projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drm/amdkfd: Turn verbose topology messages into pr_debug
[linux-block.git] / drivers / gpu / drm / amd / amdkfd / kfd_topology.c
CommitLineData
5b5c4e40
EP		 1/*
2 * Copyright 2014 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 */
22
23#include <linux/types.h>
24#include <linux/kernel.h>
25#include <linux/pci.h>
26#include <linux/errno.h>
27#include <linux/acpi.h>
28#include <linux/hash.h>
29#include <linux/cpufreq.h>
f7c826ad 30#include <linux/log2.h>
5b5c4e40
EP		 31
32#include "kfd_priv.h"
33#include "kfd_crat.h"
34#include "kfd_topology.h"
851a645e 35#include "kfd_device_queue_manager.h"
5b5c4e40 36
4f449311
HK		 37/* topology_device_list - Master list of all topology devices */
38static struct list_head topology_device_list;
174de876 39struct kfd_system_properties sys_props;
5b5c4e40
EP		 40
41static DECLARE_RWSEM(topology_lock);
42
43struct kfd_dev *kfd_device_by_id(uint32_t gpu_id)
44{
45 struct kfd_topology_device *top_dev;
46 struct kfd_dev *device = NULL;
47
48 down_read(&topology_lock);
49
50 list_for_each_entry(top_dev, &topology_device_list, list)
51 if (top_dev->gpu_id == gpu_id) {
52 device = top_dev->gpu;
53 break;
54 }
55
56 up_read(&topology_lock);
57
58 return device;
59}
60
61struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev)
62{
63 struct kfd_topology_device *top_dev;
64 struct kfd_dev *device = NULL;
65
66 down_read(&topology_lock);
67
68 list_for_each_entry(top_dev, &topology_device_list, list)
69 if (top_dev->gpu->pdev == pdev) {
70 device = top_dev->gpu;
71 break;
72 }
73
74 up_read(&topology_lock);
75
76 return device;
77}
78
5b5c4e40
EP		 79static void kfd_release_topology_device(struct kfd_topology_device *dev)
80{
81 struct kfd_mem_properties *mem;
82 struct kfd_cache_properties *cache;
83 struct kfd_iolink_properties *iolink;
84
5b5c4e40
EP		 85 list_del(&dev->list);
86
87 while (dev->mem_props.next != &dev->mem_props) {
88 mem = container_of(dev->mem_props.next,
89 struct kfd_mem_properties, list);
90 list_del(&mem->list);
91 kfree(mem);
92 }
93
94 while (dev->cache_props.next != &dev->cache_props) {
95 cache = container_of(dev->cache_props.next,
96 struct kfd_cache_properties, list);
97 list_del(&cache->list);
98 kfree(cache);
99 }
100
101 while (dev->io_link_props.next != &dev->io_link_props) {
102 iolink = container_of(dev->io_link_props.next,
103 struct kfd_iolink_properties, list);
104 list_del(&iolink->list);
105 kfree(iolink);
106 }
107
108 kfree(dev);
5b5c4e40
EP		 109}
110
4f449311 111void kfd_release_topology_device_list(struct list_head *device_list)
5b5c4e40
EP		 112{
113 struct kfd_topology_device *dev;
114
4f449311
HK		 115 while (!list_empty(device_list)) {
116 dev = list_first_entry(device_list,
117 struct kfd_topology_device, list);
5b5c4e40 118 kfd_release_topology_device(dev);
4f449311 119 }
5b5c4e40
EP		 120}
121
4f449311
HK		 122static void kfd_release_live_view(void)
123{
124 kfd_release_topology_device_list(&topology_device_list);
5b5c4e40
EP		 125 memset(&sys_props, 0, sizeof(sys_props));
126}
127
4f449311
HK		 128struct kfd_topology_device *kfd_create_topology_device(
129 struct list_head *device_list)
5b5c4e40
EP		 130{
131 struct kfd_topology_device *dev;
132
133 dev = kfd_alloc_struct(dev);
4eacc26b 134 if (!dev) {
5b5c4e40 135 pr_err("No memory to allocate a topology device");
16b9201c 136 return NULL;
5b5c4e40
EP		 137 }
138
139 INIT_LIST_HEAD(&dev->mem_props);
140 INIT_LIST_HEAD(&dev->cache_props);
141 INIT_LIST_HEAD(&dev->io_link_props);
142
4f449311 143 list_add_tail(&dev->list, device_list);
5b5c4e40
EP		 144
145 return dev;
16b9201c 146}
5b5c4e40 147
5b5c4e40
EP		 148
149#define sysfs_show_gen_prop(buffer, fmt, ...) \
150 snprintf(buffer, PAGE_SIZE, "%s"fmt, buffer, __VA_ARGS__)
151#define sysfs_show_32bit_prop(buffer, name, value) \
152 sysfs_show_gen_prop(buffer, "%s %u\n", name, value)
153#define sysfs_show_64bit_prop(buffer, name, value) \
154 sysfs_show_gen_prop(buffer, "%s %llu\n", name, value)
155#define sysfs_show_32bit_val(buffer, value) \
156 sysfs_show_gen_prop(buffer, "%u\n", value)
157#define sysfs_show_str_val(buffer, value) \
158 sysfs_show_gen_prop(buffer, "%s\n", value)
159
160static ssize_t sysprops_show(struct kobject *kobj, struct attribute *attr,
161 char *buffer)
162{
163 ssize_t ret;
164
