2 * Copyright 2016 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
22 * Authors: Christian König
24 #ifndef __AMDGPU_RING_H__
25 #define __AMDGPU_RING_H__
27 #include <drm/amdgpu_drm.h>
28 #include <drm/gpu_scheduler.h>
29 #include <drm/drm_print.h>
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
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)
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)
47 #define to_amdgpu_ring(s) container_of((s), struct amdgpu_ring, sched)
49 enum amdgpu_ring_type {
51 AMDGPU_RING_TYPE_COMPUTE,
52 AMDGPU_RING_TYPE_SDMA,
56 AMDGPU_RING_TYPE_UVD_ENC,
57 AMDGPU_RING_TYPE_VCN_DEC,
58 AMDGPU_RING_TYPE_VCN_ENC,
59 AMDGPU_RING_TYPE_VCN_JPEG
65 struct amdgpu_cs_parser;
71 struct amdgpu_fence_driver {
73 volatile uint32_t *cpu_addr;
74 /* sync_seq is protected by ring emission lock */
78 struct amdgpu_irq_src *irq_src;
80 struct timer_list fallback_timer;
81 unsigned num_fences_mask;
83 struct dma_fence **fences;
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);
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,
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,
99 int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s);
100 bool 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,
104 signed long timeout);
105 unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
111 /* provided by hw blocks that expose a ring buffer for commands */
112 struct amdgpu_ring_funcs {
113 enum amdgpu_ring_type type;
116 bool support_64bit_ptrs;
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,
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 /* Try to soft recover the ring to make the fence signal */
172 void (*soft_recovery)(struct amdgpu_ring *ring, unsigned vmid);
176 struct amdgpu_device *adev;
177 const struct amdgpu_ring_funcs *funcs;
178 struct amdgpu_fence_driver fence_drv;
179 struct drm_gpu_scheduler sched;
181 struct amdgpu_bo *ring_obj;
182 volatile uint32_t *ring;
197 struct amdgpu_bo *mqd_obj;
198 uint64_t mqd_gpu_addr;
200 uint64_t eop_gpu_addr;
206 uint64_t current_ctx;
208 unsigned cond_exe_offs;
209 u64 cond_exe_gpu_addr;
210 volatile u32 *cond_exe_cpu_addr;
212 struct dma_fence *vmid_wait;
213 bool has_compute_vm_bug;
215 atomic_t num_jobs[DRM_SCHED_PRIORITY_MAX];
216 struct mutex priority_mutex;
217 /* protected by priority_mutex */
220 #if defined(CONFIG_DEBUG_FS)
225 #define amdgpu_ring_parse_cs(r, p, ib) ((r)->funcs->parse_cs((p), (ib)))
226 #define amdgpu_ring_patch_cs_in_place(r, p, ib) ((r)->funcs->patch_cs_in_place((p), (ib)))
227 #define amdgpu_ring_test_ring(r) (r)->funcs->test_ring((r))
228 #define amdgpu_ring_test_ib(r, t) (r)->funcs->test_ib((r), (t))
229 #define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r))
230 #define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r))
231 #define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r))
232 #define amdgpu_ring_emit_ib(r, ib, vmid, c) (r)->funcs->emit_ib((r), (ib), (vmid), (c))
233 #define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r))
234 #define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr))
235 #define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags))
236 #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))
237 #define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r))
238 #define amdgpu_ring_emit_switch_buffer(r) (r)->funcs->emit_switch_buffer((r))
239 #define amdgpu_ring_emit_cntxcntl(r, d) (r)->funcs->emit_cntxcntl((r), (d))
240 #define amdgpu_ring_emit_rreg(r, d) (r)->funcs->emit_rreg((r), (d))
241 #define amdgpu_ring_emit_wreg(r, d, v) (r)->funcs->emit_wreg((r), (d), (v))
242 #define amdgpu_ring_emit_reg_wait(r, d, v, m) (r)->funcs->emit_reg_wait((r), (d), (v), (m))
243 #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))
244 #define amdgpu_ring_emit_tmz(r, b) (r)->funcs->emit_tmz((r), (b))
245 #define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib)))
246 #define amdgpu_ring_init_cond_exec(r) (r)->funcs->init_cond_exec((r))
247 #define amdgpu_ring_patch_cond_exec(r,o) (r)->funcs->patch_cond_exec((r),(o))
249 int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw);
250 void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count);
251 void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib);
252 void amdgpu_ring_commit(struct amdgpu_ring *ring);
253 void amdgpu_ring_undo(struct amdgpu_ring *ring);
254 void amdgpu_ring_priority_get(struct amdgpu_ring *ring,
255 enum drm_sched_priority priority);
256 void amdgpu_ring_priority_put(struct amdgpu_ring *ring,
257 enum drm_sched_priority priority);
258 int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring,
259 unsigned ring_size, struct amdgpu_irq_src *irq_src,
261 void amdgpu_ring_fini(struct amdgpu_ring *ring);
262 void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring,
263 uint32_t reg0, uint32_t val0,
264 uint32_t reg1, uint32_t val1);
265 bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid,
266 struct dma_fence *fence);
268 static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring)
271 while (i <= ring->buf_mask)
272 ring->ring[i++] = ring->funcs->nop;
276 static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
278 if (ring->count_dw <= 0)
279 DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
280 ring->ring[ring->wptr++ & ring->buf_mask] = v;
281 ring->wptr &= ring->ptr_mask;
285 static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring,
286 void *src, int count_dw)
288 unsigned occupied, chunk1, chunk2;
291 if (unlikely(ring->count_dw < count_dw))
292 DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n");
294 occupied = ring->wptr & ring->buf_mask;
295 dst = (void *)&ring->ring[occupied];
296 chunk1 = ring->buf_mask + 1 - occupied;
297 chunk1 = (chunk1 >= count_dw) ? count_dw: chunk1;
298 chunk2 = count_dw - chunk1;
303 memcpy(dst, src, chunk1);
307 dst = (void *)ring->ring;
308 memcpy(dst, src, chunk2);
311 ring->wptr += count_dw;
312 ring->wptr &= ring->ptr_mask;
313 ring->count_dw -= count_dw;