This source file includes following definitions.
- inc_wptr
- pm_calc_rlib_size
- pm_allocate_runlist_ib
- pm_create_runlist_ib
- pm_init
- pm_uninit
- pm_send_set_resources
- pm_send_runlist
- pm_send_query_status
- pm_send_unmap_queue
- pm_release_ib
- pm_debugfs_runlist
- pm_debugfs_hang_hws
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24 #include <linux/slab.h>
25 #include <linux/mutex.h>
26 #include "kfd_device_queue_manager.h"
27 #include "kfd_kernel_queue.h"
28 #include "kfd_priv.h"
29
30 static inline void inc_wptr(unsigned int *wptr, unsigned int increment_bytes,
31 unsigned int buffer_size_bytes)
32 {
33 unsigned int temp = *wptr + increment_bytes / sizeof(uint32_t);
34
35 WARN((temp * sizeof(uint32_t)) > buffer_size_bytes,
36 "Runlist IB overflow");
37 *wptr = temp;
38 }
39
40 static void pm_calc_rlib_size(struct packet_manager *pm,
41 unsigned int *rlib_size,
42 bool *over_subscription)
43 {
44 unsigned int process_count, queue_count, compute_queue_count;
45 unsigned int map_queue_size;
46 unsigned int max_proc_per_quantum = 1;
47 struct kfd_dev *dev = pm->dqm->dev;
48
49 process_count = pm->dqm->processes_count;
50 queue_count = pm->dqm->queue_count;
51 compute_queue_count = queue_count - pm->dqm->sdma_queue_count -
52 pm->dqm->xgmi_sdma_queue_count;
53
54
55
56
57
58
59 *over_subscription = false;
60
61 if (dev->max_proc_per_quantum > 1)
62 max_proc_per_quantum = dev->max_proc_per_quantum;
63
64 if ((process_count > max_proc_per_quantum) ||
65 compute_queue_count > get_queues_num(pm->dqm)) {
66 *over_subscription = true;
67 pr_debug("Over subscribed runlist\n");
68 }
69
70 map_queue_size = pm->pmf->map_queues_size;
71
72 *rlib_size = process_count * pm->pmf->map_process_size +
73 queue_count * map_queue_size;
74
75
76
77
78
79 if (*over_subscription)
80 *rlib_size += pm->pmf->runlist_size;
81
82 pr_debug("runlist ib size %d\n", *rlib_size);
83 }
84
85 static int pm_allocate_runlist_ib(struct packet_manager *pm,
86 unsigned int **rl_buffer,
87 uint64_t *rl_gpu_buffer,
88 unsigned int *rl_buffer_size,
89 bool *is_over_subscription)
90 {
91 int retval;
92
93 if (WARN_ON(pm->allocated))
94 return -EINVAL;
95
96 pm_calc_rlib_size(pm, rl_buffer_size, is_over_subscription);
97
98 mutex_lock(&pm->lock);
99
100 retval = kfd_gtt_sa_allocate(pm->dqm->dev, *rl_buffer_size,
101 &pm->ib_buffer_obj);
102
103 if (retval) {
104 pr_err("Failed to allocate runlist IB\n");
105 goto out;
106 }
107
108 *(void **)rl_buffer = pm->ib_buffer_obj->cpu_ptr;
109 *rl_gpu_buffer = pm->ib_buffer_obj->gpu_addr;
110
111 memset(*rl_buffer, 0, *rl_buffer_size);
112 pm->allocated = true;
113
114 out:
115 mutex_unlock(&pm->lock);
116 return retval;
117 }
118
119 static int pm_create_runlist_ib(struct packet_manager *pm,
120 struct list_head *queues,
121 uint64_t *rl_gpu_addr,
122 size_t *rl_size_bytes)
123 {
124 unsigned int alloc_size_bytes;
