Commit | Line | Data |
---|---|---|
78023016 CK |
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 | ||
b2ff0e8a | 27 | #include <drm/amdgpu_drm.h> |
1b1f42d8 | 28 | #include <drm/gpu_scheduler.h> |
61b100e9 | 29 | #include <drm/drm_print.h> |
78023016 CK |
30 | |
31 | /* max number of rings */ | |
3b17c622 | 32 | #define AMDGPU_MAX_RINGS 21 |
78023016 CK |
33 | #define AMDGPU_MAX_GFX_RINGS 1 |
34 | #define AMDGPU_MAX_COMPUTE_RINGS 8 | |
35 | #define AMDGPU_MAX_VCE_RINGS 3 | |
f7243053 | 36 | #define AMDGPU_MAX_UVD_ENC_RINGS 2 |
78023016 CK |
37 | |
38 | /* some special values for the owner field */ | |
d8d019cc FK |
39 | #define AMDGPU_FENCE_OWNER_UNDEFINED ((void *)0ul) |
40 | #define AMDGPU_FENCE_OWNER_VM ((void *)1ul) | |
41 | #define AMDGPU_FENCE_OWNER_KFD ((void *)2ul) | |
78023016 CK |
42 | |
43 | #define AMDGPU_FENCE_FLAG_64BIT (1 << 0) | |
44 | #define AMDGPU_FENCE_FLAG_INT (1 << 1) | |
d240cd9e | 45 | #define AMDGPU_FENCE_FLAG_TC_WB_ONLY (1 << 2) |
78023016 | 46 | |
0e28b10f CK |
47 | #define to_amdgpu_ring(s) container_of((s), struct amdgpu_ring, sched) |
48 | ||
78023016 CK |
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, | |
2068751d | 54 | AMDGPU_RING_TYPE_VCE, |
50c3e232 | 55 | AMDGPU_RING_TYPE_KIQ, |
cca69fe8 | 56 | AMDGPU_RING_TYPE_UVD_ENC, |
8ace845f | 57 | AMDGPU_RING_TYPE_VCN_DEC, |
8e0fce5a BZ |
58 | AMDGPU_RING_TYPE_VCN_ENC, |
59 | AMDGPU_RING_TYPE_VCN_JPEG | |
78023016 CK |
60 | }; |
61 | ||
62 | struct amdgpu_device; | |
63 | struct amdgpu_ring; | |
64 | struct amdgpu_ib; | |
65 | struct amdgpu_cs_parser; | |
b2ff0e8a | 66 | struct amdgpu_job; |
78023016 CK |
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; | |
78023016 CK |
80 | unsigned num_fences_mask; |
81 | spinlock_t lock; | |
220196b3 | 82 | struct dma_fence **fences; |
78023016 CK |
83 | }; |
84 | ||
85 | int amdgpu_fence_driver_init(struct amdgpu_device *adev); | |
86 | void amdgpu_fence_driver_fini(struct amdgpu_device *adev); | |
2f9d4084 | 87 | void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring); |
78023016 CK |
88 | |
89 | int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring, | |
90 | unsigned num_hw_submission); | |
91 | int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring, | |
92 | struct amdgpu_irq_src *irq_src, | |
93 | unsigned irq_type); | |
94 | void amdgpu_fence_driver_suspend(struct amdgpu_device *adev); | |
95 | void amdgpu_fence_driver_resume(struct amdgpu_device *adev); | |
d240cd9e MO |
96 | int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence, |
97 | unsigned flags); | |
43ca8efa | 98 | int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s); |
78023016 CK |
99 | void amdgpu_fence_process(struct amdgpu_ring *ring); |
100 | int amdgpu_fence_wait_empty(struct amdgpu_ring *ring); | |
43ca8efa | 101 | signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring, |
102 | uint32_t wait_seq, | |
103 | signed long timeout); | |
78023016 CK |
104 | unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring); |
105 | ||
106 | /* | |
107 | * Rings. | |
108 | */ | |
109 | ||
110 | /* provided by hw blocks that expose a ring buffer for commands */ | |
111 | struct amdgpu_ring_funcs { | |
21cd942e | 112 | enum amdgpu_ring_type type; |
79887142 CK |
113 | uint32_t align_mask; |
114 | u32 nop; | |
