Commit | Line | Data |
---|---|---|
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> |
520b8fb7 FK |
31 | #include <linux/dmi.h> |
32 | #include <linux/atomic.h> | |
5b5c4e40 EP |
33 | |
34 | #include "kfd_priv.h" | |
35 | #include "kfd_crat.h" | |
36 | #include "kfd_topology.h" | |
851a645e | 37 | #include "kfd_device_queue_manager.h" |
64d1c3a4 | 38 | #include "kfd_iommu.h" |
5b87245f | 39 | #include "amdgpu_amdkfd.h" |
0dee45a2 | 40 | #include "amdgpu_ras.h" |
5b5c4e40 | 41 | |
4f449311 HK |
42 | /* topology_device_list - Master list of all topology devices */ |
43 | static struct list_head topology_device_list; | |
520b8fb7 | 44 | static struct kfd_system_properties sys_props; |
5b5c4e40 EP |
45 | |
46 | static DECLARE_RWSEM(topology_lock); | |
520b8fb7 | 47 | static atomic_t topology_crat_proximity_domain; |
5b5c4e40 | 48 | |
3a87177e HK |
49 | struct kfd_topology_device *kfd_topology_device_by_proximity_domain( |
50 | uint32_t proximity_domain) | |
51 | { | |
52 | struct kfd_topology_device *top_dev; | |
53 | struct kfd_topology_device *device = NULL; | |
54 | ||
55 | down_read(&topology_lock); | |
56 | ||
57 | list_for_each_entry(top_dev, &topology_device_list, list) | |
58 | if (top_dev->proximity_domain == proximity_domain) { | |
59 | device = top_dev; | |
60 | break; | |
61 | } | |
62 | ||
63 | up_read(&topology_lock); | |
64 | ||
65 | return device; | |
66 | } | |
67 | ||
44d8cc6f | 68 | struct kfd_topology_device *kfd_topology_device_by_id(uint32_t gpu_id) |
5b5c4e40 | 69 | { |
44d8cc6f YZ |
70 | struct kfd_topology_device *top_dev = NULL; |
71 | struct kfd_topology_device *ret = NULL; | |
5b5c4e40 EP |
72 | |
73 | down_read(&topology_lock); | |
74 | ||
75 | list_for_each_entry(top_dev, &topology_device_list, list) | |
76 | if (top_dev->gpu_id == gpu_id) { | |
44d8cc6f | 77 | ret = top_dev; |
5b5c4e40 EP |
78 | break; |
79 | } | |
80 | ||
81 | up_read(&topology_lock); | |
82 | ||
44d8cc6f YZ |
83 | return ret; |
84 | } | |
85 | ||
86 | struct kfd_dev *kfd_device_by_id(uint32_t gpu_id) | |
87 | { | |
88 | struct kfd_topology_device *top_dev; | |
89 | ||
90 | top_dev = kfd_topology_device_by_id(gpu_id); | |
91 | if (!top_dev) | |
92 | return NULL; | |
93 | ||
94 | return top_dev->gpu; | |
5b5c4e40 EP |
95 | } |
96 | ||
97 | struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev) | |
98 | { | |
99 | struct kfd_topology_device *top_dev; | |
100 | struct kfd_dev *device = NULL; | |
101 | ||
102 | down_read(&topology_lock); | |
103 | ||
104 | list_for_each_entry(top_dev, &topology_device_list, list) | |
3704d56e | 105 | if (top_dev->gpu && top_dev->gpu->pdev == pdev) { |
5b5c4e40 EP |
106 | device = top_dev->gpu; |
107 | break; | |
108 | } | |
109 | ||
110 | up_read(&topology_lock); | |
111 | ||
112 | return device; | |
113 | } | |
114 | ||
1dde0ea9 FK |
115 | struct kfd_dev *kfd_device_by_kgd(const struct kgd_dev *kgd) |
116 | { | |
117 | struct kfd_topology_device *top_dev; | |
118 | struct kfd_dev *device = NULL; | |
119 | ||
120 | down_read(&topology_lock); | |
121 | ||
122 | list_for_each_entry(top_dev, &topology_device_list, list) | |
123 | if (top_dev->gpu && top_dev->gpu->kgd == kgd) { | |
124 | device = top_dev->gpu; | |
125 | break; | |
126 | } | |
127 | ||
128 | up_read(&topology_lock); | |
129 | ||
130 | return device; | |
131 | } | |
132 | ||
3a87177e | 133 | /* Called with write topology_lock acquired */ |
5b5c4e40 EP |
134 | static void kfd_release_topology_device(struct kfd_topology_device *dev) |
135 | { | |
136 | struct kfd_mem_properties *mem; | |
137 | struct kfd_cache_properties *cache; | |
138 | struct kfd_iolink_properties *iolink; | |
f4757347 | 139 | struct kfd_perf_properties *perf; |
5b5c4e40 | 140 | |
5b5c4e40 EP |
141 | list_del(&dev->list); |
142 | ||
143 | while (dev->mem_props.next != &dev->mem_props) { | |
144 | mem = container_of(dev->mem_props.next, | |
145 | struct kfd_mem_properties, list); | |
146 | list_del(&mem->list); | |
147 | kfree(mem); | |
148 | } | |
149 | ||
150 | while (dev->cache_props.next != &dev->cache_props) { | |
151 | cache = container_of(dev->cache_props.next, | |
152 | struct kfd_cache_properties, list); | |
153 | list_del(&cache->list); | |
154 | kfree(cache); | |
155 | } | |
156 | ||
157 | while (dev->io_link_props.next != &dev->io_link_props) { | |
158 | iolink = container_of(dev->io_link_props.next, | |
159 | struct kfd_iolink_properties, list); | |
160 | list_del(&iolink->list); | |
161 | kfree(iolink); | |
162 | } | |
163 | ||
f4757347 AL |
164 | while (dev->perf_props.next != &dev->perf_props) { |
165 | perf = container_of(dev->perf_props.next, | |
166 | struct kfd_perf_properties, list); | |
167 | list_del(&perf->list); | |
168 | kfree(perf); | |
169 | } | |
170 | ||
5b5c4e40 | 171 | kfree(dev); |
5b5c4e40 EP |
172 | } |
173 | ||
4f449311 | 174 | void kfd_release_topology_device_list(struct list_head *device_list) |
5b5c4e40 EP |
175 | { |
176 | struct kfd_topology_device *dev; | |
177 | ||
4f449311 HK |
178 | while (!list_empty(device_list)) { |
179 | dev = list_first_entry(device_list, | |
180 | struct kfd_topology_device, list); | |
5b5c4e40 | 181 | kfd_release_topology_device(dev); |
4f449311 | 182 | } |
5b5c4e40 EP |
183 | } |
184 | ||
4f449311 HK |
185 | static void kfd_release_live_view(void) |
186 | { | |
187 | kfd_release_topology_device_list(&topology_device_list); | |
5b5c4e40 EP |
188 | memset(&sys_props, 0, sizeof(sys_props)); |
189 | } | |
190 | ||
4f449311 HK |
191 | struct kfd_topology_device *kfd_create_topology_device( |
192 | struct list_head *device_list) | |
5b5c4e40 EP |
193 | { |
194 | struct kfd_topology_device *dev; | |
195 | ||
196 | dev = kfd_alloc_struct(dev); | |
4eacc26b | 197 | if (!dev) { |
5b5c4e40 | 198 | pr_err("No memory to allocate a topology device"); |
16b9201c | 199 | return NULL; |
5b5c4e40 EP |
200 | } |
201 | ||
202 | INIT_LIST_HEAD(&dev->mem_props); | |
203 | INIT_LIST_HEAD(&dev->cache_props); | |
204 | INIT_LIST_HEAD(&dev->io_link_props); | |
f4757347 | 205 | INIT_LIST_HEAD(&dev->perf_props); |
5b5c4e40 | 206 | |
4f449311 | 207 | list_add_tail(&dev->list, device_list); |
5b5c4e40 EP |
208 | |
209 | return dev; | |
16b9201c | 210 | } |
5b5c4e40 | 211 | |
5b5c4e40 EP |
212 | |
213 | #define sysfs_show_gen_prop(buffer, fmt, ...) \ | |
214 | snprintf(buffer, PAGE_SIZE, "%s"fmt, buffer, __VA_ARGS__) | |
215 | #define sysfs_show_32bit_prop(buffer, name, value) \ | |
216 | sysfs_show_gen_prop(buffer, "%s %u\n", name, value) | |
217 | #define sysfs_show_64bit_prop(buffer, name, value) \ | |
218 | sysfs_show_gen_prop(buffer, "%s %llu\n", name, value) | |
219 | #define sysfs_show_32bit_val(buffer, value) \ | |
220 | sysfs_show_gen_prop(buffer, "%u\n", value) | |
221 | #define sysfs_show_str_val(buffer, value) \ | |
222 | sysfs_show_gen_prop(buffer, "%s\n", value) | |
223 | ||
224 | static ssize_t sysprops_show(struct kobject *kobj, struct attribute *attr, | |
225 | char *buffer) | |
226 | { | |
227 | ssize_t ret; | |
228 | ||
229 | /* Making sure that the buffer is an empty string */ | |
230 | buffer[0] = 0; | |
231 | ||
232 | if (attr == &sys_props.attr_genid) { | |
233 | ret = sysfs_show_32bit_val(buffer, sys_props.generation_count); | |
234 | } else if (attr == &sys_props.attr_props) { | |
235 | sysfs_show_64bit_prop(buffer, "platform_oem", | |
