projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
drm/amdkfd: Fix sibling_map[] size
[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;
bc0c75a3 266 uint32_t i, j;
5b5c4e40
EP		 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);
bc0c75a3
HK		 284 for (i = 0; i < CRAT_SIBLINGMAP_SIZE; i++)
285 for (j = 0; j < sizeof(cache->sibling_map[0])*8; j++) {
286 /* Check each bit */
287 if (cache->sibling_map[i] & (1 << j))
288 ret = snprintf(buffer, PAGE_SIZE,
289 "%s%d%s", buffer, 1, ",");
290 else
291 ret = snprintf(buffer, PAGE_SIZE,
292 "%s%d%s", buffer, 0, ",");
293 }
294 /* Replace the last "," with end of line */
295 *(buffer + strlen(buffer) - 1) = 0xA;
5b5c4e40
EP		 296 return ret;
297}
298
299static const struct sysfs_ops cache_ops = {
300 .show = kfd_cache_show,
301};
302
303static struct kobj_type cache_type = {
5108d768 304 .release = kfd_topology_kobj_release,
5b5c4e40
EP		 305 .sysfs_ops = &cache_ops,
306};
307
308static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
309 char *buffer)
310{
5b5c4e40
EP		 311 struct kfd_topology_device *dev;
312 char public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE];
313 uint32_t i;
f7c826ad 314 uint32_t log_max_watch_addr;
5b5c4e40
EP		 315
316 /* Making sure that the buffer is an empty string */
317 buffer[0] = 0;
318
319 if (strcmp(attr->name, "gpu_id") == 0) {
320 dev = container_of(attr, struct kfd_topology_device,
321 attr_gpuid);
f7c826ad
AS		 322 return sysfs_show_32bit_val(buffer, dev->gpu_id);
323 }
324
325 if (strcmp(attr->name, "name") == 0) {
5b5c4e40
EP		 326 dev = container_of(attr, struct kfd_topology_device,
327 attr_name);
328 for (i = 0; i < KFD_TOPOLOGY_PUBLIC_NAME_SIZE; i++) {
329 public_name[i] =
330 (char)dev->node_props.marketing_name[i];
331 if (dev->node_props.marketing_name[i] == 0)
332 break;
333 }
334 public_name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE-1] = 0x0;
f7c826ad
AS		 335 return sysfs_show_str_val(buffer, public_name);
336 }
5b5c4e40 337
f7c826ad
AS		 338 dev = container_of(attr, struct kfd_topology_device,
339 attr_props);
340 sysfs_show_32bit_prop(buffer, "cpu_cores_count",
341 dev->node_props.cpu_cores_count);
342 sysfs_show_32bit_prop(buffer, "simd_count",
343 dev->node_props.simd_count);
175b9263
FK		 344 sysfs_show_32bit_prop(buffer, "mem_banks_count",
345 dev->node_props.mem_banks_count);
f7c826ad
AS		 346 sysfs_show_32bit_prop(buffer, "caches_count",
347 dev->node_props.caches_count);
348 sysfs_show_32bit_prop(buffer, "io_links_count",
349 dev->node_props.io_links_count);
350 sysfs_show_32bit_prop(buffer, "cpu_core_id_base",
351 dev->node_props.cpu_core_id_base);
352 sysfs_show_32bit_prop(buffer, "simd_id_base",
353 dev->node_props.simd_id_base);
f7c826ad
AS		 354 sysfs_show_32bit_prop(buffer, "max_waves_per_simd",
355 dev->node_props.max_waves_per_simd);
356 sysfs_show_32bit_prop(buffer, "lds_size_in_kb",
357 dev->node_props.lds_size_in_kb);
358 sysfs_show_32bit_prop(buffer, "gds_size_in_kb",
359 dev->node_props.gds_size_in_kb);
360 sysfs_show_32bit_prop(buffer, "wave_front_size",
361 dev->node_props.wave_front_size);
362 sysfs_show_32bit_prop(buffer, "array_count",
