * Authors: Christian König
23  */
24 #ifndef __AMDGPU_RING_H__
25 #define __AMDGPU_RING_H__
26 
27 #include <drm/amdgpu_drm.h>
28 #include <drm/gpu_scheduler.h>
29 #include <drm/drm_print.h>
30 
31 /* max number of rings */
32 #define AMDGPU_MAX_RINGS		21
33 #define AMDGPU_MAX_GFX_RINGS		1
34 #define AMDGPU_MAX_COMPUTE_RINGS	8
35 #define AMDGPU_MAX_VCE_RINGS		3
36 #define AMDGPU_MAX_UVD_ENC_RINGS	2
37 
38 /* some special values for the owner field */
39 #define AMDGPU_FENCE_OWNER_UNDEFINED	((void *)0ul)
40 #define AMDGPU_FENCE_OWNER_VM		((void *)1ul)
41 #define AMDGPU_FENCE_OWNER_KFD		((void *)2ul)
42 
43 #define AMDGPU_FENCE_FLAG_64BIT         (1 << 0)
44 #define AMDGPU_FENCE_FLAG_INT           (1 << 1)
45 #define AMDGPU_FENCE_FLAG_TC_WB_ONLY    (1 << 2)
46 
47 enum amdgpu_ring_type {
48 	AMDGPU_RING_TYPE_GFX,
49 	AMDGPU_RING_TYPE_COMPUTE,
50 	AMDGPU_RING_TYPE_SDMA,
51 	AMDGPU_RING_TYPE_UVD,
52 	AMDGPU_RING_TYPE_VCE,
53 	AMDGPU_RING_TYPE_KIQ,
54 	AMDGPU_RING_TYPE_UVD_ENC,
55 	AMDGPU_RING_TYPE_VCN_DEC,
56 	AMDGPU_RING_TYPE_VCN_ENC
57 };
58 
59 struct amdgpu_device;
60 struct amdgpu_ring;
61 struct amdgpu_ib;
62 struct amdgpu_cs_parser;
63 struct amdgpu_job;
64 
65 /*
66  * Fences.
67  */
68 struct amdgpu_fence_driver {
69 	uint64_t			gpu_addr;
70 	volatile uint32_t		*cpu_addr;
71 	/* sync_seq is protected by ring emission lock */
72 	uint32_t			sync_seq;
73 	atomic_t			last_seq;
74 	bool				initialized;
75 	struct amdgpu_irq_src		*irq_src;
76 	unsigned			irq_type;
77 	struct timer_list		fallback_timer;
78 	unsigned			num_fences_mask;
79 	spinlock_t			lock;
80 	struct dma_fence		**fences;
81 };
82 
83 int amdgpu_fence_driver_init(struct amdgpu_device *adev);
84 void amdgpu_fence_driver_fini(struct amdgpu_device *adev);
85 void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring);
86 
87 int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
88 				  unsigned num_hw_submission);
89 int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
90 				   struct amdgpu_irq_src *irq_src,
91 				   unsigned irq_type);
92 void amdgpu_fence_driver_suspend(struct amdgpu_device *adev);
93 void amdgpu_fence_driver_resume(struct amdgpu_device *adev);
94 int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence,
95 		      unsigned flags);
96 int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s);
97 void amdgpu_fence_process(struct amdgpu_ring *ring);
98 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
99 signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring,
100 				      uint32_t wait_seq,
101 				      signed long timeout);
102 unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
103 
104 /*
105  * Rings.
106  */
107 
108 /* provided by hw blocks that expose a ring buffer for commands */
109 struct amdgpu_ring_funcs {
110 	enum amdgpu_ring_type	type;
111 	uint32_t		align_mask;
112 	u32			nop;
113 	bool			support_64bit_ptrs;
114 	unsigned		vmhub;
115 
116 	/* ring read/write ptr handling */
117 	u64 (*get_rptr)(struct amdgpu_ring *ring);
118 	u64 (*get_wptr)(struct amdgpu_ring *ring);
119 	void (*set_wptr)(struct amdgpu_ring *ring);
120 	/* validating and patching of IBs */
121 	int (*parse_cs)(struct amdgpu_cs_parser *p, uint32_t ib_idx);
122 	/* constants to calculate how many DW are needed for an emit */
123 	unsigned emit_frame_size;
124 	unsigned emit_ib_size;
125 	/* command emit functions */
126 	void (*emit_ib)(struct amdgpu_ring *ring,
127 			struct amdgpu_ib *ib,
128 			unsigned vmid, bool ctx_switch);
129 	void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr,
130 			   uint64_t seq, unsigned flags);
131 	void (*emit_pipeline_sync)(struct amdgpu_ring *ring);
132 	void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vmid,
133 			      uint64_t pd_addr);
134 	void (*emit_hdp_flush)(struct amdgpu_ring *ring);
135 	void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid,
136 				uint32_t gds_base, uint32_t gds_size,
137 				uint32_t gws_base, uint32_t gws_size,
138 				uint32_t oa_base, uint32_t oa_size);
139 	/* testing functions */
140 	int (*test_ring)(struct amdgpu_ring *ring);
141 	int (*test_ib)(struct amdgpu_ring *ring, long timeout);
142 	/* insert NOP packets */
143 	void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count);
144 	void (*insert_start)(struct amdgpu_ring *ring);
145 	void (*insert_end)(struct amdgpu_ring *ring);
146 	/* pad the indirect buffer to the necessary number of dw */
147 	void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
148 	unsigned (*init_cond_exec)(struct amdgpu_ring *ring);
149 	void (*patch_cond_exec)(struct amdgpu_ring *ring, unsigned offset);