236 | sys_props.platform_oem); | |
237 | sysfs_show_64bit_prop(buffer, "platform_id", | |
238 | sys_props.platform_id); | |
239 | ret = sysfs_show_64bit_prop(buffer, "platform_rev", | |
240 | sys_props.platform_rev); | |
241 | } else { | |
242 | ret = -EINVAL; | |
243 | } | |
244 | ||
245 | return ret; | |
246 | } | |
247 | ||
5108d768 YZ |
248 | static void kfd_topology_kobj_release(struct kobject *kobj) |
249 | { | |
250 | kfree(kobj); | |
251 | } | |
252 | ||
5b5c4e40 EP |
253 | static const struct sysfs_ops sysprops_ops = { |
254 | .show = sysprops_show, | |
255 | }; | |
256 | ||
257 | static struct kobj_type sysprops_type = { | |
5108d768 | 258 | .release = kfd_topology_kobj_release, |
5b5c4e40 EP |
259 | .sysfs_ops = &sysprops_ops, |
260 | }; | |
261 | ||
262 | static ssize_t iolink_show(struct kobject *kobj, struct attribute *attr, | |
263 | char *buffer) | |
264 | { | |
265 | ssize_t ret; | |
266 | struct kfd_iolink_properties *iolink; | |
267 | ||
268 | /* Making sure that the buffer is an empty string */ | |
269 | buffer[0] = 0; | |
270 | ||
271 | iolink = container_of(attr, struct kfd_iolink_properties, attr); | |
6b855f7b HK |
272 | if (iolink->gpu && kfd_devcgroup_check_permission(iolink->gpu)) |
273 | return -EPERM; | |
5b5c4e40 EP |
274 | sysfs_show_32bit_prop(buffer, "type", iolink->iolink_type); |
275 | sysfs_show_32bit_prop(buffer, "version_major", iolink->ver_maj); | |
276 | sysfs_show_32bit_prop(buffer, "version_minor", iolink->ver_min); | |
277 | sysfs_show_32bit_prop(buffer, "node_from", iolink->node_from); | |
278 | sysfs_show_32bit_prop(buffer, "node_to", iolink->node_to); | |
279 | sysfs_show_32bit_prop(buffer, "weight", iolink->weight); | |
280 | sysfs_show_32bit_prop(buffer, "min_latency", iolink->min_latency); | |
281 | sysfs_show_32bit_prop(buffer, "max_latency", iolink->max_latency); | |
282 | sysfs_show_32bit_prop(buffer, "min_bandwidth", iolink->min_bandwidth); | |
283 | sysfs_show_32bit_prop(buffer, "max_bandwidth", iolink->max_bandwidth); | |
284 | sysfs_show_32bit_prop(buffer, "recommended_transfer_size", | |
285 | iolink->rec_transfer_size); | |
286 | ret = sysfs_show_32bit_prop(buffer, "flags", iolink->flags); | |
287 | ||
288 | return ret; | |
289 | } | |
290 | ||
291 | static const struct sysfs_ops iolink_ops = { | |
292 | .show = iolink_show, | |
293 | }; | |
294 | ||
295 | static struct kobj_type iolink_type = { | |
5108d768 | 296 | .release = kfd_topology_kobj_release, |
5b5c4e40 EP |
297 | .sysfs_ops = &iolink_ops, |
298 | }; | |
299 | ||
300 | static ssize_t mem_show(struct kobject *kobj, struct attribute *attr, | |
301 | char *buffer) | |
302 | { | |
303 | ssize_t ret; | |
304 | struct kfd_mem_properties *mem; | |
305 | ||
306 | /* Making sure that the buffer is an empty string */ | |
307 | buffer[0] = 0; | |
308 | ||
309 | mem = container_of(attr, struct kfd_mem_properties, attr); | |
6b855f7b HK |
310 | if (mem->gpu && kfd_devcgroup_check_permission(mem->gpu)) |
311 | return -EPERM; | |
5b5c4e40 EP |
312 | sysfs_show_32bit_prop(buffer, "heap_type", mem->heap_type); |
313 | sysfs_show_64bit_prop(buffer, "size_in_bytes", mem->size_in_bytes); | |
314 | sysfs_show_32bit_prop(buffer, "flags", mem->flags); | |
315 | sysfs_show_32bit_prop(buffer, "width", mem->width); | |
316 | ret = sysfs_show_32bit_prop(buffer, "mem_clk_max", mem->mem_clk_max); | |
317 | ||
318 | return ret; | |
319 | } | |
320 | ||
321 | static const struct sysfs_ops mem_ops = { | |
322 | .show = mem_show, | |
323 | }; | |
324 | ||
325 | static struct kobj_type mem_type = { | |
5108d768 | 326 | .release = kfd_topology_kobj_release, |
5b5c4e40 EP |
327 | .sysfs_ops = &mem_ops, |
328 | }; | |
329 | ||
330 | static ssize_t kfd_cache_show(struct kobject *kobj, struct attribute *attr, | |
331 | char *buffer) | |
332 | { | |
333 | ssize_t ret; | |
bc0c75a3 | 334 | uint32_t i, j; |
5b5c4e40 EP |
335 | struct kfd_cache_properties *cache; |
336 | ||
337 | /* Making sure that the buffer is an empty string */ | |
338 | buffer[0] = 0; | |
339 | ||
340 | cache = container_of(attr, struct kfd_cache_properties, attr); | |
6b855f7b HK |
341 | if (cache->gpu && kfd_devcgroup_check_permission(cache->gpu)) |
342 | return -EPERM; | |
5b5c4e40 EP |
343 | sysfs_show_32bit_prop(buffer, "processor_id_low", |
344 | cache->processor_id_low); | |
345 | sysfs_show_32bit_prop(buffer, "level", cache->cache_level); | |
346 | sysfs_show_32bit_prop(buffer, "size", cache->cache_size); | |
347 | sysfs_show_32bit_prop(buffer, "cache_line_size", cache->cacheline_size); | |
348 | sysfs_show_32bit_prop(buffer, "cache_lines_per_tag", | |
349 | cache->cachelines_per_tag); | |
350 | sysfs_show_32bit_prop(buffer, "association", cache->cache_assoc); | |
351 | sysfs_show_32bit_prop(buffer, "latency", cache->cache_latency); | |
352 | sysfs_show_32bit_prop(buffer, "type", cache->cache_type); | |
353 | snprintf(buffer, PAGE_SIZE, "%ssibling_map ", buffer); | |
bc0c75a3 HK |
354 | for (i = 0; i < CRAT_SIBLINGMAP_SIZE; i++) |
355 | for (j = 0; j < sizeof(cache->sibling_map[0])*8; j++) { | |
356 | /* Check each bit */ | |
357 | if (cache->sibling_map[i] & (1 << j)) | |
358 | ret = snprintf(buffer, PAGE_SIZE, | |
359 | "%s%d%s", buffer, 1, ","); | |
360 | else | |
361 | ret = snprintf(buffer, PAGE_SIZE, | |
362 | "%s%d%s", buffer, 0, ","); | |
363 | } | |
364 | /* Replace the last "," with end of line */ | |
365 | *(buffer + strlen(buffer) - 1) = 0xA; | |
5b5c4e40 EP |
366 | return ret; |
367 | } | |
368 | ||
369 | static const struct sysfs_ops cache_ops = { | |
370 | .show = kfd_cache_show, | |
371 | }; | |
372 | ||
373 | static struct kobj_type cache_type = { | |
5108d768 | 374 | .release = kfd_topology_kobj_release, |
5b5c4e40 EP |
375 | .sysfs_ops = &cache_ops, |
376 | }; | |
377 | ||
f4757347 AL |
378 | /****** Sysfs of Performance Counters ******/ |
379 | ||
380 | struct kfd_perf_attr { | |
381 | struct kobj_attribute attr; | |
382 | uint32_t data; | |
383 | }; | |
384 | ||
385 | static ssize_t perf_show(struct kobject *kobj, struct kobj_attribute *attrs, | |
386 | char *buf) | |
387 | { | |
388 | struct kfd_perf_attr *attr; | |
389 | ||
390 | buf[0] = 0; | |
391 | attr = container_of(attrs, struct kfd_perf_attr, attr); | |
392 | if (!attr->data) /* invalid data for PMC */ | |
393 | return 0; | |
394 | else | |
395 | return sysfs_show_32bit_val(buf, attr->data); | |
396 | } | |
397 | ||
398 | #define KFD_PERF_DESC(_name, _data) \ | |
399 | { \ | |
400 | .attr = __ATTR(_name, 0444, perf_show, NULL), \ | |
401 | .data = _data, \ | |
402 | } | |
403 | ||
404 | static struct kfd_perf_attr perf_attr_iommu[] = { | |
405 | KFD_PERF_DESC(max_concurrent, 0), | |
406 | KFD_PERF_DESC(num_counters, 0), | |
407 | KFD_PERF_DESC(counter_ids, 0), | |
408 | }; | |
409 | /****************************************/ | |
410 | ||
5b5c4e40 EP |
411 | static ssize_t node_show(struct kobject *kobj, struct attribute *attr, |
412 | char *buffer) | |
413 | { | |
5b5c4e40 | 414 | struct kfd_topology_device *dev; |
f7c826ad | 415 | uint32_t log_max_watch_addr; |
5b5c4e40 EP |
416 | |
417 | /* Making sure that the buffer is an empty string */ | |
418 | buffer[0] = 0; | |
419 | ||
420 | if (strcmp(attr->name, "gpu_id") == 0) { | |
421 | dev = container_of(attr, struct kfd_topology_device, | |