125 unsigned int *rl_buffer, rl_wptr, i;
126 int retval, proccesses_mapped;
127 struct device_process_node *cur;
128 struct qcm_process_device *qpd;
129 struct queue *q;
130 struct kernel_queue *kq;
131 bool is_over_subscription;
132
133 rl_wptr = retval = proccesses_mapped = 0;
134
135 retval = pm_allocate_runlist_ib(pm, &rl_buffer, rl_gpu_addr,
136 &alloc_size_bytes, &is_over_subscription);
137 if (retval)
138 return retval;
139
140 *rl_size_bytes = alloc_size_bytes;
141 pm->ib_size_bytes = alloc_size_bytes;
142
143 pr_debug("Building runlist ib process count: %d queues count %d\n",
144 pm->dqm->processes_count, pm->dqm->queue_count);
145
146
147 list_for_each_entry(cur, queues, list) {
148 qpd = cur->qpd;
149
150 if (proccesses_mapped >= pm->dqm->processes_count) {
151 pr_debug("Not enough space left in runlist IB\n");
152 pm_release_ib(pm);
153 return -ENOMEM;
154 }
155
156 retval = pm->pmf->map_process(pm, &rl_buffer[rl_wptr], qpd);
157 if (retval)
158 return retval;
159
160 proccesses_mapped++;
161 inc_wptr(&rl_wptr, pm->pmf->map_process_size,
162 alloc_size_bytes);
163
164 list_for_each_entry(kq, &qpd->priv_queue_list, list) {
165 if (!kq->queue->properties.is_active)
166 continue;
167
168 pr_debug("static_queue, mapping kernel q %d, is debug status %d\n",
169 kq->queue->queue, qpd->is_debug);
170
171 retval = pm->pmf->map_queues(pm,
172 &rl_buffer[rl_wptr],
173 kq->queue,
174 qpd->is_debug);
175 if (retval)
176 return retval;
177
178 inc_wptr(&rl_wptr,
179 pm->pmf->map_queues_size,
180 alloc_size_bytes);
181 }
182
183 list_for_each_entry(q, &qpd->queues_list, list) {
184 if (!q->properties.is_active)
185 continue;
186
187 pr_debug("static_queue, mapping user queue %d, is debug status %d\n",
188 q->queue, qpd->is_debug);
189
190 retval = pm->pmf->map_queues(pm,
191 &rl_buffer[rl_wptr],
192 q,
193 qpd->is_debug);
194
195 if (retval)
196 return retval;
197
198 inc_wptr(&rl_wptr,
199 pm->pmf->map_queues_size,
200 alloc_size_bytes);
201 }
202 }
203
204 pr_debug("Finished map process and queues to runlist\n");
205
206 if (is_over_subscription) {
207 if (!pm->is_over_subscription)
208 pr_warn("Runlist is getting oversubscribed. Expect reduced ROCm performance.\n");
209 retval = pm->pmf->runlist(pm, &rl_buffer[rl_wptr],
210 *rl_gpu_addr,
211 alloc_size_bytes / sizeof(uint32_t),
212 true);
213 }
214 pm->is_over_subscription = is_over_subscription;
215
216 for (i = 0; i < alloc_size_bytes / sizeof(uint32_t); i++)
217 pr_debug("0x%2X ", rl_buffer[i]);
218 pr_debug("\n");
219
220 return retval;
221 }
222
223 int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm)
224 {
225 switch (dqm->dev->device_info->asic_family) {
226 case CHIP_KAVERI:
227 case CHIP_HAWAII:
228
229 case CHIP_CARRIZO:
230 case CHIP_TONGA:
231 case CHIP_FIJI:
232 case CHIP_POLARIS10:
233 case CHIP_POLARIS11:
234 case CHIP_POLARIS12:
235 case CHIP_VEGAM:
236 pm->pmf = &kfd_vi_pm_funcs;
237 break;
238 case CHIP_VEGA10:
239 case CHIP_VEGA12:
240 case CHIP_VEGA20:
241 case CHIP_RAVEN:
242 case CHIP_ARCTURUS:
243 pm->pmf = &kfd_v9_pm_funcs;
244 break;
245 case CHIP_NAVI10:
246 pm->pmf = &kfd_v10_pm_funcs;
247 break;
248 default:
249 WARN(1, "Unexpected ASIC family %u",
250 dqm->dev->device_info->asic_family);
251 return -EINVAL;
252 }
253
254 pm->dqm = dqm;
255 mutex_init(&pm->lock);
256 pm->priv_queue = kernel_queue_init(dqm->dev, KFD_QUEUE_TYPE_HIQ);
257 if (!pm->priv_queue) {
258 mutex_destroy(&pm->lock);
259 return -ENOMEM;
260 }
261 pm->allocated = false;
262
263 return 0;
264 }
265
266 void pm_uninit(struct packet_manager *pm)
267 {
268 mutex_destroy(&pm->lock);
269 kernel_queue_uninit(pm->priv_queue);
270 }
271
272 int pm_send_set_resources(struct packet_manager *pm,
273 struct scheduling_resources *res)
274 {
275 uint32_t *buffer, size;
276 int retval = 0;
277
278 size = pm->pmf->set_resources_size;
279 mutex_lock(&pm->lock);
280 pm->priv_queue->ops.acquire_packet_buffer(pm->priv_queue,
281 size / sizeof(uint32_t),
282 (unsigned int **)&buffer);
283 if (!buffer) {
284 pr_err("Failed to allocate buffer on kernel queue\n");
285 retval = -ENOMEM;
286 goto out;
287 }
288
289 retval = pm->pmf->set_resources(pm, buffer, res);
290 if (!retval)
291 pm->priv_queue->ops.submit_packet(pm->priv_queue);
292 else
293 pm->priv_queue->ops.rollback_packet(pm->priv_queue);
294
295 out:
296 mutex_unlock(&pm->lock);
297
298 return retval;
299 }
300
301 int pm_send_runlist(struct packet_manager *pm, struct list_head *dqm_queues)
302 {
303 uint64_t rl_gpu_ib_addr;
304 uint32_t *rl_buffer;
305 size_t rl_ib_size, packet_size_dwords;
306 int retval;
307
308 retval = pm_create_runlist_ib(pm, dqm_queues, &rl_gpu_ib_addr,
309 &rl_ib_size);
310 if (retval)
311 goto fail_create_runlist_ib;
312
313 pr_debug("runlist IB address: 0x%llX\n", rl_gpu_ib_addr);
314
315 packet_size_dwords = pm->pmf->runlist_size / sizeof(uint32_t);
316 mutex_lock(&pm->lock);
317
318 retval = pm->priv_queue->ops.acquire_packet_buffer(pm->priv_queue,
319 packet_size_dwords, &rl_buffer);
320 if (retval)
321 goto fail_acquire_packet_buffer;
322
323 retval = pm->pmf->runlist(pm, rl_buffer, rl_gpu_ib_addr,
324 rl_ib_size / sizeof(uint32_t), false);
325 if (retval)
326 goto fail_create_runlist;
327
328 pm->priv_queue->ops.submit_packet(pm->priv_queue);
329
330 mutex_unlock(&pm->lock);
331
332 return retval;
333
334 fail_create_runlist:
335 pm->priv_queue->ops.rollback_packet(pm->priv_queue);
336 fail_acquire_packet_buffer:
337 mutex_unlock(&pm->lock);
338 fail_create_runlist_ib:
339 pm_release_ib(pm);
340 return retval;
341 }
342
343 int pm_send_query_status(struct packet_manager *pm, uint64_t fence_address,
344 uint32_t fence_value)
345 {
346 uint32_t *buffer, size;
347 int retval = 0;
348
349 if (WARN_ON(!fence_address))
350 return -EFAULT;
