1 /* SPDX-License-Identifier: GPL-2.0 OR MIT */
2 /*
3 * Copyright 2014-2022 Advanced Micro Devices, Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
22 *
23 */
24
25 #ifndef KFD_DEVICE_QUEUE_MANAGER_H_
26 #define KFD_DEVICE_QUEUE_MANAGER_H_
27
28 #include <linux/rwsem.h>
29 #include <linux/list.h>
30 #include <linux/mutex.h>
31 #include <linux/sched/mm.h>
32 #include "kfd_priv.h"
33 #include "kfd_mqd_manager.h"
34
35
36 #define VMID_NUM 16
37
38 #define KFD_MES_PROCESS_QUANTUM 100000
39 #define KFD_MES_GANG_QUANTUM 10000
40 #define USE_DEFAULT_GRACE_PERIOD 0xffffffff
41
42 struct device_process_node {
43 struct qcm_process_device *qpd;
44 struct list_head list;
45 };
46
47 union SQ_CMD_BITS {
48 struct {
49 uint32_t cmd:3;
50 uint32_t:1;
51 uint32_t mode:3;
52 uint32_t check_vmid:1;
53 uint32_t trap_id:3;
54 uint32_t:5;
55 uint32_t wave_id:4;
56 uint32_t simd_id:2;
57 uint32_t:2;
58 uint32_t queue_id:3;
59 uint32_t:1;
60 uint32_t vm_id:4;
61 } bitfields, bits;
62 uint32_t u32All;
63 signed int i32All;
64 float f32All;
65 };
66
67 union GRBM_GFX_INDEX_BITS {
68 struct {
69 uint32_t instance_index:8;
70 uint32_t sh_index:8;
71 uint32_t se_index:8;
72 uint32_t:5;
73 uint32_t sh_broadcast_writes:1;
74 uint32_t instance_broadcast_writes:1;
75 uint32_t se_broadcast_writes:1;
76 } bitfields, bits;
77 uint32_t u32All;
78 signed int i32All;
79 float f32All;
80 };
81
82 /**
83 * struct device_queue_manager_ops
84 *
85 * @create_queue: Queue creation routine.
86 *
87 * @destroy_queue: Queue destruction routine.
88 *
89 * @update_queue: Queue update routine.
90 *
91 * @exeute_queues: Dispatches the queues list to the H/W.
92 *
93 * @register_process: This routine associates a specific process with device.
94 *
95 * @unregister_process: destroys the associations between process to device.
96 *
97 * @initialize: Initializes the pipelines and memory module for that device.
98 *
99 * @start: Initializes the resources/modules the device needs for queues
100 * execution. This function is called on device initialization and after the
101 * system woke up after suspension.
102 *
103 * @stop: This routine stops execution of all the active queue running on the
104 * H/W and basically this function called on system suspend.
105 *
106 * @uninitialize: Destroys all the device queue manager resources allocated in
107 * initialize routine.
108 *
109 * @halt: This routine unmaps queues from runlist and set halt status to true
110 * so no more queues will be mapped to runlist until unhalt.
111 *
112 * @unhalt: This routine unset halt status to flase and maps queues back to
113 * runlist.
114 *
115 * @create_kernel_queue: Creates kernel queue. Used for debug queue.
116 *
117 * @destroy_kernel_queue: Destroys kernel queue. Used for debug queue.
118 *
119 * @set_cache_memory_policy: Sets memory policy (cached/ non cached) for the
120 * memory apertures.
121 *
122 * @process_termination: Clears all process queues belongs to that device.
123 *
124 * @evict_process_queues: Evict all active queues of a process
125 *
126 * @restore_process_queues: Restore all evicted queues of a process
127 *
128 * @get_wave_state: Retrieves context save state and optionally copies the
129 * control stack, if kept in the MQD, to the given userspace address.
130 *
131 * @reset_queues: reset queues which consume RAS poison
132 * @get_queue_checkpoint_info: Retrieves queue size information for CRIU checkpoint.
133 *
134 * @checkpoint_mqd: checkpoint queue MQD contents for CRIU.