422 | attr_gpuid); | |
6b855f7b HK |
423 | if (dev->gpu && kfd_devcgroup_check_permission(dev->gpu)) |
424 | return -EPERM; | |
f7c826ad AS |
425 | return sysfs_show_32bit_val(buffer, dev->gpu_id); |
426 | } | |
427 | ||
428 | if (strcmp(attr->name, "name") == 0) { | |
5b5c4e40 EP |
429 | dev = container_of(attr, struct kfd_topology_device, |
430 | attr_name); | |
c181159a | 431 | |
6b855f7b HK |
432 | if (dev->gpu && kfd_devcgroup_check_permission(dev->gpu)) |
433 | return -EPERM; | |
c181159a | 434 | return sysfs_show_str_val(buffer, dev->node_props.name); |
f7c826ad | 435 | } |
5b5c4e40 | 436 | |
f7c826ad AS |
437 | dev = container_of(attr, struct kfd_topology_device, |
438 | attr_props); | |
6b855f7b HK |
439 | if (dev->gpu && kfd_devcgroup_check_permission(dev->gpu)) |
440 | return -EPERM; | |
f7c826ad AS |
441 | sysfs_show_32bit_prop(buffer, "cpu_cores_count", |
442 | dev->node_props.cpu_cores_count); | |
443 | sysfs_show_32bit_prop(buffer, "simd_count", | |
444 | dev->node_props.simd_count); | |
175b9263 FK |
445 | sysfs_show_32bit_prop(buffer, "mem_banks_count", |
446 | dev->node_props.mem_banks_count); | |
f7c826ad AS |
447 | sysfs_show_32bit_prop(buffer, "caches_count", |
448 | dev->node_props.caches_count); | |
449 | sysfs_show_32bit_prop(buffer, "io_links_count", | |
450 | dev->node_props.io_links_count); | |
451 | sysfs_show_32bit_prop(buffer, "cpu_core_id_base", | |
452 | dev->node_props.cpu_core_id_base); | |
453 | sysfs_show_32bit_prop(buffer, "simd_id_base", | |
454 | dev->node_props.simd_id_base); | |
f7c826ad AS |
455 | sysfs_show_32bit_prop(buffer, "max_waves_per_simd", |
456 | dev->node_props.max_waves_per_simd); | |
457 | sysfs_show_32bit_prop(buffer, "lds_size_in_kb", | |
458 | dev->node_props.lds_size_in_kb); | |
459 | sysfs_show_32bit_prop(buffer, "gds_size_in_kb", | |
460 | dev->node_props.gds_size_in_kb); | |
29e76462 OZ |
461 | sysfs_show_32bit_prop(buffer, "num_gws", |
462 | dev->node_props.num_gws); | |
f7c826ad AS |
463 | sysfs_show_32bit_prop(buffer, "wave_front_size", |
464 | dev->node_props.wave_front_size); | |
465 | sysfs_show_32bit_prop(buffer, "array_count", | |
466 | dev->node_props.array_count); | |
467 | sysfs_show_32bit_prop(buffer, "simd_arrays_per_engine", | |
468 | dev->node_props.simd_arrays_per_engine); | |
469 | sysfs_show_32bit_prop(buffer, "cu_per_simd_array", | |
470 | dev->node_props.cu_per_simd_array); | |
471 | sysfs_show_32bit_prop(buffer, "simd_per_cu", | |
472 | dev->node_props.simd_per_cu); | |
473 | sysfs_show_32bit_prop(buffer, "max_slots_scratch_cu", | |
474 | dev->node_props.max_slots_scratch_cu); | |
475 | sysfs_show_32bit_prop(buffer, "vendor_id", | |
476 | dev->node_props.vendor_id); | |
477 | sysfs_show_32bit_prop(buffer, "device_id", | |
478 | dev->node_props.device_id); | |
479 | sysfs_show_32bit_prop(buffer, "location_id", | |
480 | dev->node_props.location_id); | |
7c9b7171 OZ |
481 | sysfs_show_32bit_prop(buffer, "drm_render_minor", |
482 | dev->node_props.drm_render_minor); | |
0c1690e3 SL |
483 | sysfs_show_64bit_prop(buffer, "hive_id", |
484 | dev->node_props.hive_id); | |
14568cf6 OZ |
485 | sysfs_show_32bit_prop(buffer, "num_sdma_engines", |
486 | dev->node_props.num_sdma_engines); | |
487 | sysfs_show_32bit_prop(buffer, "num_sdma_xgmi_engines", | |
488 | dev->node_props.num_sdma_xgmi_engines); | |
f7c826ad AS |
489 | |
490 | if (dev->gpu) { | |
491 | log_max_watch_addr = | |
492 | __ilog2_u32(dev->gpu->device_info->num_of_watch_points); | |
493 | ||
494 | if (log_max_watch_addr) { | |
495 | dev->node_props.capability |= | |
496 | HSA_CAP_WATCH_POINTS_SUPPORTED; | |
497 | ||
498 | dev->node_props.capability |= | |
499 | ((log_max_watch_addr << | |
500 | HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT) & | |
501 | HSA_CAP_WATCH_POINTS_TOTALBITS_MASK); | |
5b5c4e40 EP |
502 | } |
503 | ||
413e85d5 BG |
504 | if (dev->gpu->device_info->asic_family == CHIP_TONGA) |
505 | dev->node_props.capability |= | |
506 | HSA_CAP_AQL_QUEUE_DOUBLE_MAP; | |
507 | ||
f7c826ad | 508 | sysfs_show_32bit_prop(buffer, "max_engine_clk_fcompute", |
3a87177e | 509 | dev->node_props.max_engine_clk_fcompute); |
42e08c78 | 510 | |
f7c826ad | 511 | sysfs_show_64bit_prop(buffer, "local_mem_size", |
42e08c78 | 512 | (unsigned long long int) 0); |
f7c826ad AS |
513 | |
514 | sysfs_show_32bit_prop(buffer, "fw_version", | |
5ade6c9c | 515 | dev->gpu->mec_fw_version); |
826f5de8 AS |
516 | sysfs_show_32bit_prop(buffer, "capability", |
517 | dev->node_props.capability); | |
5ade6c9c FK |
518 | sysfs_show_32bit_prop(buffer, "sdma_fw_version", |
519 | dev->gpu->sdma_fw_version); | |
5b5c4e40 EP |
520 | } |
521 | ||
f7c826ad AS |
522 | return sysfs_show_32bit_prop(buffer, "max_engine_clk_ccompute", |
523 | cpufreq_quick_get_max(0)/1000); | |
5b5c4e40 EP |
524 | } |
525 | ||
526 | static const struct sysfs_ops node_ops = { | |
527 | .show = node_show, | |
528 | }; | |
529 | ||
530 | static struct kobj_type node_type = { | |
5108d768 | 531 | .release = kfd_topology_kobj_release, |
5b5c4e40 EP |
532 | .sysfs_ops = &node_ops, |
533 | }; | |
534 | ||
535 | static void kfd_remove_sysfs_file(struct kobject *kobj, struct attribute *attr) | |
536 | { | |
537 | sysfs_remove_file(kobj, attr); | |
538 | kobject_del(kobj); | |
539 | kobject_put(kobj); | |
540 | } | |
541 | ||
542 | static void kfd_remove_sysfs_node_entry(struct kfd_topology_device *dev) | |
543 | { | |
544 | struct kfd_iolink_properties *iolink; | |
545 | struct kfd_cache_properties *cache; | |
546 | struct kfd_mem_properties *mem; | |
f4757347 | 547 | struct kfd_perf_properties *perf; |
5b5c4e40 | 548 | |
5b5c4e40 EP |
549 | if (dev->kobj_iolink) { |
550 | list_for_each_entry(iolink, &dev->io_link_props, list) | |
551 | if (iolink->kobj) { | |
552 | kfd_remove_sysfs_file(iolink->kobj, | |
553 | &iolink->attr); | |
16b9201c | 554 | iolink->kobj = NULL; |
5b5c4e40 EP |
555 | } |
556 | kobject_del(dev->kobj_iolink); | |
557 | kobject_put(dev->kobj_iolink); | |
16b9201c | 558 | dev->kobj_iolink = NULL; |
5b5c4e40 EP |
559 | } |
560 | ||
561 | if (dev->kobj_cache) { | |
562 | list_for_each_entry(cache, &dev->cache_props, list) | |
563 | if (cache->kobj) { | |
564 | kfd_remove_sysfs_file(cache->kobj, | |
565 | &cache->attr); | |
16b9201c | 566 | cache->kobj = NULL; |
5b5c4e40 EP |
567 | } |
568 | kobject_del(dev->kobj_cache); | |
569 | kobject_put(dev->kobj_cache); | |
16b9201c | 570 | dev->kobj_cache = NULL; |
5b5c4e40 EP |
571 | } |
572 | ||
573 | if (dev->kobj_mem) { | |
574 | list_for_each_entry(mem, &dev->mem_props, list) | |
575 | if (mem->kobj) { | |
576 | kfd_remove_sysfs_file(mem->kobj, &mem->attr); | |
16b9201c | 577 | mem->kobj = NULL; |
5b5c4e40 EP |
578 | } |
579 | kobject_del(dev->kobj_mem); | |
580 | kobject_put(dev->kobj_mem); | |
16b9201c | 581 | dev->kobj_mem = NULL; |
5b5c4e40 EP |
582 | } |
583 | ||
f4757347 AL |
584 | if (dev->kobj_perf) { |
585 | list_for_each_entry(perf, &dev->perf_props, list) { | |
586 | kfree(perf->attr_group); | |
587 | perf->attr_group = NULL; | |
588 | } | |
589 | kobject_del(dev->kobj_perf); | |
590 | kobject_put(dev->kobj_perf); | |
591 | dev->kobj_perf = NULL; | |
592 | } | |
593 | ||
5b5c4e40 EP |
594 | if (dev->kobj_node) { |
595 | sysfs_remove_file(dev->kobj_node, &dev->attr_gpuid); | |
