xref: /linux/drivers/gpu/drm/xe/xe_gt_sriov_pf_debugfs.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2023-2024 Intel Corporation
4  */
5 
6 #include <linux/debugfs.h>
7 
8 #include <drm/drm_print.h>
9 #include <drm/drm_debugfs.h>
10 
11 #include "xe_bo.h"
12 #include "xe_debugfs.h"
13 #include "xe_device.h"
14 #include "xe_gt.h"
15 #include "xe_gt_debugfs.h"
16 #include "xe_gt_sriov_pf_config.h"
17 #include "xe_gt_sriov_pf_control.h"
18 #include "xe_gt_sriov_pf_debugfs.h"
19 #include "xe_gt_sriov_pf_helpers.h"
20 #include "xe_gt_sriov_pf_migration.h"
21 #include "xe_gt_sriov_pf_monitor.h"
22 #include "xe_gt_sriov_pf_policy.h"
23 #include "xe_gt_sriov_pf_service.h"
24 #include "xe_pm.h"
25 
26 /*
27  *      /sys/kernel/debug/dri/0/
28  *      ├── gt0		# d_inode->i_private = gt
29  *      │   ├── pf	# d_inode->i_private = gt
30  *      │   ├── vf1	# d_inode->i_private = VFID(1)
31  *      :   :
32  *      │   ├── vfN	# d_inode->i_private = VFID(N)
33  */
34 
35 static void *extract_priv(struct dentry *d)
36 {
37 	return d->d_inode->i_private;
38 }
39 
40 static struct xe_gt *extract_gt(struct dentry *d)
41 {
42 	return extract_priv(d->d_parent);
43 }
44 
45 static unsigned int extract_vfid(struct dentry *d)
46 {
47 	return extract_priv(d) == extract_gt(d) ? PFID : (uintptr_t)extract_priv(d);
48 }
49 
50 /*
51  *      /sys/kernel/debug/dri/0/
52  *      ├── gt0
53  *      │   ├── pf
54  *      │   │   ├── ggtt_available
55  *      │   │   ├── ggtt_provisioned
56  *      │   │   ├── contexts_provisioned
57  *      │   │   ├── doorbells_provisioned
58  *      │   │   ├── runtime_registers
59  *      │   │   ├── negotiated_versions
60  *      │   │   ├── adverse_events
61  */
62 
63 static const struct drm_info_list pf_info[] = {
64 	{
65 		"ggtt_available",
66 		.show = xe_gt_debugfs_simple_show,
67 		.data = xe_gt_sriov_pf_config_print_available_ggtt,
68 	},
69 	{
70 		"ggtt_provisioned",
71 		.show = xe_gt_debugfs_simple_show,
72 		.data = xe_gt_sriov_pf_config_print_ggtt,
73 	},
74 	{
75 		"contexts_provisioned",
76 		.show = xe_gt_debugfs_simple_show,
77 		.data = xe_gt_sriov_pf_config_print_ctxs,
78 	},
79 	{
80 		"doorbells_provisioned",
81 		.show = xe_gt_debugfs_simple_show,
82 		.data = xe_gt_sriov_pf_config_print_dbs,
83 	},
84 	{
85 		"lmem_provisioned",
86 		.show = xe_gt_debugfs_simple_show,
87 		.data = xe_gt_sriov_pf_config_print_lmem,
88 	},
89 	{
90 		"runtime_registers",
91 		.show = xe_gt_debugfs_simple_show,
92 		.data = xe_gt_sriov_pf_service_print_runtime,
93 	},
94 	{
95 		"negotiated_versions",
96 		.show = xe_gt_debugfs_simple_show,
97 		.data = xe_gt_sriov_pf_service_print_version,
98 	},
99 	{
100 		"adverse_events",
101 		.show = xe_gt_debugfs_simple_show,
102 		.data = xe_gt_sriov_pf_monitor_print_events,
103 	},
104 };
105 
106 /*
107  *      /sys/kernel/debug/dri/0/
108  *      ├── gt0
109  *      │   ├── pf
110  *      │   │   ├── reset_engine
111  *      │   │   ├── sample_period
112  *      │   │   ├── sched_if_idle
113  */
114 
115 #define DEFINE_SRIOV_GT_POLICY_DEBUGFS_ATTRIBUTE(POLICY, TYPE, FORMAT)		\
116 										\