363 dev->node_props.array_count);
364 sysfs_show_32bit_prop(buffer, "simd_arrays_per_engine",
365 dev->node_props.simd_arrays_per_engine);
366 sysfs_show_32bit_prop(buffer, "cu_per_simd_array",
367 dev->node_props.cu_per_simd_array);
368 sysfs_show_32bit_prop(buffer, "simd_per_cu",
369 dev->node_props.simd_per_cu);
370 sysfs_show_32bit_prop(buffer, "max_slots_scratch_cu",
371 dev->node_props.max_slots_scratch_cu);
372 sysfs_show_32bit_prop(buffer, "vendor_id",
373 dev->node_props.vendor_id);
374 sysfs_show_32bit_prop(buffer, "device_id",
375 dev->node_props.device_id);
376 sysfs_show_32bit_prop(buffer, "location_id",
377 dev->node_props.location_id);
378
379 if (dev->gpu) {
380 log_max_watch_addr =
381 __ilog2_u32(dev->gpu->device_info->num_of_watch_points);
382
383 if (log_max_watch_addr) {
384 dev->node_props.capability |=
385 HSA_CAP_WATCH_POINTS_SUPPORTED;
386
387 dev->node_props.capability |=
388 ((log_max_watch_addr <<
389 HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT) &
390 HSA_CAP_WATCH_POINTS_TOTALBITS_MASK);
5b5c4e40
EP		 391 }
392
f7c826ad 393 sysfs_show_32bit_prop(buffer, "max_engine_clk_fcompute",
cea405b1 394 dev->gpu->kfd2kgd->get_max_engine_clock_in_mhz(
f7c826ad 395 dev->gpu->kgd));
42e08c78 396
f7c826ad 397 sysfs_show_64bit_prop(buffer, "local_mem_size",
42e08c78 398 (unsigned long long int) 0);
f7c826ad
AS		 399
400 sysfs_show_32bit_prop(buffer, "fw_version",
cea405b1 401 dev->gpu->kfd2kgd->get_fw_version(
f7c826ad
AS		 402 dev->gpu->kgd,
403 KGD_ENGINE_MEC1));
826f5de8
AS		 404 sysfs_show_32bit_prop(buffer, "capability",
405 dev->node_props.capability);
5b5c4e40
EP		 406 }
407
f7c826ad
AS		 408 return sysfs_show_32bit_prop(buffer, "max_engine_clk_ccompute",
409 cpufreq_quick_get_max(0)/1000);
5b5c4e40
EP		 410}
411
412static const struct sysfs_ops node_ops = {
413 .show = node_show,
414};
415
416static struct kobj_type node_type = {
5108d768 417 .release = kfd_topology_kobj_release,
5b5c4e40
EP		 418 .sysfs_ops = &node_ops,
419};
420
421static void kfd_remove_sysfs_file(struct kobject *kobj, struct attribute *attr)
422{
423 sysfs_remove_file(kobj, attr);
424 kobject_del(kobj);
425 kobject_put(kobj);
426}
427
428static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev)
429{
430 struct kfd_iolink_properties *iolink;
431 struct kfd_cache_properties *cache;
432 struct kfd_mem_properties *mem;
433
5b5c4e40
EP		 434 if (dev->kobj_iolink) {
435 list_for_each_entry(iolink, &dev->io_link_props, list)
436 if (iolink->kobj) {
437 kfd_remove_sysfs_file(iolink->kobj,
438 &iolink->attr);
16b9201c 439 iolink->kobj = NULL;
5b5c4e40
EP		 440 }
441 kobject_del(dev->kobj_iolink);
442 kobject_put(dev->kobj_iolink);
16b9201c 443 dev->kobj_iolink = NULL;
5b5c4e40
EP		 444 }
445
446 if (dev->kobj_cache) {
447 list_for_each_entry(cache, &dev->cache_props, list)
448 if (cache->kobj) {
449 kfd_remove_sysfs_file(cache->kobj,
450 &cache->attr);
16b9201c 451 cache->kobj = NULL;
5b5c4e40
EP		 452 }
453 kobject_del(dev->kobj_cache);
454 kobject_put(dev->kobj_cache);
16b9201c 455 dev->kobj_cache = NULL;
5b5c4e40
EP		 456 }
457
458 if (dev->kobj_mem) {