135 */
136
137 struct device_queue_manager_ops {
138 int (*create_queue)(struct device_queue_manager *dqm,
139 struct queue *q,
140 struct qcm_process_device *qpd,
141 const struct kfd_criu_queue_priv_data *qd,
142 const void *restore_mqd,
143 const void *restore_ctl_stack);
144
145 int (*destroy_queue)(struct device_queue_manager *dqm,
146 struct qcm_process_device *qpd,
147 struct queue *q);
148
149 int (*update_queue)(struct device_queue_manager *dqm,
150 struct queue *q, struct mqd_update_info *minfo);
151
152 int (*register_process)(struct device_queue_manager *dqm,
153 struct qcm_process_device *qpd);
154
155 int (*unregister_process)(struct device_queue_manager *dqm,
156 struct qcm_process_device *qpd);
157
158 int (*initialize)(struct device_queue_manager *dqm);
159 int (*start)(struct device_queue_manager *dqm);
160 int (*stop)(struct device_queue_manager *dqm);
161 void (*uninitialize)(struct device_queue_manager *dqm);
162 int (*halt)(struct device_queue_manager *dqm);
163 int (*unhalt)(struct device_queue_manager *dqm);
164 int (*create_kernel_queue)(struct device_queue_manager *dqm,
165 struct kernel_queue *kq,
166 struct qcm_process_device *qpd);
167
168 void (*destroy_kernel_queue)(struct device_queue_manager *dqm,
169 struct kernel_queue *kq,
170 struct qcm_process_device *qpd);
171
172 bool (*set_cache_memory_policy)(struct device_queue_manager *dqm,
173 struct qcm_process_device *qpd,
174 enum cache_policy default_policy,
175 enum cache_policy alternate_policy,
176 void __user *alternate_aperture_base,
177 uint64_t alternate_aperture_size);
178
179 int (*process_termination)(struct device_queue_manager *dqm,
180 struct qcm_process_device *qpd);
181
182 int (*evict_process_queues)(struct device_queue_manager *dqm,
183 struct qcm_process_device *qpd);
184 int (*restore_process_queues)(struct device_queue_manager *dqm,
185 struct qcm_process_device *qpd);
186
187 int (*get_wave_state)(struct device_queue_manager *dqm,
188 struct queue *q,
189 void __user *ctl_stack,
190 u32 *ctl_stack_used_size,
191 u32 *save_area_used_size);
192
193 int (*reset_queues)(struct device_queue_manager *dqm,
194 uint16_t pasid);
195 void (*get_queue_checkpoint_info)(struct device_queue_manager *dqm,
196 const struct queue *q, u32 *mqd_size,
197 u32 *ctl_stack_size);
198
199 int (*checkpoint_mqd)(struct device_queue_manager *dqm,
200 const struct queue *q,
201 void *mqd,
202 void *ctl_stack);
203 };
204
205 struct device_queue_manager_asic_ops {
206 int (*update_qpd)(struct device_queue_manager *dqm,
207 struct qcm_process_device *qpd);
208 bool (*set_cache_memory_policy)(struct device_queue_manager *dqm,
209 struct qcm_process_device *qpd,
210 enum cache_policy default_policy,
211 enum cache_policy alternate_policy,
212 void __user *alternate_aperture_base,
213 uint64_t alternate_aperture_size);
214 void (*init_sdma_vm)(struct device_queue_manager *dqm,
215 struct queue *q,
216 struct qcm_process_device *qpd);
217 struct mqd_manager * (*mqd_manager_init)(enum KFD_MQD_TYPE type,
218 struct kfd_node *dev);
219 };
220
221 struct dqm_detect_hang_info {
222 int pipe_id;
223 int queue_id;
224 int xcc_id;
225 uint64_t queue_address;
226 };
227
228 /**
229 * struct device_queue_manager
230 *
231 * This struct is a base class for the kfd queues scheduler in the
232 * device level. The device base class should expose the basic operations
233 * for queue creation and queue destruction. This base class hides the
234 * scheduling mode of the driver and the specific implementation of the
235 * concrete device. This class is the only class in the queues scheduler
236 * that configures the H/W.