536fbf94 | 115 | bool support_64bit_ptrs; |
0eeb68b3 | 116 | unsigned vmhub; |
c8c1a1d2 | 117 | unsigned extra_dw; |
21cd942e | 118 | |
78023016 | 119 | /* ring read/write ptr handling */ |
536fbf94 KW |
120 | u64 (*get_rptr)(struct amdgpu_ring *ring); |
121 | u64 (*get_wptr)(struct amdgpu_ring *ring); | |
78023016 CK |
122 | void (*set_wptr)(struct amdgpu_ring *ring); |
123 | /* validating and patching of IBs */ | |
124 | int (*parse_cs)(struct amdgpu_cs_parser *p, uint32_t ib_idx); | |
9d248517 | 125 | int (*patch_cs_in_place)(struct amdgpu_cs_parser *p, uint32_t ib_idx); |
e12f3d7a CK |
126 | /* constants to calculate how many DW are needed for an emit */ |
127 | unsigned emit_frame_size; | |
128 | unsigned emit_ib_size; | |
78023016 CK |
129 | /* command emit functions */ |
130 | void (*emit_ib)(struct amdgpu_ring *ring, | |
131 | struct amdgpu_ib *ib, | |
c4f46f22 | 132 | unsigned vmid, bool ctx_switch); |
78023016 CK |
133 | void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr, |
134 | uint64_t seq, unsigned flags); | |
135 | void (*emit_pipeline_sync)(struct amdgpu_ring *ring); | |
c4f46f22 | 136 | void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vmid, |
c633c00b | 137 | uint64_t pd_addr); |
78023016 | 138 | void (*emit_hdp_flush)(struct amdgpu_ring *ring); |
78023016 CK |
139 | void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid, |
140 | uint32_t gds_base, uint32_t gds_size, | |
141 | uint32_t gws_base, uint32_t gws_size, | |
142 | uint32_t oa_base, uint32_t oa_size); | |
143 | /* testing functions */ | |
144 | int (*test_ring)(struct amdgpu_ring *ring); | |
145 | int (*test_ib)(struct amdgpu_ring *ring, long timeout); | |
146 | /* insert NOP packets */ | |
147 | void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count); | |
ef44f854 | 148 | void (*insert_start)(struct amdgpu_ring *ring); |
135d4735 | 149 | void (*insert_end)(struct amdgpu_ring *ring); |
78023016 CK |
150 | /* pad the indirect buffer to the necessary number of dw */ |
151 | void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib); | |
152 | unsigned (*init_cond_exec)(struct amdgpu_ring *ring); | |
153 | void (*patch_cond_exec)(struct amdgpu_ring *ring, unsigned offset); | |
154 | /* note usage for clock and power gating */ | |
155 | void (*begin_use)(struct amdgpu_ring *ring); | |
156 | void (*end_use)(struct amdgpu_ring *ring); | |
157 | void (*emit_switch_buffer) (struct amdgpu_ring *ring); | |
158 | void (*emit_cntxcntl) (struct amdgpu_ring *ring, uint32_t flags); | |
b6091c12 XY |
159 | void (*emit_rreg)(struct amdgpu_ring *ring, uint32_t reg); |
160 | void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val); | |
c1e877da CK |
161 | void (*emit_reg_wait)(struct amdgpu_ring *ring, uint32_t reg, |
162 | uint32_t val, uint32_t mask); | |
82853638 AD |
163 | void (*emit_reg_write_reg_wait)(struct amdgpu_ring *ring, |
164 | uint32_t reg0, uint32_t reg1, | |
165 | uint32_t ref, uint32_t mask); | |
3b4d68e9 | 166 | void (*emit_tmz)(struct amdgpu_ring *ring, bool start); |
b2ff0e8a AR |
167 | /* priority functions */ |
168 | void (*set_priority) (struct amdgpu_ring *ring, | |
1b1f42d8 | 169 | enum drm_sched_priority priority); |
7876fa4f CK |
170 | /* Try to soft recover the ring to make the fence signal */ |