165 /* Making sure that the buffer is an empty string */
166 buffer[0] = 0;
167
168 if (attr == &sys_props.attr_genid) {
169 ret = sysfs_show_32bit_val(buffer, sys_props.generation_count);
170 } else if (attr == &sys_props.attr_props) {
171 sysfs_show_64bit_prop(buffer, "platform_oem",
172 sys_props.platform_oem);
173 sysfs_show_64bit_prop(buffer, "platform_id",
174 sys_props.platform_id);
175 ret = sysfs_show_64bit_prop(buffer, "platform_rev",
176 sys_props.platform_rev);
177 } else {
178 ret = -EINVAL;
179 }
180
181 return ret;
182}
183
5108d768
YZ		 184static void kfd_topology_kobj_release(struct kobject *kobj)
185{
186 kfree(kobj);
187}
188
5b5c4e40
EP		 189static const struct sysfs_ops sysprops_ops = {
190 .show = sysprops_show,
191};
192
193static struct kobj_type sysprops_type = {
5108d768 194 .release = kfd_topology_kobj_release,
5b5c4e40
EP		 195 .sysfs_ops = &sysprops_ops,
196};
197
198static ssize_t iolink_show(struct kobject *kobj, struct attribute *attr,
199 char *buffer)
200{
201 ssize_t ret;
202 struct kfd_iolink_properties *iolink;
203
204 /* Making sure that the buffer is an empty string */
205 buffer[0] = 0;
206
207 iolink = container_of(attr, struct kfd_iolink_properties, attr);
208 sysfs_show_32bit_prop(buffer, "type", iolink->iolink_type);
209 sysfs_show_32bit_prop(buffer, "version_major", iolink->ver_maj);
210 sysfs_show_32bit_prop(buffer, "version_minor", iolink->ver_min);
211 sysfs_show_32bit_prop(buffer, "node_from", iolink->node_from);
212 sysfs_show_32bit_prop(buffer, "node_to", iolink->node_to);
213 sysfs_show_32bit_prop(buffer, "weight", iolink->weight);
214 sysfs_show_32bit_prop(buffer, "min_latency", iolink->min_latency);
215 sysfs_show_32bit_prop(buffer, "max_latency", iolink->max_latency);
216 sysfs_show_32bit_prop(buffer, "min_bandwidth", iolink->min_bandwidth);
217 sysfs_show_32bit_prop(buffer, "max_bandwidth", iolink->max_bandwidth);
218 sysfs_show_32bit_prop(buffer, "recommended_transfer_size",
219 iolink->rec_transfer_size);
220 ret = sysfs_show_32bit_prop(buffer, "flags", iolink->flags);
221
222 return ret;
223}
224
225static const struct sysfs_ops iolink_ops = {
226 .show = iolink_show,
227};
228
229static struct kobj_type iolink_type = {
5108d768 230 .release = kfd_topology_kobj_release,
5b5c4e40
EP		 231 .sysfs_ops = &iolink_ops,
232};
233
234static ssize_t mem_show(struct kobject *kobj, struct attribute *attr,
235 char *buffer)
236{
237 ssize_t ret;
238 struct kfd_mem_properties *mem;
239
240 /* Making sure that the buffer is an empty string */
241 buffer[0] = 0;
242
243 mem = container_of(attr, struct kfd_mem_properties, attr);
244 sysfs_show_32bit_prop(buffer, "heap_type", mem->heap_type);
245 sysfs_show_64bit_prop(buffer, "size_in_bytes", mem->size_in_bytes);
246 sysfs_show_32bit_prop(buffer, "flags", mem->flags);
247 sysfs_show_32bit_prop(buffer, "width", mem->width);
248 ret = sysfs_show_32bit_prop(buffer, "mem_clk_max", mem->mem_clk_max);
249
250 return ret;
251}
252
253static const struct sysfs_ops mem_ops = {
254 .show = mem_show,
255};
256
257static struct kobj_type mem_type = {
5108d768 258 .release = kfd_topology_kobj_release,
5b5c4e40
EP		 259 .sysfs_ops = &mem_ops,
260};
261
262static ssize_t kfd_cache_show(struct kobject *kobj, struct attribute *attr,
263 char *buffer)
264{
265 ssize_t ret;
266 uint32_t i;
267 struct kfd_cache_properties *cache;
268
269 /* Making sure that the buffer is an empty string */
270 buffer[0] = 0;
271
272 cache = container_of(attr, struct kfd_cache_properties, attr);
273 sysfs_show_32bit_prop(buffer, "processor_id_low",
274 cache->processor_id_low);
275 sysfs_show_32bit_prop(buffer, "level", cache->cache_level);
276 sysfs_show_32bit_prop(buffer, "size", cache->cache_size);
277 sysfs_show_32bit_prop(buffer, "cache_line_size", cache->cacheline_size);
278 sysfs_show_32bit_prop(buffer, "cache_lines_per_tag",
279 cache->cachelines_per_tag);
280 sysfs_show_32bit_prop(buffer, "association", cache->cache_assoc);
281 sysfs_show_32bit_prop(buffer, "latency", cache->cache_latency);
282 sysfs_show_32bit_prop(buffer, "type", cache->cache_type);
283 snprintf(buffer, PAGE_SIZE, "%ssibling_map ", buffer);
284 for (i = 0; i < KFD_TOPOLOGY_CPU_SIBLINGS; i++)
285 ret = snprintf(buffer, PAGE_SIZE, "%s%d%s",
286 buffer, cache->sibling_map[i],
287 (i == KFD_TOPOLOGY_CPU_SIBLINGS-1) ?