596 | sysfs_remove_file(dev->kobj_node, &dev->attr_name); | |
597 | sysfs_remove_file(dev->kobj_node, &dev->attr_props); | |
598 | kobject_del(dev->kobj_node); | |
599 | kobject_put(dev->kobj_node); | |
16b9201c | 600 | dev->kobj_node = NULL; |
5b5c4e40 EP |
601 | } |
602 | } | |
603 | ||
604 | static int kfd_build_sysfs_node_entry(struct kfd_topology_device *dev, | |
605 | uint32_t id) | |
606 | { | |
607 | struct kfd_iolink_properties *iolink; | |
608 | struct kfd_cache_properties *cache; | |
609 | struct kfd_mem_properties *mem; | |
f4757347 | 610 | struct kfd_perf_properties *perf; |
5b5c4e40 | 611 | int ret; |
f4757347 AL |
612 | uint32_t i, num_attrs; |
613 | struct attribute **attrs; | |
5b5c4e40 | 614 | |
32fa8219 FK |
615 | if (WARN_ON(dev->kobj_node)) |
616 | return -EEXIST; | |
617 | ||
5b5c4e40 EP |
618 | /* |
619 | * Creating the sysfs folders | |
620 | */ | |
5b5c4e40 EP |
621 | dev->kobj_node = kfd_alloc_struct(dev->kobj_node); |
622 | if (!dev->kobj_node) | |
623 | return -ENOMEM; | |
624 | ||
625 | ret = kobject_init_and_add(dev->kobj_node, &node_type, | |
626 | sys_props.kobj_nodes, "%d", id); | |
627 | if (ret < 0) | |
628 | return ret; | |
629 | ||
630 | dev->kobj_mem = kobject_create_and_add("mem_banks", dev->kobj_node); | |
631 | if (!dev->kobj_mem) | |
632 | return -ENOMEM; | |
633 | ||
634 | dev->kobj_cache = kobject_create_and_add("caches", dev->kobj_node); | |
635 | if (!dev->kobj_cache) | |
636 | return -ENOMEM; | |
637 | ||
638 | dev->kobj_iolink = kobject_create_and_add("io_links", dev->kobj_node); | |
639 | if (!dev->kobj_iolink) | |
640 | return -ENOMEM; | |
641 | ||
f4757347 AL |
642 | dev->kobj_perf = kobject_create_and_add("perf", dev->kobj_node); |
643 | if (!dev->kobj_perf) | |
644 | return -ENOMEM; | |
645 | ||
5b5c4e40 EP |
646 | /* |
647 | * Creating sysfs files for node properties | |
648 | */ | |
649 | dev->attr_gpuid.name = "gpu_id"; | |
650 | dev->attr_gpuid.mode = KFD_SYSFS_FILE_MODE; | |
651 | sysfs_attr_init(&dev->attr_gpuid); | |
652 | dev->attr_name.name = "name"; | |
653 | dev->attr_name.mode = KFD_SYSFS_FILE_MODE; | |
654 | sysfs_attr_init(&dev->attr_name); | |
655 | dev->attr_props.name = "properties"; | |
656 | dev->attr_props.mode = KFD_SYSFS_FILE_MODE; | |
657 | sysfs_attr_init(&dev->attr_props); | |
658 | ret = sysfs_create_file(dev->kobj_node, &dev->attr_gpuid); | |
659 | if (ret < 0) | |
660 | return ret; | |
661 | ret = sysfs_create_file(dev->kobj_node, &dev->attr_name); | |
662 | if (ret < 0) | |
663 | return ret; | |
664 | ret = sysfs_create_file(dev->kobj_node, &dev->attr_props); | |
665 | if (ret < 0) | |
666 | return ret; | |
667 | ||
668 | i = 0; | |
669 | list_for_each_entry(mem, &dev->mem_props, list) { | |
670 | mem->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); | |
671 | if (!mem->kobj) | |
672 | return -ENOMEM; | |
673 | ret = kobject_init_and_add(mem->kobj, &mem_type, | |
674 | dev->kobj_mem, "%d", i); | |
675 | if (ret < 0) | |
676 | return ret; | |
677 | ||
678 | mem->attr.name = "properties"; | |
679 | mem->attr.mode = KFD_SYSFS_FILE_MODE; | |
680 | sysfs_attr_init(&mem->attr); | |
681 | ret = sysfs_create_file(mem->kobj, &mem->attr); | |
682 | if (ret < 0) | |
683 | return ret; | |
684 | i++; | |
685 | } | |
686 | ||
687 | i = 0; | |
688 | list_for_each_entry(cache, &dev->cache_props, list) { | |
689 | cache->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); | |
690 | if (!cache->kobj) | |
691 | return -ENOMEM; | |
692 | ret = kobject_init_and_add(cache->kobj, &cache_type, | |
693 | dev->kobj_cache, "%d", i); | |
694 | if (ret < 0) | |
695 | return ret; | |
696 | ||
697 | cache->attr.name = "properties"; | |
698 | cache->attr.mode = KFD_SYSFS_FILE_MODE; | |
699 | sysfs_attr_init(&cache->attr); | |
700 | ret = sysfs_create_file(cache->kobj, &cache->attr); | |
701 | if (ret < 0) | |
702 | return ret; | |
703 | i++; | |
704 | } | |
705 | ||
706 | i = 0; | |
707 | list_for_each_entry(iolink, &dev->io_link_props, list) { | |
708 | iolink->kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL); | |
709 | if (!iolink->kobj) | |
710 | return -ENOMEM; | |
711 | ret = kobject_init_and_add(iolink->kobj, &iolink_type, | |
712 | dev->kobj_iolink, "%d", i); | |
713 | if (ret < 0) | |
714 | return ret; | |
715 | ||
716 | iolink->attr.name = "properties"; | |
717 | iolink->attr.mode = KFD_SYSFS_FILE_MODE; | |
718 | sysfs_attr_init(&iolink->attr); | |
719 | ret = sysfs_create_file(iolink->kobj, &iolink->attr); | |
720 | if (ret < 0) | |
721 | return ret; | |
722 | i++; | |
f4757347 AL |
723 | } |
724 | ||
725 | /* All hardware blocks have the same number of attributes. */ | |
3f866f5f | 726 | num_attrs = ARRAY_SIZE(perf_attr_iommu); |
f4757347 AL |
727 | list_for_each_entry(perf, &dev->perf_props, list) { |
728 | perf->attr_group = kzalloc(sizeof(struct kfd_perf_attr) | |
729 | * num_attrs + sizeof(struct attribute_group), | |
730 | GFP_KERNEL); | |
731 | if (!perf->attr_group) | |
732 | return -ENOMEM; | |
733 | ||
734 | attrs = (struct attribute **)(perf->attr_group + 1); | |
735 | if (!strcmp(perf->block_name, "iommu")) { | |
736 | /* Information of IOMMU's num_counters and counter_ids is shown | |
737 | * under /sys/bus/event_source/devices/amd_iommu. We don't | |
738 | * duplicate here. | |
739 | */ | |
740 | perf_attr_iommu[0].data = perf->max_concurrent; | |
741 | for (i = 0; i < num_attrs; i++) | |
742 | attrs[i] = &perf_attr_iommu[i].attr.attr; | |
743 | } | |
744 | perf->attr_group->name = perf->block_name; | |
745 | perf->attr_group->attrs = attrs; | |
746 | ret = sysfs_create_group(dev->kobj_perf, perf->attr_group); | |
747 | if (ret < 0) | |
748 | return ret; | |
749 | } | |
5b5c4e40 EP |
750 | |
751 | return 0; | |
752 | } | |
753 | ||
3a87177e | 754 | /* Called with write topology lock acquired */ |
5b5c4e40 EP |
755 | static int kfd_build_sysfs_node_tree(void) |
756 | { | |
757 | struct kfd_topology_device *dev; | |
758 | int ret; | |
759 | uint32_t i = 0; | |
760 | ||
761 | list_for_each_entry(dev, &topology_device_list, list) { | |
8dfead6c | 762 | ret = kfd_build_sysfs_node_entry(dev, i); |
5b5c4e40 EP |
763 | if (ret < 0) |
764 | return ret; | |
765 | i++; | |
766 | } | |
767 | ||
768 | return 0; | |
769 | } | |
770 | ||
3a87177e | 771 | /* Called with write topology lock acquired */ |
5b5c4e40 EP |
772 | static void kfd_remove_sysfs_node_tree(void) |
773 | { | |
774 | struct kfd_topology_device *dev; | |
775 | ||
776 | list_for_each_entry(dev, &topology_device_list, list) | |
777 | kfd_remove_sysfs_node_entry(dev); | |
778 | } | |
779 | ||
780 | static int kfd_topology_update_sysfs(void) | |
781 | { | |
782 | int ret; | |
783 | ||
784 | pr_info("Creating topology SYSFS entries\n"); | |
4eacc26b | 785 | if (!sys_props.kobj_topology) { |
5b5c4e40 EP |
786 | sys_props.kobj_topology = |
787 | kfd_alloc_struct(sys_props.kobj_topology); | |
788 | if (!sys_props.kobj_topology) | |
789 | return -ENOMEM; | |
790 | ||
791 | ret = kobject_init_and_add(sys_props.kobj_topology, | |
792 | &sysprops_type, &kfd_device->kobj, | |
793 | "topology"); | |
794 | if (ret < 0) | |
795 | return ret; | |
796 | ||
797 | sys_props.kobj_nodes = kobject_create_and_add("nodes", | |
798 | sys_props.kobj_topology); | |
799 | if (!sys_props.kobj_nodes) | |
800 | return -ENOMEM; | |