171 | void (*soft_recovery)(struct amdgpu_ring *ring, unsigned vmid); | |
78023016 CK |
172 | }; |
173 | ||
174 | struct amdgpu_ring { | |
175 | struct amdgpu_device *adev; | |
176 | const struct amdgpu_ring_funcs *funcs; | |
177 | struct amdgpu_fence_driver fence_drv; | |
1b1f42d8 | 178 | struct drm_gpu_scheduler sched; |
78023016 CK |
179 | |
180 | struct amdgpu_bo *ring_obj; | |
181 | volatile uint32_t *ring; | |
182 | unsigned rptr_offs; | |
536fbf94 KW |
183 | u64 wptr; |
184 | u64 wptr_old; | |
78023016 CK |
185 | unsigned ring_size; |
186 | unsigned max_dw; | |
187 | int count_dw; | |
188 | uint64_t gpu_addr; | |
536fbf94 KW |
189 | uint64_t ptr_mask; |
190 | uint32_t buf_mask; | |
78023016 | 191 | bool ready; |
78023016 CK |
192 | u32 idx; |
193 | u32 me; | |
194 | u32 pipe; | |
195 | u32 queue; | |
196 | struct amdgpu_bo *mqd_obj; | |
f3972b53 | 197 | uint64_t mqd_gpu_addr; |
59a82d7d | 198 | void *mqd_ptr; |
34534610 | 199 | uint64_t eop_gpu_addr; |
78023016 CK |
200 | u32 doorbell_index; |
201 | bool use_doorbell; | |
2ffe31de | 202 | bool use_pollmem; |
78023016 CK |
203 | unsigned wptr_offs; |
204 | unsigned fence_offs; | |
205 | uint64_t current_ctx; | |
78023016 CK |
206 | char name[16]; |
207 | unsigned cond_exe_offs; | |
208 | u64 cond_exe_gpu_addr; | |
209 | volatile u32 *cond_exe_cpu_addr; | |
4789c463 | 210 | unsigned vm_inv_eng; |
3af81440 | 211 | struct dma_fence *vmid_wait; |
dd684d31 | 212 | bool has_compute_vm_bug; |
b2ff0e8a | 213 | |
1b1f42d8 | 214 | atomic_t num_jobs[DRM_SCHED_PRIORITY_MAX]; |
b2ff0e8a AR |
215 | struct mutex priority_mutex; |
216 | /* protected by priority_mutex */ | |
217 | int priority; | |
218 | ||
78023016 CK |
219 | #if defined(CONFIG_DEBUG_FS) |
220 | struct dentry *ent; | |
221 | #endif | |
222 | }; | |
223 | ||
0a7845db HR |
224 | #define amdgpu_ring_parse_cs(r, p, ib) ((r)->funcs->parse_cs((p), (ib))) |
225 | #define amdgpu_ring_patch_cs_in_place(r, p, ib) ((r)->funcs->patch_cs_in_place((p), (ib))) | |
226 | #define amdgpu_ring_test_ring(r) (r)->funcs->test_ring((r)) | |
227 | #define amdgpu_ring_test_ib(r, t) (r)->funcs->test_ib((r), (t)) | |
228 | #define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r)) | |
229 | #define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r)) | |
230 | #define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r)) | |
231 | #define amdgpu_ring_emit_ib(r, ib, vmid, c) (r)->funcs->emit_ib((r), (ib), (vmid), (c)) | |
232 | #define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r)) | |
233 | #define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr)) | |
234 | #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags)) | |
235 | #define amdgpu_ring_emit_gds_switch(r, v, db, ds, wb, ws, ab, as) (r)->funcs->emit_gds_switch((r), (v), (db), (ds), (wb), (ws), (ab), (as)) | |
236 | #define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r)) | |
237 | #define amdgpu_ring_emit_switch_buffer(r) (r)->funcs->emit_switch_buffer((r)) | |
238 | #define amdgpu_ring_emit_cntxcntl(r, d) (r)->funcs->emit_cntxcntl((r), (d)) | |
239 | #define amdgpu_ring_emit_rreg(r, d) (r)->funcs->emit_rreg((r), (d)) | |
240 | #define amdgpu_ring_emit_wreg(r, d, v) (r)->funcs->emit_wreg((r), (d), (v)) | |
241 | #define amdgpu_ring_emit_reg_wait(r, d, v, m) (r)->funcs->emit_reg_wait((r), (d), (v), (m)) | |
