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 #define to_amdgpu_ring(s) container_of((s), struct amdgpu_ring, sched)
48
49 enum amdgpu_ring_type {
50 AMDGPU_RING_TYPE_GFX,
51 AMDGPU_RING_TYPE_COMPUTE,
52 AMDGPU_RING_TYPE_SDMA,
53 AMDGPU_RING_TYPE_UVD,
54 AMDGPU_RING_TYPE_VCE,
55 AMDGPU_RING_TYPE_KIQ,
56 AMDGPU_RING_TYPE_UVD_ENC,
57 AMDGPU_RING_TYPE_VCN_DEC,
58 AMDGPU_RING_TYPE_VCN_ENC,
59 AMDGPU_RING_TYPE_VCN_JPEG
60 };
61
62 struct amdgpu_device;
63 struct amdgpu_ring;
64 struct amdgpu_ib;
65 struct amdgpu_cs_parser;
66 struct amdgpu_job;
67
68 /*
69 * Fences.
70 */
71 struct amdgpu_fence_driver {
72 uint64_t gpu_addr;
73 volatile uint32_t *cpu_addr;
74 /* sync_seq is protected by ring emission lock */
75 uint32_t sync_seq;
76 atomic_t last_seq;
77 bool initialized;
78 struct amdgpu_irq_src *irq_src;
79 unsigned irq_type;
80 struct timer_list fallback_timer;
81 unsigned num_fences_mask;
82 spinlock_t lock;
83 struct dma_fence **fences;
84 };
85
86 int amdgpu_fence_driver_init(struct amdgpu_device *adev);
87 void amdgpu_fence_driver_fini(struct amdgpu_device *adev);
88 void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring);
89
90 int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
91 unsigned num_hw_submission);
92 int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring,
93 struct amdgpu_irq_src *irq_src,
94 unsigned irq_type);
95 void amdgpu_fence_driver_suspend(struct amdgpu_device *adev);
96 void amdgpu_fence_driver_resume(struct amdgpu_device *adev);
97 int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence,
98 unsigned flags);
99 int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s);
100 void amdgpu_fence_process(struct amdgpu_ring *ring);
101 int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
102 signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring,
103 uint32_t wait_seq,
104 signed long timeout);
105 unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
106
107 /*
108 * Rings.
109 */
110
111 /* provided by hw blocks that expose a ring buffer for commands */
112 struct amdgpu_ring_funcs {
113 enum amdgpu_ring_type type;
114 uint32_t align_mask;
115 u32 nop;
116 bool support_64bit_ptrs;
117 unsigned vmhub;
118 unsigned extra_dw;
119
120 /* ring read/write ptr handling */
121 u64 (*get_rptr)(struct amdgpu_ring *ring);
122 u64 (*get_wptr)(struct amdgpu_ring *ring);
123 void (*set_wptr)(struct amdgpu_ring *ring);
124 /* validating and patching of IBs */
125 int (*parse_cs)(struct amdgpu_cs_parser *p, uint32_t ib_idx);
126 int (*patch_cs_in_place)(struct amdgpu_cs_parser *p, uint32_t ib_idx);
127 /* constants to calculate how many DW are needed for an emit */
128 unsigned emit_frame_size;
129 unsigned emit_ib_size;
130 /* command emit functions */
131 void (*emit_ib)(struct amdgpu_ring *ring,
132 struct amdgpu_ib *ib,
133 unsigned vmid, bool ctx_switch);
134 void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr,
135 uint64_t seq, unsigned flags);
136 void (*emit_pipeline_sync)(struct amdgpu_ring *ring);
137 void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vmid,
138 uint64_t pd_addr);
139 void (*emit_hdp_flush)(struct amdgpu_ring *ring);
140 void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid,
141 uint32_t gds_base, uint32_t gds_size,
142 uint32_t gws_base, uint32_t gws_size,
143 uint32_t oa_base, uint32_t oa_size);
144 /* testing functions */
145 int (*test_ring)(struct amdgpu_ring *ring);
146 int (*test_ib)(struct amdgpu_ring *ring, long timeout);
147 /* insert NOP packets */
148 void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count);
149 void (*insert_start)(struct amdgpu_ring *ring);
150 void (*insert_end)(struct amdgpu_ring *ring);
151 /* pad the indirect buffer to the necessary number of dw */
152 void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
153 unsigned (*init_cond_exec)(struct amdgpu_ring *ring);
154 void (*patch_cond_exec)(struct amdgpu_ring *ring, unsigned offset);
155 /* note usage for clock and power gating */
156 void (*begin_use)(struct amdgpu_ring *ring);
157 void (*end_use)(struct amdgpu_ring *ring);
158 void (*emit_switch_buffer) (struct amdgpu_ring *ring);
159 void (*emit_cntxcntl) (struct amdgpu_ring *ring, uint32_t flags);