288 "\n" : ",");
289
290 return ret;
291}
292
293static const struct sysfs_ops cache_ops = {
294 .show = kfd_cache_show,
295};
296
297static struct kobj_type cache_type = {
5108d768 298 .release = kfd_topology_kobj_release,
5b5c4e40
EP		 299 .sysfs_ops = &cache_ops,
300};
301
302static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
303 char *buffer)
304{
5b5c4e40
EP		 305 struct kfd_topology_device *dev;
306 char public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE];
307 uint32_t i;
f7c826ad 308 uint32_t log_max_watch_addr;
5b5c4e40
EP		 309
310 /* Making sure that the buffer is an empty string */
311 buffer[0] = 0;
312
313 if (strcmp(attr->name, "gpu_id") == 0) {
314 dev = container_of(attr, struct kfd_topology_device,
315 attr_gpuid);
f7c826ad
AS		 316 return sysfs_show_32bit_val(buffer, dev->gpu_id);
317 }
318
319 if (strcmp(attr->name, "name") == 0) {
5b5c4e40
EP		 320 dev = container_of(attr, struct kfd_topology_device,
321 attr_name);
322 for (i = 0; i < KFD_TOPOLOGY_PUBLIC_NAME_SIZE; i++) {
323 public_name[i] =
324 (char)dev->node_props.marketing_name[i];
325 if (dev->node_props.marketing_name[i] == 0)
326 break;
327 }
328 public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE-1] = 0x0;
f7c826ad
AS		 329 return sysfs_show_str_val(buffer, public_name);
330 }
5b5c4e40 331
f7c826ad
AS		 332 dev = container_of(attr, struct kfd_topology_device,
333 attr_props);
334 sysfs_show_32bit_prop(buffer, "cpu_cores_count",
335 dev->node_props.cpu_cores_count);
336 sysfs_show_32bit_prop(buffer, "simd_count",
337 dev->node_props.simd_count);
338
339 if (dev->mem_bank_count < dev->node_props.mem_banks_count) {
79775b62 340 pr_info_once("mem_banks_count truncated from %d to %d\n",
f7c826ad
AS		 341 dev->node_props.mem_banks_count,
342 dev->mem_bank_count);
343 sysfs_show_32bit_prop(buffer, "mem_banks_count",
344 dev->mem_bank_count);
345 } else {
346 sysfs_show_32bit_prop(buffer, "mem_banks_count",
347 dev->node_props.mem_banks_count);
348 }
f1386fbc 349
f7c826ad
AS		 350 sysfs_show_32bit_prop(buffer, "caches_count",
351 dev->node_props.caches_count);
352 sysfs_show_32bit_prop(buffer, "io_links_count",
353 dev->node_props.io_links_count);
354 sysfs_show_32bit_prop(buffer, "cpu_core_id_base",
355 dev->node_props.cpu_core_id_base);
356 sysfs_show_32bit_prop(buffer, "simd_id_base",
357 dev->node_props.simd_id_base);
f7c826ad
AS		 358 sysfs_show_32bit_prop(buffer, "max_waves_per_simd",
359 dev->node_props.max_waves_per_simd);
360 sysfs_show_32bit_prop(buffer, "lds_size_in_kb",
361 dev->node_props.lds_size_in_kb);
362 sysfs_show_32bit_prop(buffer, "gds_size_in_kb",
363 dev->node_props.gds_size_in_kb);
364 sysfs_show_32bit_prop(buffer, "wave_front_size",
365 dev->node_props.wave_front_size);
366 sysfs_show_32bit_prop(buffer, "array_count",
367 dev->node_props.array_count);
368 sysfs_show_32bit_prop(buffer, "simd_arrays_per_engine",
369 dev->node_props.simd_arrays_per_engine);
370 sysfs_show_32bit_prop(buffer, "cu_per_simd_array",
371 dev->node_props.cu_per_simd_array);
372 sysfs_show_32bit_prop(buffer, "simd_per_cu",
373 dev->node_props.simd_per_cu);
374 sysfs_show_32bit_prop(buffer, "max_slots_scratch_cu",
375 dev->node_props.max_slots_scratch_cu);
376 sysfs_show_32bit_prop(buffer, "vendor_id",
377 dev->node_props.vendor_id);
378 sysfs_show_32bit_prop(buffer, "device_id",
379 dev->node_props.device_id);
380 sysfs_show_32bit_prop(buffer, "location_id",
381 dev->node_props.location_id);
382
383 if (dev->gpu) {
384 log_max_watch_addr =
385 __ilog2_u32(dev->gpu->device_info->num_of_watch_points);
386
387 if (log_max_watch_addr) {