459 list_for_each_entry(mem, &dev->mem_props, list)
460 if (mem->kobj) {
461 kfd_remove_sysfs_file(mem->kobj, &mem->attr);
16b9201c 462 mem->kobj = NULL;
5b5c4e40
EP		 463 }
464 kobject_del(dev->kobj_mem);
465 kobject_put(dev->kobj_mem);
16b9201c 466 dev->kobj_mem = NULL;
5b5c4e40
EP		 467 }
468
469 if (dev->kobj_node) {
470 sysfs_remove_file(dev->kobj_node, &dev->attr_gpuid);
471 sysfs_remove_file(dev->kobj_node, &dev->attr_name);
472 sysfs_remove_file(dev->kobj_node, &dev->attr_props);
473 kobject_del(dev->kobj_node);
474 kobject_put(dev->kobj_node);
16b9201c 475 dev->kobj_node = NULL;
5b5c4e40
EP		 476 }
477}
478
479static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev,
480 uint32_t id)
481{
482 struct kfd_iolink_properties *iolink;
483 struct kfd_cache_properties *cache;
484 struct kfd_mem_properties *mem;
485 int ret;
486 uint32_t i;
487
32fa8219
FK		 488 if (WARN_ON(dev->kobj_node))
489 return -EEXIST;
490
5b5c4e40
EP		 491 /*
492 * Creating the sysfs folders
493 */
5b5c4e40
EP		 494 dev->kobj_node = kfd_alloc_struct(dev->kobj_node);
495 if (!dev->kobj_node)
496 return -ENOMEM;
497
498 ret = kobject_init_and_add(dev->kobj_node, &node_type,
499 sys_props.kobj_nodes, "%d", id);
500 if (ret < 0)
501 return ret;
502
503 dev->kobj_mem = kobject_create_and_add("mem_banks", dev->kobj_node);
504 if (!dev->kobj_mem)
505 return -ENOMEM;
506
507 dev->kobj_cache = kobject_create_and_add("caches", dev->kobj_node);
508 if (!dev->kobj_cache)
509 return -ENOMEM;
510
511 dev->kobj_iolink = kobject_create_and_add("io_links", dev->kobj_node);
512 if (!dev->kobj_iolink)
513 return -ENOMEM;
514
515 /*
516 * Creating sysfs files for node properties
517 */
518 dev->attr_gpuid.name = "gpu_id";
519 dev->attr_gpuid.mode = KFD_SYSFS_FILE_MODE;
520 sysfs_attr_init(&dev->attr_gpuid);
521 dev->attr_name.name = "name";
522 dev->attr_name.mode = KFD_SYSFS_FILE_MODE;
523 sysfs_attr_init(&dev->attr_name);
524 dev->attr_props.name = "properties";
525 dev->attr_props.mode = KFD_SYSFS_FILE_MODE;
526 sysfs_attr_init(&dev->attr_props);
527 ret = sysfs_create_file(dev->kobj_node, &dev->attr_gpuid);
528 if (ret < 0)
529 return ret;
530 ret = sysfs_create_file(dev->kobj_node, &dev->attr_name);
531 if (ret < 0)
532 return ret;
533 ret = sysfs_create_file(dev->kobj_node, &dev->attr_props);
534 if (ret < 0)
535 return ret;
536
537 i = 0;
538 list_for_each_entry(mem, &dev->mem_props, list) {
539 mem->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
540 if (!mem->kobj)
541 return -ENOMEM;
542 ret = kobject_init_and_add(mem->kobj, &mem_type,
543 dev->kobj_mem, "%d", i);
544 if (ret < 0)
545 return ret;
546
547 mem->attr.name = "properties";
548 mem->attr.mode = KFD_SYSFS_FILE_MODE;
549 sysfs_attr_init(&mem->attr);
550 ret = sysfs_create_file(mem->kobj, &mem->attr);
551 if (ret < 0)
552 return ret;
553 i++;
554 }
555
556 i = 0;
557 list_for_each_entry(cache, &dev->cache_props, list) {
558 cache->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
559 if (!cache->kobj)
560 return -ENOMEM;
561 ret = kobject_init_and_add(cache->kobj, &cache_type,
562 dev->kobj_cache, "%d", i);
563 if (ret < 0)
564 return ret;
565
566 cache->attr.name = "properties";