117 static int POLICY##_set(void *data, u64 val)					\
118 {										\
119 	struct xe_gt *gt = extract_gt(data);					\
120 	struct xe_device *xe = gt_to_xe(gt);					\
121 	int err;								\
122 										\
123 	if (val > (TYPE)~0ull)							\
124 		return -EOVERFLOW;						\
125 										\
126 	xe_pm_runtime_get(xe);							\
127 	err = xe_gt_sriov_pf_policy_set_##POLICY(gt, val);			\
128 	xe_pm_runtime_put(xe);							\
129 										\
130 	return err;								\
131 }										\
132 										\
133 static int POLICY##_get(void *data, u64 *val)					\
134 {										\
135 	struct xe_gt *gt = extract_gt(data);					\
136 										\
137 	*val = xe_gt_sriov_pf_policy_get_##POLICY(gt);				\
138 	return 0;								\
139 }										\
140 										\
141 DEFINE_DEBUGFS_ATTRIBUTE(POLICY##_fops, POLICY##_get, POLICY##_set, FORMAT)
142 
143 DEFINE_SRIOV_GT_POLICY_DEBUGFS_ATTRIBUTE(reset_engine, bool, "%llu\n");
144 DEFINE_SRIOV_GT_POLICY_DEBUGFS_ATTRIBUTE(sched_if_idle, bool, "%llu\n");
145 DEFINE_SRIOV_GT_POLICY_DEBUGFS_ATTRIBUTE(sample_period, u32, "%llu\n");
146 
147 static void pf_add_policy_attrs(struct xe_gt *gt, struct dentry *parent)
148 {
149 	xe_gt_assert(gt, gt == extract_gt(parent));
150 	xe_gt_assert(gt, PFID == extract_vfid(parent));
151 
152 	debugfs_create_file_unsafe("reset_engine", 0644, parent, parent, &reset_engine_fops);
153 	debugfs_create_file_unsafe("sched_if_idle", 0644, parent, parent, &sched_if_idle_fops);
154 	debugfs_create_file_unsafe("sample_period_ms", 0644, parent, parent, &sample_period_fops);
155 }
156 
157 /*
158  *      /sys/kernel/debug/dri/0/
159  *      ├── gt0
160  *      │   ├── pf
161  *      │   │   ├── ggtt_spare
162  *      │   │   ├── lmem_spare
163  *      │   │   ├── doorbells_spare
164  *      │   │   ├── contexts_spare
165  *      │   │   ├── exec_quantum_ms
166  *      │   │   ├── preempt_timeout_us
167  *      │   ├── vf1
168  *      │   │   ├── ggtt_quota
169  *      │   │   ├── lmem_quota
170  *      │   │   ├── doorbells_quota
171  *      │   │   ├── contexts_quota
172  *      │   │   ├── exec_quantum_ms
173  *      │   │   ├── preempt_timeout_us
174  */
175 
176 #define DEFINE_SRIOV_GT_CONFIG_DEBUGFS_ATTRIBUTE(CONFIG, TYPE, FORMAT)		\
177 										\
178 static int CONFIG##_set(void *data, u64 val)					\
179 {										\
180 	struct xe_gt *gt = extract_gt(data);					\
181 	unsigned int vfid = extract_vfid(data);					\
182 	struct xe_device *xe = gt_to_xe(gt);					\
183 	int err;								\
184 										\
185 	if (val > (TYPE)~0ull)							\
186 		return -EOVERFLOW;						\
187 										\
188 	xe_pm_runtime_get(xe);							\
189 	err = xe_gt_sriov_pf_config_set_##CONFIG(gt, vfid, val);		\
190 	xe_pm_runtime_put(xe);							\
191 										\
192 	return err;								\
193 }										\
194 										\
195 static int CONFIG##_get(void *data, u64 *val)					\
196 {										\
197 	struct xe_gt *gt = extract_gt(data);					\
198 	unsigned int vfid = extract_vfid(data);					\
199 										\
200 	*val = xe_gt_sriov_pf_config_get_##CONFIG(gt, vfid);			\
201 	return 0;								\
202 }										\
203 										\
204 DEFINE_DEBUGFS_ATTRIBUTE(CONFIG##_fops, CONFIG##_get, CONFIG##_set, FORMAT)