388 dev->node_props.capability |=
389 HSA_CAP_WATCH_POINTS_SUPPORTED;
390
391 dev->node_props.capability |=
392 ((log_max_watch_addr <<
393 HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT) &
394 HSA_CAP_WATCH_POINTS_TOTALBITS_MASK);
5b5c4e40
EP		 395 }
396
f7c826ad 397 sysfs_show_32bit_prop(buffer, "max_engine_clk_fcompute",
cea405b1 398 dev->gpu->kfd2kgd->get_max_engine_clock_in_mhz(
f7c826ad 399 dev->gpu->kgd));
42e08c78 400
f7c826ad 401 sysfs_show_64bit_prop(buffer, "local_mem_size",
42e08c78 402 (unsigned long long int) 0);
f7c826ad
AS		 403
404 sysfs_show_32bit_prop(buffer, "fw_version",
cea405b1 405 dev->gpu->kfd2kgd->get_fw_version(
f7c826ad
AS		 406 dev->gpu->kgd,
407 KGD_ENGINE_MEC1));
826f5de8
AS		 408 sysfs_show_32bit_prop(buffer, "capability",
409 dev->node_props.capability);
5b5c4e40
EP		 410 }
411
f7c826ad
AS		 412 return sysfs_show_32bit_prop(buffer, "max_engine_clk_ccompute",
413 cpufreq_quick_get_max(0)/1000);
5b5c4e40
EP		 414}
415
416static const struct sysfs_ops node_ops = {
417 .show = node_show,
418};
419
420static struct kobj_type node_type = {
5108d768 421 .release = kfd_topology_kobj_release,
5b5c4e40
EP		 422 .sysfs_ops = &node_ops,
423};
424
425static void kfd_remove_sysfs_file(struct kobject *kobj, struct attribute *attr)
426{
427 sysfs_remove_file(kobj, attr);
428 kobject_del(kobj);
429 kobject_put(kobj);
430}
431
432static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev)
433{
434 struct kfd_iolink_properties *iolink;
435 struct kfd_cache_properties *cache;
436 struct kfd_mem_properties *mem;
437
5b5c4e40
EP		 438 if (dev->kobj_iolink) {
439 list_for_each_entry(iolink, &dev->io_link_props, list)
440 if (iolink->kobj) {
441 kfd_remove_sysfs_file(iolink->kobj,
442 &iolink->attr);
16b9201c 443 iolink->kobj = NULL;
5b5c4e40
EP		 444 }
445 kobject_del(dev->kobj_iolink);
446 kobject_put(dev->kobj_iolink);
16b9201c 447 dev->kobj_iolink = NULL;
5b5c4e40
EP		 448 }
449
450 if (dev->kobj_cache) {
451 list_for_each_entry(cache, &dev->cache_props, list)
452 if (cache->kobj) {
453 kfd_remove_sysfs_file(cache->kobj,
454 &cache->attr);
16b9201c 455 cache->kobj = NULL;
5b5c4e40
EP		 456 }
457 kobject_del(dev->kobj_cache);
458 kobject_put(dev->kobj_cache);
16b9201c 459 dev->kobj_cache = NULL;
5b5c4e40
EP		 460 }
461
462 if (dev->kobj_mem) {
463 list_for_each_entry(mem, &dev->mem_props, list)
464 if (mem->kobj) {
465 kfd_remove_sysfs_file(mem->kobj, &mem->attr);
16b9201c 466 mem->kobj = NULL;
5b5c4e40
EP		 467 }
468 kobject_del(dev->kobj_mem);
469 kobject_put(dev->kobj_mem);
16b9201c 470 dev->kobj_mem = NULL;
5b5c4e40
EP		 471 }
472
473 if (dev->kobj_node) {
474 sysfs_remove_file(dev->kobj_node, &dev->attr_gpuid);
475 sysfs_remove_file(dev->kobj_node, &dev->attr_name);
476 sysfs_remove_file(dev->kobj_node, &dev->attr_props);
477 kobject_del(dev->kobj_node);
478 kobject_put(dev->kobj_node);
16b9201c 479 dev->kobj_node = NULL;
5b5c4e40
EP		 480 }
481}
482
483static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev,
484 uint32_t id)
485{
486 struct kfd_iolink_properties *iolink;
487 struct kfd_cache_properties *cache;
488 struct kfd_mem_properties *mem;
489 int ret;
490 uint32_t i;
491
32fa8219
FK		 492 if (WARN_ON(dev->kobj_node))
493 return -EEXIST;
494
5b5c4e40
EP		 495 /*
496 * Creating the sysfs folders
497 */
5b5c4e40
EP		 498 dev->kobj_node = kfd_alloc_struct(dev->kobj_node);
499 if (!dev->kobj_node)
500 return -ENOMEM;
501
502 ret = kobject_init_and_add(dev->kobj_node, &node_type,
503 sys_props.kobj_nodes, "%d", id);
504 if (ret < 0)
505 return ret;
506
507 dev->kobj_mem = kobject_create_and_add("mem_banks", dev->kobj_node);