801 | ||
802 | sys_props.attr_genid.name = "generation_id"; | |
803 | sys_props.attr_genid.mode = KFD_SYSFS_FILE_MODE; | |
804 | sysfs_attr_init(&sys_props.attr_genid); | |
805 | ret = sysfs_create_file(sys_props.kobj_topology, | |
806 | &sys_props.attr_genid); | |
807 | if (ret < 0) | |
808 | return ret; | |
809 | ||
810 | sys_props.attr_props.name = "system_properties"; | |
811 | sys_props.attr_props.mode = KFD_SYSFS_FILE_MODE; | |
812 | sysfs_attr_init(&sys_props.attr_props); | |
813 | ret = sysfs_create_file(sys_props.kobj_topology, | |
814 | &sys_props.attr_props); | |
815 | if (ret < 0) | |
816 | return ret; | |
817 | } | |
818 | ||
819 | kfd_remove_sysfs_node_tree(); | |
820 | ||
821 | return kfd_build_sysfs_node_tree(); | |
822 | } | |
823 | ||
824 | static void kfd_topology_release_sysfs(void) | |
825 | { | |
826 | kfd_remove_sysfs_node_tree(); | |
827 | if (sys_props.kobj_topology) { | |
828 | sysfs_remove_file(sys_props.kobj_topology, | |
829 | &sys_props.attr_genid); | |
830 | sysfs_remove_file(sys_props.kobj_topology, | |
831 | &sys_props.attr_props); | |
832 | if (sys_props.kobj_nodes) { | |
833 | kobject_del(sys_props.kobj_nodes); | |
834 | kobject_put(sys_props.kobj_nodes); | |
16b9201c | 835 | sys_props.kobj_nodes = NULL; |
5b5c4e40 EP |
836 | } |
837 | kobject_del(sys_props.kobj_topology); | |
838 | kobject_put(sys_props.kobj_topology); | |
16b9201c | 839 | sys_props.kobj_topology = NULL; |
5b5c4e40 EP |
840 | } |
841 | } | |
842 | ||
4f449311 HK |
843 | /* Called with write topology_lock acquired */ |
844 | static void kfd_topology_update_device_list(struct list_head *temp_list, | |
845 | struct list_head *master_list) | |
846 | { | |
847 | while (!list_empty(temp_list)) { | |
848 | list_move_tail(temp_list->next, master_list); | |
849 | sys_props.num_devices++; | |
850 | } | |
851 | } | |
852 | ||
520b8fb7 FK |
853 | static void kfd_debug_print_topology(void) |
854 | { | |
855 | struct kfd_topology_device *dev; | |
856 | ||
857 | down_read(&topology_lock); | |
858 | ||
859 | dev = list_last_entry(&topology_device_list, | |
860 | struct kfd_topology_device, list); | |
861 | if (dev) { | |
862 | if (dev->node_props.cpu_cores_count && | |
863 | dev->node_props.simd_count) { | |
864 | pr_info("Topology: Add APU node [0x%0x:0x%0x]\n", | |
865 | dev->node_props.device_id, | |
866 | dev->node_props.vendor_id); | |
867 | } else if (dev->node_props.cpu_cores_count) | |
868 | pr_info("Topology: Add CPU node\n"); | |
869 | else if (dev->node_props.simd_count) | |
870 | pr_info("Topology: Add dGPU node [0x%0x:0x%0x]\n", | |
871 | dev->node_props.device_id, | |
872 | dev->node_props.vendor_id); | |
873 | } | |
874 | up_read(&topology_lock); | |
875 | } | |
876 | ||
877 | /* Helper function for intializing platform_xx members of | |
878 | * kfd_system_properties. Uses OEM info from the last CPU/APU node. | |
879 | */ | |
880 | static void kfd_update_system_properties(void) | |
881 | { | |
882 | struct kfd_topology_device *dev; | |
883 | ||
884 | down_read(&topology_lock); | |
885 | dev = list_last_entry(&topology_device_list, | |
886 | struct kfd_topology_device, list); | |
887 | if (dev) { | |
888 | sys_props.platform_id = | |
889 | (*((uint64_t *)dev->oem_id)) & CRAT_OEMID_64BIT_MASK; | |
890 | sys_props.platform_oem = *((uint64_t *)dev->oem_table_id); | |
891 | sys_props.platform_rev = dev->oem_revision; | |
892 | } | |
893 | up_read(&topology_lock); | |
894 | } | |
895 | ||
896 | static void find_system_memory(const struct dmi_header *dm, | |
897 | void *private) | |
898 | { | |
899 | struct kfd_mem_properties *mem; | |
900 | u16 mem_width, mem_clock; | |
901 | struct kfd_topology_device *kdev = | |
902 | (struct kfd_topology_device *)private; | |
903 | const u8 *dmi_data = (const u8 *)(dm + 1); | |
904 | ||
905 | if (dm->type == DMI_ENTRY_MEM_DEVICE && dm->length >= 0x15) { | |
906 | mem_width = (u16)(*(const u16 *)(dmi_data + 0x6)); | |
907 | mem_clock = (u16)(*(const u16 *)(dmi_data + 0x11)); | |
908 | list_for_each_entry(mem, &kdev->mem_props, list) { | |
909 | if (mem_width != 0xFFFF && mem_width != 0) | |
910 | mem->width = mem_width; | |
911 | if (mem_clock != 0) | |
912 | mem->mem_clk_max = mem_clock; | |
913 | } | |
914 | } | |
915 | } | |
f4757347 AL |
916 | |
917 | /* | |
918 | * Performance counters information is not part of CRAT but we would like to | |
919 | * put them in the sysfs under topology directory for Thunk to get the data. | |
920 | * This function is called before updating the sysfs. | |
921 | */ | |
922 | static int kfd_add_perf_to_topology(struct kfd_topology_device *kdev) | |
923 | { | |
64d1c3a4 FK |
924 | /* These are the only counters supported so far */ |
925 | return kfd_iommu_add_perf_counters(kdev); | |
f4757347 AL |
926 | } |
927 | ||
520b8fb7 FK |
928 | /* kfd_add_non_crat_information - Add information that is not currently |
929 | * defined in CRAT but is necessary for KFD topology | |
930 | * @dev - topology device to which addition info is added | |
931 | */ | |
932 | static void kfd_add_non_crat_information(struct kfd_topology_device *kdev) | |
933 | { | |
934 | /* Check if CPU only node. */ | |
935 | if (!kdev->gpu) { | |
936 | /* Add system memory information */ | |
937 | dmi_walk(find_system_memory, kdev); | |
938 | } | |
939 | /* TODO: For GPU node, rearrange code from kfd_topology_add_device */ | |
940 | } | |
941 | ||
b441093e HK |
942 | /* kfd_is_acpi_crat_invalid - CRAT from ACPI is valid only for AMD APU devices. |
943 | * Ignore CRAT for all other devices. AMD APU is identified if both CPU | |
944 | * and GPU cores are present. | |
945 | * @device_list - topology device list created by parsing ACPI CRAT table. | |
946 | * @return - TRUE if invalid, FALSE is valid. | |
947 | */ | |
948 | static bool kfd_is_acpi_crat_invalid(struct list_head *device_list) | |
949 | { | |
950 | struct kfd_topology_device *dev; | |
951 | ||
952 | list_for_each_entry(dev, device_list, list) { | |
953 | if (dev->node_props.cpu_cores_count && | |
954 | dev->node_props.simd_count) | |
955 | return false; | |
956 | } | |
957 | pr_info("Ignoring ACPI CRAT on non-APU system\n"); | |
958 | return true; | |
959 | } | |
960 | ||
5b5c4e40 EP |
961 | int kfd_topology_init(void) |
962 | { | |
16b9201c | 963 | void *crat_image = NULL; |
5b5c4e40 EP |
964 | size_t image_size = 0; |
965 | int ret; | |
4f449311 | 966 | struct list_head temp_topology_device_list; |
520b8fb7 FK |
967 | int cpu_only_node = 0; |
968 | struct kfd_topology_device *kdev; | |
969 | int proximity_domain; | |
5b5c4e40 | 970 | |
4f449311 HK |
971 | /* topology_device_list - Master list of all topology devices |
972 | * temp_topology_device_list - temporary list created while parsing CRAT | |
973 | * or VCRAT. Once parsing is complete the contents of list is moved to | |
974 | * topology_device_list | |
5b5c4e40 | 975 | */ |
4f449311 HK |
976 | |
977 | /* Initialize the head for the both the lists */ | |
5b5c4e40 | 978 | INIT_LIST_HEAD(&topology_device_list); |
4f449311 | 979 | INIT_LIST_HEAD(&temp_topology_device_list); |
5b5c4e40 | 980 | init_rwsem(&topology_lock); |
5b5c4e40 EP |
981 | |
982 | memset(&sys_props, 0, sizeof(sys_props)); | |
983 | ||
520b8fb7 FK |
984 | /* Proximity domains in ACPI CRAT tables start counting at |