205 
206 DEFINE_SRIOV_GT_CONFIG_DEBUGFS_ATTRIBUTE(ggtt, u64, "%llu\n");
207 DEFINE_SRIOV_GT_CONFIG_DEBUGFS_ATTRIBUTE(lmem, u64, "%llu\n");
208 DEFINE_SRIOV_GT_CONFIG_DEBUGFS_ATTRIBUTE(ctxs, u32, "%llu\n");
209 DEFINE_SRIOV_GT_CONFIG_DEBUGFS_ATTRIBUTE(dbs, u32, "%llu\n");
210 DEFINE_SRIOV_GT_CONFIG_DEBUGFS_ATTRIBUTE(exec_quantum, u32, "%llu\n");
211 DEFINE_SRIOV_GT_CONFIG_DEBUGFS_ATTRIBUTE(preempt_timeout, u32, "%llu\n");
212 
213 /*
214  *      /sys/kernel/debug/dri/0/
215  *      ├── gt0
216  *      │   ├── pf
217  *      │   │   ├── threshold_cat_error_count
218  *      │   │   ├── threshold_doorbell_time_us
219  *      │   │   ├── threshold_engine_reset_count
220  *      │   │   ├── threshold_guc_time_us
221  *      │   │   ├── threshold_irq_time_us
222  *      │   │   ├── threshold_page_fault_count
223  *      │   ├── vf1
224  *      │   │   ├── threshold_cat_error_count
225  *      │   │   ├── threshold_doorbell_time_us
226  *      │   │   ├── threshold_engine_reset_count
227  *      │   │   ├── threshold_guc_time_us
228  *      │   │   ├── threshold_irq_time_us
229  *      │   │   ├── threshold_page_fault_count
230  */
231 
232 static int set_threshold(void *data, u64 val, enum xe_guc_klv_threshold_index index)
233 {
234 	struct xe_gt *gt = extract_gt(data);
235 	unsigned int vfid = extract_vfid(data);
236 	struct xe_device *xe = gt_to_xe(gt);
237 	int err;
238 
239 	if (val > (u32)~0ull)
240 		return -EOVERFLOW;
241 
242 	xe_pm_runtime_get(xe);
243 	err = xe_gt_sriov_pf_config_set_threshold(gt, vfid, index, val);
244 	xe_pm_runtime_put(xe);
245 
246 	return err;
247 }
248 
249 static int get_threshold(void *data, u64 *val, enum xe_guc_klv_threshold_index index)
250 {
251 	struct xe_gt *gt = extract_gt(data);
252 	unsigned int vfid = extract_vfid(data);
253 
254 	*val = xe_gt_sriov_pf_config_get_threshold(gt, vfid, index);
255 	return 0;
256 }
257 
258 #define DEFINE_SRIOV_GT_THRESHOLD_DEBUGFS_ATTRIBUTE(THRESHOLD, INDEX)		\
259 										\
260 static int THRESHOLD##_set(void *data, u64 val)					\
261 {										\
262 	return set_threshold(data, val, INDEX);					\
263 }										\
264 										\
265 static int THRESHOLD##_get(void *data, u64 *val)				\
266 {										\
267 	return get_threshold(data, val, INDEX);					\
268 }										\
269 										\
270 DEFINE_DEBUGFS_ATTRIBUTE(THRESHOLD##_fops, THRESHOLD##_get, THRESHOLD##_set, "%llu\n")
271 
272 /* generate all threshold attributes */
273 #define define_threshold_attribute(TAG, NAME, ...) \
274 	DEFINE_SRIOV_GT_THRESHOLD_DEBUGFS_ATTRIBUTE(NAME, MAKE_XE_GUC_KLV_THRESHOLD_INDEX(TAG));
275 MAKE_XE_GUC_KLV_THRESHOLDS_SET(define_threshold_attribute)
276 #undef define_threshold_attribute
277 
278 static void pf_add_config_attrs(struct xe_gt *gt, struct dentry *parent, unsigned int vfid)
279 {
280 	xe_gt_assert(gt, gt == extract_gt(parent));
281 	xe_gt_assert(gt, vfid == extract_vfid(parent));
282 
283 	if (!xe_gt_is_media_type(gt)) {
284 		debugfs_create_file_unsafe(vfid ? "ggtt_quota" : "ggtt_spare",
285 					   0644, parent, parent, &ggtt_fops);
286 		if (IS_DGFX(gt_to_xe(gt)))