508 if (!dev->kobj_mem)
509 return -ENOMEM;
510
511 dev->kobj_cache = kobject_create_and_add("caches", dev->kobj_node);
512 if (!dev->kobj_cache)
513 return -ENOMEM;
514
515 dev->kobj_iolink = kobject_create_and_add("io_links", dev->kobj_node);
516 if (!dev->kobj_iolink)
517 return -ENOMEM;
518
519 /*
520 * Creating sysfs files for node properties
521 */
522 dev->attr_gpuid.name = "gpu_id";
523 dev->attr_gpuid.mode = KFD_SYSFS_FILE_MODE;
524 sysfs_attr_init(&dev->attr_gpuid);
525 dev->attr_name.name = "name";
526 dev->attr_name.mode = KFD_SYSFS_FILE_MODE;
527 sysfs_attr_init(&dev->attr_name);
528 dev->attr_props.name = "properties";
529 dev->attr_props.mode = KFD_SYSFS_FILE_MODE;
530 sysfs_attr_init(&dev->attr_props);
531 ret = sysfs_create_file(dev->kobj_node, &dev->attr_gpuid);
532 if (ret < 0)
533 return ret;
534 ret = sysfs_create_file(dev->kobj_node, &dev->attr_name);
535 if (ret < 0)
536 return ret;
537 ret = sysfs_create_file(dev->kobj_node, &dev->attr_props);
538 if (ret < 0)
539 return ret;
540
541 i = 0;
542 list_for_each_entry(mem, &dev->mem_props, list) {
543 mem->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
544 if (!mem->kobj)
545 return -ENOMEM;
546 ret = kobject_init_and_add(mem->kobj, &mem_type,
547 dev->kobj_mem, "%d", i);
548 if (ret < 0)
549 return ret;
550
551 mem->attr.name = "properties";
552 mem->attr.mode = KFD_SYSFS_FILE_MODE;
553 sysfs_attr_init(&mem->attr);
554 ret = sysfs_create_file(mem->kobj, &mem->attr);
555 if (ret < 0)
556 return ret;
557 i++;
558 }
559
560 i = 0;
561 list_for_each_entry(cache, &dev->cache_props, list) {
562 cache->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
563 if (!cache->kobj)
564 return -ENOMEM;
565 ret = kobject_init_and_add(cache->kobj, &cache_type,
566 dev->kobj_cache, "%d", i);
567 if (ret < 0)
568 return ret;
569
570 cache->attr.name = "properties";
571 cache->attr.mode = KFD_SYSFS_FILE_MODE;
572 sysfs_attr_init(&cache->attr);
573 ret = sysfs_create_file(cache->kobj, &cache->attr);
574 if (ret < 0)
575 return ret;
576 i++;
577 }
578
579 i = 0;
580 list_for_each_entry(iolink, &dev->io_link_props, list) {
581 iolink->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
582 if (!iolink->kobj)
583 return -ENOMEM;
584 ret = kobject_init_and_add(iolink->kobj, &iolink_type,
585 dev->kobj_iolink, "%d", i);
586 if (ret < 0)
587 return ret;
588
589 iolink->attr.name = "properties";
590 iolink->attr.mode = KFD_SYSFS_FILE_MODE;
591 sysfs_attr_init(&iolink->attr);
592 ret = sysfs_create_file(iolink->kobj, &iolink->attr);
593 if (ret < 0)
594 return ret;
595 i++;
596}
597
598 return 0;
599}
600
601static int kfd_build_sysfs_node_tree(void)
602{
603 struct kfd_topology_device *dev;
604 int ret;
605 uint32_t i = 0;
606
607 list_for_each_entry(dev, &topology_device_list, list) {
8dfead6c 608 ret = kfd_build_sysfs_node_entry(dev, i);
5b5c4e40
EP		 609 if (ret < 0)
610 return ret;
611 i++;
612 }
613
614 return 0;
615}
616
617static void kfd_remove_sysfs_node_tree(void)
618{
619 struct kfd_topology_device *dev;
620
621 list_for_each_entry(dev, &topology_device_list, list)
622 kfd_remove_sysfs_node_entry(dev);
623}
624
625static int kfd_topology_update_sysfs(void)
626{
627 int ret;
628
629 pr_info("Creating topology SYSFS entries\n");
4eacc26b 630 if (!sys_props.kobj_topology) {
5b5c4e40
EP		 631 sys_props.kobj_topology =
632 kfd_alloc_struct(sys_props.kobj_topology);
633 if (!sys_props.kobj_topology)
634 return -ENOMEM;
635
636 ret = kobject_init_and_add(sys_props.kobj_topology,
637 &sysprops_type, &kfd_device->kobj,
638 "topology");
639 if (ret < 0)
640 return ret;
641
