1/* $NetBSD: amdgpu_ring.h,v 1.3 2021/12/19 10:59:01 riastradh Exp $ */ 2 3/* 4 * Copyright 2016 Advanced Micro Devices, Inc. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the "Software"), 8 * to deal in the Software without restriction, including without limitation 9 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 10 * and/or sell copies of the Software, and to permit persons to whom the 11 * Software is furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 22 * OTHER DEALINGS IN THE SOFTWARE. 23 * 24 * Authors: Christian K��nig 25 */ 26#ifndef __AMDGPU_RING_H__ 27#define __AMDGPU_RING_H__ 28 29#include <linux/idr.h> 30 31#include <drm/amdgpu_drm.h> 32#include <drm/gpu_scheduler.h> 33#include <drm/drm_print.h> 34 35/* max number of rings */ 36#define AMDGPU_MAX_RINGS 28 37#define AMDGPU_MAX_GFX_RINGS 2 38#define AMDGPU_MAX_COMPUTE_RINGS 8 39#define AMDGPU_MAX_VCE_RINGS 3 40#define AMDGPU_MAX_UVD_ENC_RINGS 2 41 42/* some special values for the owner field */ 43#define AMDGPU_FENCE_OWNER_UNDEFINED ((void *)0ul) 44#define AMDGPU_FENCE_OWNER_VM ((void *)1ul) 45#define AMDGPU_FENCE_OWNER_KFD ((void *)2ul) 46 47#define AMDGPU_FENCE_FLAG_64BIT (1 << 0) 48#define AMDGPU_FENCE_FLAG_INT (1 << 1) 49#define AMDGPU_FENCE_FLAG_TC_WB_ONLY (1 << 2) 50 51#define to_amdgpu_ring(s) container_of((s), struct amdgpu_ring, sched) 52 53enum amdgpu_ring_type { 54 AMDGPU_RING_TYPE_GFX, 55 AMDGPU_RING_TYPE_COMPUTE, 56 AMDGPU_RING_TYPE_SDMA, 57 AMDGPU_RING_TYPE_UVD, 58 AMDGPU_RING_TYPE_VCE, 59 AMDGPU_RING_TYPE_KIQ, 60 AMDGPU_RING_TYPE_UVD_ENC, 61 AMDGPU_RING_TYPE_VCN_DEC, 62 AMDGPU_RING_TYPE_VCN_ENC, 63 AMDGPU_RING_TYPE_VCN_JPEG 64}; 65 66struct amdgpu_device; 67struct amdgpu_ring; 68struct amdgpu_ib; 69struct amdgpu_cs_parser; 70struct amdgpu_job; 71 72/* 73 * Fences. 74 */ 75struct amdgpu_fence_driver { 76 uint64_t gpu_addr; 77 volatile uint32_t *cpu_addr; 78 /* sync_seq is protected by ring emission lock */ 79 uint32_t sync_seq; 80 atomic_t last_seq; 81 bool initialized; 82 struct amdgpu_irq_src *irq_src; 83 unsigned irq_type; 84 struct timer_list fallback_timer; 85 unsigned num_fences_mask; 86 spinlock_t lock; 87 struct dma_fence **fences; 88}; 89 90int amdgpu_fence_driver_init(struct amdgpu_device *adev); 91void amdgpu_fence_driver_fini(struct amdgpu_device *adev); 92void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring); 93 94int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring, 95 unsigned num_hw_submission); 96int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring, 97 struct amdgpu_irq_src *irq_src, 98 unsigned irq_type); 99void amdgpu_fence_driver_suspend(struct amdgpu_device *adev); 100void amdgpu_fence_driver_resume(struct amdgpu_device *adev); 101int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence, 102 unsigned flags); 103int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s); 104bool amdgpu_fence_process(struct amdgpu_ring *ring); 105int amdgpu_fence_wait_empty(struct amdgpu_ring *ring); 106signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring, 107 uint32_t wait_seq, 108 signed long timeout); 109unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring); 110 111/* 112 * Rings. 113 */ 114 115/* provided by hw blocks that expose a ring buffer for commands */ 116struct amdgpu_ring_funcs { 117 enum amdgpu_ring_type type; 118 uint32_t align_mask; 119 u32 nop; 120 bool support_64bit_ptrs; 121 bool no_user_fence; 122 unsigned vmhub; 123 unsigned extra_dw; 124 125 /* ring read/write ptr handling */ 126 u64 (*get_rptr)(struct amdgpu_ring *ring); 127 u64 (*get_wptr)(struct amdgpu_ring *ring); 128 void (*set_wptr)(struct amdgpu_ring *ring); 129 /* validating and patching of IBs */ 130 int (*parse_cs)(struct amdgpu_cs_parser *p, uint32_t ib_idx); 131 int (*patch_cs_in_place)(struct amdgpu_cs_parser *p, uint32_t ib_idx); 132 /* constants to calculate how many DW are needed for an emit */ 133 unsigned emit_frame_size; 134 unsigned