242 | #define amdgpu_ring_emit_reg_write_reg_wait(r, d0, d1, v, m) (r)->funcs->emit_reg_write_reg_wait((r), (d0), (d1), (v), (m)) | |
243 | #define amdgpu_ring_emit_tmz(r, b) (r)->funcs->emit_tmz((r), (b)) | |
244 | #define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib))) | |
245 | #define amdgpu_ring_init_cond_exec(r) (r)->funcs->init_cond_exec((r)) | |
246 | #define amdgpu_ring_patch_cond_exec(r,o) (r)->funcs->patch_cond_exec((r),(o)) | |
247 | ||
78023016 CK |
248 | int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw); |
249 | void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count); | |
250 | void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib); | |
251 | void amdgpu_ring_commit(struct amdgpu_ring *ring); | |
252 | void amdgpu_ring_undo(struct amdgpu_ring *ring); | |
b2ff0e8a | 253 | void amdgpu_ring_priority_get(struct amdgpu_ring *ring, |
1b1f42d8 | 254 | enum drm_sched_priority priority); |
b2ff0e8a | 255 | void amdgpu_ring_priority_put(struct amdgpu_ring *ring, |
1b1f42d8 | 256 | enum drm_sched_priority priority); |
78023016 | 257 | int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, |
79887142 CK |
258 | unsigned ring_size, struct amdgpu_irq_src *irq_src, |
259 | unsigned irq_type); | |
78023016 | 260 | void amdgpu_ring_fini(struct amdgpu_ring *ring); |
82853638 AD |
261 | void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring, |
262 | uint32_t reg0, uint32_t val0, | |
263 | uint32_t reg1, uint32_t val1); | |
7876fa4f CK |
264 | bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid, |
265 | struct dma_fence *fence); | |
82853638 | 266 | |
c79ecfbf ML |
267 | static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring) |
268 | { | |
269 | int i = 0; | |
e09706f4 | 270 | while (i <= ring->buf_mask) |
c79ecfbf ML |
271 | ring->ring[i++] = ring->funcs->nop; |
272 | ||
273 | } | |
78023016 | 274 | |
e8110b1c CK |
275 | static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v) |
276 | { | |
277 | if (ring->count_dw <= 0) | |
278 | DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n"); | |
279 | ring->ring[ring->wptr++ & ring->buf_mask] = v; | |
280 | ring->wptr &= ring->ptr_mask; | |
281 | ring->count_dw--; | |
282 | } | |
283 | ||
284 | static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring, | |
285 | void *src, int count_dw) | |
286 | { | |
287 | unsigned occupied, chunk1, chunk2; | |
288 | void *dst; | |
289 | ||
369421cb | 290 | if (unlikely(ring->count_dw < count_dw)) |
e8110b1c | 291 | DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n"); |
e8110b1c CK |
292 | |
293 | occupied = ring->wptr & ring->buf_mask; | |
294 | dst = (void *)&ring->ring[occupied]; | |
295 | chunk1 = ring->buf_mask + 1 - occupied; | |
296 | chunk1 = (chunk1 >= count_dw) ? count_dw: chunk1; | |
297 | chunk2 = count_dw - chunk1; | |
298 | chunk1 <<= 2; | |
299 | chunk2 <<= 2; | |
300 | ||
301 | if (chunk1) | |
302 | memcpy(dst, src, chunk1); | |
303 | ||
304 | if (chunk2) { | |
305 | src += chunk1; | |
306 | dst = (void *)ring->ring; | |
307 | memcpy(dst, src, chunk2); | |
308 | } | |
309 | ||
310 | ring->wptr += count_dw; | |
311 | ring->wptr &= ring->ptr_mask; | |
312 | ring->count_dw -= count_dw; | |
313 | } | |
314 | ||
78023016 | 315 | #endif |