642 sys_props.kobj_nodes = kobject_create_and_add("nodes",
643 sys_props.kobj_topology);
644 if (!sys_props.kobj_nodes)
645 return -ENOMEM;
646
647 sys_props.attr_genid.name = "generation_id";
648 sys_props.attr_genid.mode = KFD_SYSFS_FILE_MODE;
649 sysfs_attr_init(&sys_props.attr_genid);
650 ret = sysfs_create_file(sys_props.kobj_topology,
651 &sys_props.attr_genid);
652 if (ret < 0)
653 return ret;
654
655 sys_props.attr_props.name = "system_properties";
656 sys_props.attr_props.mode = KFD_SYSFS_FILE_MODE;
657 sysfs_attr_init(&sys_props.attr_props);
658 ret = sysfs_create_file(sys_props.kobj_topology,
659 &sys_props.attr_props);
660 if (ret < 0)
661 return ret;
662 }
663
664 kfd_remove_sysfs_node_tree();
665
666 return kfd_build_sysfs_node_tree();
667}
668
669static void kfd_topology_release_sysfs(void)
670{
671 kfd_remove_sysfs_node_tree();
672 if (sys_props.kobj_topology) {
673 sysfs_remove_file(sys_props.kobj_topology,
674 &sys_props.attr_genid);
675 sysfs_remove_file(sys_props.kobj_topology,
676 &sys_props.attr_props);
677 if (sys_props.kobj_nodes) {
678 kobject_del(sys_props.kobj_nodes);
679 kobject_put(sys_props.kobj_nodes);
16b9201c 680 sys_props.kobj_nodes = NULL;
5b5c4e40
EP		 681 }
682 kobject_del(sys_props.kobj_topology);
683 kobject_put(sys_props.kobj_topology);
16b9201c 684 sys_props.kobj_topology = NULL;
5b5c4e40
EP		 685 }
686}
687
4f449311
HK		 688/* Called with write topology_lock acquired */
689static void kfd_topology_update_device_list(struct list_head *temp_list,
690 struct list_head *master_list)
691{
692 while (!list_empty(temp_list)) {
693 list_move_tail(temp_list->next, master_list);
694 sys_props.num_devices++;
695 }
696}
697
5b5c4e40
EP		 698int kfd_topology_init(void)
699{
16b9201c 700 void *crat_image = NULL;
5b5c4e40
EP		 701 size_t image_size = 0;
702 int ret;
4f449311 703 struct list_head temp_topology_device_list;
5b5c4e40 704
4f449311
HK		 705 /* topology_device_list - Master list of all topology devices
706 * temp_topology_device_list - temporary list created while parsing CRAT
707 * or VCRAT. Once parsing is complete the contents of list is moved to
708 * topology_device_list
5b5c4e40 709 */
4f449311
HK		 710
711 /* Initialize the head for the both the lists */
5b5c4e40 712 INIT_LIST_HEAD(&topology_device_list);
4f449311 713 INIT_LIST_HEAD(&temp_topology_device_list);
5b5c4e40 714 init_rwsem(&topology_lock);
5b5c4e40
EP		 715
716 memset(&sys_props, 0, sizeof(sys_props));
717
718 /*
719 * Get the CRAT image from the ACPI
720 */
8e05247d
HK		 721 ret = kfd_create_crat_image_acpi(&crat_image, &image_size);
722 if (!ret) {
4f449311
HK		 723 ret = kfd_parse_crat_table(crat_image,
724 &temp_topology_device_list, 0);
8e05247d 725 if (ret)
5b5c4e40 726 goto err;
5b5c4e40 727 } else if (ret == -ENODATA) {
8e05247d 728 /* TODO: Create fake CRAT table */
5b5c4e40 729 ret = 0;
8e05247d 730 goto err;
5b5c4e40
EP		 731 } else {
732 pr_err("Couldn't get CRAT table size from ACPI\n");
8e05247d 733 goto err;
5b5c4e40
EP		 734 }
735
8e05247d 736 down_write(&topology_lock);
4f449311
HK		 737 kfd_topology_update_device_list(&temp_topology_device_list,
738 &topology_device_list);
8e05247d
HK		 739 ret = kfd_topology_update_sysfs();
740 up_write(&topology_lock);
741
4f449311
HK		 742 if (!ret) {
743 sys_props.generation_count++;
8e05247d 744 pr_info("Finished initializing topology\n");
4f449311 745 } else
8e05247d
HK		 746 pr_err("Failed to update topology in sysfs ret=%d\n", ret);
747
5b5c4e40 748err:
8e05247d 749 kfd_destroy_crat_image(crat_image);
5b5c4e40
EP		 750 return ret;
751}
752
753void kfd_topology_shutdown(void)