567 cache->attr.mode = KFD_SYSFS_FILE_MODE;
568 sysfs_attr_init(&cache->attr);
569 ret = sysfs_create_file(cache->kobj, &cache->attr);
570 if (ret < 0)
571 return ret;
572 i++;
573 }
574
575 i = 0;
576 list_for_each_entry(iolink, &dev->io_link_props, list) {
577 iolink->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
578 if (!iolink->kobj)
579 return -ENOMEM;
580 ret = kobject_init_and_add(iolink->kobj, &iolink_type,
581 dev->kobj_iolink, "%d", i);
582 if (ret < 0)
583 return ret;
584
585 iolink->attr.name = "properties";
586 iolink->attr.mode = KFD_SYSFS_FILE_MODE;
587 sysfs_attr_init(&iolink->attr);
588 ret = sysfs_create_file(iolink->kobj, &iolink->attr);
589 if (ret < 0)
590 return ret;
591 i++;
592}
593
594 return 0;
595}
596
597static int kfd_build_sysfs_node_tree(void)
598{
599 struct kfd_topology_device *dev;
600 int ret;
601 uint32_t i = 0;
602
603 list_for_each_entry(dev, &topology_device_list, list) {
8dfead6c 604 ret = kfd_build_sysfs_node_entry(dev, i);
5b5c4e40
EP		 605 if (ret < 0)
606 return ret;
607 i++;
608 }
609
610 return 0;
611}
612
613static void kfd_remove_sysfs_node_tree(void)
614{
615 struct kfd_topology_device *dev;
616
617 list_for_each_entry(dev, &topology_device_list, list)
618 kfd_remove_sysfs_node_entry(dev);
619}
620
621static int kfd_topology_update_sysfs(void)
622{
623 int ret;
624
625 pr_info("Creating topology SYSFS entries\n");
4eacc26b 626 if (!sys_props.kobj_topology) {
5b5c4e40
EP		 627 sys_props.kobj_topology =
628 kfd_alloc_struct(sys_props.kobj_topology);
629 if (!sys_props.kobj_topology)
630 return -ENOMEM;
631
632 ret = kobject_init_and_add(sys_props.kobj_topology,
633 &sysprops_type, &kfd_device->kobj,
634 "topology");
635 if (ret < 0)
636 return ret;
637
638 sys_props.kobj_nodes = kobject_create_and_add("nodes",
639 sys_props.kobj_topology);
640 if (!sys_props.kobj_nodes)
641 return -ENOMEM;
642
643 sys_props.attr_genid.name = "generation_id";
644 sys_props.attr_genid.mode = KFD_SYSFS_FILE_MODE;
645 sysfs_attr_init(&sys_props.attr_genid);
646 ret = sysfs_create_file(sys_props.kobj_topology,
647 &sys_props.attr_genid);
648 if (ret < 0)
649 return ret;
650
651 sys_props.attr_props.name = "system_properties";
652 sys_props.attr_props.mode = KFD_SYSFS_FILE_MODE;
653 sysfs_attr_init(&sys_props.attr_props);
654 ret = sysfs_create_file(sys_props.kobj_topology,
655 &sys_props.attr_props);
656 if (ret < 0)
657 return ret;
658 }
659
660 kfd_remove_sysfs_node_tree();
661
662 return kfd_build_sysfs_node_tree();
663}
664
665static void kfd_topology_release_sysfs(void)
666{
667 kfd_remove_sysfs_node_tree();
668 if (sys_props.kobj_topology) {
669 sysfs_remove_file(sys_props.kobj_topology,
670 &sys_props.attr_genid);
671 sysfs_remove_file(sys_props.kobj_topology,
672 &sys_props.attr_props);
673 if (sys_props.kobj_nodes) {
674 kobject_del(sys_props.kobj_nodes);
675 kobject_put(sys_props.kobj_nodes);
16b9201c 676 sys_props.kobj_nodes = NULL;
5b5c4e40
EP		 677 }
678 kobject_del(sys_props.kobj_topology);
679 kobject_put(sys_props.kobj_topology);
16b9201c 680 sys_props.kobj_topology = NULL;
5b5c4e40
EP		 681 }
682}
683
4f449311