351
352 size = pm->pmf->query_status_size;
353 mutex_lock(&pm->lock);
354 pm->priv_queue->ops.acquire_packet_buffer(pm->priv_queue,
355 size / sizeof(uint32_t), (unsigned int **)&buffer);
356 if (!buffer) {
357 pr_err("Failed to allocate buffer on kernel queue\n");
358 retval = -ENOMEM;
359 goto out;
360 }
361
362 retval = pm->pmf->query_status(pm, buffer, fence_address, fence_value);
363 if (!retval)
364 pm->priv_queue->ops.submit_packet(pm->priv_queue);
365 else
366 pm->priv_queue->ops.rollback_packet(pm->priv_queue);
367
368 out:
369 mutex_unlock(&pm->lock);
370 return retval;
371 }
372
373 int pm_send_unmap_queue(struct packet_manager *pm, enum kfd_queue_type type,
374 enum kfd_unmap_queues_filter filter,
375 uint32_t filter_param, bool reset,
376 unsigned int sdma_engine)
377 {
378 uint32_t *buffer, size;
379 int retval = 0;
380
381 size = pm->pmf->unmap_queues_size;
382 mutex_lock(&pm->lock);
383 pm->priv_queue->ops.acquire_packet_buffer(pm->priv_queue,
384 size / sizeof(uint32_t), (unsigned int **)&buffer);
385 if (!buffer) {
386 pr_err("Failed to allocate buffer on kernel queue\n");
387 retval = -ENOMEM;
388 goto out;
389 }
390
391 retval = pm->pmf->unmap_queues(pm, buffer, type, filter, filter_param,
392 reset, sdma_engine);
393 if (!retval)
394 pm->priv_queue->ops.submit_packet(pm->priv_queue);
395 else
396 pm->priv_queue->ops.rollback_packet(pm->priv_queue);
397
398 out:
399 mutex_unlock(&pm->lock);
400 return retval;
401 }
402
403 void pm_release_ib(struct packet_manager *pm)
404 {
405 mutex_lock(&pm->lock);
406 if (pm->allocated) {
407 kfd_gtt_sa_free(pm->dqm->dev, pm->ib_buffer_obj);
408 pm->allocated = false;
409 }
410 mutex_unlock(&pm->lock);
411 }
412
413 #if defined(CONFIG_DEBUG_FS)
414
415 int pm_debugfs_runlist(struct seq_file *m, void *data)
416 {
417 struct packet_manager *pm = data;
418
419 mutex_lock(&pm->lock);
420
421 if (!pm->allocated) {
422 seq_puts(m, " No active runlist\n");
423 goto out;
424 }
425
426 seq_hex_dump(m, " ", DUMP_PREFIX_OFFSET, 32, 4,
427 pm->ib_buffer_obj->cpu_ptr, pm->ib_size_bytes, false);
428
429 out:
430 mutex_unlock(&pm->lock);
431 return 0;
432 }
433
434 int pm_debugfs_hang_hws(struct packet_manager *pm)
435 {
436 uint32_t *buffer, size;
437 int r = 0;
438
439 size = pm->pmf->query_status_size;
440 mutex_lock(&pm->lock);
441 pm->priv_queue->ops.acquire_packet_buffer(pm->priv_queue,
442 size / sizeof(uint32_t), (unsigned int **)&buffer);
443 if (!buffer) {
444 pr_err("Failed to allocate buffer on kernel queue\n");
445 r = -ENOMEM;
446 goto out;
447 }
448 memset(buffer, 0x55, size);
449 pm->priv_queue->ops.submit_packet(pm->priv_queue);
450
451 pr_info("Submitting %x %x %x %x %x %x %x to HIQ to hang the HWS.",
452 buffer[0], buffer[1], buffer[2], buffer[3],
453 buffer[4], buffer[5], buffer[6]);
454 out:
455 mutex_unlock(&pm->lock);
456 return r;
457 }
458
459
460 #endif