754{
4f449311 755 down_write(&topology_lock);
5b5c4e40
EP		 756 kfd_topology_release_sysfs();
757 kfd_release_live_view();
4f449311 758 up_write(&topology_lock);
5b5c4e40
EP		 759}
760
761static void kfd_debug_print_topology(void)
762{
763 struct kfd_topology_device *dev;
764 uint32_t i = 0;
765
766 pr_info("DEBUG PRINT OF TOPOLOGY:");
767 list_for_each_entry(dev, &topology_device_list, list) {
768 pr_info("Node: %d\n", i);
769 pr_info("\tGPU assigned: %s\n", (dev->gpu ? "yes" : "no"));
770 pr_info("\tCPU count: %d\n", dev->node_props.cpu_cores_count);
8e05247d 771 pr_info("\tSIMD count: %d\n", dev->node_props.simd_count);
5b5c4e40
EP		 772 i++;
773 }
774}
775
776static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu)
777{
778 uint32_t hashout;
779 uint32_t buf[7];
585f0e6c 780 uint64_t local_mem_size;
5b5c4e40 781 int i;
0504cccf 782 struct kfd_local_mem_info local_mem_info;
5b5c4e40
EP		 783
784 if (!gpu)
785 return 0;
786
0504cccf
HK		 787 gpu->kfd2kgd->get_local_mem_info(gpu->kgd, &local_mem_info);
788
789 local_mem_size = local_mem_info.local_mem_size_private +
790 local_mem_info.local_mem_size_public;
585f0e6c 791
5b5c4e40
EP		 792 buf[0] = gpu->pdev->devfn;
793 buf[1] = gpu->pdev->subsystem_vendor;
794 buf[2] = gpu->pdev->subsystem_device;
795 buf[3] = gpu->pdev->device;
796 buf[4] = gpu->pdev->bus->number;
585f0e6c
EC		 797 buf[5] = lower_32_bits(local_mem_size);
798 buf[6] = upper_32_bits(local_mem_size);
5b5c4e40
EP		 799
800 for (i = 0, hashout = 0; i < 7; i++)
801 hashout ^= hash_32(buf[i], KFD_GPU_ID_HASH_WIDTH);
802
803 return hashout;
804}
805
806static struct kfd_topology_device *kfd_assign_gpu(struct kfd_dev *gpu)
807{
808 struct kfd_topology_device *dev;
16b9201c 809 struct kfd_topology_device *out_dev = NULL;
5b5c4e40 810
5b5c4e40 811 list_for_each_entry(dev, &topology_device_list, list)
4eacc26b 812 if (!dev->gpu && (dev->node_props.simd_count > 0)) {
5b5c4e40
EP		 813 dev->gpu = gpu;
814 out_dev = dev;
815 break;
816 }
817
818 return out_dev;
819}
820
821static void kfd_notify_gpu_change(uint32_t gpu_id, int arrival)
822{
823 /*
824 * TODO: Generate an event for thunk about the arrival/removal
825 * of the GPU
826 */
827}
828
829int kfd_topology_add_device(struct kfd_dev *gpu)
830{
831 uint32_t gpu_id;
832 struct kfd_topology_device *dev;
f7ce2fad 833 struct kfd_cu_info cu_info;
4f449311
HK		 834 int res = 0;
835 struct list_head temp_topology_device_list;
836
837 INIT_LIST_HEAD(&temp_topology_device_list);
5b5c4e40 838
5b5c4e40
EP		 839 gpu_id = kfd_generate_gpu_id(gpu);
840
79775b62 841 pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id);
5b5c4e40 842
5b5c4e40
EP		 843 /*
844 * Try to assign the GPU to existing topology device (generated from
845 * CRAT table
846 */
847 dev = kfd_assign_gpu(gpu);
848 if (!dev) {
849 pr_info("GPU was not found in the current topology. Extending.\n");
850 kfd_debug_print_topology();
4f449311 851 dev = kfd_create_topology_device(&temp_topology_device_list);
5b5c4e40
EP		 852 if (!dev) {
853 res = -ENOMEM;
854 goto err;
855 }
4f449311 856
5b5c4e40
EP		 857 dev->gpu = gpu;
858
859 /*
860 * TODO: Make a call to retrieve topology information from the
861 * GPU vBIOS
862 */
863
4f449311
HK		 864 down_write(&topology_lock);
865 kfd_topology_update_device_list(&temp_topology_device_list,
866 &topology_device_list);
867
8eabaf54
KR		 868 /* Update the SYSFS tree, since we added another topology
869 * device
5b5c4e40
EP		 870 */
871 if (kfd_topology_update_sysfs() < 0)
872 kfd_topology_release_sysfs();
873
4f449311
HK		 874 up_write(&topology_lock);
875
5b5c4e40
EP		 876 }
877
