1 /* 1 /*
2 * Copyright 2014 Advanced Micro Devices, Inc. 2 * Copyright 2014 Advanced Micro Devices, Inc.
3 * 3 *
4 * Permission is hereby granted, free of charg 4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated docume 5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction 6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, pub 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to 8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to 9 * Software is furnished to do so, subject to the following conditions:
10 * 10 *
11 * The above copyright notice and this permiss 11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the S 12 * all copies or substantial portions of the Software.
13 * 13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT W 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE W 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINF 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIA 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CO 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE. 20 * OTHER DEALINGS IN THE SOFTWARE.
21 */ 21 */
22 22
23 #ifndef KFD_IOCTL_H_INCLUDED 23 #ifndef KFD_IOCTL_H_INCLUDED
24 #define KFD_IOCTL_H_INCLUDED 24 #define KFD_IOCTL_H_INCLUDED
25 25
26 #include <drm/drm.h> 26 #include <drm/drm.h>
27 #include <linux/ioctl.h> 27 #include <linux/ioctl.h>
28 28
29 /* 29 /*
30 * - 1.1 - initial version 30 * - 1.1 - initial version
31 * - 1.3 - Add SMI events support 31 * - 1.3 - Add SMI events support
32 * - 1.4 - Indicate new SRAM EDC bit in device 32 * - 1.4 - Indicate new SRAM EDC bit in device properties
33 * - 1.5 - Add SVM API 33 * - 1.5 - Add SVM API
34 * - 1.6 - Query clear flags in SVM get_attr A 34 * - 1.6 - Query clear flags in SVM get_attr API
35 * - 1.7 - Checkpoint Restore (CRIU) API 35 * - 1.7 - Checkpoint Restore (CRIU) API
36 * - 1.8 - CRIU - Support for SDMA transfers w 36 * - 1.8 - CRIU - Support for SDMA transfers with GTT BOs
37 * - 1.9 - Add available memory ioctl 37 * - 1.9 - Add available memory ioctl
38 * - 1.10 - Add SMI profiler event log 38 * - 1.10 - Add SMI profiler event log
39 * - 1.11 - Add unified memory for ctx save/re 39 * - 1.11 - Add unified memory for ctx save/restore area
40 * - 1.12 - Add DMA buf export ioctl 40 * - 1.12 - Add DMA buf export ioctl
41 * - 1.13 - Add debugger API 41 * - 1.13 - Add debugger API
42 * - 1.14 - Update kfd_event_data 42 * - 1.14 - Update kfd_event_data
43 * - 1.15 - Enable managing mappings in comput 43 * - 1.15 - Enable managing mappings in compute VMs with GEM_VA ioctl
44 * - 1.16 - Add contiguous VRAM allocation fla 44 * - 1.16 - Add contiguous VRAM allocation flag
45 */ 45 */
46 #define KFD_IOCTL_MAJOR_VERSION 1 46 #define KFD_IOCTL_MAJOR_VERSION 1
47 #define KFD_IOCTL_MINOR_VERSION 16 47 #define KFD_IOCTL_MINOR_VERSION 16
48 48
49 struct kfd_ioctl_get_version_args { 49 struct kfd_ioctl_get_version_args {
50 __u32 major_version; /* from KFD */ 50 __u32 major_version; /* from KFD */
51 __u32 minor_version; /* from KFD */ 51 __u32 minor_version; /* from KFD */
52 }; 52 };
53 53
54 /* For kfd_ioctl_create_queue_args.queue_type. 54 /* For kfd_ioctl_create_queue_args.queue_type. */
55 #define KFD_IOC_QUEUE_TYPE_COMPUTE 55 #define KFD_IOC_QUEUE_TYPE_COMPUTE 0x0
56 #define KFD_IOC_QUEUE_TYPE_SDMA 56 #define KFD_IOC_QUEUE_TYPE_SDMA 0x1
57 #define KFD_IOC_QUEUE_TYPE_COMPUTE_AQL 57 #define KFD_IOC_QUEUE_TYPE_COMPUTE_AQL 0x2
58 #define KFD_IOC_QUEUE_TYPE_SDMA_XGMI 58 #define KFD_IOC_QUEUE_TYPE_SDMA_XGMI 0x3
59 59
60 #define KFD_MAX_QUEUE_PERCENTAGE 100 60 #define KFD_MAX_QUEUE_PERCENTAGE 100
61 #define KFD_MAX_QUEUE_PRIORITY 15 61 #define KFD_MAX_QUEUE_PRIORITY 15
62 62
63 struct kfd_ioctl_create_queue_args { 63 struct kfd_ioctl_create_queue_args {
64 __u64 ring_base_address; /* to 64 __u64 ring_base_address; /* to KFD */
65 __u64 write_pointer_address; /* fro 65 __u64 write_pointer_address; /* from KFD */
66 __u64 read_pointer_address; /* fro 66 __u64 read_pointer_address; /* from KFD */
67 __u64 doorbell_offset; /* from KFD */ 67 __u64 doorbell_offset; /* from KFD */
68 68
69 __u32 ring_size; /* to 69 __u32 ring_size; /* to KFD */
70 __u32 gpu_id; /* to KFD */ 70 __u32 gpu_id; /* to KFD */
71 __u32 queue_type; /* to 71 __u32 queue_type; /* to KFD */
72 __u32 queue_percentage; /* to KFD */ 72 __u32 queue_percentage; /* to KFD */
73 __u32 queue_priority; /* to KFD */ 73 __u32 queue_priority; /* to KFD */
74 __u32 queue_id; /* from KFD */ 74 __u32 queue_id; /* from KFD */
75 75
76 __u64 eop_buffer_address; /* to 76 __u64 eop_buffer_address; /* to KFD */
77 __u64 eop_buffer_size; /* to KFD */ 77 __u64 eop_buffer_size; /* to KFD */
78 __u64 ctx_save_restore_address; /* to 78 __u64 ctx_save_restore_address; /* to KFD */
79 __u32 ctx_save_restore_size; /* to 79 __u32 ctx_save_restore_size; /* to KFD */
80 __u32 ctl_stack_size; /* to 80 __u32 ctl_stack_size; /* to KFD */
81 }; 81 };
82 82
83 struct kfd_ioctl_destroy_queue_args { 83 struct kfd_ioctl_destroy_queue_args {
84 __u32 queue_id; /* to KFD */ 84 __u32 queue_id; /* to KFD */
85 __u32 pad; 85 __u32 pad;
86 }; 86 };
87 87
88 struct kfd_ioctl_update_queue_args { 88 struct kfd_ioctl_update_queue_args {
89 __u64 ring_base_address; /* to 89 __u64 ring_base_address; /* to KFD */
90 90
91 __u32 queue_id; /* to KFD */ 91 __u32 queue_id; /* to KFD */
92 __u32 ring_size; /* to 92 __u32 ring_size; /* to KFD */
93 __u32 queue_percentage; /* to KFD */ 93 __u32 queue_percentage; /* to KFD */
94 __u32 queue_priority; /* to KFD */ 94 __u32 queue_priority; /* to KFD */
95 }; 95 };
96 96
97 struct kfd_ioctl_set_cu_mask_args { 97 struct kfd_ioctl_set_cu_mask_args {
98 __u32 queue_id; /* to KFD */ 98 __u32 queue_id; /* to KFD */
99 __u32 num_cu_mask; /* to 99 __u32 num_cu_mask; /* to KFD */
100 __u64 cu_mask_ptr; /* to 100 __u64 cu_mask_ptr; /* to KFD */
101 }; 101 };
102 102
103 struct kfd_ioctl_get_queue_wave_state_args { 103 struct kfd_ioctl_get_queue_wave_state_args {
104 __u64 ctl_stack_address; /* to 104 __u64 ctl_stack_address; /* to KFD */
105 __u32 ctl_stack_used_size; /* fro 105 __u32 ctl_stack_used_size; /* from KFD */
106 __u32 save_area_used_size; /* fro 106 __u32 save_area_used_size; /* from KFD */
107 __u32 queue_id; /* to 107 __u32 queue_id; /* to KFD */
108 __u32 pad; 108 __u32 pad;
109 }; 109 };
110 110
111 struct kfd_ioctl_get_available_memory_args { 111 struct kfd_ioctl_get_available_memory_args {
112 __u64 available; /* from KFD */ 112 __u64 available; /* from KFD */
113 __u32 gpu_id; /* to KFD */ 113 __u32 gpu_id; /* to KFD */
114 __u32 pad; 114 __u32 pad;
115 }; 115 };
116 116
117 struct kfd_dbg_device_info_entry { 117 struct kfd_dbg_device_info_entry {
118 __u64 exception_status; 118 __u64 exception_status;
119 __u64 lds_base; 119 __u64 lds_base;
120 __u64 lds_limit; 120 __u64 lds_limit;
121 __u64 scratch_base; 121 __u64 scratch_base;
122 __u64 scratch_limit; 122 __u64 scratch_limit;
123 __u64 gpuvm_base; 123 __u64 gpuvm_base;
124 __u64 gpuvm_limit; 124 __u64 gpuvm_limit;
125 __u32 gpu_id; 125 __u32 gpu_id;
126 __u32 location_id; 126 __u32 location_id;
127 __u32 vendor_id; 127 __u32 vendor_id;
128 __u32 device_id; 128 __u32 device_id;
129 __u32 revision_id; 129 __u32 revision_id;
130 __u32 subsystem_vendor_id; 130 __u32 subsystem_vendor_id;
131 __u32 subsystem_device_id; 131 __u32 subsystem_device_id;
132 __u32 fw_version; 132 __u32 fw_version;
133 __u32 gfx_target_version; 133 __u32 gfx_target_version;
134 __u32 simd_count; 134 __u32 simd_count;
135 __u32 max_waves_per_simd; 135 __u32 max_waves_per_simd;
136 __u32 array_count; 136 __u32 array_count;
137 __u32 simd_arrays_per_engine; 137 __u32 simd_arrays_per_engine;
138 __u32 num_xcc; 138 __u32 num_xcc;
139 __u32 capability; 139 __u32 capability;
140 __u32 debug_prop; 140 __u32 debug_prop;
141 }; 141 };
142 142
143 /* For kfd_ioctl_set_memory_policy_args.defaul 143 /* For kfd_ioctl_set_memory_policy_args.default_policy and alternate_policy */
144 #define KFD_IOC_CACHE_POLICY_COHERENT 0 144 #define KFD_IOC_CACHE_POLICY_COHERENT 0
145 #define KFD_IOC_CACHE_POLICY_NONCOHERENT 1 145 #define KFD_IOC_CACHE_POLICY_NONCOHERENT 1
146 146
147 struct kfd_ioctl_set_memory_policy_args { 147 struct kfd_ioctl_set_memory_policy_args {
148 __u64 alternate_aperture_base; /* to 148 __u64 alternate_aperture_base; /* to KFD */
149 __u64 alternate_aperture_size; /* to 149 __u64 alternate_aperture_size; /* to KFD */
150 150
151 __u32 gpu_id; /* to 151 __u32 gpu_id; /* to KFD */
152 __u32 default_policy; /* to 152 __u32 default_policy; /* to KFD */
153 __u32 alternate_policy; /* to 153 __u32 alternate_policy; /* to KFD */
154 __u32 pad; 154 __u32 pad;
155 }; 155 };
156 156
157 /* 157 /*
158 * All counters are monotonic. They are used f 158 * All counters are monotonic. They are used for profiling of compute jobs.
159 * The profiling is done by userspace. 159 * The profiling is done by userspace.
160 * 160 *
161 * In case of GPU reset, the counter should no 161 * In case of GPU reset, the counter should not be affected.
162 */ 162 */
163 163
164 struct kfd_ioctl_get_clock_counters_args { 164 struct kfd_ioctl_get_clock_counters_args {
165 __u64 gpu_clock_counter; /* fro 165 __u64 gpu_clock_counter; /* from KFD */
166 __u64 cpu_clock_counter; /* fro 166 __u64 cpu_clock_counter; /* from KFD */
167 __u64 system_clock_counter; /* fro 167 __u64 system_clock_counter; /* from KFD */
168 __u64 system_clock_freq; /* fro 168 __u64 system_clock_freq; /* from KFD */
169 169
170 __u32 gpu_id; /* to KFD */ 170 __u32 gpu_id; /* to KFD */
171 __u32 pad; 171 __u32 pad;
172 }; 172 };
173 173
174 struct kfd_process_device_apertures { 174 struct kfd_process_device_apertures {
175 __u64 lds_base; /* from KFD */ 175 __u64 lds_base; /* from KFD */
176 __u64 lds_limit; /* fro 176 __u64 lds_limit; /* from KFD */
177 __u64 scratch_base; /* fro 177 __u64 scratch_base; /* from KFD */
178 __u64 scratch_limit; /* fro 178 __u64 scratch_limit; /* from KFD */
179 __u64 gpuvm_base; /* fro 179 __u64 gpuvm_base; /* from KFD */
180 __u64 gpuvm_limit; /* fro 180 __u64 gpuvm_limit; /* from KFD */
181 __u32 gpu_id; /* from KFD */ 181 __u32 gpu_id; /* from KFD */
182 __u32 pad; 182 __u32 pad;
183 }; 183 };
184 184
185 /* 185 /*
186 * AMDKFD_IOC_GET_PROCESS_APERTURES is depreca 186 * AMDKFD_IOC_GET_PROCESS_APERTURES is deprecated. Use
187 * AMDKFD_IOC_GET_PROCESS_APERTURES_NEW instea 187 * AMDKFD_IOC_GET_PROCESS_APERTURES_NEW instead, which supports an
188 * unlimited number of GPUs. 188 * unlimited number of GPUs.
189 */ 189 */
190 #define NUM_OF_SUPPORTED_GPUS 7 190 #define NUM_OF_SUPPORTED_GPUS 7
191 struct kfd_ioctl_get_process_apertures_args { 191 struct kfd_ioctl_get_process_apertures_args {
192 struct kfd_process_device_apertures 192 struct kfd_process_device_apertures
193 process_apertures[NUM_ 193 process_apertures[NUM_OF_SUPPORTED_GPUS];/* from KFD */
194 194
195 /* from KFD, should be in the range [1 195 /* from KFD, should be in the range [1 - NUM_OF_SUPPORTED_GPUS] */
196 __u32 num_of_nodes; 196 __u32 num_of_nodes;
197 __u32 pad; 197 __u32 pad;
198 }; 198 };
199 199
200 struct kfd_ioctl_get_process_apertures_new_arg 200 struct kfd_ioctl_get_process_apertures_new_args {
201 /* User allocated. Pointer to struct k 201 /* User allocated. Pointer to struct kfd_process_device_apertures
202 * filled in by Kernel 202 * filled in by Kernel
203 */ 203 */
204 __u64 kfd_process_device_apertures_ptr 204 __u64 kfd_process_device_apertures_ptr;
205 /* to KFD - indicates amount of memory 205 /* to KFD - indicates amount of memory present in
206 * kfd_process_device_apertures_ptr 206 * kfd_process_device_apertures_ptr
207 * from KFD - Number of entries filled 207 * from KFD - Number of entries filled by KFD.