emit_ib_size; 135 /* command emit functions */ 136 void (*emit_ib)(struct amdgpu_ring *ring, 137 struct amdgpu_job *job, 138 struct amdgpu_ib *ib, 139 uint32_t flags); 140 void (*emit_fence)(struct amdgpu_ring *ring, uint64_t addr, 141 uint64_t seq, unsigned flags); 142 void (*emit_pipeline_sync)(struct amdgpu_ring *ring); 143 void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vmid, 144 uint64_t pd_addr); 145 void (*emit_hdp_flush)(struct amdgpu_ring *ring); 146 void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid, 147 uint32_t gds_base, uint32_t gds_size, 148 uint32_t gws_base, uint32_t gws_size, 149 uint32_t oa_base, uint32_t oa_size); 150 /* testing functions */ 151 int (*test_ring)(struct amdgpu_ring *ring); 152 int (*test_ib)(struct amdgpu_ring *ring, long timeout); 153 /* insert NOP packets */ 154 void (*insert_nop)(struct amdgpu_ring *ring, uint32_t count); 155 void (*insert_start)(struct amdgpu_ring *ring); 156 void (*insert_end)(struct amdgpu_ring *ring); 157 /* pad the indirect buffer to the necessary number of dw */ 158 void (*pad_ib)(struct amdgpu_ring *ring, struct amdgpu_ib *ib); 159 unsigned (*init_cond_exec)(struct amdgpu_ring *ring); 160 void (*patch_cond_exec)(struct amdgpu_ring *ring, unsigned offset); 161 /* note usage for clock and power gating */ 162 void (*begin_use)(struct amdgpu_ring *ring); 163 void (*end_use)(struct amdgpu_ring *ring); 164 void (*emit_switch_buffer) (struct amdgpu_ring *ring); 165 void (*emit_cntxcntl) (struct amdgpu_ring *ring, uint32_t flags); 166 void (*emit_rreg)(struct amdgpu_ring *ring, uint32_t reg); 167 void (*emit_wreg)(struct amdgpu_ring *ring, uint32_t reg, uint32_t val); 168 void (*emit_reg_wait)(struct amdgpu_ring *ring, uint32_t reg, 169 uint32_t val, uint32_t mask); 170 void (*emit_reg_write_reg_wait)(struct amdgpu_ring *ring, 171 uint32_t reg0, uint32_t reg1, 172 uint32_t ref, uint32_t mask); 173 void (*emit_tmz)(struct amdgpu_ring *ring, bool start); 174 /* priority functions */ 175 void (*set_priority) (struct amdgpu_ring *ring, 176 enum drm_sched_priority priority); 177 /* Try to soft recover the ring to make the fence signal */ 178 void (*soft_recovery)(struct amdgpu_ring *ring, unsigned vmid); 179 int (*preempt_ib)(struct amdgpu_ring *ring); 180}; 181 182struct amdgpu_ring { 183 struct amdgpu_device *adev; 184 const struct amdgpu_ring_funcs *funcs; 185 struct amdgpu_fence_driver fence_drv; 186 struct drm_gpu_scheduler sched; 187 188 struct amdgpu_bo *ring_obj; 189 volatile uint32_t *ring; 190 unsigned rptr_offs; 191 u64 wptr; 192 u64 wptr_old; 193 unsigned ring_size; 194 unsigned max_dw; 195 int count_dw; 196 uint64_t gpu_addr; 197 uint64_t ptr_mask; 198 uint32_t buf_mask; 199 u32 idx; 200 u32 me; 201 u32 pipe; 202 u32 queue; 203 struct amdgpu_bo *mqd_obj; 204 uint64_t mqd_gpu_addr; 205 void *mqd_ptr; 206 uint64_t eop_gpu_addr; 207 u32 doorbell_index; 208 bool use_doorbell; 209 bool use_pollmem; 210 unsigned wptr_offs; 211 unsigned fence_offs; 212 uint64_t current_ctx; 213 char name[16]; 214 u32 trail_seq; 215 unsigned trail_fence_offs; 216 u64 trail_fence_gpu_addr; 217 volatile u32 *trail_fence_cpu_addr; 218 unsigned cond_exe_offs; 219 u64 cond_exe_gpu_addr; 220 volatile u32 *cond_exe_cpu_addr; 221 unsigned vm_inv_eng; 222 struct dma_fence *vmid_wait; 223 bool has_compute_vm_bug; 224 225 atomic_t num_jobs[DRM_SCHED_PRIORITY_MAX]; 226 struct mutex priority_mutex; 227 /* protected by priority_mutex */ 228 int priority; 229 230#if defined(CONFIG_DEBUG_FS) 231 struct dentry *ent; 232#endif 233}; 234 235#define amdgpu_ring_parse_cs(r, p, ib) ((r)->funcs->parse_cs((p), (ib))) 236#define amdgpu_ring_patch_cs_in_place(r, p, ib) ((r)->funcs->patch_cs_in_place((p), (ib))) 237#define amdgpu_ring_test_ring(r) (r)->funcs->test_ring((r)) 238#define amdgpu_ring_test_ib(r, t) (r)->funcs->test_ib((r), (t)) 239#define amdgpu_ring_get_rptr(r) (r)->funcs->get_rptr((r)) 240#define