HK		 684/* Called with write topology_lock acquired */
685static void kfd_topology_update_device_list(struct list_head *temp_list,
686 struct list_head *master_list)
687{
688 while (!list_empty(temp_list)) {
689 list_move_tail(temp_list->next, master_list);
690 sys_props.num_devices++;
691 }
692}
693
5b5c4e40
EP		 694int kfd_topology_init(void)
695{
16b9201c 696 void *crat_image = NULL;
5b5c4e40
EP		 697 size_t image_size = 0;
698 int ret;
4f449311 699 struct list_head temp_topology_device_list;
5b5c4e40 700
4f449311
HK		 701 /* topology_device_list - Master list of all topology devices
702 * temp_topology_device_list - temporary list created while parsing CRAT
703 * or VCRAT. Once parsing is complete the contents of list is moved to
704 * topology_device_list
5b5c4e40 705 */
4f449311
HK		 706
707 /* Initialize the head for the both the lists */
5b5c4e40 708 INIT_LIST_HEAD(&topology_device_list);
4f449311 709 INIT_LIST_HEAD(&temp_topology_device_list);
5b5c4e40 710 init_rwsem(&topology_lock);
5b5c4e40
EP		 711
712 memset(&sys_props, 0, sizeof(sys_props));
713
714 /*
715 * Get the CRAT image from the ACPI
716 */
8e05247d
HK		 717 ret = kfd_create_crat_image_acpi(&crat_image, &image_size);
718 if (!ret) {
4f449311
HK		 719 ret = kfd_parse_crat_table(crat_image,
720 &temp_topology_device_list, 0);
8e05247d 721 if (ret)
5b5c4e40 722 goto err;
5b5c4e40 723 } else if (ret == -ENODATA) {
8e05247d 724 /* TODO: Create fake CRAT table */
5b5c4e40 725 ret = 0;
8e05247d 726 goto err;
5b5c4e40
EP		 727 } else {
728 pr_err("Couldn't get CRAT table size from ACPI\n");
8e05247d 729 goto err;
5b5c4e40
EP		 730 }
731
8e05247d 732 down_write(&topology_lock);
4f449311
HK		 733 kfd_topology_update_device_list(&temp_topology_device_list,
734 &topology_device_list);
8e05247d
HK		 735 ret = kfd_topology_update_sysfs();
736 up_write(&topology_lock);
737
4f449311
HK		 738 if (!ret) {
739 sys_props.generation_count++;
8e05247d 740 pr_info("Finished initializing topology\n");
4f449311 741 } else
8e05247d
HK		 742 pr_err("Failed to update topology in sysfs ret=%d\n", ret);
743
5b5c4e40 744err:
8e05247d 745 kfd_destroy_crat_image(crat_image);
5b5c4e40
EP		 746 return ret;
747}
748
749void kfd_topology_shutdown(void)
750{
4f449311 751 down_write(&topology_lock);
5b5c4e40
EP		 752 kfd_topology_release_sysfs();
753 kfd_release_live_view();
4f449311 754 up_write(&topology_lock);
5b5c4e40
EP		 755}
756
757static void kfd_debug_print_topology(void)
758{
759 struct kfd_topology_device *dev;
760 uint32_t i = 0;
761
762 pr_info("DEBUG PRINT OF TOPOLOGY:");
763 list_for_each_entry(dev, &topology_device_list, list) {
764 pr_info("Node: %d\n", i);
765 pr_info("\tGPU assigned: %s\n", (dev->gpu ? "yes" : "no"));
766 pr_info("\tCPU count: %d\n", dev->node_props.cpu_cores_count);
8e05247d 767 pr_info("\tSIMD count: %d\n", dev->node_props.simd_count);
5b5c4e40
EP		 768 i++;
769 }
770}
771
772static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu)
773{
774 uint32_t hashout;
775 uint32_t buf[7];
585f0e6c 776 uint64_t local_mem_size;
5b5c4e40 777 int i;
0504cccf 778 struct kfd_local_mem_info local_mem_info;
5b5c4e40
EP		 779
780 if (!gpu)