208 */ 208 */
209 __u32 num_of_nodes; 209 __u32 num_of_nodes;
210 __u32 pad; 210 __u32 pad;
211 }; 211 };
212 212
213 #define MAX_ALLOWED_NUM_POINTS 100 213 #define MAX_ALLOWED_NUM_POINTS 100
214 #define MAX_ALLOWED_AW_BUFF_SIZE 4096 214 #define MAX_ALLOWED_AW_BUFF_SIZE 4096
215 #define MAX_ALLOWED_WAC_BUFF_SIZE 128 215 #define MAX_ALLOWED_WAC_BUFF_SIZE 128
216 216
217 struct kfd_ioctl_dbg_register_args { 217 struct kfd_ioctl_dbg_register_args {
218 __u32 gpu_id; /* to KFD */ 218 __u32 gpu_id; /* to KFD */
219 __u32 pad; 219 __u32 pad;
220 }; 220 };
221 221
222 struct kfd_ioctl_dbg_unregister_args { 222 struct kfd_ioctl_dbg_unregister_args {
223 __u32 gpu_id; /* to KFD */ 223 __u32 gpu_id; /* to KFD */
224 __u32 pad; 224 __u32 pad;
225 }; 225 };
226 226
227 struct kfd_ioctl_dbg_address_watch_args { 227 struct kfd_ioctl_dbg_address_watch_args {
228 __u64 content_ptr; /* a p 228 __u64 content_ptr; /* a pointer to the actual content */
229 __u32 gpu_id; /* to KFD */ 229 __u32 gpu_id; /* to KFD */
230 __u32 buf_size_in_bytes; /*incl 230 __u32 buf_size_in_bytes; /*including gpu_id and buf_size */
231 }; 231 };
232 232
233 struct kfd_ioctl_dbg_wave_control_args { 233 struct kfd_ioctl_dbg_wave_control_args {
234 __u64 content_ptr; /* a p 234 __u64 content_ptr; /* a pointer to the actual content */
235 __u32 gpu_id; /* to KFD */ 235 __u32 gpu_id; /* to KFD */
236 __u32 buf_size_in_bytes; /*incl 236 __u32 buf_size_in_bytes; /*including gpu_id and buf_size */
237 }; 237 };
238 238
239 #define KFD_INVALID_FD 0xffffffff 239 #define KFD_INVALID_FD 0xffffffff
240 240
241 /* Matching HSA_EVENTTYPE */ 241 /* Matching HSA_EVENTTYPE */
242 #define KFD_IOC_EVENT_SIGNAL 242 #define KFD_IOC_EVENT_SIGNAL 0
243 #define KFD_IOC_EVENT_NODECHANGE 243 #define KFD_IOC_EVENT_NODECHANGE 1
244 #define KFD_IOC_EVENT_DEVICESTATECHANGE 244 #define KFD_IOC_EVENT_DEVICESTATECHANGE 2
245 #define KFD_IOC_EVENT_HW_EXCEPTION 245 #define KFD_IOC_EVENT_HW_EXCEPTION 3
246 #define KFD_IOC_EVENT_SYSTEM_EVENT 246 #define KFD_IOC_EVENT_SYSTEM_EVENT 4
247 #define KFD_IOC_EVENT_DEBUG_EVENT 247 #define KFD_IOC_EVENT_DEBUG_EVENT 5
248 #define KFD_IOC_EVENT_PROFILE_EVENT 248 #define KFD_IOC_EVENT_PROFILE_EVENT 6
249 #define KFD_IOC_EVENT_QUEUE_EVENT 249 #define KFD_IOC_EVENT_QUEUE_EVENT 7
250 #define KFD_IOC_EVENT_MEMORY 250 #define KFD_IOC_EVENT_MEMORY 8
251 251
252 #define KFD_IOC_WAIT_RESULT_COMPLETE 252 #define KFD_IOC_WAIT_RESULT_COMPLETE 0
253 #define KFD_IOC_WAIT_RESULT_TIMEOUT 253 #define KFD_IOC_WAIT_RESULT_TIMEOUT 1
254 #define KFD_IOC_WAIT_RESULT_FAIL 254 #define KFD_IOC_WAIT_RESULT_FAIL 2
255 255
256 #define KFD_SIGNAL_EVENT_LIMIT 256 #define KFD_SIGNAL_EVENT_LIMIT 4096
257 257
258 /* For kfd_event_data.hw_exception_data.reset_ 258 /* For kfd_event_data.hw_exception_data.reset_type. */
259 #define KFD_HW_EXCEPTION_WHOLE_GPU_RESET 259 #define KFD_HW_EXCEPTION_WHOLE_GPU_RESET 0
260 #define KFD_HW_EXCEPTION_PER_ENGINE_RESET 260 #define KFD_HW_EXCEPTION_PER_ENGINE_RESET 1
261 261
262 /* For kfd_event_data.hw_exception_data.reset_ 262 /* For kfd_event_data.hw_exception_data.reset_cause. */
263 #define KFD_HW_EXCEPTION_GPU_HANG 0 263 #define KFD_HW_EXCEPTION_GPU_HANG 0
264 #define KFD_HW_EXCEPTION_ECC 1 264 #define KFD_HW_EXCEPTION_ECC 1
265 265
266 /* For kfd_hsa_memory_exception_data.ErrorType 266 /* For kfd_hsa_memory_exception_data.ErrorType */
267 #define KFD_MEM_ERR_NO_RAS 0 267 #define KFD_MEM_ERR_NO_RAS 0
268 #define KFD_MEM_ERR_SRAM_ECC 1 268 #define KFD_MEM_ERR_SRAM_ECC 1
269 #define KFD_MEM_ERR_POISON_CONSUMED 2 269 #define KFD_MEM_ERR_POISON_CONSUMED 2
270 #define KFD_MEM_ERR_GPU_HANG 3 270 #define KFD_MEM_ERR_GPU_HANG 3
271 271
272 struct kfd_ioctl_create_event_args { 272 struct kfd_ioctl_create_event_args {
273 __u64 event_page_offset; /* fro 273 __u64 event_page_offset; /* from KFD */
274 __u32 event_trigger_data; /* fro 274 __u32 event_trigger_data; /* from KFD - signal events only */
275 __u32 event_type; /* to 275 __u32 event_type; /* to KFD */
276 __u32 auto_reset; /* to 276 __u32 auto_reset; /* to KFD */
277 __u32 node_id; /* to KFD - on 277 __u32 node_id; /* to KFD - only valid for certain
278 278 event types */
279 __u32 event_id; /* from KFD */ 279 __u32 event_id; /* from KFD */
280 __u32 event_slot_index; /* from KFD */ 280 __u32 event_slot_index; /* from KFD */
281 }; 281 };
282 282
283 struct kfd_ioctl_destroy_event_args { 283 struct kfd_ioctl_destroy_event_args {
284 __u32 event_id; /* to KFD */ 284 __u32 event_id; /* to KFD */
285 __u32 pad; 285 __u32 pad;
286 }; 286 };
287 287
288 struct kfd_ioctl_set_event_args { 288 struct kfd_ioctl_set_event_args {
289 __u32 event_id; /* to KFD */ 289 __u32 event_id; /* to KFD */
290 __u32 pad; 290 __u32 pad;
291 }; 291 };
292 292
293 struct kfd_ioctl_reset_event_args { 293 struct kfd_ioctl_reset_event_args {
294 __u32 event_id; /* to KFD */ 294 __u32 event_id; /* to KFD */
295 __u32 pad; 295 __u32 pad;
296 }; 296 };
297 297
298 struct kfd_memory_exception_failure { 298 struct kfd_memory_exception_failure {
299 __u32 NotPresent; /* Page not pr 299 __u32 NotPresent; /* Page not present or supervisor privilege */
300 __u32 ReadOnly; /* Write access to a r 300 __u32 ReadOnly; /* Write access to a read-only page */
301 __u32 NoExecute; /* Execute acc 301 __u32 NoExecute; /* Execute access to a page marked NX */
302 __u32 imprecise; /* Can't deter 302 __u32 imprecise; /* Can't determine the exact fault address */
303 }; 303 };
304 304
305 /* memory exception data */ 305 /* memory exception data */
306 struct kfd_hsa_memory_exception_data { 306 struct kfd_hsa_memory_exception_data {
307 struct kfd_memory_exception_failure fa 307 struct kfd_memory_exception_failure failure;
308 __u64 va; 308 __u64 va;
309 __u32 gpu_id; 309 __u32 gpu_id;
310 __u32 ErrorType; /* 0 = no RAS error, 310 __u32 ErrorType; /* 0 = no RAS error,
311 * 1 = ECC_SRAM, 311 * 1 = ECC_SRAM,
312 * 2 = Link_SYNFLOOD 312 * 2 = Link_SYNFLOOD (poison),
313 * 3 = GPU hang (not 313 * 3 = GPU hang (not attributable to a specific cause),
314 * other values reser 314 * other values reserved
315 */ 315 */
316 }; 316 };
317 317
318 /* hw exception data */ 318 /* hw exception data */
319 struct kfd_hsa_hw_exception_data { 319 struct kfd_hsa_hw_exception_data {
320 __u32 reset_type; 320 __u32 reset_type;
321 __u32 reset_cause; 321 __u32 reset_cause;
322 __u32 memory_lost; 322 __u32 memory_lost;
323 __u32 gpu_id; 323 __u32 gpu_id;
324 }; 324 };
325 325
326 /* hsa signal event data */ 326 /* hsa signal event data */
327 struct kfd_hsa_signal_event_data { 327 struct kfd_hsa_signal_event_data {
328 __u64 last_event_age; /* to and from 328 __u64 last_event_age; /* to and from KFD */
329 }; 329 };
330 330
331 /* Event data */ 331 /* Event data */
332 struct kfd_event_data { 332 struct kfd_event_data {
333 union { 333 union {
334 /* From KFD */ 334 /* From KFD */
335 struct kfd_hsa_memory_exceptio 335 struct kfd_hsa_memory_exception_data memory_exception_data;
336 struct kfd_hsa_hw_exception_da 336 struct kfd_hsa_hw_exception_data hw_exception_data;
337 /* To and From KFD */ 337 /* To and From KFD */
338 struct kfd_hsa_signal_event_da 338 struct kfd_hsa_signal_event_data signal_event_data;
339 }; 339 };
340 __u64 kfd_event_data_ext; /* poi 340 __u64 kfd_event_data_ext; /* pointer to an extension structure
341 for 341 for future exception types */
342 __u32 event_id; /* to KFD */ 342 __u32 event_id; /* to KFD */
343 __u32 pad; 343 __u32 pad;
344 }; 344 };
345 345
346 struct kfd_ioctl_wait_events_args { 346 struct kfd_ioctl_wait_events_args {
347 __u64 events_ptr; /* poi 347 __u64 events_ptr; /* pointed to struct
348 kfd 348 kfd_event_data array, to KFD */
349 __u32 num_events; /* to 349 __u32 num_events; /* to KFD */
350 __u32 wait_for_all; /* to 350 __u32 wait_for_all; /* to KFD */
351 __u32 timeout; /* to KFD */ 351 __u32 timeout; /* to KFD */
352 __u32 wait_result; /* fro 352 __u32 wait_result; /* from KFD */
353 }; 353 };
354 354
355 struct kfd_ioctl_set_scratch_backing_va_args { 355 struct kfd_ioctl_set_scratch_backing_va_args {
356 __u64 va_addr; /* to KFD */ 356 __u64 va_addr; /* to KFD */
357 __u32 gpu_id; /* to KFD */ 357 __u32 gpu_id; /* to KFD */
358 __u32 pad; 358 __u32 pad;
359 }; 359 };
360 360
361 struct kfd_ioctl_get_tile_config_args { 361 struct kfd_ioctl_get_tile_config_args {
362 /* to KFD: pointer to tile array */ 362 /* to KFD: pointer to tile array */
363 __u64 tile_config_ptr; 363 __u64 tile_config_ptr;
364 /* to KFD: pointer to macro tile array 364 /* to KFD: pointer to macro tile array */
365 __u64 macro_tile_config_ptr; 365 __u64 macro_tile_config_ptr;
366 /* to KFD: array size allocated by use 366 /* to KFD: array size allocated by user mode
367 * from KFD: array size filled by kern 367 * from KFD: array size filled by kernel
368 */ 368 */
369 __u32 num_tile_configs; 369 __u32 num_tile_configs;
370 /* to KFD: array size allocated by use 370 /* to KFD: array size allocated by user mode
371 * from KFD: array size filled by kern 371 * from KFD: array size filled by kernel
372 */ 372 */
373 __u32 num_macro_tile_configs; 373 __u32 num_macro_tile_configs;
374 374
375 __u32 gpu_id; /* to KFD */ 375 __u32 gpu_id; /* to KFD */
376 __u32 gb_addr_config; /* from KFD */ 376 __u32 gb_addr_config; /* from KFD */
377 __u32 num_banks; /* fro 377 __u32 num_banks; /* from KFD */
378 __u32 num_ranks; /* fro 378 __u32 num_ranks; /* from KFD */
379 /* struct size can be extended later i 379 /* struct size can be extended later if needed
380 * without breaking ABI compatibility 380 * without breaking ABI compatibility
381 */ 381 */
382 }; 382 };
383 383
384 struct kfd_ioctl_set_trap_handler_args { 384 struct kfd_ioctl_set_trap_handler_args {
385 __u64 tba_addr; /* to KFD */ 385 __u64 tba_addr; /* to KFD */
386 __u64 tma_addr; /* to KFD */ 386 __u64 tma_addr; /* to KFD */
387 __u32 gpu_id; /* to KFD */ 387 __u32 gpu_id; /* to KFD */
388 __u32 pad; 388 __u32 pad;
389 }; 389 };
390 390
391 struct kfd_ioctl_acquire_vm_args { 391 struct kfd_ioctl_acquire_vm_args {
392 __u32 drm_fd; /* to KFD */ 392 __u32 drm_fd; /* to KFD */
393 __u32 gpu_id; /* to KFD */ 393 __u32 gpu_id; /* to KFD */
394 }; 394 };
395 395
396 /* Allocation flags: memory types */ 396 /* Allocation flags: memory types */
397 #define KFD_IOC_ALLOC_MEM_FLAGS_VRAM 397 #define KFD_IOC_ALLOC_MEM_FLAGS_VRAM (1 << 0)
398 #define KFD_IOC_ALLOC_MEM_FLAGS_GTT 398 #define KFD_IOC_ALLOC_MEM_FLAGS_GTT (1 << 1)
399 #define KFD_IOC_ALLOC_MEM_FLAGS_USERPTR 399 #define KFD_IOC_ALLOC_MEM_FLAGS_USERPTR (1 << 2)
400 #define KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL 400 #define KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL (1 << 3)
401 #define KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP 401 #define KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP (1 << 4)
402 /* Allocation flags: attributes/access options 402 /* Allocation flags: attributes/access options */
403 #define KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE 403 #define KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE (1 << 31)
404 #define KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE 404 #define KFD_IOC_ALLOC_MEM_FLAGS_EXECUTABLE (1 << 30)
405 #define KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC 405 #define KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC (1 << 29)
406 #define KFD_IOC_ALLOC_MEM_FLAGS_NO_SUBSTITUTE 406 #define KFD_IOC_ALLOC_MEM_FLAGS_NO_SUBSTITUTE (1 << 28)
407 #define KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM 407 #define KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM (1 << 27)
408 #define KFD_IOC_ALLOC_MEM_FLAGS_COHERENT 408 #define KFD_IOC_ALLOC_MEM_FLAGS_COHERENT (1 << 26)
409 #define KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED 409 #define KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED (1 << 25)
410 #define KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT 410 #define KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT (1 << 24)
411 #define KFD_IOC_ALLOC_MEM_FLAGS_CONTIGUOUS 411 #define KFD_IOC_ALLOC_MEM_FLAGS_CONTIGUOUS (1 << 23)
412 412
413 /* Allocate memory for later SVM (shared virtu 413 /* Allocate memory for later SVM (shared virtual memory) mapping.
414 * 414 *
415 * @va_addr: virtual address of the memory 415 * @va_addr: virtual address of the memory to be allocated
416 * all later mappings on all GPU 416 * all later mappings on all GPUs will use this address
417 * @size: size in bytes 417 * @size: size in bytes
418 * @handle: buffer handle returned to use 418 * @handle: buffer handle returned to user mode, used to refer to
419 * this allocation for mapping, 419 * this allocation for mapping, unmapping and freeing
420 * @mmap_offset: for CPU-mapping the allocatio 420 * @mmap_offset: for CPU-mapping the allocation by mmapping a render node
421 * for userptrs this is overload 421 * for userptrs this is overloaded to specify the CPU address
422 * @gpu_id: device identifier 422 * @gpu_id: device identifier
423 * @flags: memory type and attributes. S 423 * @flags: memory type and attributes. See KFD_IOC_ALLOC_MEM_FLAGS above
424 */ 424 */
425 struct kfd_ioctl_alloc_memory_of_gpu_args { 425 struct kfd_ioctl_alloc_memory_of_gpu_args {
426 __u64 va_addr; /* to KFD */ 426 __u64 va_addr; /* to KFD */
427 __u64 size; /* to KFD */ 427 __u64 size; /* to KFD */
428 __u64 handle; /* from KFD */ 428 __u64 handle; /* from KFD */
429 __u64 mmap_offset; /* to KFD (use 429 __u64 mmap_offset; /* to KFD (userptr), from KFD (mmap offset) */
430 __u32 gpu_id; /* to KFD */ 430 __u32 gpu_id; /* to KFD */
431 __u32 flags; 431 __u32 flags;
432 }; 432 };
433 433
434 /* Free memory allocated with kfd_ioctl_alloc_ 434 /* Free memory allocated with kfd_ioctl_alloc_memory_of_gpu
435 * 435 *
436 * @handle: memory handle returned by alloc 436 * @handle: memory handle returned by alloc
437 */ 437 */
438 struct kfd_ioctl_free_memory_of_gpu_args { 438 struct kfd_ioctl_free_memory_of_gpu_args {
439 __u64 handle; /* to KFD */ 439 __u64 handle; /* to KFD */
440 }; 440 };
441 441
442 /* Map memory to one or more GPUs 442 /* Map memory to one or more GPUs
443 * 443 *
444 * @handle: memory handle retur 444 * @handle: memory handle returned by alloc
445 * @device_ids_array_ptr: array of gpu_ids (_ 445 * @device_ids_array_ptr: array of gpu_ids (__u32 per device)
446 * @n_devices: number of devices i 446 * @n_devices: number of devices in the array
447 * @n_success: number of devices m 447 * @n_success: number of devices mapped successfully
448 * 448 *
449 * @n_success returns information to the calle 449 * @n_success returns information to the caller how many devices from
450 * the start of the array have mapped the buff 450 * the start of the array have mapped the buffer successfully. It can
451 * be passed into a subsequent retry call to s 451 * be passed into a subsequent retry call to skip those devices. For
452 * the first call the caller should initialize 452 * the first call the caller should initialize it to 0.
453 * 453 *
454 * If the ioctl completes with return code 0 ( 454 * If the ioctl completes with return code 0 (success), n_success ==
455 * n_devices. 455 * n_devices.