287 			debugfs_create_file_unsafe(vfid ? "lmem_quota" : "lmem_spare",
288 						   0644, parent, parent, &lmem_fops);
289 	}
290 	debugfs_create_file_unsafe(vfid ? "doorbells_quota" : "doorbells_spare",
291 				   0644, parent, parent, &dbs_fops);
292 	debugfs_create_file_unsafe(vfid ? "contexts_quota" : "contexts_spare",
293 				   0644, parent, parent, &ctxs_fops);
294 	debugfs_create_file_unsafe("exec_quantum_ms", 0644, parent, parent,
295 				   &exec_quantum_fops);
296 	debugfs_create_file_unsafe("preempt_timeout_us", 0644, parent, parent,
297 				   &preempt_timeout_fops);
298 
299 	/* register all threshold attributes */
300 #define register_threshold_attribute(TAG, NAME, ...) \
301 	debugfs_create_file_unsafe("threshold_" #NAME, 0644, parent, parent, \
302 				   &NAME##_fops);
303 	MAKE_XE_GUC_KLV_THRESHOLDS_SET(register_threshold_attribute)
304 #undef register_threshold_attribute
305 }
306 
307 /*
308  *      /sys/kernel/debug/dri/0/
309  *      ├── gt0
310  *      │   ├── vf1
311  *      │   │   ├── control { stop, pause, resume }
312  */
313 
314 static const struct {
315 	const char *cmd;
316 	int (*fn)(struct xe_gt *gt, unsigned int vfid);
317 } control_cmds[] = {
318 	{ "stop", xe_gt_sriov_pf_control_stop_vf },
319 	{ "pause", xe_gt_sriov_pf_control_pause_vf },
320 	{ "resume", xe_gt_sriov_pf_control_resume_vf },
321 #ifdef CONFIG_DRM_XE_DEBUG_SRIOV
322 	{ "restore!", xe_gt_sriov_pf_migration_restore_guc_state },
323 #endif
324 };
325 
326 static ssize_t control_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
327 {
328 	struct dentry *dent = file_dentry(file);
329 	struct dentry *parent = dent->d_parent;
330 	struct xe_gt *gt = extract_gt(parent);
331 	struct xe_device *xe = gt_to_xe(gt);
332 	unsigned int vfid = extract_vfid(parent);
333 	int ret = -EINVAL;
334 	char cmd[32];
335 	size_t n;
336 
337 	xe_gt_assert(gt, vfid);
338 	xe_gt_sriov_pf_assert_vfid(gt, vfid);
339 
340 	if (*pos)
341 		return -ESPIPE;
342 
343 	if (count > sizeof(cmd) - 1)
344 		return -EINVAL;
345 
346 	ret = simple_write_to_buffer(cmd, sizeof(cmd) - 1, pos, buf, count);
347 	if (ret < 0)
348 		return ret;
349 	cmd[ret] = '\0';
350 
351 	for (n = 0; n < ARRAY_SIZE(control_cmds); n++) {
352 		xe_gt_assert(gt, sizeof(cmd) > strlen(control_cmds[n].cmd));
353 
354 		if (sysfs_streq(cmd, control_cmds[n].cmd)) {
355 			xe_pm_runtime_get(xe);
356 			ret = control_cmds[n].fn ? (*control_cmds[n].fn)(gt, vfid) : 0;
357 			xe_pm_runtime_put(xe);
358 			break;
359 		}
360 	}
361 
362 	return (ret < 0) ? ret : count;
363 }
364 
365 static ssize_t control_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
366 {
367 	char help[128];
368 	size_t n;
369 
370 	help[0] = '\0';
371 	for (n = 0; n < ARRAY_SIZE(control_cmds); n++) {
372 		strlcat(help, control_cmds[n].cmd, sizeof(help));
373 		strlcat(help, "\n", sizeof(help));
374 	}
375 
376 	return simple_read_from_buffer(buf, count, ppos, help, strlen(help));
377 }
378 
379 static const struct file_operations control_ops = {
380 	.owner		= THIS_MODULE,
381 	.open		= simple_open,
382 	.write		= control_write,
383 	.read		= control_read,
384 	.llseek		= default_llseek,