781 return 0;
782
0504cccf
HK		 783 gpu->kfd2kgd->get_local_mem_info(gpu->kgd, &local_mem_info);
784
785 local_mem_size = local_mem_info.local_mem_size_private +
786 local_mem_info.local_mem_size_public;
585f0e6c 787
5b5c4e40
EP		 788 buf[0] = gpu->pdev->devfn;
789 buf[1] = gpu->pdev->subsystem_vendor;
790 buf[2] = gpu->pdev->subsystem_device;
791 buf[3] = gpu->pdev->device;
792 buf[4] = gpu->pdev->bus->number;
585f0e6c
EC		 793 buf[5] = lower_32_bits(local_mem_size);
794 buf[6] = upper_32_bits(local_mem_size);
5b5c4e40
EP		 795
796 for (i = 0, hashout = 0; i < 7; i++)
797 hashout ^= hash_32(buf[i], KFD_GPU_ID_HASH_WIDTH);
798
799 return hashout;
800}
801
802static struct kfd_topology_device *kfd_assign_gpu(struct kfd_dev *gpu)
803{
804 struct kfd_topology_device *dev;
16b9201c 805 struct kfd_topology_device *out_dev = NULL;
5b5c4e40 806
5b5c4e40 807 list_for_each_entry(dev, &topology_device_list, list)
4eacc26b 808 if (!dev->gpu && (dev->node_props.simd_count > 0)) {
5b5c4e40
EP		 809 dev->gpu = gpu;
810 out_dev = dev;
811 break;
812 }
813
814 return out_dev;
815}
816
817static void kfd_notify_gpu_change(uint32_t gpu_id, int arrival)
818{
819 /*
820 * TODO: Generate an event for thunk about the arrival/removal
821 * of the GPU
822 */
823}
824
825int kfd_topology_add_device(struct kfd_dev *gpu)
826{
827 uint32_t gpu_id;
828 struct kfd_topology_device *dev;
f7ce2fad 829 struct kfd_cu_info cu_info;
4f449311
HK		 830 int res = 0;
831 struct list_head temp_topology_device_list;
832
833 INIT_LIST_HEAD(&temp_topology_device_list);
5b5c4e40 834
5b5c4e40
EP		 835 gpu_id = kfd_generate_gpu_id(gpu);
836
79775b62 837 pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id);
5b5c4e40 838
5b5c4e40
EP		 839 /*
840 * Try to assign the GPU to existing topology device (generated from
841 * CRAT table
842 */
843 dev = kfd_assign_gpu(gpu);
844 if (!dev) {
845 pr_info("GPU was not found in the current topology. Extending.\n");
846 kfd_debug_print_topology();
4f449311 847 dev = kfd_create_topology_device(&temp_topology_device_list);
5b5c4e40
EP		 848 if (!dev) {
849 res = -ENOMEM;
850 goto err;
851 }
4f449311 852
5b5c4e40
EP		 853 dev->gpu = gpu;
854
855 /*
856 * TODO: Make a call to retrieve topology information from the
857 * GPU vBIOS
858 */
859
4f449311
HK		 860 down_write(&topology_lock);
861 kfd_topology_update_device_list(&temp_topology_device_list,
862 &topology_device_list);
863
8eabaf54
KR		 864 /* Update the SYSFS tree, since we added another topology
865 * device
5b5c4e40
EP		 866 */
867 if (kfd_topology_update_sysfs() < 0)
868 kfd_topology_release_sysfs();
869
4f449311
HK		 870 up_write(&topology_lock);
871
5b5c4e40
EP		 872 }
873
874 dev->gpu_id = gpu_id;
875 gpu->id = gpu_id;
f7ce2fad
FC		 876 dev->gpu->kfd2kgd->get_cu_info(dev->gpu->kgd, &cu_info);
877 dev->node_props.simd_count = dev->node_props.simd_per_cu *
878 cu_info.cu_active_number;
5b5c4e40
EP		 879 dev->node_props.vendor_id = gpu->pdev->vendor;
880 dev->node_props.device_id = gpu->pdev->device;
d63f0ba2
HK		 881 dev->node_props.location_id = PCI_DEVID(gpu->pdev->bus->number,
882 gpu->pdev->devfn);
5b5c4e40
EP		 883 /*