456 */ 456 */
457 struct kfd_ioctl_map_memory_to_gpu_args { 457 struct kfd_ioctl_map_memory_to_gpu_args {
458 __u64 handle; /* to 458 __u64 handle; /* to KFD */
459 __u64 device_ids_array_ptr; /* to 459 __u64 device_ids_array_ptr; /* to KFD */
460 __u32 n_devices; /* to 460 __u32 n_devices; /* to KFD */
461 __u32 n_success; /* to/ 461 __u32 n_success; /* to/from KFD */
462 }; 462 };
463 463
464 /* Unmap memory from one or more GPUs 464 /* Unmap memory from one or more GPUs
465 * 465 *
466 * same arguments as for mapping 466 * same arguments as for mapping
467 */ 467 */
468 struct kfd_ioctl_unmap_memory_from_gpu_args { 468 struct kfd_ioctl_unmap_memory_from_gpu_args {
469 __u64 handle; /* to 469 __u64 handle; /* to KFD */
470 __u64 device_ids_array_ptr; /* to 470 __u64 device_ids_array_ptr; /* to KFD */
471 __u32 n_devices; /* to 471 __u32 n_devices; /* to KFD */
472 __u32 n_success; /* to/ 472 __u32 n_success; /* to/from KFD */
473 }; 473 };
474 474
475 /* Allocate GWS for specific queue 475 /* Allocate GWS for specific queue
476 * 476 *
477 * @queue_id: queue's id that GWS is alloca 477 * @queue_id: queue's id that GWS is allocated for
478 * @num_gws: how many GWS to allocate 478 * @num_gws: how many GWS to allocate
479 * @first_gws: index of the first GWS alloca 479 * @first_gws: index of the first GWS allocated.
480 * only support contiguous GWS a 480 * only support contiguous GWS allocation
481 */ 481 */
482 struct kfd_ioctl_alloc_queue_gws_args { 482 struct kfd_ioctl_alloc_queue_gws_args {
483 __u32 queue_id; /* to KFD */ 483 __u32 queue_id; /* to KFD */
484 __u32 num_gws; /* to KFD */ 484 __u32 num_gws; /* to KFD */
485 __u32 first_gws; /* from KFD */ 485 __u32 first_gws; /* from KFD */
486 __u32 pad; 486 __u32 pad;
487 }; 487 };
488 488
489 struct kfd_ioctl_get_dmabuf_info_args { 489 struct kfd_ioctl_get_dmabuf_info_args {
490 __u64 size; /* from KFD */ 490 __u64 size; /* from KFD */
491 __u64 metadata_ptr; /* to KFD */ 491 __u64 metadata_ptr; /* to KFD */
492 __u32 metadata_size; /* to KFD (spa 492 __u32 metadata_size; /* to KFD (space allocated by user)
493 * from KFD (a 493 * from KFD (actual metadata size)
494 */ 494 */
495 __u32 gpu_id; /* from KFD */ 495 __u32 gpu_id; /* from KFD */
496 __u32 flags; /* from KFD (K 496 __u32 flags; /* from KFD (KFD_IOC_ALLOC_MEM_FLAGS) */
497 __u32 dmabuf_fd; /* to KFD */ 497 __u32 dmabuf_fd; /* to KFD */
498 }; 498 };
499 499
500 struct kfd_ioctl_import_dmabuf_args { 500 struct kfd_ioctl_import_dmabuf_args {
501 __u64 va_addr; /* to KFD */ 501 __u64 va_addr; /* to KFD */
502 __u64 handle; /* from KFD */ 502 __u64 handle; /* from KFD */
503 __u32 gpu_id; /* to KFD */ 503 __u32 gpu_id; /* to KFD */
504 __u32 dmabuf_fd; /* to KFD */ 504 __u32 dmabuf_fd; /* to KFD */
505 }; 505 };
506 506
507 struct kfd_ioctl_export_dmabuf_args { 507 struct kfd_ioctl_export_dmabuf_args {
508 __u64 handle; /* to KFD */ 508 __u64 handle; /* to KFD */
509 __u32 flags; /* to KFD */ 509 __u32 flags; /* to KFD */
510 __u32 dmabuf_fd; /* from KFD */ 510 __u32 dmabuf_fd; /* from KFD */
511 }; 511 };
512 512
513 /* 513 /*
514 * KFD SMI(System Management Interface) events 514 * KFD SMI(System Management Interface) events
515 */ 515 */
516 enum kfd_smi_event { 516 enum kfd_smi_event {
517 KFD_SMI_EVENT_NONE = 0, /* not used */ 517 KFD_SMI_EVENT_NONE = 0, /* not used */
518 KFD_SMI_EVENT_VMFAULT = 1, /* event st 518 KFD_SMI_EVENT_VMFAULT = 1, /* event start counting at 1 */
519 KFD_SMI_EVENT_THERMAL_THROTTLE = 2, 519 KFD_SMI_EVENT_THERMAL_THROTTLE = 2,
520 KFD_SMI_EVENT_GPU_PRE_RESET = 3, 520 KFD_SMI_EVENT_GPU_PRE_RESET = 3,
521 KFD_SMI_EVENT_GPU_POST_RESET = 4, 521 KFD_SMI_EVENT_GPU_POST_RESET = 4,
522 KFD_SMI_EVENT_MIGRATE_START = 5, 522 KFD_SMI_EVENT_MIGRATE_START = 5,
523 KFD_SMI_EVENT_MIGRATE_END = 6, 523 KFD_SMI_EVENT_MIGRATE_END = 6,
524 KFD_SMI_EVENT_PAGE_FAULT_START = 7, 524 KFD_SMI_EVENT_PAGE_FAULT_START = 7,
525 KFD_SMI_EVENT_PAGE_FAULT_END = 8, 525 KFD_SMI_EVENT_PAGE_FAULT_END = 8,
526 KFD_SMI_EVENT_QUEUE_EVICTION = 9, 526 KFD_SMI_EVENT_QUEUE_EVICTION = 9,
527 KFD_SMI_EVENT_QUEUE_RESTORE = 10, 527 KFD_SMI_EVENT_QUEUE_RESTORE = 10,
528 KFD_SMI_EVENT_UNMAP_FROM_GPU = 11, 528 KFD_SMI_EVENT_UNMAP_FROM_GPU = 11,
529 529
530 /* 530 /*
531 * max event number, as a flag bit to 531 * max event number, as a flag bit to get events from all processes,
532 * this requires super user permission 532 * this requires super user permission, otherwise will not be able to
533 * receive event from any process. Wit 533 * receive event from any process. Without this flag to receive events
534 * from same process. 534 * from same process.
535 */ 535 */
536 KFD_SMI_EVENT_ALL_PROCESS = 64 536 KFD_SMI_EVENT_ALL_PROCESS = 64
537 }; 537 };
538 538
539 enum KFD_MIGRATE_TRIGGERS { 539 enum KFD_MIGRATE_TRIGGERS {
540 KFD_MIGRATE_TRIGGER_PREFETCH, 540 KFD_MIGRATE_TRIGGER_PREFETCH,
541 KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU, 541 KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU,
542 KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU, 542 KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU,
543 KFD_MIGRATE_TRIGGER_TTM_EVICTION 543 KFD_MIGRATE_TRIGGER_TTM_EVICTION
544 }; 544 };
545 545
546 enum KFD_QUEUE_EVICTION_TRIGGERS { 546 enum KFD_QUEUE_EVICTION_TRIGGERS {
547 KFD_QUEUE_EVICTION_TRIGGER_SVM, 547 KFD_QUEUE_EVICTION_TRIGGER_SVM,
548 KFD_QUEUE_EVICTION_TRIGGER_USERPTR, 548 KFD_QUEUE_EVICTION_TRIGGER_USERPTR,
549 KFD_QUEUE_EVICTION_TRIGGER_TTM, 549 KFD_QUEUE_EVICTION_TRIGGER_TTM,
550 KFD_QUEUE_EVICTION_TRIGGER_SUSPEND, 550 KFD_QUEUE_EVICTION_TRIGGER_SUSPEND,
551 KFD_QUEUE_EVICTION_CRIU_CHECKPOINT, 551 KFD_QUEUE_EVICTION_CRIU_CHECKPOINT,
552 KFD_QUEUE_EVICTION_CRIU_RESTORE 552 KFD_QUEUE_EVICTION_CRIU_RESTORE
553 }; 553 };
554 554
555 enum KFD_SVM_UNMAP_TRIGGERS { 555 enum KFD_SVM_UNMAP_TRIGGERS {
556 KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY, 556 KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY,
557 KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY_MIGRA 557 KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY_MIGRATE,
558 KFD_SVM_UNMAP_TRIGGER_UNMAP_FROM_CPU 558 KFD_SVM_UNMAP_TRIGGER_UNMAP_FROM_CPU
559 }; 559 };
560 560
561 #define KFD_SMI_EVENT_MASK_FROM_INDEX(i) (1ULL 561 #define KFD_SMI_EVENT_MASK_FROM_INDEX(i) (1ULL << ((i) - 1))
562 #define KFD_SMI_EVENT_MSG_SIZE 96 562 #define KFD_SMI_EVENT_MSG_SIZE 96
563 563
564 struct kfd_ioctl_smi_events_args { 564 struct kfd_ioctl_smi_events_args {
565 __u32 gpuid; /* to KFD */ 565 __u32 gpuid; /* to KFD */
566 __u32 anon_fd; /* from KFD */ 566 __u32 anon_fd; /* from KFD */
567 }; 567 };
568 568
569 /********************************************* 569 /**************************************************************************************************
570 * CRIU IOCTLs (Checkpoint Restore In Userspac 570 * CRIU IOCTLs (Checkpoint Restore In Userspace)
571 * 571 *
572 * When checkpointing a process, the userspace 572 * When checkpointing a process, the userspace application will perform:
573 * 1. PROCESS_INFO op to determine current pro 573 * 1. PROCESS_INFO op to determine current process information. This pauses execution and evicts
574 * all the queues. 574 * all the queues.
575 * 2. CHECKPOINT op to checkpoint process cont 575 * 2. CHECKPOINT op to checkpoint process contents (BOs, queues, events, svm-ranges)
576 * 3. UNPAUSE op to un-evict all the queues 576 * 3. UNPAUSE op to un-evict all the queues
577 * 577 *
578 * When restoring a process, the CRIU userspac 578 * When restoring a process, the CRIU userspace application will perform:
579 * 579 *
580 * 1. RESTORE op to restore process contents 580 * 1. RESTORE op to restore process contents
581 * 2. RESUME op to start the process 581 * 2. RESUME op to start the process
582 * 582 *
583 * Note: Queues are forced into an evicted sta 583 * Note: Queues are forced into an evicted state after a successful PROCESS_INFO. User
584 * application needs to perform an UNPAUSE ope 584 * application needs to perform an UNPAUSE operation after calling PROCESS_INFO.
585 */ 585 */
586 586
587 enum kfd_criu_op { 587 enum kfd_criu_op {
588 KFD_CRIU_OP_PROCESS_INFO, 588 KFD_CRIU_OP_PROCESS_INFO,
589 KFD_CRIU_OP_CHECKPOINT, 589 KFD_CRIU_OP_CHECKPOINT,
590 KFD_CRIU_OP_UNPAUSE, 590 KFD_CRIU_OP_UNPAUSE,
591 KFD_CRIU_OP_RESTORE, 591 KFD_CRIU_OP_RESTORE,
592 KFD_CRIU_OP_RESUME, 592 KFD_CRIU_OP_RESUME,
593 }; 593 };
594 594
595 /** 595 /**
596 * kfd_ioctl_criu_args - Arguments perform CRI 596 * kfd_ioctl_criu_args - Arguments perform CRIU operation
597 * @devices: [in/out] User pointer 597 * @devices: [in/out] User pointer to memory location for devices information.
598 * This is an array of ty 598 * This is an array of type kfd_criu_device_bucket.
599 * @bos: [in/out] User pointer 599 * @bos: [in/out] User pointer to memory location for BOs information
600 * This is an array of ty 600 * This is an array of type kfd_criu_bo_bucket.
601 * @priv_data: [in/out] User pointer 601 * @priv_data: [in/out] User pointer to memory location for private data
602 * @priv_data_size: [in/out] Size of priv_ 602 * @priv_data_size: [in/out] Size of priv_data in bytes
603 * @num_devices: [in/out] Number of GPU 603 * @num_devices: [in/out] Number of GPUs used by process. Size of @devices array.
604 * @num_bos [in/out] Number of BOs 604 * @num_bos [in/out] Number of BOs used by process. Size of @bos array.
605 * @num_objects: [in/out] Number of obj 605 * @num_objects: [in/out] Number of objects used by process. Objects are opaque to
606 * user applicat 606 * user application.
607 * @pid: [in/out] PID of the pr 607 * @pid: [in/out] PID of the process being checkpointed
608 * @op [in] Type of operation 608 * @op [in] Type of operation (kfd_criu_op)
609 * 609 *
610 * Return: 0 on success, -errno on failure 610 * Return: 0 on success, -errno on failure
611 */ 611 */
612 struct kfd_ioctl_criu_args { 612 struct kfd_ioctl_criu_args {
613 __u64 devices; /* Used during 613 __u64 devices; /* Used during ops: CHECKPOINT, RESTORE */
614 __u64 bos; /* Used during 614 __u64 bos; /* Used during ops: CHECKPOINT, RESTORE */
615 __u64 priv_data; /* Used during 615 __u64 priv_data; /* Used during ops: CHECKPOINT, RESTORE */
616 __u64 priv_data_size; /* Used during 616 __u64 priv_data_size; /* Used during ops: PROCESS_INFO, RESTORE */
617 __u32 num_devices; /* Used during 617 __u32 num_devices; /* Used during ops: PROCESS_INFO, RESTORE */
618 __u32 num_bos; /* Used during 618 __u32 num_bos; /* Used during ops: PROCESS_INFO, RESTORE */
619 __u32 num_objects; /* Used during 619 __u32 num_objects; /* Used during ops: PROCESS_INFO, RESTORE */
620 __u32 pid; /* Used during 620 __u32 pid; /* Used during ops: PROCESS_INFO, RESUME */
621 __u32 op; 621 __u32 op;
622 }; 622 };
623 623
624 struct kfd_criu_device_bucket { 624 struct kfd_criu_device_bucket {
625 __u32 user_gpu_id; 625 __u32 user_gpu_id;
626 __u32 actual_gpu_id; 626 __u32 actual_gpu_id;
627 __u32 drm_fd; 627 __u32 drm_fd;
628 __u32 pad; 628 __u32 pad;
629 }; 629 };
630 630
631 struct kfd_criu_bo_bucket { 631 struct kfd_criu_bo_bucket {
632 __u64 addr; 632 __u64 addr;
633 __u64 size; 633 __u64 size;
634 __u64 offset; 634 __u64 offset;
635 __u64 restored_offset; /* During re 635 __u64 restored_offset; /* During restore, updated offset for BO */
636 __u32 gpu_id; /* This is t 636 __u32 gpu_id; /* This is the user_gpu_id */
637 __u32 alloc_flags; 637 __u32 alloc_flags;
638 __u32 dmabuf_fd; 638 __u32 dmabuf_fd;
639 __u32 pad; 639 __u32 pad;
640 }; 640 };
641 641
642 /* CRIU IOCTLs - END */ 642 /* CRIU IOCTLs - END */
643 /********************************************* 643 /**************************************************************************************************/
644 644
645 /* Register offset inside the remapped mmio pa 645 /* Register offset inside the remapped mmio page
646 */ 646 */
647 enum kfd_mmio_remap { 647 enum kfd_mmio_remap {
648 KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL = 0, 648 KFD_MMIO_REMAP_HDP_MEM_FLUSH_CNTL = 0,
649 KFD_MMIO_REMAP_HDP_REG_FLUSH_CNTL = 4, 649 KFD_MMIO_REMAP_HDP_REG_FLUSH_CNTL = 4,
650 }; 650 };
651 651
652 /* Guarantee host access to memory */ 652 /* Guarantee host access to memory */
653 #define KFD_IOCTL_SVM_FLAG_HOST_ACCESS 0x00000 653 #define KFD_IOCTL_SVM_FLAG_HOST_ACCESS 0x00000001
654 /* Fine grained coherency between all devices 654 /* Fine grained coherency between all devices with access */
655 #define KFD_IOCTL_SVM_FLAG_COHERENT 0x00000 655 #define KFD_IOCTL_SVM_FLAG_COHERENT 0x00000002
656 /* Use any GPU in same hive as preferred devic 656 /* Use any GPU in same hive as preferred device */
657 #define KFD_IOCTL_SVM_FLAG_HIVE_LOCAL 0x00000 657 #define KFD_IOCTL_SVM_FLAG_HIVE_LOCAL 0x00000004
658 /* GPUs only read, allows replication */ 658 /* GPUs only read, allows replication */
659 #define KFD_IOCTL_SVM_FLAG_GPU_RO 0x00000 659 #define KFD_IOCTL_SVM_FLAG_GPU_RO 0x00000008
660 /* Allow execution on GPU */ 660 /* Allow execution on GPU */
661 #define KFD_IOCTL_SVM_FLAG_GPU_EXEC 0x00000 661 #define KFD_IOCTL_SVM_FLAG_GPU_EXEC 0x00000010
662 /* GPUs mostly read, may allow similar optimiz 662 /* GPUs mostly read, may allow similar optimizations as RO, but writes fault */
663 #define KFD_IOCTL_SVM_FLAG_GPU_READ_MOSTLY 663 #define KFD_IOCTL_SVM_FLAG_GPU_READ_MOSTLY 0x00000020
664 /* Keep GPU memory mapping always valid as if 664 /* Keep GPU memory mapping always valid as if XNACK is disable */
665 #define KFD_IOCTL_SVM_FLAG_GPU_ALWAYS_MAPPED 665 #define KFD_IOCTL_SVM_FLAG_GPU_ALWAYS_MAPPED 0x00000040
666 /* Fine grained coherency between all devices 666 /* Fine grained coherency between all devices using device-scope atomics */
667 #define KFD_IOCTL_SVM_FLAG_EXT_COHERENT 667 #define KFD_IOCTL_SVM_FLAG_EXT_COHERENT 0x00000080
668 668
669 /** 669 /**
670 * kfd_ioctl_svm_op - SVM ioctl operations 670 * kfd_ioctl_svm_op - SVM ioctl operations
671 * 671 *
672 * @KFD_IOCTL_SVM_OP_SET_ATTR: Modify one or m 672 * @KFD_IOCTL_SVM_OP_SET_ATTR: Modify one or more attributes
673 * @KFD_IOCTL_SVM_OP_GET_ATTR: Query one or mo 673 * @KFD_IOCTL_SVM_OP_GET_ATTR: Query one or more attributes
674 */ 674 */
675 enum kfd_ioctl_svm_op { 675 enum kfd_ioctl_svm_op {
676 KFD_IOCTL_SVM_OP_SET_ATTR, 676 KFD_IOCTL_SVM_OP_SET_ATTR,
677 KFD_IOCTL_SVM_OP_GET_ATTR 677 KFD_IOCTL_SVM_OP_GET_ATTR
678 }; 678 };
679 679
680 /** kfd_ioctl_svm_location - Enum for preferre 680 /** kfd_ioctl_svm_location - Enum for preferred and prefetch locations
681 * 681 *
682 * GPU IDs are used to specify GPUs as preferr 682 * GPU IDs are used to specify GPUs as preferred and prefetch locations.