985 | * 0. The same should be true for virtual CRAT tables created | |
986 | * at this stage. GPUs added later in kfd_topology_add_device | |
987 | * use a counter. | |
988 | */ | |
989 | proximity_domain = 0; | |
990 | ||
5b5c4e40 | 991 | /* |
520b8fb7 | 992 | * Get the CRAT image from the ACPI. If ACPI doesn't have one |
b441093e | 993 | * or if ACPI CRAT is invalid create a virtual CRAT. |
520b8fb7 FK |
994 | * NOTE: The current implementation expects all AMD APUs to have |
995 | * CRAT. If no CRAT is available, it is assumed to be a CPU | |
5b5c4e40 | 996 | */ |
8e05247d HK |
997 | ret = kfd_create_crat_image_acpi(&crat_image, &image_size); |
998 | if (!ret) { | |
4f449311 | 999 | ret = kfd_parse_crat_table(crat_image, |
520b8fb7 FK |
1000 | &temp_topology_device_list, |
1001 | proximity_domain); | |
b441093e HK |
1002 | if (ret || |
1003 | kfd_is_acpi_crat_invalid(&temp_topology_device_list)) { | |
520b8fb7 FK |
1004 | kfd_release_topology_device_list( |
1005 | &temp_topology_device_list); | |
1006 | kfd_destroy_crat_image(crat_image); | |
1007 | crat_image = NULL; | |
1008 | } | |
1009 | } | |
1010 | ||
1011 | if (!crat_image) { | |
1012 | ret = kfd_create_crat_image_virtual(&crat_image, &image_size, | |
1013 | COMPUTE_UNIT_CPU, NULL, | |
1014 | proximity_domain); | |
1015 | cpu_only_node = 1; | |
1016 | if (ret) { | |
1017 | pr_err("Error creating VCRAT table for CPU\n"); | |
1018 | return ret; | |
1019 | } | |
1020 | ||
1021 | ret = kfd_parse_crat_table(crat_image, | |
1022 | &temp_topology_device_list, | |
1023 | proximity_domain); | |
1024 | if (ret) { | |
1025 | pr_err("Error parsing VCRAT table for CPU\n"); | |
5b5c4e40 | 1026 | goto err; |
520b8fb7 | 1027 | } |
5b5c4e40 EP |
1028 | } |
1029 | ||
f4757347 AL |
1030 | kdev = list_first_entry(&temp_topology_device_list, |
1031 | struct kfd_topology_device, list); | |
1032 | kfd_add_perf_to_topology(kdev); | |
1033 | ||
8e05247d | 1034 | down_write(&topology_lock); |
4f449311 HK |
1035 | kfd_topology_update_device_list(&temp_topology_device_list, |
1036 | &topology_device_list); | |
520b8fb7 | 1037 | atomic_set(&topology_crat_proximity_domain, sys_props.num_devices-1); |
8e05247d HK |
1038 | ret = kfd_topology_update_sysfs(); |
1039 | up_write(&topology_lock); | |
1040 | ||
4f449311 HK |
1041 | if (!ret) { |
1042 | sys_props.generation_count++; | |
520b8fb7 FK |
1043 | kfd_update_system_properties(); |
1044 | kfd_debug_print_topology(); | |
8e05247d | 1045 | pr_info("Finished initializing topology\n"); |
4f449311 | 1046 | } else |
8e05247d HK |
1047 | pr_err("Failed to update topology in sysfs ret=%d\n", ret); |
1048 | ||
520b8fb7 FK |
1049 | /* For nodes with GPU, this information gets added |
1050 | * when GPU is detected (kfd_topology_add_device). | |
1051 | */ | |
1052 | if (cpu_only_node) { | |
1053 | /* Add additional information to CPU only node created above */ | |
1054 | down_write(&topology_lock); | |
1055 | kdev = list_first_entry(&topology_device_list, | |
1056 | struct kfd_topology_device, list); | |
1057 | up_write(&topology_lock); | |
1058 | kfd_add_non_crat_information(kdev); | |
1059 | } | |
1060 | ||
5b5c4e40 | 1061 | err: |
8e05247d | 1062 | kfd_destroy_crat_image(crat_image); |
5b5c4e40 EP |
1063 | return ret; |
1064 | } | |
1065 | ||
1066 | void kfd_topology_shutdown(void) | |
1067 | { | |
4f449311 | 1068 | down_write(&topology_lock); |
5b5c4e40 EP |
1069 | kfd_topology_release_sysfs(); |
1070 | kfd_release_live_view(); | |
4f449311 | 1071 | up_write(&topology_lock); |
5b5c4e40 EP |
1072 | } |
1073 | ||
5b5c4e40 EP |
1074 | static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu) |
1075 | { | |
1076 | uint32_t hashout; | |
1077 | uint32_t buf[7]; | |
585f0e6c | 1078 | uint64_t local_mem_size; |
5b5c4e40 | 1079 | int i; |
0504cccf | 1080 | struct kfd_local_mem_info local_mem_info; |
5b5c4e40 EP |
1081 | |
1082 | if (!gpu) | |
1083 | return 0; | |
1084 | ||
7cd52c91 | 1085 | amdgpu_amdkfd_get_local_mem_info(gpu->kgd, &local_mem_info); |
0504cccf HK |
1086 | |
1087 | local_mem_size = local_mem_info.local_mem_size_private + | |
1088 | local_mem_info.local_mem_size_public; | |
585f0e6c | 1089 | |
5b5c4e40 | 1090 | buf[0] = gpu->pdev->devfn; |
46096058 AL |
1091 | buf[1] = gpu->pdev->subsystem_vendor | |
1092 | (gpu->pdev->subsystem_device << 16); | |
1093 | buf[2] = pci_domain_nr(gpu->pdev->bus); | |
5b5c4e40 EP |
1094 | buf[3] = gpu->pdev->device; |
1095 | buf[4] = gpu->pdev->bus->number; | |
585f0e6c EC |
1096 | buf[5] = lower_32_bits(local_mem_size); |
1097 | buf[6] = upper_32_bits(local_mem_size); | |
5b5c4e40 EP |
1098 | |
1099 | for (i = 0, hashout = 0; i < 7; i++) | |
1100 | hashout ^= hash_32(buf[i], KFD_GPU_ID_HASH_WIDTH); | |
1101 | ||
1102 | return hashout; | |
1103 | } | |
3a87177e HK |
1104 | /* kfd_assign_gpu - Attach @gpu to the correct kfd topology device. If |
1105 | * the GPU device is not already present in the topology device | |
1106 | * list then return NULL. This means a new topology device has to | |
1107 | * be created for this GPU. | |
3a87177e | 1108 | */ |
5b5c4e40 EP |
1109 | static struct kfd_topology_device *kfd_assign_gpu(struct kfd_dev *gpu) |
1110 | { | |
1111 | struct kfd_topology_device *dev; | |
16b9201c | 1112 | struct kfd_topology_device *out_dev = NULL; |
171bc67e HK |
1113 | struct kfd_mem_properties *mem; |
1114 | struct kfd_cache_properties *cache; | |
1115 | struct kfd_iolink_properties *iolink; | |
5b5c4e40 | 1116 | |
3a87177e | 1117 | down_write(&topology_lock); |
b8fe0524 FK |
1118 | list_for_each_entry(dev, &topology_device_list, list) { |
1119 | /* Discrete GPUs need their own topology device list | |
1120 | * entries. Don't assign them to CPU/APU nodes. | |
1121 | */ | |
1122 | if (!gpu->device_info->needs_iommu_device && | |
1123 | dev->node_props.cpu_cores_count) | |
1124 | continue; | |
1125 | ||
4eacc26b | 1126 | if (!dev->gpu && (dev->node_props.simd_count > 0)) { |
5b5c4e40 EP |
1127 | dev->gpu = gpu; |
1128 | out_dev = dev; | |
171bc67e HK |
1129 | |
1130 | list_for_each_entry(mem, &dev->mem_props, list) | |
1131 | mem->gpu = dev->gpu; | |
1132 | list_for_each_entry(cache, &dev->cache_props, list) | |
1133 | cache->gpu = dev->gpu; | |
1134 | list_for_each_entry(iolink, &dev->io_link_props, list) | |
1135 | iolink->gpu = dev->gpu; | |
5b5c4e40 EP |
1136 | break; |
1137 | } | |
b8fe0524 | 1138 | } |
3a87177e | 1139 | up_write(&topology_lock); |
5b5c4e40 EP |
1140 | return out_dev; |
1141 | } | |
1142 | ||
1143 | static void kfd_notify_gpu_change(uint32_t gpu_id, int arrival) | |
1144 | { | |
1145 | /* | |
1146 | * TODO: Generate an event for thunk about the arrival/removal | |
1147 | * of the GPU | |
1148 | */ | |
1149 | } | |
1150 | ||
3a87177e HK |
1151 | /* kfd_fill_mem_clk_max_info - Since CRAT doesn't have memory clock info, |
1152 | * patch this after CRAT parsing. | |
1153 | */ | |
1154 | static void kfd_fill_mem_clk_max_info(struct kfd_topology_device *dev) | |
1155 | { | |
1156 | struct kfd_mem_properties *mem; | |
1157 | struct kfd_local_mem_info local_mem_info; | |
1158 | ||
1159 | if (!dev) | |
1160 | return; | |
1161 | ||
1162 | /* Currently, amdgpu driver (amdgpu_mc) deals only with GPUs with | |
1163 | * single bank of VRAM local memory. | |
1164 | * for dGPUs - VCRAT reports only one bank of Local Memory | |