884 * TODO: Retrieve max engine clock values from KGD
885 */
886
7639a8c4
BG		 887 if (dev->gpu->device_info->asic_family == CHIP_CARRIZO) {
888 dev->node_props.capability |= HSA_CAP_DOORBELL_PACKET_TYPE;
42aa8793 889 pr_debug("Adding doorbell packet type capability\n");
7639a8c4
BG		 890 }
891
4f449311 892 if (!res)
5b5c4e40 893 kfd_notify_gpu_change(gpu_id, 1);
4f449311 894err:
5b5c4e40
EP		 895 return res;
896}
897
898int kfd_topology_remove_device(struct kfd_dev *gpu)
899{
4f449311 900 struct kfd_topology_device *dev, *tmp;
5b5c4e40
EP		 901 uint32_t gpu_id;
902 int res = -ENODEV;
903
5b5c4e40
EP		 904 down_write(&topology_lock);
905
4f449311 906 list_for_each_entry_safe(dev, tmp, &topology_device_list, list)
5b5c4e40
EP		 907 if (dev->gpu == gpu) {
908 gpu_id = dev->gpu_id;
909 kfd_remove_sysfs_node_entry(dev);
910 kfd_release_topology_device(dev);
4f449311 911 sys_props.num_devices--;
5b5c4e40
EP		 912 res = 0;
913 if (kfd_topology_update_sysfs() < 0)
914 kfd_topology_release_sysfs();
915 break;
916 }
917
918 up_write(&topology_lock);
919
174de876 920 if (!res)
5b5c4e40
EP		 921 kfd_notify_gpu_change(gpu_id, 0);
922
923 return res;
924}
925
6d82eb0e
HK		 926/* kfd_topology_enum_kfd_devices - Enumerate through all devices in KFD
927 * topology. If GPU device is found @idx, then valid kfd_dev pointer is
928 * returned through @kdev
929 * Return - 0: On success (@kdev will be NULL for non GPU nodes)
930 * -1: If end of list
5b5c4e40 931 */
6d82eb0e 932int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev)
5b5c4e40
EP		 933{
934
935 struct kfd_topology_device *top_dev;
5b5c4e40
EP		 936 uint8_t device_idx = 0;
937
6d82eb0e 938 *kdev = NULL;
5b5c4e40
EP		 939 down_read(&topology_lock);
940
941 list_for_each_entry(top_dev, &topology_device_list, list) {
942 if (device_idx == idx) {
6d82eb0e
HK		 943 *kdev = top_dev->gpu;
944 up_read(&topology_lock);
945 return 0;
5b5c4e40
EP		 946 }
947
948 device_idx++;
949 }
950
951 up_read(&topology_lock);
952
6d82eb0e 953 return -1;
5b5c4e40
EP		 954
955}
851a645e
FK		 956
957#if defined(CONFIG_DEBUG_FS)
958
959int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data)
960{
961 struct kfd_topology_device *dev;
962 unsigned int i = 0;
963 int r = 0;
964
965 down_read(&topology_lock);
966
967 list_for_each_entry(dev, &topology_device_list, list) {
968 if (!dev->gpu) {
969 i++;
970 continue;
971 }
972
973 seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id);
974 r = dqm_debugfs_hqds(m, dev->gpu->dqm);
975 if (r)
976 break;
977 }
978
979 up_read(&topology_lock);
980
981 return r;
982}
983
984int kfd_debugfs_rls_by_device(struct seq_file *m, void *data)
985{
986 struct kfd_topology_device *dev;
987 unsigned int i = 0;
988 int r = 0;
989
990 down_read(&topology_lock);
991
992 list_for_each_entry(dev, &topology_device_list, list) {
993 if (!dev->gpu) {
994 i++;
995 continue;
996 }
997
998 seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id);
999 r = pm_debugfs_runlist(m, &dev->gpu->dqm->packets);
1000 if (r)
1001 break;
1002 }
1003
1004 up_read(&topology_lock);
1005
1006 return r;
1007}
1008
1009#endif
Linux 6.x block layer and io_uring tree(s)