683 * Below definitions are used for system memor 683 * Below definitions are used for system memory or for leaving the preferred
684 * location unspecified. 684 * location unspecified.
685 */ 685 */
686 enum kfd_ioctl_svm_location { 686 enum kfd_ioctl_svm_location {
687 KFD_IOCTL_SVM_LOCATION_SYSMEM = 0, 687 KFD_IOCTL_SVM_LOCATION_SYSMEM = 0,
688 KFD_IOCTL_SVM_LOCATION_UNDEFINED = 0xf 688 KFD_IOCTL_SVM_LOCATION_UNDEFINED = 0xffffffff
689 }; 689 };
690 690
691 /** 691 /**
692 * kfd_ioctl_svm_attr_type - SVM attribute typ 692 * kfd_ioctl_svm_attr_type - SVM attribute types
693 * 693 *
694 * @KFD_IOCTL_SVM_ATTR_PREFERRED_LOC: gpuid of 694 * @KFD_IOCTL_SVM_ATTR_PREFERRED_LOC: gpuid of the preferred location, 0 for
695 * system m 695 * system memory
696 * @KFD_IOCTL_SVM_ATTR_PREFETCH_LOC: gpuid of 696 * @KFD_IOCTL_SVM_ATTR_PREFETCH_LOC: gpuid of the prefetch location, 0 for
697 * system me 697 * system memory. Setting this triggers an
698 * immediate 698 * immediate prefetch (migration).
699 * @KFD_IOCTL_SVM_ATTR_ACCESS: 699 * @KFD_IOCTL_SVM_ATTR_ACCESS:
700 * @KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE: 700 * @KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
701 * @KFD_IOCTL_SVM_ATTR_NO_ACCESS: specify memo 701 * @KFD_IOCTL_SVM_ATTR_NO_ACCESS: specify memory access for the gpuid given
702 * by the attri 702 * by the attribute value
703 * @KFD_IOCTL_SVM_ATTR_SET_FLAGS: bitmask of f 703 * @KFD_IOCTL_SVM_ATTR_SET_FLAGS: bitmask of flags to set (see
704 * KFD_IOCTL_SV 704 * KFD_IOCTL_SVM_FLAG_...)
705 * @KFD_IOCTL_SVM_ATTR_CLR_FLAGS: bitmask of f 705 * @KFD_IOCTL_SVM_ATTR_CLR_FLAGS: bitmask of flags to clear
706 * @KFD_IOCTL_SVM_ATTR_GRANULARITY: migration 706 * @KFD_IOCTL_SVM_ATTR_GRANULARITY: migration granularity
707 * (log2 num 707 * (log2 num pages)
708 */ 708 */
709 enum kfd_ioctl_svm_attr_type { 709 enum kfd_ioctl_svm_attr_type {
710 KFD_IOCTL_SVM_ATTR_PREFERRED_LOC, 710 KFD_IOCTL_SVM_ATTR_PREFERRED_LOC,
711 KFD_IOCTL_SVM_ATTR_PREFETCH_LOC, 711 KFD_IOCTL_SVM_ATTR_PREFETCH_LOC,
712 KFD_IOCTL_SVM_ATTR_ACCESS, 712 KFD_IOCTL_SVM_ATTR_ACCESS,
713 KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE, 713 KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE,
714 KFD_IOCTL_SVM_ATTR_NO_ACCESS, 714 KFD_IOCTL_SVM_ATTR_NO_ACCESS,
715 KFD_IOCTL_SVM_ATTR_SET_FLAGS, 715 KFD_IOCTL_SVM_ATTR_SET_FLAGS,
716 KFD_IOCTL_SVM_ATTR_CLR_FLAGS, 716 KFD_IOCTL_SVM_ATTR_CLR_FLAGS,
717 KFD_IOCTL_SVM_ATTR_GRANULARITY 717 KFD_IOCTL_SVM_ATTR_GRANULARITY
718 }; 718 };
719 719
720 /** 720 /**
721 * kfd_ioctl_svm_attribute - Attributes as pai 721 * kfd_ioctl_svm_attribute - Attributes as pairs of type and value
722 * 722 *
723 * The meaning of the @value depends on the at 723 * The meaning of the @value depends on the attribute type.
724 * 724 *
725 * @type: attribute type (see enum @kfd_ioctl_ 725 * @type: attribute type (see enum @kfd_ioctl_svm_attr_type)
726 * @value: attribute value 726 * @value: attribute value
727 */ 727 */
728 struct kfd_ioctl_svm_attribute { 728 struct kfd_ioctl_svm_attribute {
729 __u32 type; 729 __u32 type;
730 __u32 value; 730 __u32 value;
731 }; 731 };
732 732
733 /** 733 /**
734 * kfd_ioctl_svm_args - Arguments for SVM ioct 734 * kfd_ioctl_svm_args - Arguments for SVM ioctl
735 * 735 *
736 * @op specifies the operation to perform (see 736 * @op specifies the operation to perform (see enum
737 * @kfd_ioctl_svm_op). @start_addr and @size 737 * @kfd_ioctl_svm_op). @start_addr and @size are common for all
738 * operations. 738 * operations.
739 * 739 *
740 * A variable number of attributes can be give 740 * A variable number of attributes can be given in @attrs.
741 * @nattr specifies the number of attributes. 741 * @nattr specifies the number of attributes. New attributes can be
742 * added in the future without breaking the AB 742 * added in the future without breaking the ABI. If unknown attributes
743 * are given, the function returns -EINVAL. 743 * are given, the function returns -EINVAL.
744 * 744 *
745 * @KFD_IOCTL_SVM_OP_SET_ATTR sets attributes 745 * @KFD_IOCTL_SVM_OP_SET_ATTR sets attributes for a virtual address
746 * range. It may overlap existing virtual addr 746 * range. It may overlap existing virtual address ranges. If it does,
747 * the existing ranges will be split such that 747 * the existing ranges will be split such that the attribute changes
748 * only apply to the specified address range. 748 * only apply to the specified address range.
749 * 749 *
750 * @KFD_IOCTL_SVM_OP_GET_ATTR returns the inte 750 * @KFD_IOCTL_SVM_OP_GET_ATTR returns the intersection of attributes
751 * over all memory in the given range and retu 751 * over all memory in the given range and returns the result as the
752 * attribute value. If different pages have di 752 * attribute value. If different pages have different preferred or
753 * prefetch locations, 0xffffffff will be retu 753 * prefetch locations, 0xffffffff will be returned for
754 * @KFD_IOCTL_SVM_ATTR_PREFERRED_LOC or 754 * @KFD_IOCTL_SVM_ATTR_PREFERRED_LOC or
755 * @KFD_IOCTL_SVM_ATTR_PREFETCH_LOC resepctive 755 * @KFD_IOCTL_SVM_ATTR_PREFETCH_LOC resepctively. For
756 * @KFD_IOCTL_SVM_ATTR_SET_FLAGS, flags of all 756 * @KFD_IOCTL_SVM_ATTR_SET_FLAGS, flags of all pages will be
757 * aggregated by bitwise AND. That means, a fl 757 * aggregated by bitwise AND. That means, a flag will be set in the
758 * output, if that flag is set for all pages i 758 * output, if that flag is set for all pages in the range. For
759 * @KFD_IOCTL_SVM_ATTR_CLR_FLAGS, flags of all 759 * @KFD_IOCTL_SVM_ATTR_CLR_FLAGS, flags of all pages will be
760 * aggregated by bitwise NOR. That means, a fl 760 * aggregated by bitwise NOR. That means, a flag will be set in the
761 * output, if that flag is clear for all pages 761 * output, if that flag is clear for all pages in the range.
762 * The minimum migration granularity throughou 762 * The minimum migration granularity throughout the range will be
763 * returned for @KFD_IOCTL_SVM_ATTR_GRANULARIT 763 * returned for @KFD_IOCTL_SVM_ATTR_GRANULARITY.
764 * 764 *
765 * Querying of accessibility attributes works 765 * Querying of accessibility attributes works by initializing the
766 * attribute type to @KFD_IOCTL_SVM_ATTR_ACCES 766 * attribute type to @KFD_IOCTL_SVM_ATTR_ACCESS and the value to the
767 * GPUID being queried. Multiple attributes ca 767 * GPUID being queried. Multiple attributes can be given to allow
768 * querying multiple GPUIDs. The ioctl functio 768 * querying multiple GPUIDs. The ioctl function overwrites the
769 * attribute type to indicate the access for t 769 * attribute type to indicate the access for the specified GPU.
770 */ 770 */
771 struct kfd_ioctl_svm_args { 771 struct kfd_ioctl_svm_args {
772 __u64 start_addr; 772 __u64 start_addr;
773 __u64 size; 773 __u64 size;
774 __u32 op; 774 __u32 op;
775 __u32 nattr; 775 __u32 nattr;
776 /* Variable length array of attributes 776 /* Variable length array of attributes */
777 struct kfd_ioctl_svm_attribute attrs[] 777 struct kfd_ioctl_svm_attribute attrs[];
778 }; 778 };
779 779
780 /** 780 /**
781 * kfd_ioctl_set_xnack_mode_args - Arguments f 781 * kfd_ioctl_set_xnack_mode_args - Arguments for set_xnack_mode
782 * 782 *
783 * @xnack_enabled: [in/out] Whether to e 783 * @xnack_enabled: [in/out] Whether to enable XNACK mode for this process
784 * 784 *
785 * @xnack_enabled indicates whether recoverabl 785 * @xnack_enabled indicates whether recoverable page faults should be
786 * enabled for the current process. 0 means di 786 * enabled for the current process. 0 means disabled, positive means
787 * enabled, negative means leave unchanged. If 787 * enabled, negative means leave unchanged. If enabled, virtual address
788 * translations on GFXv9 and later AMD GPUs ca 788 * translations on GFXv9 and later AMD GPUs can return XNACK and retry
789 * the access until a valid PTE is available. 789 * the access until a valid PTE is available. This is used to implement
790 * device page faults. 790 * device page faults.
791 * 791 *
792 * On output, @xnack_enabled returns the (new) 792 * On output, @xnack_enabled returns the (new) current mode (0 or
793 * positive). Therefore, a negative input valu 793 * positive). Therefore, a negative input value can be used to query
794 * the current mode without changing it. 794 * the current mode without changing it.
795 * 795 *
796 * The XNACK mode fundamentally changes the wa 796 * The XNACK mode fundamentally changes the way SVM managed memory works
797 * in the driver, with subtle effects on appli 797 * in the driver, with subtle effects on application performance and
798 * functionality. 798 * functionality.
799 * 799 *
800 * Enabling XNACK mode requires shader program 800 * Enabling XNACK mode requires shader programs to be compiled
801 * differently. Furthermore, not all GPUs supp 801 * differently. Furthermore, not all GPUs support changing the mode
802 * per-process. Therefore changing the mode is 802 * per-process. Therefore changing the mode is only allowed while no
803 * user mode queues exist in the process. This 803 * user mode queues exist in the process. This ensure that no shader
804 * code is running that may be compiled for th 804 * code is running that may be compiled for the wrong mode. And GPUs
805 * that cannot change to the requested mode wi 805 * that cannot change to the requested mode will prevent the XNACK
806 * mode from occurring. All GPUs used by the p 806 * mode from occurring. All GPUs used by the process must be in the
807 * same XNACK mode. 807 * same XNACK mode.
808 * 808 *
809 * GFXv8 or older GPUs do not support 48 bit v 809 * GFXv8 or older GPUs do not support 48 bit virtual addresses or SVM.
810 * Therefore those GPUs are not considered for 810 * Therefore those GPUs are not considered for the XNACK mode switch.