385 };
386 
387 /*
388  *      /sys/kernel/debug/dri/0/
389  *      ├── gt0
390  *      │   ├── vf1
391  *      │   │   ├── guc_state
392  */
393 static ssize_t guc_state_read(struct file *file, char __user *buf,
394 			      size_t count, loff_t *pos)
395 {
396 	struct dentry *dent = file_dentry(file);
397 	struct dentry *parent = dent->d_parent;
398 	struct xe_gt *gt = extract_gt(parent);
399 	unsigned int vfid = extract_vfid(parent);
400 
401 	return xe_gt_sriov_pf_migration_read_guc_state(gt, vfid, buf, count, pos);
402 }
403 
404 static ssize_t guc_state_write(struct file *file, const char __user *buf,
405 			       size_t count, loff_t *pos)
406 {
407 	struct dentry *dent = file_dentry(file);
408 	struct dentry *parent = dent->d_parent;
409 	struct xe_gt *gt = extract_gt(parent);
410 	unsigned int vfid = extract_vfid(parent);
411 
412 	if (*pos)
413 		return -EINVAL;
414 
415 	return xe_gt_sriov_pf_migration_write_guc_state(gt, vfid, buf, count);
416 }
417 
418 static const struct file_operations guc_state_ops = {
419 	.owner		= THIS_MODULE,
420 	.read		= guc_state_read,
421 	.write		= guc_state_write,
422 	.llseek		= default_llseek,
423 };
424 
425 /*
426  *      /sys/kernel/debug/dri/0/
427  *      ├── gt0
428  *      │   ├── vf1
429  *      │   │   ├── config_blob
430  */
431 static ssize_t config_blob_read(struct file *file, char __user *buf,
432 				size_t count, loff_t *pos)
433 {
434 	struct dentry *dent = file_dentry(file);
435 	struct dentry *parent = dent->d_parent;
436 	struct xe_gt *gt = extract_gt(parent);
437 	unsigned int vfid = extract_vfid(parent);
438 	ssize_t ret;
439 	void *tmp;
440 
441 	ret = xe_gt_sriov_pf_config_save(gt, vfid, NULL, 0);
442 	if (!ret)
443 		return -ENODATA;
444 	if (ret < 0)
445 		return ret;
446 
447 	tmp = kzalloc(ret, GFP_KERNEL);
448 	if (!tmp)
449 		return -ENOMEM;
450 
451 	ret = xe_gt_sriov_pf_config_save(gt, vfid, tmp, ret);
452 	if (ret > 0)
453 		ret = simple_read_from_buffer(buf, count, pos, tmp, ret);
454 
455 	kfree(tmp);
456 	return ret;
457 }
458 
459 static ssize_t config_blob_write(struct file *file, const char __user *buf,
460 				 size_t count, loff_t *pos)
461 {
462 	struct dentry *dent = file_dentry(file);
463 	struct dentry *parent = dent->d_parent;
464 	struct xe_gt *gt = extract_gt(parent);
465 	unsigned int vfid = extract_vfid(parent);
466 	ssize_t ret;
467 	void *tmp;
468 
469 	if (*pos)
470 		return -EINVAL;
471 
472 	if (!count)
473 		return -ENODATA;
474 
475 	if (count > SZ_4K)
476 		return -EINVAL;
477 
478 	tmp = kzalloc(count, GFP_KERNEL);
479 	if (!tmp)
480 		return -ENOMEM;
481 
482 	if (copy_from_user(tmp, buf, count)) {
483 		ret = -EFAULT;
484 	} else {
485 		ret = xe_gt_sriov_pf_config_restore(gt, vfid, tmp, count);
486 		if (!ret)
487 			ret = count;
488 	}
489 	kfree(tmp);
490 	return ret;
491 }
492 
493 static const struct file_operations config_blob_ops = {
494 	.owner		= THIS_MODULE,
495 	.read		= config_blob_read,
496 	.write		= config_blob_write,
497 	.llseek		= default_llseek,
498 };
499 
500 /**
501  * xe_gt_sriov_pf_debugfs_register - Register SR-IOV PF specific entries in GT debugfs.