150 	/* note usage for clock and power gating */
151 	void (*begin_use)(struct amdgpu_ring *ring);
152 	void (*end_use)(struct amdgpu_ring *ring);
153 	void (*emit_switch_buffer) (struct amdgpu_ring *ring);
154 	void (*emit_cntxcntl) (struct amdgpu_ring *ring, uint32_t flags);
155 	void (*emit_rreg)(struct amdgpu_ring *ring, uint32_t reg);
156 	void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val);
157 	void (*emit_reg_wait)(struct amdgpu_ring *ring, uint32_t reg,
158 			      uint32_t val, uint32_t mask);
159 	void (*emit_reg_write_reg_wait)(struct amdgpu_ring *ring,
160 					uint32_t reg0, uint32_t reg1,
161 					uint32_t ref, uint32_t mask);
162 	void (*emit_tmz)(struct amdgpu_ring *ring, bool start);
163 	/* priority functions */
164 	void (*set_priority) (struct amdgpu_ring *ring,
165 			      enum drm_sched_priority priority);
166 };
167 
168 struct amdgpu_ring {
169 	struct amdgpu_device		*adev;
170 	const struct amdgpu_ring_funcs	*funcs;
171 	struct amdgpu_fence_driver	fence_drv;
172 	struct drm_gpu_scheduler	sched;
173 	struct list_head		lru_list;
174 
175 	struct amdgpu_bo	*ring_obj;
176 	volatile uint32_t	*ring;
177 	unsigned		rptr_offs;
178 	u64			wptr;
179 	u64			wptr_old;
180 	unsigned		ring_size;
181 	unsigned		max_dw;
182 	int			count_dw;
183 	uint64_t		gpu_addr;
184 	uint64_t		ptr_mask;
185 	uint32_t		buf_mask;
186 	bool			ready;
187 	u32			idx;
188 	u32			me;
189 	u32			pipe;
190 	u32			queue;
191 	struct amdgpu_bo	*mqd_obj;
192 	uint64_t                mqd_gpu_addr;
193 	void                    *mqd_ptr;
194 	uint64_t                eop_gpu_addr;
195 	u32			doorbell_index;
196 	bool			use_doorbell;
197 	bool			use_pollmem;
198 	unsigned		wptr_offs;
199 	unsigned		fence_offs;
200 	uint64_t		current_ctx;
201 	char			name[16];
202 	unsigned		cond_exe_offs;
203 	u64			cond_exe_gpu_addr;
204 	volatile u32		*cond_exe_cpu_addr;
205 	unsigned		vm_inv_eng;
206 	struct dma_fence	*vmid_wait;
207 	bool			has_compute_vm_bug;
208 
209 	atomic_t		num_jobs[DRM_SCHED_PRIORITY_MAX];
210 	struct mutex		priority_mutex;
211 	/* protected by priority_mutex */
212 	int			priority;
213 
214 #if defined(CONFIG_DEBUG_FS)
215 	struct dentry *ent;
216 #endif
217 };
218 
219 int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw);
220 void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
221 void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
222 void amdgpu_ring_commit(struct amdgpu_ring *ring);
223 void amdgpu_ring_undo(struct amdgpu_ring *ring);
224 void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
225 			      enum drm_sched_priority priority);
226 void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
227 			      enum drm_sched_priority priority);
228 int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
229 		     unsigned ring_size, struct amdgpu_irq_src *irq_src,
230 		     unsigned irq_type);
231 void amdgpu_ring_fini(struct amdgpu_ring *ring);
232 int amdgpu_ring_lru_get(struct amdgpu_device *adev, int type,
233 			int *blacklist, int num_blacklist,
234 			bool lru_pipe_order, struct amdgpu_ring **ring);
235 void amdgpu_ring_lru_touch(struct amdgpu_device *adev, struct amdgpu_ring *ring);
236 void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring,
237 						uint32_t reg0, uint32_t val0,
238 						uint32_t reg1, uint32_t val1);
239 
240 static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring)
241 {
242 	int i = 0;
243 	while (i <= ring->buf_mask)
244 		ring->ring[i++] = ring->funcs->nop;
245 
246 }
247 
248 static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
249 {
250 	if (ring->count_dw <= 0)
251 		DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
252 	ring->ring[ring->wptr++ & ring->buf_mask] = v;
253 	ring->wptr &= ring->ptr_mask;
254 	ring->count_dw--;
255 }
256 
257 static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring,
258 					      void *src, int count_dw)
259 {
260 	unsigned occupied, chunk1, chunk2;
261 	void *dst;
262 
263 	if (unlikely(ring->count_dw < count_dw))
264 		DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
265 
266 	occupied = ring->wptr & ring->buf_mask;
267 	dst = (void *)&ring->ring[occupied];
268 	chunk1 = ring->buf_mask + 1 - occupied;
269 	chunk1 = (chunk1 >= count_dw) ? count_dw: chunk1;
270 	chunk2 = count_dw - chunk1;
271 	chunk1 <<= 2;
272 	chunk2 <<= 2;
273 
274 	if (chunk1)
275 		memcpy(dst, src, chunk1);
276 
277 	if (chunk2) {
278 		src += chunk1;
279 		dst = (void *)ring->ring;
280 		memcpy(dst, src, chunk2);
281 	}
282 
283 	ring->wptr += count_dw;
284 	ring->wptr &= ring->ptr_mask;
285 	ring->count_dw -= count_dw;
286 }
287 
288 #endif
289