811 * 811 *
812 * Return: 0 on success, -errno on failure 812 * Return: 0 on success, -errno on failure
813 */ 813 */
814 struct kfd_ioctl_set_xnack_mode_args { 814 struct kfd_ioctl_set_xnack_mode_args {
815 __s32 xnack_enabled; 815 __s32 xnack_enabled;
816 }; 816 };
817 817
818 /* Wave launch override modes */ 818 /* Wave launch override modes */
819 enum kfd_dbg_trap_override_mode { 819 enum kfd_dbg_trap_override_mode {
820 KFD_DBG_TRAP_OVERRIDE_OR = 0, 820 KFD_DBG_TRAP_OVERRIDE_OR = 0,
821 KFD_DBG_TRAP_OVERRIDE_REPLACE = 1 821 KFD_DBG_TRAP_OVERRIDE_REPLACE = 1
822 }; 822 };
823 823
824 /* Wave launch overrides */ 824 /* Wave launch overrides */
825 enum kfd_dbg_trap_mask { 825 enum kfd_dbg_trap_mask {
826 KFD_DBG_TRAP_MASK_FP_INVALID = 1, 826 KFD_DBG_TRAP_MASK_FP_INVALID = 1,
827 KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL = 827 KFD_DBG_TRAP_MASK_FP_INPUT_DENORMAL = 2,
828 KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO = 828 KFD_DBG_TRAP_MASK_FP_DIVIDE_BY_ZERO = 4,
829 KFD_DBG_TRAP_MASK_FP_OVERFLOW = 8, 829 KFD_DBG_TRAP_MASK_FP_OVERFLOW = 8,
830 KFD_DBG_TRAP_MASK_FP_UNDERFLOW = 16, 830 KFD_DBG_TRAP_MASK_FP_UNDERFLOW = 16,
831 KFD_DBG_TRAP_MASK_FP_INEXACT = 32, 831 KFD_DBG_TRAP_MASK_FP_INEXACT = 32,
832 KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO = 832 KFD_DBG_TRAP_MASK_INT_DIVIDE_BY_ZERO = 64,
833 KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH = 833 KFD_DBG_TRAP_MASK_DBG_ADDRESS_WATCH = 128,
834 KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION 834 KFD_DBG_TRAP_MASK_DBG_MEMORY_VIOLATION = 256,
835 KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START = 835 KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_START = (1 << 30),
836 KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END = ( 836 KFD_DBG_TRAP_MASK_TRAP_ON_WAVE_END = (1 << 31)
837 }; 837 };
838 838
839 /* Wave launch modes */ 839 /* Wave launch modes */
840 enum kfd_dbg_trap_wave_launch_mode { 840 enum kfd_dbg_trap_wave_launch_mode {
841 KFD_DBG_TRAP_WAVE_LAUNCH_MODE_NORMAL = 841 KFD_DBG_TRAP_WAVE_LAUNCH_MODE_NORMAL = 0,
842 KFD_DBG_TRAP_WAVE_LAUNCH_MODE_HALT = 1 842 KFD_DBG_TRAP_WAVE_LAUNCH_MODE_HALT = 1,
843 KFD_DBG_TRAP_WAVE_LAUNCH_MODE_DEBUG = 843 KFD_DBG_TRAP_WAVE_LAUNCH_MODE_DEBUG = 3
844 }; 844 };
845 845
846 /* Address watch modes */ 846 /* Address watch modes */
847 enum kfd_dbg_trap_address_watch_mode { 847 enum kfd_dbg_trap_address_watch_mode {
848 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_READ = 848 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_READ = 0,
849 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_NONREA 849 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_NONREAD = 1,
850 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_ATOMIC 850 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_ATOMIC = 2,
851 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_ALL = 851 KFD_DBG_TRAP_ADDRESS_WATCH_MODE_ALL = 3
852 }; 852 };
853 853
854 /* Additional wave settings */ 854 /* Additional wave settings */
855 enum kfd_dbg_trap_flags { 855 enum kfd_dbg_trap_flags {
856 KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP = 1, 856 KFD_DBG_TRAP_FLAG_SINGLE_MEM_OP = 1,
857 KFD_DBG_TRAP_FLAG_SINGLE_ALU_OP = 2, 857 KFD_DBG_TRAP_FLAG_SINGLE_ALU_OP = 2,
858 }; 858 };
859 859
860 /* Trap exceptions */ 860 /* Trap exceptions */
861 enum kfd_dbg_trap_exception_code { 861 enum kfd_dbg_trap_exception_code {
862 EC_NONE = 0, 862 EC_NONE = 0,
863 /* per queue */ 863 /* per queue */
864 EC_QUEUE_WAVE_ABORT = 1, 864 EC_QUEUE_WAVE_ABORT = 1,
865 EC_QUEUE_WAVE_TRAP = 2, 865 EC_QUEUE_WAVE_TRAP = 2,
866 EC_QUEUE_WAVE_MATH_ERROR = 3, 866 EC_QUEUE_WAVE_MATH_ERROR = 3,
867 EC_QUEUE_WAVE_ILLEGAL_INSTRUCTION = 4, 867 EC_QUEUE_WAVE_ILLEGAL_INSTRUCTION = 4,
868 EC_QUEUE_WAVE_MEMORY_VIOLATION = 5, 868 EC_QUEUE_WAVE_MEMORY_VIOLATION = 5,
869 EC_QUEUE_WAVE_APERTURE_VIOLATION = 6, 869 EC_QUEUE_WAVE_APERTURE_VIOLATION = 6,
870 EC_QUEUE_PACKET_DISPATCH_DIM_INVALID = 870 EC_QUEUE_PACKET_DISPATCH_DIM_INVALID = 16,
871 EC_QUEUE_PACKET_DISPATCH_GROUP_SEGMENT 871 EC_QUEUE_PACKET_DISPATCH_GROUP_SEGMENT_SIZE_INVALID = 17,
872 EC_QUEUE_PACKET_DISPATCH_CODE_INVALID 872 EC_QUEUE_PACKET_DISPATCH_CODE_INVALID = 18,
873 EC_QUEUE_PACKET_RESERVED = 19, 873 EC_QUEUE_PACKET_RESERVED = 19,
874 EC_QUEUE_PACKET_UNSUPPORTED = 20, 874 EC_QUEUE_PACKET_UNSUPPORTED = 20,
875 EC_QUEUE_PACKET_DISPATCH_WORK_GROUP_SI 875 EC_QUEUE_PACKET_DISPATCH_WORK_GROUP_SIZE_INVALID = 21,
876 EC_QUEUE_PACKET_DISPATCH_REGISTER_INVA 876 EC_QUEUE_PACKET_DISPATCH_REGISTER_INVALID = 22,
877 EC_QUEUE_PACKET_VENDOR_UNSUPPORTED = 2 877 EC_QUEUE_PACKET_VENDOR_UNSUPPORTED = 23,
878 EC_QUEUE_PREEMPTION_ERROR = 30, 878 EC_QUEUE_PREEMPTION_ERROR = 30,
879 EC_QUEUE_NEW = 31, 879 EC_QUEUE_NEW = 31,
880 /* per device */ 880 /* per device */
881 EC_DEVICE_QUEUE_DELETE = 32, 881 EC_DEVICE_QUEUE_DELETE = 32,
882 EC_DEVICE_MEMORY_VIOLATION = 33, 882 EC_DEVICE_MEMORY_VIOLATION = 33,
883 EC_DEVICE_RAS_ERROR = 34, 883 EC_DEVICE_RAS_ERROR = 34,
884 EC_DEVICE_FATAL_HALT = 35, 884 EC_DEVICE_FATAL_HALT = 35,
885 EC_DEVICE_NEW = 36, 885 EC_DEVICE_NEW = 36,
886 /* per process */ 886 /* per process */
887 EC_PROCESS_RUNTIME = 48, 887 EC_PROCESS_RUNTIME = 48,
888 EC_PROCESS_DEVICE_REMOVE = 49, 888 EC_PROCESS_DEVICE_REMOVE = 49,
889 EC_MAX 889 EC_MAX
890 }; 890 };
891 891
892 /* Mask generated by ecode in kfd_dbg_trap_exc 892 /* Mask generated by ecode in kfd_dbg_trap_exception_code */
893 #define KFD_EC_MASK(ecode) (1ULL << (ecod 893 #define KFD_EC_MASK(ecode) (1ULL << (ecode - 1))
894 894
895 /* Masks for exception code type checks below 895 /* Masks for exception code type checks below */
896 #define KFD_EC_MASK_QUEUE (KFD_EC_MASK(E 896 #define KFD_EC_MASK_QUEUE (KFD_EC_MASK(EC_QUEUE_WAVE_ABORT) | \
897 KFD_EC_MASK(E 897 KFD_EC_MASK(EC_QUEUE_WAVE_TRAP) | \
898 KFD_EC_MASK(E 898 KFD_EC_MASK(EC_QUEUE_WAVE_MATH_ERROR) | \
899 KFD_EC_MASK(E 899 KFD_EC_MASK(EC_QUEUE_WAVE_ILLEGAL_INSTRUCTION) | \
900 KFD_EC_MASK(E 900 KFD_EC_MASK(EC_QUEUE_WAVE_MEMORY_VIOLATION) | \
901 KFD_EC_MASK(E 901 KFD_EC_MASK(EC_QUEUE_WAVE_APERTURE_VIOLATION) | \
902 KFD_EC_MASK(E 902 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_DIM_INVALID) | \
903 KFD_EC_MASK(E 903 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_GROUP_SEGMENT_SIZE_INVALID) | \
904 KFD_EC_MASK(E 904 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_CODE_INVALID) | \
905 KFD_EC_MASK(E 905 KFD_EC_MASK(EC_QUEUE_PACKET_RESERVED) | \
906 KFD_EC_MASK(E 906 KFD_EC_MASK(EC_QUEUE_PACKET_UNSUPPORTED) | \
907 KFD_EC_MASK(E 907 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_WORK_GROUP_SIZE_INVALID) | \
908 KFD_EC_MASK(E 908 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_REGISTER_INVALID) | \
909 KFD_EC_MASK(E 909 KFD_EC_MASK(EC_QUEUE_PACKET_VENDOR_UNSUPPORTED) | \
910 KFD_EC_MASK(E 910 KFD_EC_MASK(EC_QUEUE_PREEMPTION_ERROR) | \
911 KFD_EC_MASK(E 911 KFD_EC_MASK(EC_QUEUE_NEW))
912 #define KFD_EC_MASK_DEVICE (KFD_EC_MASK(E 912 #define KFD_EC_MASK_DEVICE (KFD_EC_MASK(EC_DEVICE_QUEUE_DELETE) | \
913 KFD_EC_MASK(E 913 KFD_EC_MASK(EC_DEVICE_RAS_ERROR) | \
914 KFD_EC_MASK(E 914 KFD_EC_MASK(EC_DEVICE_FATAL_HALT) | \
915 KFD_EC_MASK(E 915 KFD_EC_MASK(EC_DEVICE_MEMORY_VIOLATION) | \
916 KFD_EC_MASK(E 916 KFD_EC_MASK(EC_DEVICE_NEW))
917 #define KFD_EC_MASK_PROCESS (KFD_EC_MASK(E 917 #define KFD_EC_MASK_PROCESS (KFD_EC_MASK(EC_PROCESS_RUNTIME) | \
918 KFD_EC_MASK(E 918 KFD_EC_MASK(EC_PROCESS_DEVICE_REMOVE))
919 #define KFD_EC_MASK_PACKET (KFD_EC_MASK(E 919 #define KFD_EC_MASK_PACKET (KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_DIM_INVALID) | \
920 KFD_EC_MASK(E 920 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_GROUP_SEGMENT_SIZE_INVALID) | \
921 KFD_EC_MASK(E 921 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_CODE_INVALID) | \
922 KFD_EC_MASK(E 922 KFD_EC_MASK(EC_QUEUE_PACKET_RESERVED) | \
923 KFD_EC_MASK(E 923 KFD_EC_MASK(EC_QUEUE_PACKET_UNSUPPORTED) | \
924 KFD_EC_MASK(E 924 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_WORK_GROUP_SIZE_INVALID) | \
925 KFD_EC_MASK(E 925 KFD_EC_MASK(EC_QUEUE_PACKET_DISPATCH_REGISTER_INVALID) | \
926 KFD_EC_MASK(E 926 KFD_EC_MASK(EC_QUEUE_PACKET_VENDOR_UNSUPPORTED))
927 927
928 /* Checks for exception code types for KFD sea 928 /* Checks for exception code types for KFD search */
929 #define KFD_DBG_EC_IS_VALID(ecode) (ecode > EC 929 #define KFD_DBG_EC_IS_VALID(ecode) (ecode > EC_NONE && ecode < EC_MAX)
930 #define KFD_DBG_EC_TYPE_IS_QUEUE(ecode) 930 #define KFD_DBG_EC_TYPE_IS_QUEUE(ecode) \
931 (KFD_DBG_EC_IS_VALID(e 931 (KFD_DBG_EC_IS_VALID(ecode) && !!(KFD_EC_MASK(ecode) & KFD_EC_MASK_QUEUE))
932 #define KFD_DBG_EC_TYPE_IS_DEVICE(ecode) 932 #define KFD_DBG_EC_TYPE_IS_DEVICE(ecode) \
933 (KFD_DBG_EC_IS_VALID(e 933 (KFD_DBG_EC_IS_VALID(ecode) && !!(KFD_EC_MASK(ecode) & KFD_EC_MASK_DEVICE))
934 #define KFD_DBG_EC_TYPE_IS_PROCESS(ecode) 934 #define KFD_DBG_EC_TYPE_IS_PROCESS(ecode) \
935 (KFD_DBG_EC_IS_VALID(e 935 (KFD_DBG_EC_IS_VALID(ecode) && !!(KFD_EC_MASK(ecode) & KFD_EC_MASK_PROCESS))
936 #define KFD_DBG_EC_TYPE_IS_PACKET(ecode) 936 #define KFD_DBG_EC_TYPE_IS_PACKET(ecode) \
937 (KFD_DBG_EC_IS_VALID(e 937 (KFD_DBG_EC_IS_VALID(ecode) && !!(KFD_EC_MASK(ecode) & KFD_EC_MASK_PACKET))
938 938
939 939
940 /* Runtime enable states */ 940 /* Runtime enable states */
941 enum kfd_dbg_runtime_state { 941 enum kfd_dbg_runtime_state {
942 DEBUG_RUNTIME_STATE_DISABLED = 0, 942 DEBUG_RUNTIME_STATE_DISABLED = 0,
943 DEBUG_RUNTIME_STATE_ENABLED = 1, 943 DEBUG_RUNTIME_STATE_ENABLED = 1,
944 DEBUG_RUNTIME_STATE_ENABLED_BUSY = 2, 944 DEBUG_RUNTIME_STATE_ENABLED_BUSY = 2,
945 DEBUG_RUNTIME_STATE_ENABLED_ERROR = 3 945 DEBUG_RUNTIME_STATE_ENABLED_ERROR = 3
946 }; 946 };
947 947
948 /* Runtime enable status */ 948 /* Runtime enable status */
949 struct kfd_runtime_info { 949 struct kfd_runtime_info {
950 __u64 r_debug; 950 __u64 r_debug;
951 __u32 runtime_state; 951 __u32 runtime_state;
952 __u32 ttmp_setup; 952 __u32 ttmp_setup;
953 }; 953 };
954 954
955 /* Enable modes for runtime enable */ 955 /* Enable modes for runtime enable */
956 #define KFD_RUNTIME_ENABLE_MODE_ENABLE_MASK 956 #define KFD_RUNTIME_ENABLE_MODE_ENABLE_MASK 1
957 #define KFD_RUNTIME_ENABLE_MODE_TTMP_SAVE_MASK 957 #define KFD_RUNTIME_ENABLE_MODE_TTMP_SAVE_MASK 2
958 958
959 /** 959 /**
960 * kfd_ioctl_runtime_enable_args - Arguments f 960 * kfd_ioctl_runtime_enable_args - Arguments for runtime enable
961 * 961 *
962 * Coordinates debug exception signalling and 962 * Coordinates debug exception signalling and debug device enablement with runtime.
963 * 963 *
964 * @r_debug - pointer to user struct for shari 964 * @r_debug - pointer to user struct for sharing information between ROCr and the debuggger
965 * @mode_mask - mask to set mode 965 * @mode_mask - mask to set mode
966 * KFD_RUNTIME_ENABLE_MODE_ENABLE_MASK - 966 * KFD_RUNTIME_ENABLE_MODE_ENABLE_MASK - enable runtime for debugging, otherwise disable
967 * KFD_RUNTIME_ENABLE_MODE_TTMP_SAVE_MASK 967 * KFD_RUNTIME_ENABLE_MODE_TTMP_SAVE_MASK - enable trap temporary setup (ignore on disable)
968 * @capabilities_mask - mask to notify runtime 968 * @capabilities_mask - mask to notify runtime on what KFD supports
969 * 969 *
970 * Return - 0 on SUCCESS. 970 * Return - 0 on SUCCESS.
971 * - EBUSY if runtime enable call alrea 971 * - EBUSY if runtime enable call already pending.
972 * - EEXIST if user queues already acti 972 * - EEXIST if user queues already active prior to call.
973 * If process is debug enabled, runti 973 * If process is debug enabled, runtime enable will enable debug devices and
974 * wait for debugger process to send 974 * wait for debugger process to send runtime exception EC_PROCESS_RUNTIME
975 * to unblock - see kfd_ioctl_dbg_tra 975 * to unblock - see kfd_ioctl_dbg_trap_args.
976 * 976 *
977 */ 977 */
978 struct kfd_ioctl_runtime_enable_args { 978 struct kfd_ioctl_runtime_enable_args {
979 __u64 r_debug; 979 __u64 r_debug;
980 __u32 mode_mask; 980 __u32 mode_mask;
981 __u32 capabilities_mask; 981 __u32 capabilities_mask;
982 }; 982 };
983 983
984 /* Queue information */ 984 /* Queue information */
985 struct kfd_queue_snapshot_entry { 985 struct kfd_queue_snapshot_entry {
986 __u64 exception_status; 986 __u64 exception_status;
987 __u64 ring_base_address; 987 __u64 ring_base_address;
988 __u64 write_pointer_address; 988 __u64 write_pointer_address;
989 __u64 read_pointer_address; 989 __u64 read_pointer_address;
990 __u64 ctx_save_restore_address; 990 __u64 ctx_save_restore_address;
991 __u32 queue_id; 991 __u32 queue_id;
992 __u32 gpu_id; 992 __u32 gpu_id;
993 __u32 ring_size; 993 __u32 ring_size;
994 __u32 queue_type; 994 __u32 queue_type;
995 __u32 ctx_save_restore_area_size; 995 __u32 ctx_save_restore_area_size;
996 __u32 reserved; 996 __u32 reserved;
997 }; 997 };
998 998
999 /* Queue status return for suspend/resume */ 999 /* Queue status return for suspend/resume */
1000 #define KFD_DBG_QUEUE_ERROR_BIT 30 1000 #define KFD_DBG_QUEUE_ERROR_BIT 30
1001 #define KFD_DBG_QUEUE_INVALID_BIT 31 1001 #define KFD_DBG_QUEUE_INVALID_BIT 31
1002 #define KFD_DBG_QUEUE_ERROR_MASK (1 << 1002 #define KFD_DBG_QUEUE_ERROR_MASK (1 << KFD_DBG_QUEUE_ERROR_BIT)
1003 #define KFD_DBG_QUEUE_INVALID_MASK (1 << 1003 #define KFD_DBG_QUEUE_INVALID_MASK (1 << KFD_DBG_QUEUE_INVALID_BIT)
1004 1004
1005 /* Context save area header information */ 1005 /* Context save area header information */
1006 struct kfd_context_save_area_header { 1006 struct kfd_context_save_area_header {
1007 struct { 1007 struct {
1008 __u32 control_stack_offset; 1008 __u32 control_stack_offset;
1009 __u32 control_stack_size; 1009 __u32 control_stack_size;
1010 __u32 wave_state_offset; 1010 __u32 wave_state_offset;
1011 __u32 wave_state_size; 1011 __u32 wave_state_size;
1012 } wave_state; 1012 } wave_state;
1013 __u32 debug_offset; 1013 __u32 debug_offset;
1014 __u32 debug_size; 1014 __u32 debug_size;
1015 __u64 err_payload_addr; 1015 __u64 err_payload_addr;
1016 __u32 err_event_id; 1016 __u32 err_event_id;
1017 __u32 reserved1; 1017 __u32 reserved1;
1018 }; 1018 };
1019 1019
1020 /* 1020 /*
1021 * Debug operations 1021 * Debug operations
1022 * 1022 *
1023 * For specifics on usage and return values, 1023 * For specifics on usage and return values, see documentation per operation
1024 * below. Otherwise, generic error returns a 1024 * below. Otherwise, generic error returns apply:
1025 * - ESRCH if the process to debug does 1025 * - ESRCH if the process to debug does not exist.
1026 * 1026 *
1027 * - EINVAL (with KFD_IOC_DBG_TRAP_ENABL 1027 * - EINVAL (with KFD_IOC_DBG_TRAP_ENABLE exempt) if operation
1028 * KFD_IOC_DBG_TRAP_ENABLE has 1028 * KFD_IOC_DBG_TRAP_ENABLE has not succeeded prior.