1165 | * for APUs - If CRAT from ACPI reports more than one bank, then | |
1166 | * all the banks will report the same mem_clk_max information | |
1167 | */ | |
7cd52c91 | 1168 | amdgpu_amdkfd_get_local_mem_info(dev->gpu->kgd, &local_mem_info); |
3a87177e HK |
1169 | |
1170 | list_for_each_entry(mem, &dev->mem_props, list) | |
1171 | mem->mem_clk_max = local_mem_info.mem_clk_max; | |
1172 | } | |
1173 | ||
1174 | static void kfd_fill_iolink_non_crat_info(struct kfd_topology_device *dev) | |
1175 | { | |
d35f00d8 EH |
1176 | struct kfd_iolink_properties *link, *cpu_link; |
1177 | struct kfd_topology_device *cpu_dev; | |
1178 | uint32_t cap; | |
1179 | uint32_t cpu_flag = CRAT_IOLINK_FLAGS_ENABLED; | |
1180 | uint32_t flag = CRAT_IOLINK_FLAGS_ENABLED; | |
3a87177e HK |
1181 | |
1182 | if (!dev || !dev->gpu) | |
1183 | return; | |
1184 | ||
d35f00d8 EH |
1185 | pcie_capability_read_dword(dev->gpu->pdev, |
1186 | PCI_EXP_DEVCAP2, &cap); | |
1187 | ||
1188 | if (!(cap & (PCI_EXP_DEVCAP2_ATOMIC_COMP32 | | |
1189 | PCI_EXP_DEVCAP2_ATOMIC_COMP64))) | |
1190 | cpu_flag |= CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT | | |
1191 | CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT; | |
1192 | ||
1193 | if (!dev->gpu->pci_atomic_requested || | |
1194 | dev->gpu->device_info->asic_family == CHIP_HAWAII) | |
1195 | flag |= CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT | | |
1196 | CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT; | |
1197 | ||
1198 | /* GPU only creates direct links so apply flags setting to all */ | |
1199 | list_for_each_entry(link, &dev->io_link_props, list) { | |
1200 | link->flags = flag; | |
1201 | cpu_dev = kfd_topology_device_by_proximity_domain( | |
1202 | link->node_to); | |
1203 | if (cpu_dev) { | |
1204 | list_for_each_entry(cpu_link, | |
1205 | &cpu_dev->io_link_props, list) | |
1206 | if (cpu_link->node_to == link->node_from) | |
1207 | cpu_link->flags = cpu_flag; | |
1208 | } | |
1209 | } | |
3a87177e HK |
1210 | } |
1211 | ||
5b5c4e40 EP |
1212 | int kfd_topology_add_device(struct kfd_dev *gpu) |
1213 | { | |
1214 | uint32_t gpu_id; | |
1215 | struct kfd_topology_device *dev; | |
f7ce2fad | 1216 | struct kfd_cu_info cu_info; |
4f449311 HK |
1217 | int res = 0; |
1218 | struct list_head temp_topology_device_list; | |
3a87177e HK |
1219 | void *crat_image = NULL; |
1220 | size_t image_size = 0; | |
1221 | int proximity_domain; | |
0dee45a2 | 1222 | struct amdgpu_ras *ctx; |
4f449311 HK |
1223 | |
1224 | INIT_LIST_HEAD(&temp_topology_device_list); | |
5b5c4e40 | 1225 | |
5b5c4e40 EP |
1226 | gpu_id = kfd_generate_gpu_id(gpu); |
1227 | ||
79775b62 | 1228 | pr_debug("Adding new GPU (ID: 0x%x) to topology\n", gpu_id); |
5b5c4e40 | 1229 | |
3a87177e HK |
1230 | proximity_domain = atomic_inc_return(&topology_crat_proximity_domain); |
1231 | ||
1232 | /* Check to see if this gpu device exists in the topology_device_list. | |
1233 | * If so, assign the gpu to that device, | |
1234 | * else create a Virtual CRAT for this gpu device and then parse that | |
1235 | * CRAT to create a new topology device. Once created assign the gpu to | |
1236 | * that topology device | |
5b5c4e40 EP |
1237 | */ |
1238 | dev = kfd_assign_gpu(gpu); | |
1239 | if (!dev) { | |
3a87177e HK |
1240 | res = kfd_create_crat_image_virtual(&crat_image, &image_size, |
1241 | COMPUTE_UNIT_GPU, gpu, | |
1242 | proximity_domain); | |
1243 | if (res) { | |
1244 | pr_err("Error creating VCRAT for GPU (ID: 0x%x)\n", | |
1245 | gpu_id); | |
1246 | return res; | |
1247 | } | |
1248 | res = kfd_parse_crat_table(crat_image, | |
1249 | &temp_topology_device_list, | |
1250 | proximity_domain); | |
1251 | if (res) { | |
1252 | pr_err("Error parsing VCRAT for GPU (ID: 0x%x)\n", | |
1253 | gpu_id); | |
5b5c4e40 EP |
1254 | goto err; |
1255 | } | |
4f449311 | 1256 | |
4f449311 HK |
1257 | down_write(&topology_lock); |
1258 | kfd_topology_update_device_list(&temp_topology_device_list, | |
1259 | &topology_device_list); | |
1260 | ||
8eabaf54 KR |
1261 | /* Update the SYSFS tree, since we added another topology |
1262 | * device | |
5b5c4e40 | 1263 | */ |
3a87177e | 1264 | res = kfd_topology_update_sysfs(); |
4f449311 HK |
1265 | up_write(&topology_lock); |
1266 | ||
3a87177e HK |
1267 | if (!res) |
1268 | sys_props.generation_count++; | |
1269 | else | |
1270 | pr_err("Failed to update GPU (ID: 0x%x) to sysfs topology. res=%d\n", | |
1271 | gpu_id, res); | |
1272 | dev = kfd_assign_gpu(gpu); | |
1273 | if (WARN_ON(!dev)) { | |
1274 | res = -ENODEV; | |
1275 | goto err; | |
1276 | } | |
5b5c4e40 EP |
1277 | } |
1278 | ||
1279 | dev->gpu_id = gpu_id; | |
1280 | gpu->id = gpu_id; | |
3a87177e HK |
1281 | |
1282 | /* TODO: Move the following lines to function | |
1283 | * kfd_add_non_crat_information | |
1284 | */ | |
1285 | ||
1286 | /* Fill-in additional information that is not available in CRAT but | |
1287 | * needed for the topology | |
1288 | */ | |
1289 | ||
7cd52c91 | 1290 | amdgpu_amdkfd_get_cu_info(dev->gpu->kgd, &cu_info); |
c181159a YZ |
1291 | |
1292 | strncpy(dev->node_props.name, gpu->device_info->asic_name, | |
1293 | KFD_TOPOLOGY_PUBLIC_NAME_SIZE); | |
1294 | ||
3a87177e HK |
1295 | dev->node_props.simd_arrays_per_engine = |
1296 | cu_info.num_shader_arrays_per_engine; | |
1297 | ||
5b5c4e40 EP |
1298 | dev->node_props.vendor_id = gpu->pdev->vendor; |
1299 | dev->node_props.device_id = gpu->pdev->device; | |
babe2ef3 | 1300 | dev->node_props.location_id = pci_dev_id(gpu->pdev); |
3a87177e | 1301 | dev->node_props.max_engine_clk_fcompute = |
7cd52c91 | 1302 | amdgpu_amdkfd_get_max_engine_clock_in_mhz(dev->gpu->kgd); |
3a87177e HK |
1303 | dev->node_props.max_engine_clk_ccompute = |
1304 | cpufreq_quick_get_max(0) / 1000; | |
7c9b7171 OZ |
1305 | dev->node_props.drm_render_minor = |
1306 | gpu->shared_resources.drm_render_minor; | |
3a87177e | 1307 | |
0c1690e3 | 1308 | dev->node_props.hive_id = gpu->hive_id; |
14568cf6 OZ |
1309 | dev->node_props.num_sdma_engines = gpu->device_info->num_sdma_engines; |
1310 | dev->node_props.num_sdma_xgmi_engines = | |
1311 | gpu->device_info->num_xgmi_sdma_engines; | |
29e76462 OZ |
1312 | dev->node_props.num_gws = (hws_gws_support && |
1313 | dev->gpu->dqm->sched_policy != KFD_SCHED_POLICY_NO_HWS) ? | |
1314 | amdgpu_amdkfd_get_num_gws(dev->gpu->kgd) : 0; | |
0c1690e3 | 1315 | |
3a87177e HK |
1316 | kfd_fill_mem_clk_max_info(dev); |
1317 | kfd_fill_iolink_non_crat_info(dev); | |
1318 | ||
1319 | switch (dev->gpu->device_info->asic_family) { | |
1320 | case CHIP_KAVERI: | |
1321 | case CHIP_HAWAII: | |
1322 | case CHIP_TONGA: | |
1323 | dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_PRE_1_0 << | |
1324 | HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & | |
1325 | HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); | |
1326 | break; | |
1327 | case CHIP_CARRIZO: | |
1328 | case CHIP_FIJI: | |
1329 | case CHIP_POLARIS10: | |
1330 | case CHIP_POLARIS11: | |
846a44d7 | 1331 | case CHIP_POLARIS12: |
ed81cd6e | 1332 | case CHIP_VEGAM: |
42aa8793 | 1333 | pr_debug("Adding doorbell packet type capability\n"); |
3a87177e HK |
1334 | dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_1_0 << |
1335 | HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & | |
1336 | HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); | |
1337 | break; | |
389056e5 | 1338 | case CHIP_VEGA10: |
846a44d7 | 1339 | case CHIP_VEGA12: |
22a3a294 | 1340 | case CHIP_VEGA20: |
389056e5 | 1341 | case CHIP_RAVEN: |
f5d843d4 | 1342 | case CHIP_RENOIR: |
49adcf8a | 1343 | case CHIP_ARCTURUS: |
14328aa5 | 1344 | case CHIP_NAVI10: |
0e94b564 | 1345 | case CHIP_NAVI12: |
8099ae40 | 1346 | case CHIP_NAVI14: |
389056e5 FK |
1347 | dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_2_0 << |
1348 | HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & | |
1349 | HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); | |
1350 | break; | |
3a87177e HK |
1351 | default: |
1352 | WARN(1, "Unexpected ASIC family %u", | |
1353 | dev->gpu->device_info->asic_family); | |
7639a8c4 BG |
1354 | } |
1355 | ||
1ae99eab OZ |
1356 | /* |
1357 | * Overwrite ATS capability according to needs_iommu_device to fix | |
1358 | * potential missing corresponding bit in CRAT of BIOS. | |
1359 | */ | |
1360 | if (dev->gpu->device_info->needs_iommu_device) | |
1361 | dev->node_props.capability |= HSA_CAP_ATS_PRESENT; | |
1362 | else | |
1363 | dev->node_props.capability &= ~HSA_CAP_ATS_PRESENT; | |
1364 | ||
3a87177e HK |
1365 | /* Fix errors in CZ CRAT. |
1366 | * simd_count: Carrizo CRAT reports wrong simd_count, probably | |
1367 | * because it doesn't consider masked out CUs | |
70f372bf | 1368 | * max_waves_per_simd: Carrizo reports wrong max_waves_per_simd |
3a87177e | 1369 | */ |
70f372bf | 1370 | if (dev->gpu->device_info->asic_family == CHIP_CARRIZO) { |
3a87177e HK |
1371 | dev->node_props.simd_count = |
1372 | cu_info.simd_per_cu * cu_info.cu_active_number; | |
70f372bf | 1373 | dev->node_props.max_waves_per_simd = 10; |
70f372bf | 1374 | } |
3a87177e | 1375 | |
0dee45a2 EH |
1376 | ctx = amdgpu_ras_get_context((struct amdgpu_device *)(dev->gpu->kgd)); |
1377 | if (ctx) { | |
1378 | /* kfd only concerns sram ecc on GFX/SDMA and HBM ecc on UMC */ | |
1379 | dev->node_props.capability |= | |
1380 | (((ctx->features & BIT(AMDGPU_RAS_BLOCK__SDMA)) != 0) || | |
1381 | ((ctx->features & BIT(AMDGPU_RAS_BLOCK__GFX)) != 0)) ? | |
1382 | HSA_CAP_SRAM_EDCSUPPORTED : 0; | |
1383 | dev->node_props.capability |= ((ctx->features & BIT(AMDGPU_RAS_BLOCK__UMC)) != 0) ? | |
1384 | HSA_CAP_MEM_EDCSUPPORTED : 0; | |
1385 | ||
1386 | dev->node_props.capability |= (ctx->features != 0) ? | |
1387 | HSA_CAP_RASEVENTNOTIFY : 0; | |
1388 | } | |
1389 | ||
3a87177e HK |
1390 | kfd_debug_print_topology(); |
1391 | ||
4f449311 | 1392 | if (!res) |
5b5c4e40 | 1393 | kfd_notify_gpu_change(gpu_id, 1); |
4f449311 | 1394 | err: |
3a87177e | 1395 | kfd_destroy_crat_image(crat_image); |
5b5c4e40 EP |
1396 | return res; |
1397 | } | |
1398 | ||
1399 | int kfd_topology_remove_device(struct kfd_dev *gpu) | |
1400 | { | |
4f449311 | 1401 | struct kfd_topology_device *dev, *tmp; |
5b5c4e40 EP |
1402 | uint32_t gpu_id; |
1403 | int res = -ENODEV; | |
1404 | ||
5b5c4e40 EP |
1405 | down_write(&topology_lock); |
1406 | ||
4f449311 | 1407 | list_for_each_entry_safe(dev, tmp, &topology_device_list, list) |
5b5c4e40 EP |
1408 | if (dev->gpu == gpu) { |
1409 | gpu_id = dev->gpu_id; | |
1410 | kfd_remove_sysfs_node_entry(dev); | |
1411 | kfd_release_topology_device(dev); | |
4f449311 | 1412 | sys_props.num_devices--; |
5b5c4e40 EP |
1413 | res = 0; |
1414 | if (kfd_topology_update_sysfs() < 0) | |
1415 | kfd_topology_release_sysfs(); | |
1416 | break; | |
1417 | } | |
1418 | ||
1419 | up_write(&topology_lock); | |
1420 | ||
174de876 | 1421 | if (!res) |
5b5c4e40 EP |
1422 | kfd_notify_gpu_change(gpu_id, 0); |
1423 | ||
1424 | return res; | |
1425 | } | |
1426 | ||
6d82eb0e HK |
1427 | /* kfd_topology_enum_kfd_devices - Enumerate through all devices in KFD |
1428 | * topology. If GPU device is found @idx, then valid kfd_dev pointer is | |
1429 | * returned through @kdev | |
1430 | * Return - 0: On success (@kdev will be NULL for non GPU nodes) | |
1431 | * -1: If end of list | |
5b5c4e40 | 1432 | */ |
6d82eb0e | 1433 | int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev) |
5b5c4e40 EP |
1434 | { |
1435 | ||
1436 | struct kfd_topology_device *top_dev; | |
5b5c4e40 EP |
1437 | uint8_t device_idx = 0; |
1438 | ||
6d82eb0e | 1439 | *kdev = NULL; |
5b5c4e40 EP |
1440 | down_read(&topology_lock); |
1441 | ||
1442 | list_for_each_entry(top_dev, &topology_device_list, list) { | |
1443 | if (device_idx == idx) { | |
6d82eb0e HK |
1444 | *kdev = top_dev->gpu; |
1445 | up_read(&topology_lock); | |
1446 | return 0; | |
5b5c4e40 EP |
1447 | } |
1448 | ||
1449 | device_idx++; | |
1450 | } | |
1451 | ||
1452 | up_read(&topology_lock); | |
1453 | ||
6d82eb0e | 1454 | return -1; |
5b5c4e40 EP |
1455 | |
1456 | } | |
851a645e | 1457 | |
520b8fb7 FK |
1458 | static int kfd_cpumask_to_apic_id(const struct cpumask *cpumask) |
1459 | { | |
520b8fb7 FK |
1460 | int first_cpu_of_numa_node; |
1461 | ||
1462 | if (!cpumask || cpumask == cpu_none_mask) | |
1463 | return -1; | |
1464 | first_cpu_of_numa_node = cpumask_first(cpumask); | |
1465 | if (first_cpu_of_numa_node >= nr_cpu_ids) | |
1466 | return -1; | |
df1dd4f4 FK |
1467 | #ifdef CONFIG_X86_64 |
1468 | return cpu_data(first_cpu_of_numa_node).apicid; | |
1469 | #else | |
1470 | return first_cpu_of_numa_node; | |
1471 | #endif | |
520b8fb7 FK |
1472 | } |
1473 | ||
1474 | /* kfd_numa_node_to_apic_id - Returns the APIC ID of the first logical processor | |
1475 | * of the given NUMA node (numa_node_id) | |
1476 | * Return -1 on failure | |
1477 | */ | |
1478 | int kfd_numa_node_to_apic_id(int numa_node_id) | |
1479 | { | |
1480 | if (numa_node_id == -1) { | |
1481 | pr_warn("Invalid NUMA Node. Use online CPU mask\n"); | |
1482 | return kfd_cpumask_to_apic_id(cpu_online_mask); | |
1483 | } | |
1484 | return kfd_cpumask_to_apic_id(cpumask_of_node(numa_node_id)); | |
1485 | } | |
1486 | ||
851a645e FK |
1487 | #if defined(CONFIG_DEBUG_FS) |
1488 | ||
1489 | int kfd_debugfs_hqds_by_device(struct seq_file *m, void *data) | |
1490 | { | |
1491 | struct kfd_topology_device *dev; | |
1492 | unsigned int i = 0; | |
1493 | int r = 0; | |
1494 | ||
1495 | down_read(&topology_lock); | |
1496 | ||
1497 | list_for_each_entry(dev, &topology_device_list, list) { | |
1498 | if (!dev->gpu) { | |
1499 | i++; | |
1500 | continue; | |
1501 | } | |
1502 | ||
1503 | seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id); | |
1504 | r = dqm_debugfs_hqds(m, dev->gpu->dqm); | |
1505 | if (r) | |
1506 | break; | |
1507 | } | |
1508 | ||
1509 | up_read(&topology_lock); | |
1510 | ||
1511 | return r; | |
1512 | } | |
1513 | ||
1514 | int kfd_debugfs_rls_by_device(struct seq_file *m, void *data) | |
1515 | { | |
1516 | struct kfd_topology_device *dev; | |
1517 | unsigned int i = 0; | |
1518 | int r = 0; | |
1519 | ||
1520 | down_read(&topology_lock); | |
1521 | ||
1522 | list_for_each_entry(dev, &topology_device_list, list) { | |
1523 | if (!dev->gpu) { | |
1524 | i++; | |
1525 | continue; | |
1526 | } | |
1527 | ||
1528 | seq_printf(m, "Node %u, gpu_id %x:\n", i++, dev->gpu->id); | |
1529 | r = pm_debugfs_runlist(m, &dev->gpu->dqm->packets); | |
1530 | if (r) | |
1531 | break; | |
1532 | } | |
1533 | ||
1534 | up_read(&topology_lock); | |
1535 | ||
1536 | return r; | |
1537 | } | |
1538 | ||
1539 | #endif |