237 *
238 */
239
240 struct device_queue_manager {
241 struct device_queue_manager_ops ops;
242 struct device_queue_manager_asic_ops asic_ops;
243
244 struct mqd_manager *mqd_mgrs[KFD_MQD_TYPE_MAX];
245 struct packet_manager packet_mgr;
246 struct kfd_node *dev;
247 struct mutex lock_hidden; /* use dqm_lock/unlock(dqm) */
248 struct list_head queues;
249 unsigned int saved_flags;
250 unsigned int processes_count;
251 unsigned int active_queue_count;
252 unsigned int active_cp_queue_count;
253 unsigned int gws_queue_count;
254 unsigned int total_queue_count;
255 unsigned int next_pipe_to_allocate;
256 unsigned int *allocated_queues;
257 DECLARE_BITMAP(sdma_bitmap, KFD_MAX_SDMA_QUEUES);
258 DECLARE_BITMAP(xgmi_sdma_bitmap, KFD_MAX_SDMA_QUEUES);
259 /* the pasid mapping for each kfd vmid */
260 uint16_t vmid_pasid[VMID_NUM];
261 uint64_t pipelines_addr;
262 uint64_t fence_gpu_addr;
263 uint64_t *fence_addr;
264 struct kfd_mem_obj *fence_mem;
265 bool active_runlist;
266 int sched_policy;
267 uint32_t trap_debug_vmid;
268
269 /* hw exception */
270 bool is_hws_hang;
271 bool is_resetting;
272 struct work_struct hw_exception_work;
273 struct kfd_mem_obj hiq_sdma_mqd;
274 bool sched_running;
275 bool sched_halt;
276
277 /* used for GFX 9.4.3 only */
278 uint32_t current_logical_xcc_start;
279
280 uint32_t wait_times;
281
282 wait_queue_head_t destroy_wait;
283
284 /* for per-queue reset support */
285 struct dqm_detect_hang_info *detect_hang_info;
286 size_t detect_hang_info_size;
287 int detect_hang_count;
288 };
289
290 void device_queue_manager_init_cik(
291 struct device_queue_manager_asic_ops *asic_ops);
292 void device_queue_manager_init_vi(
293 struct device_queue_manager_asic_ops *asic_ops);
294 void device_queue_manager_init_v9(
295 struct device_queue_manager_asic_ops *asic_ops);
296 void device_queue_manager_init_v10(
297 struct device_queue_manager_asic_ops *asic_ops);
298 void device_queue_manager_init_v11(
299 struct device_queue_manager_asic_ops *asic_ops);
300 void device_queue_manager_init_v12(
301 struct device_queue_manager_asic_ops *asic_ops);
302 void program_sh_mem_settings(struct device_queue_manager *dqm,
303 struct qcm_process_device *qpd);
304 unsigned int get_cp_queues_num(struct device_queue_manager *dqm);
305 unsigned int get_queues_per_pipe(struct device_queue_manager *dqm);
306 unsigned int get_pipes_per_mec(struct device_queue_manager *dqm);
307 unsigned int get_num_sdma_queues(struct device_queue_manager *dqm);
308 unsigned int get_num_xgmi_sdma_queues(struct device_queue_manager *dqm);
309 int reserve_debug_trap_vmid(struct device_queue_manager *dqm,
310 struct qcm_process_device *qpd);
311 int release_debug_trap_vmid(struct device_queue_manager *dqm,
312 struct qcm_process_device *qpd);
313 int suspend_queues(struct kfd_process *p,
314 uint32_t num_queues,
315 uint32_t grace_period,
316 uint64_t exception_clear_mask,
317 uint32_t *usr_queue_id_array);
318 int resume_queues(struct kfd_process *p,
319 uint32_t num_queues,
320 uint32_t *usr_queue_id_array);
321 void set_queue_snapshot_entry(struct queue *q,
322 uint64_t exception_clear_mask,
323 struct kfd_queue_snapshot_entry *qss_entry);
324 int debug_lock_and_unmap(struct device_queue_manager *dqm);
325 int debug_map_and_unlock(struct device_queue_manager *dqm);
326 int debug_refresh_runlist(struct device_queue_manager *dqm);
327 bool kfd_dqm_is_queue_in_process(struct device_queue_manager *dqm,
328 struct qcm_process_device *qpd,
329 int doorbell_off, u32 *queue_format);
330
get_sh_mem_bases_32(struct kfd_process_device * pdd)331 static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
332 {
333 return (pdd->lds_base >> 16) & 0xFF;
334 }
335
336 static inline unsigned int
get_sh_mem_bases_nybble_64(struct kfd_process_device * pdd)337 get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd)
338 {
339 return (pdd->lds_base >> 60) & 0x0E;
340 }
341
342 /* The DQM lock can be taken in MMU notifiers. Make sure no reclaim-FS
343 * happens while holding this lock anywhere to prevent deadlocks when
344 * an MMU notifier runs in reclaim-FS context.
345 */
dqm_lock(struct device_queue_manager * dqm)346 static inline void dqm_lock(struct device_queue_manager *dqm)
347 {
348 mutex_lock(&dqm->lock_hidden);
349 dqm->saved_flags = memalloc_noreclaim_save();
350 }
dqm_unlock(struct device_queue_manager * dqm)351 static inline void dqm_unlock(struct device_queue_manager *dqm)
352 {
353 memalloc_noreclaim_restore(dqm->saved_flags);
354 mutex_unlock(&dqm->lock_hidden);
355 }
356
read_sdma_queue_counter(uint64_t __user * q_rptr,uint64_t * val)357 static inline int read_sdma_queue_counter(uint64_t __user *q_rptr, uint64_t *val)
358 {
359 /* SDMA activity counter is stored at queue's RPTR + 0x8 location. */
360 return get_user(*val, q_rptr + 1);
361 }
362 #endif /* KFD_DEVICE_QUEUE_MANAGER_H_ */
363