1029 * Also returns this error if G 1029 * Also returns this error if GPU hardware scheduling is not supported.
1030 * 1030 *
1031 * - EPERM (with KFD_IOC_DBG_TRAP_DISABL 1031 * - EPERM (with KFD_IOC_DBG_TRAP_DISABLE exempt) if target process is not
1032 * PTRACE_ATTACHED. KFD_IOC_DB 1032 * PTRACE_ATTACHED. KFD_IOC_DBG_TRAP_DISABLE is exempt to allow
1033 * clean up of debug mode as lo 1033 * clean up of debug mode as long as process is debug enabled.
1034 * 1034 *
1035 * - EACCES if any DBG_HW_OP (debug hard 1035 * - EACCES if any DBG_HW_OP (debug hardware operation) is requested when
1036 * AMDKFD_IOC_RUNTIME_ENABLE ha 1036 * AMDKFD_IOC_RUNTIME_ENABLE has not succeeded prior.
1037 * 1037 *
1038 * - ENODEV if any GPU does not support 1038 * - ENODEV if any GPU does not support debugging on a DBG_HW_OP call.
1039 * 1039 *
1040 * - Other errors may be returned when a 1040 * - Other errors may be returned when a DBG_HW_OP occurs while the GPU
1041 * is in a fatal state. 1041 * is in a fatal state.
1042 * 1042 *
1043 */ 1043 */
1044 enum kfd_dbg_trap_operations { 1044 enum kfd_dbg_trap_operations {
1045 KFD_IOC_DBG_TRAP_ENABLE = 0, 1045 KFD_IOC_DBG_TRAP_ENABLE = 0,
1046 KFD_IOC_DBG_TRAP_DISABLE = 1, 1046 KFD_IOC_DBG_TRAP_DISABLE = 1,
1047 KFD_IOC_DBG_TRAP_SEND_RUNTIME_EVENT = 1047 KFD_IOC_DBG_TRAP_SEND_RUNTIME_EVENT = 2,
1048 KFD_IOC_DBG_TRAP_SET_EXCEPTIONS_ENABL 1048 KFD_IOC_DBG_TRAP_SET_EXCEPTIONS_ENABLED = 3,
1049 KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_OVER 1049 KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_OVERRIDE = 4, /* DBG_HW_OP */
1050 KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE 1050 KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE = 5, /* DBG_HW_OP */
1051 KFD_IOC_DBG_TRAP_SUSPEND_QUEUES = 6, 1051 KFD_IOC_DBG_TRAP_SUSPEND_QUEUES = 6, /* DBG_HW_OP */
1052 KFD_IOC_DBG_TRAP_RESUME_QUEUES = 7, 1052 KFD_IOC_DBG_TRAP_RESUME_QUEUES = 7, /* DBG_HW_OP */
1053 KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WAT 1053 KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WATCH = 8, /* DBG_HW_OP */
1054 KFD_IOC_DBG_TRAP_CLEAR_NODE_ADDRESS_W 1054 KFD_IOC_DBG_TRAP_CLEAR_NODE_ADDRESS_WATCH = 9, /* DBG_HW_OP */
1055 KFD_IOC_DBG_TRAP_SET_FLAGS = 10, 1055 KFD_IOC_DBG_TRAP_SET_FLAGS = 10,
1056 KFD_IOC_DBG_TRAP_QUERY_DEBUG_EVENT = 1056 KFD_IOC_DBG_TRAP_QUERY_DEBUG_EVENT = 11,
1057 KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO 1057 KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO = 12,
1058 KFD_IOC_DBG_TRAP_GET_QUEUE_SNAPSHOT = 1058 KFD_IOC_DBG_TRAP_GET_QUEUE_SNAPSHOT = 13,
1059 KFD_IOC_DBG_TRAP_GET_DEVICE_SNAPSHOT 1059 KFD_IOC_DBG_TRAP_GET_DEVICE_SNAPSHOT = 14
1060 }; 1060 };
1061 1061
1062 /** 1062 /**
1063 * kfd_ioctl_dbg_trap_enable_args 1063 * kfd_ioctl_dbg_trap_enable_args
1064 * 1064 *
1065 * Arguments for KFD_IOC_DBG_TRAP_ENABLE. 1065 * Arguments for KFD_IOC_DBG_TRAP_ENABLE.
1066 * 1066 *
1067 * Enables debug session for target proce 1067 * Enables debug session for target process. Call @op KFD_IOC_DBG_TRAP_DISABLE in
1068 * kfd_ioctl_dbg_trap_args to disable deb 1068 * kfd_ioctl_dbg_trap_args to disable debug session.
1069 * 1069 *
1070 * @exception_mask (IN) - exceptions 1070 * @exception_mask (IN) - exceptions to raise to the debugger
1071 * @rinfo_ptr (IN) - pointer to 1071 * @rinfo_ptr (IN) - pointer to runtime info buffer (see kfd_runtime_info)
1072 * @rinfo_size (IN/OUT) - size of run 1072 * @rinfo_size (IN/OUT) - size of runtime info buffer in bytes
1073 * @dbg_fd (IN) - fd the KFD 1073 * @dbg_fd (IN) - fd the KFD will nofify the debugger with of raised
1074 * exceptions 1074 * exceptions set in exception_mask.
1075 * 1075 *
1076 * Generic errors apply (see kfd_dbg_trap 1076 * Generic errors apply (see kfd_dbg_trap_operations).
1077 * Return - 0 on SUCCESS. 1077 * Return - 0 on SUCCESS.
1078 * Copies KFD saved kfd_runtime_ 1078 * Copies KFD saved kfd_runtime_info to @rinfo_ptr on enable.
1079 * Size of kfd_runtime saved by 1079 * Size of kfd_runtime saved by the KFD returned to @rinfo_size.
1080 * - EBADF if KFD cannot get a ref 1080 * - EBADF if KFD cannot get a reference to dbg_fd.
1081 * - EFAULT if KFD cannot copy run 1081 * - EFAULT if KFD cannot copy runtime info to rinfo_ptr.
1082 * - EINVAL if target process is a 1082 * - EINVAL if target process is already debug enabled.
1083 * 1083 *
1084 */ 1084 */
1085 struct kfd_ioctl_dbg_trap_enable_args { 1085 struct kfd_ioctl_dbg_trap_enable_args {
1086 __u64 exception_mask; 1086 __u64 exception_mask;
1087 __u64 rinfo_ptr; 1087 __u64 rinfo_ptr;
1088 __u32 rinfo_size; 1088 __u32 rinfo_size;
1089 __u32 dbg_fd; 1089 __u32 dbg_fd;
1090 }; 1090 };
1091 1091
1092 /** 1092 /**
1093 * kfd_ioctl_dbg_trap_send_runtime_event_args 1093 * kfd_ioctl_dbg_trap_send_runtime_event_args
1094 * 1094 *
1095 * 1095 *
1096 * Arguments for KFD_IOC_DBG_TRAP_SEND_RU 1096 * Arguments for KFD_IOC_DBG_TRAP_SEND_RUNTIME_EVENT.
1097 * Raises exceptions to runtime. 1097 * Raises exceptions to runtime.
1098 * 1098 *
1099 * @exception_mask (IN) - exceptions to r 1099 * @exception_mask (IN) - exceptions to raise to runtime
1100 * @gpu_id (IN) - target device i 1100 * @gpu_id (IN) - target device id
1101 * @queue_id (IN) - target queue id 1101 * @queue_id (IN) - target queue id
1102 * 1102 *
1103 * Generic errors apply (see kfd_dbg_trap 1103 * Generic errors apply (see kfd_dbg_trap_operations).
1104 * Return - 0 on SUCCESS. 1104 * Return - 0 on SUCCESS.
1105 * - ENODEV if gpu_id not found. 1105 * - ENODEV if gpu_id not found.
1106 * If exception_mask contains EC 1106 * If exception_mask contains EC_PROCESS_RUNTIME, unblocks pending
1107 * AMDKFD_IOC_RUNTIME_ENABLE cal 1107 * AMDKFD_IOC_RUNTIME_ENABLE call - see kfd_ioctl_runtime_enable_args.
1108 * All other exceptions are rais 1108 * All other exceptions are raised to runtime through err_payload_addr.
1109 * See kfd_context_save_area_hea 1109 * See kfd_context_save_area_header.
1110 */ 1110 */
1111 struct kfd_ioctl_dbg_trap_send_runtime_event_ 1111 struct kfd_ioctl_dbg_trap_send_runtime_event_args {
1112 __u64 exception_mask; 1112 __u64 exception_mask;
1113 __u32 gpu_id; 1113 __u32 gpu_id;
1114 __u32 queue_id; 1114 __u32 queue_id;
1115 }; 1115 };
1116 1116
1117 /** 1117 /**
1118 * kfd_ioctl_dbg_trap_set_exceptions_enabled_ 1118 * kfd_ioctl_dbg_trap_set_exceptions_enabled_args
1119 * 1119 *
1120 * Arguments for KFD_IOC_SET_EXCEPTIONS_E 1120 * Arguments for KFD_IOC_SET_EXCEPTIONS_ENABLED
1121 * Set new exceptions to be raised to the 1121 * Set new exceptions to be raised to the debugger.
1122 * 1122 *
1123 * @exception_mask (IN) - new exceptions 1123 * @exception_mask (IN) - new exceptions to raise the debugger
1124 * 1124 *
1125 * Generic errors apply (see kfd_dbg_trap 1125 * Generic errors apply (see kfd_dbg_trap_operations).
1126 * Return - 0 on SUCCESS. 1126 * Return - 0 on SUCCESS.
1127 */ 1127 */
1128 struct kfd_ioctl_dbg_trap_set_exceptions_enab 1128 struct kfd_ioctl_dbg_trap_set_exceptions_enabled_args {
1129 __u64 exception_mask; 1129 __u64 exception_mask;
1130 }; 1130 };
1131 1131
1132 /** 1132 /**
1133 * kfd_ioctl_dbg_trap_set_wave_launch_overrid 1133 * kfd_ioctl_dbg_trap_set_wave_launch_override_args
1134 * 1134 *
1135 * Arguments for KFD_IOC_DBG_TRAP_SET_WAV 1135 * Arguments for KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_OVERRIDE
1136 * Enable HW exceptions to raise trap. 1136 * Enable HW exceptions to raise trap.
1137 * 1137 *
1138 * @override_mode (IN) - see k 1138 * @override_mode (IN) - see kfd_dbg_trap_override_mode
1139 * @enable_mask (IN/OUT) - refer 1139 * @enable_mask (IN/OUT) - reference kfd_dbg_trap_mask.
1140 * IN is 1140 * IN is the override modes requested to be enabled.
1141 * OUT i 1141 * OUT is referenced in Return below.
1142 * @support_request_mask (IN/OUT) - refer 1142 * @support_request_mask (IN/OUT) - reference kfd_dbg_trap_mask.
1143 * IN is 1143 * IN is the override modes requested for support check.
1144 * OUT i 1144 * OUT is referenced in Return below.
1145 * 1145 *
1146 * Generic errors apply (see kfd_dbg_trap 1146 * Generic errors apply (see kfd_dbg_trap_operations).
1147 * Return - 0 on SUCCESS. 1147 * Return - 0 on SUCCESS.
1148 * Previous enablement is return 1148 * Previous enablement is returned in @enable_mask.
1149 * Actual override support is re 1149 * Actual override support is returned in @support_request_mask.
1150 * - EINVAL if override mode is no 1150 * - EINVAL if override mode is not supported.
1151 * - EACCES if trap support reques 1151 * - EACCES if trap support requested is not actually supported.
1152 * i.e. enable_mask (IN) is not 1152 * i.e. enable_mask (IN) is not a subset of support_request_mask (OUT).
1153 * Otherwise it is considered a 1153 * Otherwise it is considered a generic error (see kfd_dbg_trap_operations).
1154 */ 1154 */
1155 struct kfd_ioctl_dbg_trap_set_wave_launch_ove 1155 struct kfd_ioctl_dbg_trap_set_wave_launch_override_args {
1156 __u32 override_mode; 1156 __u32 override_mode;
1157 __u32 enable_mask; 1157 __u32 enable_mask;
1158 __u32 support_request_mask; 1158 __u32 support_request_mask;
1159 __u32 pad; 1159 __u32 pad;
1160 }; 1160 };
1161 1161
1162 /** 1162 /**
1163 * kfd_ioctl_dbg_trap_set_wave_launch_mode_ar 1163 * kfd_ioctl_dbg_trap_set_wave_launch_mode_args
1164 * 1164 *
1165 * Arguments for KFD_IOC_DBG_TRAP_SET_WAV 1165 * Arguments for KFD_IOC_DBG_TRAP_SET_WAVE_LAUNCH_MODE
1166 * Set wave launch mode. 1166 * Set wave launch mode.
1167 * 1167 *
1168 * @mode (IN) - see kfd_dbg_trap_wave_lau 1168 * @mode (IN) - see kfd_dbg_trap_wave_launch_mode
1169 * 1169 *
1170 * Generic errors apply (see kfd_dbg_trap 1170 * Generic errors apply (see kfd_dbg_trap_operations).
1171 * Return - 0 on SUCCESS. 1171 * Return - 0 on SUCCESS.
1172 */ 1172 */
1173 struct kfd_ioctl_dbg_trap_set_wave_launch_mod 1173 struct kfd_ioctl_dbg_trap_set_wave_launch_mode_args {
1174 __u32 launch_mode; 1174 __u32 launch_mode;
1175 __u32 pad; 1175 __u32 pad;
1176 }; 1176 };
1177 1177
1178 /** 1178 /**
1179 * kfd_ioctl_dbg_trap_suspend_queues_ags 1179 * kfd_ioctl_dbg_trap_suspend_queues_ags
1180 * 1180 *
1181 * Arguments for KFD_IOC_DBG_TRAP_SUSPEND 1181 * Arguments for KFD_IOC_DBG_TRAP_SUSPEND_QUEUES
1182 * Suspend queues. 1182 * Suspend queues.
1183 * 1183 *
1184 * @exception_mask (IN) - raised excepti 1184 * @exception_mask (IN) - raised exceptions to clear
1185 * @queue_array_ptr (IN) - pointer to arr 1185 * @queue_array_ptr (IN) - pointer to array of queue ids (u32 per queue id)
1186 * to suspend 1186 * to suspend
1187 * @num_queues (IN) - number of queu 1187 * @num_queues (IN) - number of queues to suspend in @queue_array_ptr
1188 * @grace_period (IN) - wave time allo 1188 * @grace_period (IN) - wave time allowance before preemption
1189 * per 1K GPU clo 1189 * per 1K GPU clock cycle unit
1190 * 1190 *
1191 * Generic errors apply (see kfd_dbg_trap 1191 * Generic errors apply (see kfd_dbg_trap_operations).
1192 * Destruction of a suspended queue is bl 1192 * Destruction of a suspended queue is blocked until the queue is
1193 * resumed. This allows the debugger to 1193 * resumed. This allows the debugger to access queue information and
1194 * the its context save area without runn 1194 * the its context save area without running into a race condition on
1195 * queue destruction. 1195 * queue destruction.
1196 * Automatically copies per queue context 1196 * Automatically copies per queue context save area header information
1197 * into the save area base 1197 * into the save area base
1198 * (see kfd_queue_snapshot_entry and kfd_ 1198 * (see kfd_queue_snapshot_entry and kfd_context_save_area_header).
1199 * 1199 *
1200 * Return - Number of queues suspended on 1200 * Return - Number of queues suspended on SUCCESS.
1201 * . KFD_DBG_QUEUE_ERROR_MASK and 1201 * . KFD_DBG_QUEUE_ERROR_MASK and KFD_DBG_QUEUE_INVALID_MASK masked
1202 * for each queue id in @queue_a 1202 * for each queue id in @queue_array_ptr array reports unsuccessful
1203 * suspend reason. 1203 * suspend reason.
1204 * KFD_DBG_QUEUE_ERROR_MASK = HW 1204 * KFD_DBG_QUEUE_ERROR_MASK = HW failure.
1205 * KFD_DBG_QUEUE_INVALID_MASK = 1205 * KFD_DBG_QUEUE_INVALID_MASK = queue does not exist, is new or
1206 * is being destroyed. 1206 * is being destroyed.
1207 */ 1207 */
1208 struct kfd_ioctl_dbg_trap_suspend_queues_args 1208 struct kfd_ioctl_dbg_trap_suspend_queues_args {
1209 __u64 exception_mask; 1209 __u64 exception_mask;
1210 __u64 queue_array_ptr; 1210 __u64 queue_array_ptr;
1211 __u32 num_queues; 1211 __u32 num_queues;
1212 __u32 grace_period; 1212 __u32 grace_period;
1213 }; 1213 };
1214 1214
1215 /** 1215 /**
1216 * kfd_ioctl_dbg_trap_resume_queues_args 1216 * kfd_ioctl_dbg_trap_resume_queues_args
1217 * 1217 *
1218 * Arguments for KFD_IOC_DBG_TRAP_RESUME_ 1218 * Arguments for KFD_IOC_DBG_TRAP_RESUME_QUEUES
1219 * Resume queues. 1219 * Resume queues.
1220 * 1220 *
1221 * @queue_array_ptr (IN) - pointer to arr 1221 * @queue_array_ptr (IN) - pointer to array of queue ids (u32 per queue id)
1222 * to resume 1222 * to resume
1223 * @num_queues (IN) - number of queu 1223 * @num_queues (IN) - number of queues to resume in @queue_array_ptr
1224 * 1224 *
1225 * Generic errors apply (see kfd_dbg_trap 1225 * Generic errors apply (see kfd_dbg_trap_operations).