878 dev->gpu_id = gpu_id;
879 gpu->id = gpu_id;
f7ce2fad
FC		 880 dev->gpu->kfd2kgd->get_cu_info(dev->gpu->kgd, &cu_info);
881 dev->node_props.simd_count = dev->node_props.simd_per_cu *
882 cu_info.cu_active_number;
5b5c4e40
EP		 883 dev->node_props.vendor_id = gpu->pdev->vendor;
884 dev->node_props.device_id = gpu->pdev->device;
d63f0ba2
HK		 885 dev->node_props.location_id = PCI_DEVID(gpu->pdev->bus->number,
886 gpu->pdev->devfn);
5b5c4e40
EP		 887 /*
888 * TODO: Retrieve max engine clock values from KGD
889 */
890
7639a8c4
BG		 891 if (dev->gpu->device_info->asic_family == CHIP_CARRIZO) {
892 dev->node_props.capability |= HSA_CAP_DOORBELL_PACKET_TYPE;
42aa8793 893 pr_debug("Adding doorbell packet type capability\n");
7639a8c4
BG		 894 }
895
4f449311 896 if (!res)
5b5c4e40 897 kfd_notify_gpu_change(gpu_id, 1);
4f449311 898err:
5b5c4e40
EP		 899 return res;
900}
901
902int kfd_topology_remove_device(struct kfd_dev *gpu)
903{
4f449311 904 struct kfd_topology_device *dev, *tmp;
5b5c4e40
EP		 905 uint32_t gpu_id;
906 int res = -ENODEV;
907
5b5c4e40
EP		 908 down_write(&topology_lock);
909
4f449311 910 list_for_each_entry_safe(dev, tmp, &topology_device_list, list)
5b5c4e40
EP		 911 if (dev->gpu == gpu) {
912 gpu_id = dev->gpu_id;
913 kfd_remove_sysfs_node_entry(dev);
914 kfd_release_topology_device(dev);
4f449311 915 sys_props.num_devices--;
5b5c4e40
EP		 916 res = 0;
917 if (kfd_topology_update_sysfs() < 0)
918 kfd_topology_release_sysfs();
919 break;
920 }
921
922 up_write(&topology_lock);
923
174de876 924 if (!res)
5b5c4e40
EP		 925 kfd_notify_gpu_change(gpu_id, 0);
926
927 return res;
928}
929
6d82eb0e
HK		 930/* kfd_topology_enum_kfd_devices - Enumerate through all devices in KFD
931 * topology. If GPU device is found @idx, then valid kfd_dev pointer is
932 * returned through @kdev
933 * Return - 0: On success (@kdev will be NULL for non GPU nodes)
934 * -1: If end of list
5b5c4e40 935 */
6d82eb0e 936int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev)
5b5c4e40
EP		 937{
938
939 struct kfd_topology_device *top_dev;
5b5c4e40
EP		 940 uint8_t device_idx = 0;
941
6d82eb0e 942 *kdev = NULL;
5b5c4e40
EP		 943 down_read(&topology_lock);
944
945 list_for_each_entry(top_dev, &topology_device_list, list) {
946 if (device_idx == idx) {
6d82eb0e
HK		 947 *kdev = top_dev->gpu;
948 up_read(&topology_lock);
949 return 0;
5b5c4e40
EP		 950 }
951
952 device_idx++;
953 }
954
955 up_read(&topology_lock);
956
6d82eb0e 957 return -1;
5b5c4e40
EP		 958
959}
851a645e
FK		 960
961#if defined(CONFIG_DEBUG_FS)
962
963int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data)
964{
965 struct kfd_topology_device *dev;
966 unsigned int i = 0;
967 int r = 0;
968
969 down_read(&topology_lock);
970
971 list_for_each_entry(dev, &topology_device_list, list) {
972 if (!dev->gpu) {
973 i++;
974 continue;
975 }
976
977 seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id);
978 r = dqm_debugfs_hqds(m, dev->gpu->dqm);
979 if (r)
980 break;
981 }
982
983 up_read(&topology_lock);
984
985 return r;
986}
987
988int kfd_debugfs_rls_by_device(struct seq_file *m, void *data)
989{
990 struct kfd_topology_device *dev;
991 unsigned int i = 0;
992 int r = 0;
993
994 down_read(&topology_lock);
995
996 list_for_each_entry(dev, &topology_device_list, list) {
997 if (!dev->gpu) {
998 i++;
999 continue;
1000 }
1001
1002 seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id);
1003 r = pm_debugfs_runlist(m, &dev->gpu->dqm->packets);
1004 if (r)
1005 break;
1006 }
1007
1008 up_read(&topology_lock);
1009
1010 return r;
1011}
1012
1013#endif
Linux 6.x block layer and io_uring tree(s)