502  * @gt: the &xe_gt to register
503  * @root: the &dentry that represents the GT directory
504  *
505  * Register SR-IOV PF entries that are GT related and must be shown under GT debugfs.
506  */
507 void xe_gt_sriov_pf_debugfs_register(struct xe_gt *gt, struct dentry *root)
508 {
509 	struct xe_device *xe = gt_to_xe(gt);
510 	struct drm_minor *minor = xe->drm.primary;
511 	int n, totalvfs = xe_sriov_pf_get_totalvfs(xe);
512 	struct dentry *pfdentry;
513 	struct dentry *vfdentry;
514 	char buf[14]; /* should be enough up to "vf%u\0" for 2^32 - 1 */
515 
516 	xe_gt_assert(gt, IS_SRIOV_PF(xe));
517 	xe_gt_assert(gt, root->d_inode->i_private == gt);
518 
519 	/*
520 	 *      /sys/kernel/debug/dri/0/
521 	 *      ├── gt0
522 	 *      │   ├── pf
523 	 */
524 	pfdentry = debugfs_create_dir("pf", root);
525 	if (IS_ERR(pfdentry))
526 		return;
527 	pfdentry->d_inode->i_private = gt;
528 
529 	drm_debugfs_create_files(pf_info, ARRAY_SIZE(pf_info), pfdentry, minor);
530 	pf_add_policy_attrs(gt, pfdentry);
531 	pf_add_config_attrs(gt, pfdentry, PFID);
532 
533 	for (n = 1; n <= totalvfs; n++) {
534 		/*
535 		 *      /sys/kernel/debug/dri/0/
536 		 *      ├── gt0
537 		 *      │   ├── vf1
538 		 *      │   ├── vf2
539 		 */
540 		snprintf(buf, sizeof(buf), "vf%u", n);
541 		vfdentry = debugfs_create_dir(buf, root);
542 		if (IS_ERR(vfdentry))
543 			break;
544 		vfdentry->d_inode->i_private = (void *)(uintptr_t)n;
545 
546 		pf_add_config_attrs(gt, vfdentry, VFID(n));
547 		debugfs_create_file("control", 0600, vfdentry, NULL, &control_ops);
548 
549 		/* for testing/debugging purposes only! */
550 		if (IS_ENABLED(CONFIG_DRM_XE_DEBUG)) {
551 			debugfs_create_file("guc_state",
552 					    IS_ENABLED(CONFIG_DRM_XE_DEBUG_SRIOV) ? 0600 : 0400,
553 					    vfdentry, NULL, &guc_state_ops);
554 			debugfs_create_file("config_blob",
555 					    IS_ENABLED(CONFIG_DRM_XE_DEBUG_SRIOV) ? 0600 : 0400,
556 					    vfdentry, NULL, &config_blob_ops);
557 		}
558 	}
559 }
560