1226 * Return - Number of queues resumed on S 1226 * Return - Number of queues resumed on SUCCESS.
1227 * KFD_DBG_QUEUE_ERROR_MASK and 1227 * KFD_DBG_QUEUE_ERROR_MASK and KFD_DBG_QUEUE_INVALID_MASK mask
1228 * for each queue id in @queue_a 1228 * for each queue id in @queue_array_ptr array reports unsuccessful
1229 * resume reason. 1229 * resume reason.
1230 * KFD_DBG_QUEUE_ERROR_MASK = HW 1230 * KFD_DBG_QUEUE_ERROR_MASK = HW failure.
1231 * KFD_DBG_QUEUE_INVALID_MASK = 1231 * KFD_DBG_QUEUE_INVALID_MASK = queue does not exist.
1232 */ 1232 */
1233 struct kfd_ioctl_dbg_trap_resume_queues_args 1233 struct kfd_ioctl_dbg_trap_resume_queues_args {
1234 __u64 queue_array_ptr; 1234 __u64 queue_array_ptr;
1235 __u32 num_queues; 1235 __u32 num_queues;
1236 __u32 pad; 1236 __u32 pad;
1237 }; 1237 };
1238 1238
1239 /** 1239 /**
1240 * kfd_ioctl_dbg_trap_set_node_address_watch_ 1240 * kfd_ioctl_dbg_trap_set_node_address_watch_args
1241 * 1241 *
1242 * Arguments for KFD_IOC_DBG_TRAP_SET_NOD 1242 * Arguments for KFD_IOC_DBG_TRAP_SET_NODE_ADDRESS_WATCH
1243 * Sets address watch for device. 1243 * Sets address watch for device.
1244 * 1244 *
1245 * @address (IN) - watch address to set 1245 * @address (IN) - watch address to set
1246 * @mode (IN) - see kfd_dbg_trap_addr 1246 * @mode (IN) - see kfd_dbg_trap_address_watch_mode
1247 * @mask (IN) - watch address mask 1247 * @mask (IN) - watch address mask
1248 * @gpu_id (IN) - target gpu to set wat 1248 * @gpu_id (IN) - target gpu to set watch point
1249 * @id (OUT) - watch id allocated 1249 * @id (OUT) - watch id allocated
1250 * 1250 *
1251 * Generic errors apply (see kfd_dbg_trap 1251 * Generic errors apply (see kfd_dbg_trap_operations).
1252 * Return - 0 on SUCCESS. 1252 * Return - 0 on SUCCESS.
1253 * Allocated watch ID returned t 1253 * Allocated watch ID returned to @id.
1254 * - ENODEV if gpu_id not found. 1254 * - ENODEV if gpu_id not found.
1255 * - ENOMEM if watch IDs can be al 1255 * - ENOMEM if watch IDs can be allocated
1256 */ 1256 */
1257 struct kfd_ioctl_dbg_trap_set_node_address_wa 1257 struct kfd_ioctl_dbg_trap_set_node_address_watch_args {
1258 __u64 address; 1258 __u64 address;
1259 __u32 mode; 1259 __u32 mode;
1260 __u32 mask; 1260 __u32 mask;
1261 __u32 gpu_id; 1261 __u32 gpu_id;
1262 __u32 id; 1262 __u32 id;
1263 }; 1263 };
1264 1264
1265 /** 1265 /**
1266 * kfd_ioctl_dbg_trap_clear_node_address_watc 1266 * kfd_ioctl_dbg_trap_clear_node_address_watch_args
1267 * 1267 *
1268 * Arguments for KFD_IOC_DBG_TRAP_CLEAR_N 1268 * Arguments for KFD_IOC_DBG_TRAP_CLEAR_NODE_ADDRESS_WATCH
1269 * Clear address watch for device. 1269 * Clear address watch for device.
1270 * 1270 *
1271 * @gpu_id (IN) - target device to clea 1271 * @gpu_id (IN) - target device to clear watch point
1272 * @id (IN) - allocated watch id to 1272 * @id (IN) - allocated watch id to clear
1273 * 1273 *
1274 * Generic errors apply (see kfd_dbg_trap 1274 * Generic errors apply (see kfd_dbg_trap_operations).
1275 * Return - 0 on SUCCESS. 1275 * Return - 0 on SUCCESS.
1276 * - ENODEV if gpu_id not found. 1276 * - ENODEV if gpu_id not found.
1277 * - EINVAL if watch ID has not be 1277 * - EINVAL if watch ID has not been allocated.
1278 */ 1278 */
1279 struct kfd_ioctl_dbg_trap_clear_node_address_ 1279 struct kfd_ioctl_dbg_trap_clear_node_address_watch_args {
1280 __u32 gpu_id; 1280 __u32 gpu_id;
1281 __u32 id; 1281 __u32 id;
1282 }; 1282 };
1283 1283
1284 /** 1284 /**
1285 * kfd_ioctl_dbg_trap_set_flags_args 1285 * kfd_ioctl_dbg_trap_set_flags_args
1286 * 1286 *
1287 * Arguments for KFD_IOC_DBG_TRAP_SET_FLA 1287 * Arguments for KFD_IOC_DBG_TRAP_SET_FLAGS
1288 * Sets flags for wave behaviour. 1288 * Sets flags for wave behaviour.
1289 * 1289 *
1290 * @flags (IN/OUT) - IN = flags to enable 1290 * @flags (IN/OUT) - IN = flags to enable, OUT = flags previously enabled
1291 * 1291 *
1292 * Generic errors apply (see kfd_dbg_trap 1292 * Generic errors apply (see kfd_dbg_trap_operations).
1293 * Return - 0 on SUCCESS. 1293 * Return - 0 on SUCCESS.
1294 * - EACCESS if any debug device d 1294 * - EACCESS if any debug device does not allow flag options.
1295 */ 1295 */
1296 struct kfd_ioctl_dbg_trap_set_flags_args { 1296 struct kfd_ioctl_dbg_trap_set_flags_args {
1297 __u32 flags; 1297 __u32 flags;
1298 __u32 pad; 1298 __u32 pad;
1299 }; 1299 };
1300 1300
1301 /** 1301 /**
1302 * kfd_ioctl_dbg_trap_query_debug_event_args 1302 * kfd_ioctl_dbg_trap_query_debug_event_args
1303 * 1303 *
1304 * Arguments for KFD_IOC_DBG_TRAP_QUERY_D 1304 * Arguments for KFD_IOC_DBG_TRAP_QUERY_DEBUG_EVENT
1305 * 1305 *
1306 * Find one or more raised exceptions. Th 1306 * Find one or more raised exceptions. This function can return multiple
1307 * exceptions from a single queue or a si 1307 * exceptions from a single queue or a single device with one call. To find
1308 * all raised exceptions, this function m 1308 * all raised exceptions, this function must be called repeatedly until it
1309 * returns -EAGAIN. Returned exceptions c 1309 * returns -EAGAIN. Returned exceptions can optionally be cleared by
1310 * setting the corresponding bit in the @ 1310 * setting the corresponding bit in the @exception_mask input parameter.
1311 * However, clearing an exception prevent 1311 * However, clearing an exception prevents retrieving further information
1312 * about it with KFD_IOC_DBG_TRAP_QUERY_E 1312 * about it with KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO.
1313 * 1313 *
1314 * @exception_mask (IN/OUT) - exception t 1314 * @exception_mask (IN/OUT) - exception to clear (IN) and raised (OUT)
1315 * @gpu_id (OUT) - gpu id of e 1315 * @gpu_id (OUT) - gpu id of exceptions raised
1316 * @queue_id (OUT) - queue id of 1316 * @queue_id (OUT) - queue id of exceptions raised
1317 * 1317 *
1318 * Generic errors apply (see kfd_dbg_trap 1318 * Generic errors apply (see kfd_dbg_trap_operations).
1319 * Return - 0 on raised exception found 1319 * Return - 0 on raised exception found
1320 * Raised exceptions found are r 1320 * Raised exceptions found are returned in @exception mask
1321 * with reported source id retur 1321 * with reported source id returned in @gpu_id or @queue_id.
1322 * - EAGAIN if no raised exception 1322 * - EAGAIN if no raised exception has been found
1323 */ 1323 */
1324 struct kfd_ioctl_dbg_trap_query_debug_event_a 1324 struct kfd_ioctl_dbg_trap_query_debug_event_args {
1325 __u64 exception_mask; 1325 __u64 exception_mask;
1326 __u32 gpu_id; 1326 __u32 gpu_id;
1327 __u32 queue_id; 1327 __u32 queue_id;
1328 }; 1328 };
1329 1329
1330 /** 1330 /**
1331 * kfd_ioctl_dbg_trap_query_exception_info_ar 1331 * kfd_ioctl_dbg_trap_query_exception_info_args
1332 * 1332 *
1333 * Arguments KFD_IOC_DBG_TRAP_QUERY_EXCEP 1333 * Arguments KFD_IOC_DBG_TRAP_QUERY_EXCEPTION_INFO
1334 * Get additional info on raised exceptio 1334 * Get additional info on raised exception.
1335 * 1335 *
1336 * @info_ptr (IN) - pointer to 1336 * @info_ptr (IN) - pointer to exception info buffer to copy to
1337 * @info_size (IN/OUT) - exception 1337 * @info_size (IN/OUT) - exception info buffer size (bytes)
1338 * @source_id (IN) - target gpu 1338 * @source_id (IN) - target gpu or queue id
1339 * @exception_code (IN) - target exc 1339 * @exception_code (IN) - target exception
1340 * @clear_exception (IN) - clear rais 1340 * @clear_exception (IN) - clear raised @exception_code exception
1341 * (0 = false 1341 * (0 = false, 1 = true)
1342 * 1342 *
1343 * Generic errors apply (see kfd_dbg_trap 1343 * Generic errors apply (see kfd_dbg_trap_operations).
1344 * Return - 0 on SUCCESS. 1344 * Return - 0 on SUCCESS.
1345 * If @exception_code is EC_DEVI 1345 * If @exception_code is EC_DEVICE_MEMORY_VIOLATION, copy @info_size(OUT)
1346 * bytes of memory exception dat 1346 * bytes of memory exception data to @info_ptr.
1347 * If @exception_code is EC_PROC 1347 * If @exception_code is EC_PROCESS_RUNTIME, copy saved
1348 * kfd_runtime_info to @info_ptr 1348 * kfd_runtime_info to @info_ptr.
1349 * Actual required @info_ptr siz 1349 * Actual required @info_ptr size (bytes) is returned in @info_size.
1350 */ 1350 */
1351 struct kfd_ioctl_dbg_trap_query_exception_inf 1351 struct kfd_ioctl_dbg_trap_query_exception_info_args {
1352 __u64 info_ptr; 1352 __u64 info_ptr;
1353 __u32 info_size; 1353 __u32 info_size;
1354 __u32 source_id; 1354 __u32 source_id;
1355 __u32 exception_code; 1355 __u32 exception_code;
1356 __u32 clear_exception; 1356 __u32 clear_exception;
1357 }; 1357 };
1358 1358
1359 /** 1359 /**
1360 * kfd_ioctl_dbg_trap_get_queue_snapshot_args 1360 * kfd_ioctl_dbg_trap_get_queue_snapshot_args
1361 * 1361 *
1362 * Arguments KFD_IOC_DBG_TRAP_GET_QUEUE_S 1362 * Arguments KFD_IOC_DBG_TRAP_GET_QUEUE_SNAPSHOT
1363 * Get queue information. 1363 * Get queue information.
1364 * 1364 *
1365 * @exception_mask (IN) - exception 1365 * @exception_mask (IN) - exceptions raised to clear
1366 * @snapshot_buf_ptr (IN) - queue sna 1366 * @snapshot_buf_ptr (IN) - queue snapshot entry buffer (see kfd_queue_snapshot_entry)
1367 * @num_queues (IN/OUT) - number of 1367 * @num_queues (IN/OUT) - number of queue snapshot entries
1368 * The debugger specifies the size of 1368 * The debugger specifies the size of the array allocated in @num_queues.
1369 * KFD returns the number of queues t 1369 * KFD returns the number of queues that actually existed. If this is
1370 * larger than the size specified by 1370 * larger than the size specified by the debugger, KFD will not overflow
1371 * the array allocated by the debugge 1371 * the array allocated by the debugger.
1372 * 1372 *
1373 * @entry_size (IN/OUT) - size per 1373 * @entry_size (IN/OUT) - size per entry in bytes
1374 * The debugger specifies sizeof(stru 1374 * The debugger specifies sizeof(struct kfd_queue_snapshot_entry) in
1375 * @entry_size. KFD returns the numbe 1375 * @entry_size. KFD returns the number of bytes actually populated per
1376 * entry. The debugger should use the 1376 * entry. The debugger should use the KFD_IOCTL_MINOR_VERSION to determine,
1377 * which fields in struct kfd_queue_s 1377 * which fields in struct kfd_queue_snapshot_entry are valid. This allows
1378 * growing the ABI in a backwards com 1378 * growing the ABI in a backwards compatible manner.
1379 * Note that entry_size(IN) should st 1379 * Note that entry_size(IN) should still be used to stride the snapshot buffer in the
1380 * event that it's larger than actual 1380 * event that it's larger than actual kfd_queue_snapshot_entry.
1381 * 1381 *
1382 * Generic errors apply (see kfd_dbg_trap 1382 * Generic errors apply (see kfd_dbg_trap_operations).
1383 * Return - 0 on SUCCESS. 1383 * Return - 0 on SUCCESS.
1384 * Copies @num_queues(IN) queue 1384 * Copies @num_queues(IN) queue snapshot entries of size @entry_size(IN)
1385 * into @snapshot_buf_ptr if @nu 1385 * into @snapshot_buf_ptr if @num_queues(IN) > 0.
1386 * Otherwise return @num_queues( 1386 * Otherwise return @num_queues(OUT) queue snapshot entries that exist.
1387 */ 1387 */
1388 struct kfd_ioctl_dbg_trap_queue_snapshot_args 1388 struct kfd_ioctl_dbg_trap_queue_snapshot_args {
1389 __u64 exception_mask; 1389 __u64 exception_mask;
1390 __u64 snapshot_buf_ptr; 1390 __u64 snapshot_buf_ptr;
1391 __u32 num_queues; 1391 __u32 num_queues;
1392 __u32 entry_size; 1392 __u32 entry_size;
1393 }; 1393 };
1394 1394
1395 /** 1395 /**
1396 * kfd_ioctl_dbg_trap_get_device_snapshot_arg 1396 * kfd_ioctl_dbg_trap_get_device_snapshot_args
1397 * 1397 *
1398 * Arguments for KFD_IOC_DBG_TRAP_GET_DEV 1398 * Arguments for KFD_IOC_DBG_TRAP_GET_DEVICE_SNAPSHOT
1399 * Get device information. 1399 * Get device information.
1400 * 1400 *
1401 * @exception_mask (IN) - exception 1401 * @exception_mask (IN) - exceptions raised to clear
1402 * @snapshot_buf_ptr (IN) - pointer t 1402 * @snapshot_buf_ptr (IN) - pointer to snapshot buffer (see kfd_dbg_device_info_entry)
1403 * @num_devices (IN/OUT) - number of 1403 * @num_devices (IN/OUT) - number of debug devices to snapshot
1404 * The debugger specifies the size of 1404 * The debugger specifies the size of the array allocated in @num_devices.
1405 * KFD returns the number of devices 1405 * KFD returns the number of devices that actually existed. If this is
1406 * larger than the size specified by 1406 * larger than the size specified by the debugger, KFD will not overflow
1407 * the array allocated by the debugge 1407 * the array allocated by the debugger.
1408 * 1408 *
1409 * @entry_size (IN/OUT) - size per 1409 * @entry_size (IN/OUT) - size per entry in bytes
1410 * The debugger specifies sizeof(stru 1410 * The debugger specifies sizeof(struct kfd_dbg_device_info_entry) in
1411 * @entry_size. KFD returns the numbe 1411 * @entry_size. KFD returns the number of bytes actually populated. The
1412 * debugger should use KFD_IOCTL_MINO 1412 * debugger should use KFD_IOCTL_MINOR_VERSION to determine, which fields
1413 * in struct kfd_dbg_device_info_entr 1413 * in struct kfd_dbg_device_info_entry are valid. This allows growing the
1414 * ABI in a backwards compatible mann 1414 * ABI in a backwards compatible manner.
1415 * Note that entry_size(IN) should st 1415 * Note that entry_size(IN) should still be used to stride the snapshot buffer in the
1416 * event that it's larger than actual 1416 * event that it's larger than actual kfd_dbg_device_info_entry.
1417 * 1417 *
1418 * Generic errors apply (see kfd_dbg_trap 1418 * Generic errors apply (see kfd_dbg_trap_operations).
1419 * Return - 0 on SUCCESS. 1419 * Return - 0 on SUCCESS.
1420 * Copies @num_devices(IN) devic 1420 * Copies @num_devices(IN) device snapshot entries of size @entry_size(IN)
1421 * into @snapshot_buf_ptr if @nu 1421 * into @snapshot_buf_ptr if @num_devices(IN) > 0.