160 void (*emit_rreg)(struct amdgpu_ring *ring, uint32_t reg);
161 void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val);
162 void (*emit_reg_wait)(struct amdgpu_ring *ring, uint32_t reg,
163 uint32_t val, uint32_t mask);
164 void (*emit_reg_write_reg_wait)(struct amdgpu_ring *ring,
165 uint32_t reg0, uint32_t reg1,
166 uint32_t ref, uint32_t mask);
167 void (*emit_tmz)(struct amdgpu_ring *ring, bool start);
168 /* priority functions */
169 void (*set_priority) (struct amdgpu_ring *ring,
170 enum drm_sched_priority priority);
171 };
172
173 struct amdgpu_ring {
174 struct amdgpu_device *adev;
175 const struct amdgpu_ring_funcs *funcs;
176 struct amdgpu_fence_driver fence_drv;
177 struct drm_gpu_scheduler sched;
178 struct list_head lru_list;
179
180 struct amdgpu_bo *ring_obj;
181 volatile uint32_t *ring;
182 unsigned rptr_offs;
183 u64 wptr;
184 u64 wptr_old;
185 unsigned ring_size;
186 unsigned max_dw;
187 int count_dw;
188 uint64_t gpu_addr;
189 uint64_t ptr_mask;
190 uint32_t buf_mask;
191 bool ready;
192 u32 idx;
193 u32 me;
194 u32 pipe;
195 u32 queue;
196 struct amdgpu_bo *mqd_obj;
197 uint64_t mqd_gpu_addr;
198 void *mqd_ptr;
199 uint64_t eop_gpu_addr;
200 u32 doorbell_index;
201 bool use_doorbell;
202 bool use_pollmem;
203 unsigned wptr_offs;
204 unsigned fence_offs;
205 uint64_t current_ctx;
206 char name[16];
207 unsigned cond_exe_offs;
208 u64 cond_exe_gpu_addr;
209 volatile u32 *cond_exe_cpu_addr;
210 unsigned vm_inv_eng;
211 struct dma_fence *vmid_wait;
212 bool has_compute_vm_bug;
213
214 atomic_t num_jobs[DRM_SCHED_PRIORITY_MAX];
215 struct mutex priority_mutex;
216 /* protected by priority_mutex */
217 int priority;
218
219 #if defined(CONFIG_DEBUG_FS)
220 struct dentry *ent;
221 #endif
222 };
223
224 int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw);
225 void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
226 void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
227 void amdgpu_ring_commit(struct amdgpu_ring *ring);
228 void amdgpu_ring_undo(struct amdgpu_ring *ring);
229 void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
230 enum drm_sched_priority priority);
231 void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
232 enum drm_sched_priority priority);
233 int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
234 unsigned ring_size, struct amdgpu_irq_src *irq_src,
235 unsigned irq_type);
236 void amdgpu_ring_fini(struct amdgpu_ring *ring);
237 int amdgpu_ring_lru_get(struct amdgpu_device *adev, int type,
238 int *blacklist, int num_blacklist,
239 bool lru_pipe_order, struct amdgpu_ring **ring);
240 void amdgpu_ring_lru_touch(struct amdgpu_device *adev, struct amdgpu_ring *ring);
241 void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring,
242 uint32_t reg0, uint32_t val0,
243 uint32_t reg1, uint32_t val1);
244
amdgpu_ring_clear_ring(struct amdgpu_ring * ring)245 static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring)
246 {
247 int i = 0;
248 while (i <= ring->buf_mask)
249 ring->ring[i++] = ring->funcs->nop;
250
251 }
252
amdgpu_ring_write(struct amdgpu_ring * ring,uint32_t v)253 static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
254 {
255 if (ring->count_dw <= 0)
256 DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
257 ring->ring[ring->wptr++ & ring->buf_mask] = v;
258 ring->wptr &= ring->ptr_mask;
259 ring->count_dw--;
260 }
261
amdgpu_ring_write_multiple(struct amdgpu_ring * ring,void * src,int count_dw)262 static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring,
263 void *src, int count_dw)
264 {
265 unsigned occupied, chunk1, chunk2;
266 void *dst;
267
268 if (unlikely(ring->count_dw < count_dw))
269 DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
270
271 occupied = ring->wptr & ring->buf_mask;
272 dst = (void *)&ring->ring[occupied];
273 chunk1 = ring->buf_mask + 1 - occupied;
274 chunk1 = (chunk1 >= count_dw) ? count_dw: chunk1;
275 chunk2 = count_dw - chunk1;
276 chunk1 <<= 2;
277 chunk2 <<= 2;
278
279 if (chunk1)
280 memcpy(dst, src, chunk1);
281
282 if (chunk2) {
283 src += chunk1;
284 dst = (void *)ring->ring;
285 memcpy(dst, src, chunk2);
286 }
287
288 ring->wptr += count_dw;
289 ring->wptr &= ring->ptr_mask;
290 ring->count_dw -= count_dw;
291 }
292
293 #endif
294