amdgpu_ring_get_wptr(r) (r)->funcs->get_wptr((r)) 241#define amdgpu_ring_set_wptr(r) (r)->funcs->set_wptr((r)) 242#define amdgpu_ring_emit_ib(r, job, ib, flags) ((r)->funcs->emit_ib((r), (job), (ib), (flags))) 243#define amdgpu_ring_emit_pipeline_sync(r) (r)->funcs->emit_pipeline_sync((r)) 244#define amdgpu_ring_emit_vm_flush(r, vmid, addr) (r)->funcs->emit_vm_flush((r), (vmid), (addr)) 245#define amdgpu_ring_emit_fence(r, addr, seq, flags) (r)->funcs->emit_fence((r), (addr), (seq), (flags)) 246#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)) 247#define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r)) 248#define amdgpu_ring_emit_switch_buffer(r) (r)->funcs->emit_switch_buffer((r)) 249#define amdgpu_ring_emit_cntxcntl(r, d) (r)->funcs->emit_cntxcntl((r), (d)) 250#define amdgpu_ring_emit_rreg(r, d) (r)->funcs->emit_rreg((r), (d)) 251#define amdgpu_ring_emit_wreg(r, d, v) (r)->funcs->emit_wreg((r), (d), (v)) 252#define amdgpu_ring_emit_reg_wait(r, d, v, m) (r)->funcs->emit_reg_wait((r), (d), (v), (m)) 253#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)) 254#define amdgpu_ring_emit_tmz(r, b) (r)->funcs->emit_tmz((r), (b)) 255#define amdgpu_ring_pad_ib(r, ib) ((r)->funcs->pad_ib((r), (ib))) 256#define amdgpu_ring_init_cond_exec(r) (r)->funcs->init_cond_exec((r)) 257#define amdgpu_ring_patch_cond_exec(r,o) (r)->funcs->patch_cond_exec((r),(o)) 258#define amdgpu_ring_preempt_ib(r) (r)->funcs->preempt_ib(r) 259 260int amdgpu_ring_alloc(struct amdgpu_ring *ring, unsigned ndw); 261void amdgpu_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count); 262void amdgpu_ring_generic_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib); 263void amdgpu_ring_commit(struct amdgpu_ring *ring); 264void amdgpu_ring_undo(struct amdgpu_ring *ring); 265void amdgpu_ring_priority_get(struct amdgpu_ring *ring, 266 enum drm_sched_priority priority); 267void amdgpu_ring_priority_put(struct amdgpu_ring *ring, 268 enum drm_sched_priority priority); 269int amdgpu_ring_init(struct amdgpu_device *adev, struct amdgpu_ring *ring, 270 unsigned ring_size, struct amdgpu_irq_src *irq_src, 271 unsigned irq_type); 272void amdgpu_ring_fini(struct amdgpu_ring *ring); 273void amdgpu_ring_emit_reg_write_reg_wait_helper(struct amdgpu_ring *ring, 274 uint32_t reg0, uint32_t val0, 275 uint32_t reg1, uint32_t val1); 276bool amdgpu_ring_soft_recovery(struct amdgpu_ring *ring, unsigned int vmid, 277 struct dma_fence *fence); 278 279static inline void amdgpu_ring_set_preempt_cond_exec(struct amdgpu_ring *ring, 280 bool cond_exec) 281{ 282 *ring->cond_exe_cpu_addr = cond_exec; 283} 284 285static inline void amdgpu_ring_clear_ring(struct amdgpu_ring *ring) 286{ 287 int i = 0; 288 while (i <= ring->buf_mask) 289 ring->ring[i++] = ring->funcs->nop; 290 291} 292 293static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v) 294{ 295 if (ring->count_dw <= 0) 296 DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n"); 297 ring->ring[ring->wptr++ & ring->buf_mask] = v; 298 ring->wptr &= ring->ptr_mask; 299 ring->count_dw--; 300} 301 302static inline void amdgpu_ring_write_multiple(struct amdgpu_ring *ring, 303 void *src, int count_dw) 304{ 305 unsigned occupied, chunk1, chunk2; 306 void *dst; 307 308 if (unlikely(ring->count_dw < count_dw)) 309 DRM_ERROR("amdgpu: writing more dwords to the ring than expected!\n"); 310 311 occupied = ring->wptr & ring->buf_mask; 312 dst = __UNVOLATILE(&ring->ring[occupied]); 313 chunk1 = ring->buf_mask + 1 - occupied; 314 chunk1 = (chunk1 >= count_dw) ? count_dw: chunk1; 315 chunk2 = count_dw - chunk1; 316 chunk1 <<= 2; 317 chunk2 <<= 2; 318 319 if (chunk1) 320 memcpy(dst, src, chunk1); 321 322 if (chunk2) { 323 src += chunk1; 324 dst = __UNVOLATILE(ring->ring); 325 memcpy(dst, src, chunk2); 326 } 327 328 ring->wptr += count_dw; 329 ring->wptr &= ring->ptr_mask; 330 ring->count_dw -= count_dw; 331} 332 333int amdgpu_ring_test_helper(struct amdgpu_ring *ring); 334 335#endif 336