1422 * Otherwise return @num_devices 1422 * Otherwise return @num_devices(OUT) queue snapshot entries that exist.
1423 */ 1423 */
1424 struct kfd_ioctl_dbg_trap_device_snapshot_arg 1424 struct kfd_ioctl_dbg_trap_device_snapshot_args {
1425 __u64 exception_mask; 1425 __u64 exception_mask;
1426 __u64 snapshot_buf_ptr; 1426 __u64 snapshot_buf_ptr;
1427 __u32 num_devices; 1427 __u32 num_devices;
1428 __u32 entry_size; 1428 __u32 entry_size;
1429 }; 1429 };
1430 1430
1431 /** 1431 /**
1432 * kfd_ioctl_dbg_trap_args 1432 * kfd_ioctl_dbg_trap_args
1433 * 1433 *
1434 * Arguments to debug target process. 1434 * Arguments to debug target process.
1435 * 1435 *
1436 * @pid - target process to debug 1436 * @pid - target process to debug
1437 * @op - debug operation (see kfd_dbg_tr 1437 * @op - debug operation (see kfd_dbg_trap_operations)
1438 * 1438 *
1439 * @op determines which union struct args 1439 * @op determines which union struct args to use.
1440 * Refer to kern docs for each kfd_ioctl_ 1440 * Refer to kern docs for each kfd_ioctl_dbg_trap_*_args struct.
1441 */ 1441 */
1442 struct kfd_ioctl_dbg_trap_args { 1442 struct kfd_ioctl_dbg_trap_args {
1443 __u32 pid; 1443 __u32 pid;
1444 __u32 op; 1444 __u32 op;
1445 1445
1446 union { 1446 union {
1447 struct kfd_ioctl_dbg_trap_ena 1447 struct kfd_ioctl_dbg_trap_enable_args enable;
1448 struct kfd_ioctl_dbg_trap_sen 1448 struct kfd_ioctl_dbg_trap_send_runtime_event_args send_runtime_event;
1449 struct kfd_ioctl_dbg_trap_set 1449 struct kfd_ioctl_dbg_trap_set_exceptions_enabled_args set_exceptions_enabled;
1450 struct kfd_ioctl_dbg_trap_set 1450 struct kfd_ioctl_dbg_trap_set_wave_launch_override_args launch_override;
1451 struct kfd_ioctl_dbg_trap_set 1451 struct kfd_ioctl_dbg_trap_set_wave_launch_mode_args launch_mode;
1452 struct kfd_ioctl_dbg_trap_sus 1452 struct kfd_ioctl_dbg_trap_suspend_queues_args suspend_queues;
1453 struct kfd_ioctl_dbg_trap_res 1453 struct kfd_ioctl_dbg_trap_resume_queues_args resume_queues;
1454 struct kfd_ioctl_dbg_trap_set 1454 struct kfd_ioctl_dbg_trap_set_node_address_watch_args set_node_address_watch;
1455 struct kfd_ioctl_dbg_trap_cle 1455 struct kfd_ioctl_dbg_trap_clear_node_address_watch_args clear_node_address_watch;
1456 struct kfd_ioctl_dbg_trap_set 1456 struct kfd_ioctl_dbg_trap_set_flags_args set_flags;
1457 struct kfd_ioctl_dbg_trap_que 1457 struct kfd_ioctl_dbg_trap_query_debug_event_args query_debug_event;
1458 struct kfd_ioctl_dbg_trap_que 1458 struct kfd_ioctl_dbg_trap_query_exception_info_args query_exception_info;
1459 struct kfd_ioctl_dbg_trap_que 1459 struct kfd_ioctl_dbg_trap_queue_snapshot_args queue_snapshot;
1460 struct kfd_ioctl_dbg_trap_dev 1460 struct kfd_ioctl_dbg_trap_device_snapshot_args device_snapshot;
1461 }; 1461 };
1462 }; 1462 };
1463 1463
1464 #define AMDKFD_IOCTL_BASE 'K' 1464 #define AMDKFD_IOCTL_BASE 'K'
1465 #define AMDKFD_IO(nr) _IO(A 1465 #define AMDKFD_IO(nr) _IO(AMDKFD_IOCTL_BASE, nr)
1466 #define AMDKFD_IOR(nr, type) _IOR( 1466 #define AMDKFD_IOR(nr, type) _IOR(AMDKFD_IOCTL_BASE, nr, type)
1467 #define AMDKFD_IOW(nr, type) _IOW( 1467 #define AMDKFD_IOW(nr, type) _IOW(AMDKFD_IOCTL_BASE, nr, type)
1468 #define AMDKFD_IOWR(nr, type) _IOWR 1468 #define AMDKFD_IOWR(nr, type) _IOWR(AMDKFD_IOCTL_BASE, nr, type)
1469 1469
1470 #define AMDKFD_IOC_GET_VERSION 1470 #define AMDKFD_IOC_GET_VERSION \
1471 AMDKFD_IOR(0x01, struct kfd_i 1471 AMDKFD_IOR(0x01, struct kfd_ioctl_get_version_args)
1472 1472
1473 #define AMDKFD_IOC_CREATE_QUEUE 1473 #define AMDKFD_IOC_CREATE_QUEUE \
1474 AMDKFD_IOWR(0x02, struct kfd_ 1474 AMDKFD_IOWR(0x02, struct kfd_ioctl_create_queue_args)
1475 1475
1476 #define AMDKFD_IOC_DESTROY_QUEUE 1476 #define AMDKFD_IOC_DESTROY_QUEUE \
1477 AMDKFD_IOWR(0x03, struct kfd_ 1477 AMDKFD_IOWR(0x03, struct kfd_ioctl_destroy_queue_args)
1478 1478
1479 #define AMDKFD_IOC_SET_MEMORY_POLICY 1479 #define AMDKFD_IOC_SET_MEMORY_POLICY \
1480 AMDKFD_IOW(0x04, struct kfd_i 1480 AMDKFD_IOW(0x04, struct kfd_ioctl_set_memory_policy_args)
1481 1481
1482 #define AMDKFD_IOC_GET_CLOCK_COUNTERS 1482 #define AMDKFD_IOC_GET_CLOCK_COUNTERS \
1483 AMDKFD_IOWR(0x05, struct kfd_ 1483 AMDKFD_IOWR(0x05, struct kfd_ioctl_get_clock_counters_args)
1484 1484
1485 #define AMDKFD_IOC_GET_PROCESS_APERTURES 1485 #define AMDKFD_IOC_GET_PROCESS_APERTURES \
1486 AMDKFD_IOR(0x06, struct kfd_i 1486 AMDKFD_IOR(0x06, struct kfd_ioctl_get_process_apertures_args)
1487 1487
1488 #define AMDKFD_IOC_UPDATE_QUEUE 1488 #define AMDKFD_IOC_UPDATE_QUEUE \
1489 AMDKFD_IOW(0x07, struct kfd_i 1489 AMDKFD_IOW(0x07, struct kfd_ioctl_update_queue_args)
1490 1490
1491 #define AMDKFD_IOC_CREATE_EVENT 1491 #define AMDKFD_IOC_CREATE_EVENT \
1492 AMDKFD_IOWR(0x08, struct kfd_ 1492 AMDKFD_IOWR(0x08, struct kfd_ioctl_create_event_args)
1493 1493
1494 #define AMDKFD_IOC_DESTROY_EVENT 1494 #define AMDKFD_IOC_DESTROY_EVENT \
1495 AMDKFD_IOW(0x09, struct kfd_i 1495 AMDKFD_IOW(0x09, struct kfd_ioctl_destroy_event_args)
1496 1496
1497 #define AMDKFD_IOC_SET_EVENT 1497 #define AMDKFD_IOC_SET_EVENT \
1498 AMDKFD_IOW(0x0A, struct kfd_i 1498 AMDKFD_IOW(0x0A, struct kfd_ioctl_set_event_args)
1499 1499
1500 #define AMDKFD_IOC_RESET_EVENT 1500 #define AMDKFD_IOC_RESET_EVENT \
1501 AMDKFD_IOW(0x0B, struct kfd_i 1501 AMDKFD_IOW(0x0B, struct kfd_ioctl_reset_event_args)
1502 1502
1503 #define AMDKFD_IOC_WAIT_EVENTS 1503 #define AMDKFD_IOC_WAIT_EVENTS \
1504 AMDKFD_IOWR(0x0C, struct kfd_ 1504 AMDKFD_IOWR(0x0C, struct kfd_ioctl_wait_events_args)
1505 1505
1506 #define AMDKFD_IOC_DBG_REGISTER_DEPRECATED 1506 #define AMDKFD_IOC_DBG_REGISTER_DEPRECATED \
1507 AMDKFD_IOW(0x0D, struct kfd_i 1507 AMDKFD_IOW(0x0D, struct kfd_ioctl_dbg_register_args)
1508 1508
1509 #define AMDKFD_IOC_DBG_UNREGISTER_DEPRECATED 1509 #define AMDKFD_IOC_DBG_UNREGISTER_DEPRECATED \
1510 AMDKFD_IOW(0x0E, struct kfd_i 1510 AMDKFD_IOW(0x0E, struct kfd_ioctl_dbg_unregister_args)
1511 1511
1512 #define AMDKFD_IOC_DBG_ADDRESS_WATCH_DEPRECAT 1512 #define AMDKFD_IOC_DBG_ADDRESS_WATCH_DEPRECATED \
1513 AMDKFD_IOW(0x0F, struct kfd_i 1513 AMDKFD_IOW(0x0F, struct kfd_ioctl_dbg_address_watch_args)
1514 1514
1515 #define AMDKFD_IOC_DBG_WAVE_CONTROL_DEPRECATE 1515 #define AMDKFD_IOC_DBG_WAVE_CONTROL_DEPRECATED \
1516 AMDKFD_IOW(0x10, struct kfd_i 1516 AMDKFD_IOW(0x10, struct kfd_ioctl_dbg_wave_control_args)
1517 1517
1518 #define AMDKFD_IOC_SET_SCRATCH_BACKING_VA 1518 #define AMDKFD_IOC_SET_SCRATCH_BACKING_VA \
1519 AMDKFD_IOWR(0x11, struct kfd_ 1519 AMDKFD_IOWR(0x11, struct kfd_ioctl_set_scratch_backing_va_args)
1520 1520
1521 #define AMDKFD_IOC_GET_TILE_CONFIG 1521 #define AMDKFD_IOC_GET_TILE_CONFIG \
1522 AMDKFD_IOWR(0x12, struct kfd_ 1522 AMDKFD_IOWR(0x12, struct kfd_ioctl_get_tile_config_args)
1523 1523
1524 #define AMDKFD_IOC_SET_TRAP_HANDLER 1524 #define AMDKFD_IOC_SET_TRAP_HANDLER \
1525 AMDKFD_IOW(0x13, struct kfd_i 1525 AMDKFD_IOW(0x13, struct kfd_ioctl_set_trap_handler_args)
1526 1526
1527 #define AMDKFD_IOC_GET_PROCESS_APERTURES_NEW 1527 #define AMDKFD_IOC_GET_PROCESS_APERTURES_NEW \
1528 AMDKFD_IOWR(0x14, 1528 AMDKFD_IOWR(0x14, \
1529 struct kfd_ioctl_get_ 1529 struct kfd_ioctl_get_process_apertures_new_args)
1530 1530
1531 #define AMDKFD_IOC_ACQUIRE_VM 1531 #define AMDKFD_IOC_ACQUIRE_VM \
1532 AMDKFD_IOW(0x15, struct kfd_i 1532 AMDKFD_IOW(0x15, struct kfd_ioctl_acquire_vm_args)
1533 1533
1534 #define AMDKFD_IOC_ALLOC_MEMORY_OF_GPU 1534 #define AMDKFD_IOC_ALLOC_MEMORY_OF_GPU \
1535 AMDKFD_IOWR(0x16, struct kfd_ 1535 AMDKFD_IOWR(0x16, struct kfd_ioctl_alloc_memory_of_gpu_args)
1536 1536
1537 #define AMDKFD_IOC_FREE_MEMORY_OF_GPU 1537 #define AMDKFD_IOC_FREE_MEMORY_OF_GPU \
1538 AMDKFD_IOW(0x17, struct kfd_i 1538 AMDKFD_IOW(0x17, struct kfd_ioctl_free_memory_of_gpu_args)
1539 1539
1540 #define AMDKFD_IOC_MAP_MEMORY_TO_GPU 1540 #define AMDKFD_IOC_MAP_MEMORY_TO_GPU \
1541 AMDKFD_IOWR(0x18, struct kfd_ 1541 AMDKFD_IOWR(0x18, struct kfd_ioctl_map_memory_to_gpu_args)
1542 1542
1543 #define AMDKFD_IOC_UNMAP_MEMORY_FROM_GPU 1543 #define AMDKFD_IOC_UNMAP_MEMORY_FROM_GPU \
1544 AMDKFD_IOWR(0x19, struct kfd_ 1544 AMDKFD_IOWR(0x19, struct kfd_ioctl_unmap_memory_from_gpu_args)
1545 1545
1546 #define AMDKFD_IOC_SET_CU_MASK \ 1546 #define AMDKFD_IOC_SET_CU_MASK \
1547 AMDKFD_IOW(0x1A, struct kfd_i 1547 AMDKFD_IOW(0x1A, struct kfd_ioctl_set_cu_mask_args)
1548 1548
1549 #define AMDKFD_IOC_GET_QUEUE_WAVE_STATE 1549 #define AMDKFD_IOC_GET_QUEUE_WAVE_STATE \
1550 AMDKFD_IOWR(0x1B, struct kfd_ 1550 AMDKFD_IOWR(0x1B, struct kfd_ioctl_get_queue_wave_state_args)
1551 1551
1552 #define AMDKFD_IOC_GET_DMABUF_INFO 1552 #define AMDKFD_IOC_GET_DMABUF_INFO \
1553 AMDKFD_IOWR(0x1C, struct kfd_ 1553 AMDKFD_IOWR(0x1C, struct kfd_ioctl_get_dmabuf_info_args)
1554 1554
1555 #define AMDKFD_IOC_IMPORT_DMABUF 1555 #define AMDKFD_IOC_IMPORT_DMABUF \
1556 AMDKFD_IOWR(0x1D, struct kfd_ 1556 AMDKFD_IOWR(0x1D, struct kfd_ioctl_import_dmabuf_args)
1557 1557
1558 #define AMDKFD_IOC_ALLOC_QUEUE_GWS 1558 #define AMDKFD_IOC_ALLOC_QUEUE_GWS \
1559 AMDKFD_IOWR(0x1E, struct kfd_ 1559 AMDKFD_IOWR(0x1E, struct kfd_ioctl_alloc_queue_gws_args)
1560 1560
1561 #define AMDKFD_IOC_SMI_EVENTS 1561 #define AMDKFD_IOC_SMI_EVENTS \
1562 AMDKFD_IOWR(0x1F, struct kfd_ 1562 AMDKFD_IOWR(0x1F, struct kfd_ioctl_smi_events_args)
1563 1563
1564 #define AMDKFD_IOC_SVM AMDKFD_IOWR(0x20, str 1564 #define AMDKFD_IOC_SVM AMDKFD_IOWR(0x20, struct kfd_ioctl_svm_args)
1565 1565
1566 #define AMDKFD_IOC_SET_XNACK_MODE 1566 #define AMDKFD_IOC_SET_XNACK_MODE \
1567 AMDKFD_IOWR(0x21, struct kfd_ 1567 AMDKFD_IOWR(0x21, struct kfd_ioctl_set_xnack_mode_args)
1568 1568
1569 #define AMDKFD_IOC_CRIU_OP 1569 #define AMDKFD_IOC_CRIU_OP \
1570 AMDKFD_IOWR(0x22, struct kfd_ 1570 AMDKFD_IOWR(0x22, struct kfd_ioctl_criu_args)
1571 1571
1572 #define AMDKFD_IOC_AVAILABLE_MEMORY 1572 #define AMDKFD_IOC_AVAILABLE_MEMORY \
1573 AMDKFD_IOWR(0x23, struct kfd_ 1573 AMDKFD_IOWR(0x23, struct kfd_ioctl_get_available_memory_args)
1574 1574
1575 #define AMDKFD_IOC_EXPORT_DMABUF 1575 #define AMDKFD_IOC_EXPORT_DMABUF \
1576 AMDKFD_IOWR(0x24, struct kfd_ 1576 AMDKFD_IOWR(0x24, struct kfd_ioctl_export_dmabuf_args)
1577 1577
1578 #define AMDKFD_IOC_RUNTIME_ENABLE 1578 #define AMDKFD_IOC_RUNTIME_ENABLE \
1579 AMDKFD_IOWR(0x25, struct kfd_ 1579 AMDKFD_IOWR(0x25, struct kfd_ioctl_runtime_enable_args)
1580 1580
1581 #define AMDKFD_IOC_DBG_TRAP 1581 #define AMDKFD_IOC_DBG_TRAP \
1582 AMDKFD_IOWR(0x26, struct kfd_ 1582 AMDKFD_IOWR(0x26, struct kfd_ioctl_dbg_trap_args)
1583 1583
1584 #define AMDKFD_COMMAND_START 0x01 1584 #define AMDKFD_COMMAND_START 0x01
1585 #define AMDKFD_COMMAND_END 0x27 1585 #define AMDKFD_COMMAND_END 0x27
1586 1586
1587 #endif 1587 #endif
1588 1588
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.