xref: /linux/drivers/gpu/drm/xe/xe_debugfs.c (revision d6112dddbf354d21ff2fcd49338df68782492c73)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2022 Intel Corporation
4  */
5 
6 #include "xe_debugfs.h"
7 
8 #include <linux/debugfs.h>
9 #include <linux/fault-inject.h>
10 #include <linux/string_helpers.h>
11 
12 #include <drm/drm_debugfs.h>
13 
14 #include "regs/xe_pmt.h"
15 #include "xe_bo.h"
16 #include "xe_device.h"
17 #include "xe_force_wake.h"
18 #include "xe_gt_debugfs.h"
19 #include "xe_gt_printk.h"
20 #include "xe_guc_ads.h"
21 #include "xe_mmio.h"
22 #include "xe_pm.h"
23 #include "xe_psmi.h"
24 #include "xe_pxp_debugfs.h"
25 #include "xe_sriov.h"
26 #include "xe_sriov_pf_debugfs.h"
27 #include "xe_sriov_vf.h"
28 #include "xe_step.h"
29 #include "xe_tile_debugfs.h"
30 #include "xe_vsec.h"
31 #include "xe_wa.h"
32 
33 #ifdef CONFIG_DRM_XE_DEBUG
34 #include "xe_bo_evict.h"
35 #include "xe_migrate.h"
36 #include "xe_vm.h"
37 #endif
38 
39 DECLARE_FAULT_ATTR(gt_reset_failure);
40 DECLARE_FAULT_ATTR(inject_csc_hw_error);
41 
read_residency_counter(struct xe_device * xe,struct xe_mmio * mmio,u32 offset,const char * name,struct drm_printer * p)42 static void read_residency_counter(struct xe_device *xe, struct xe_mmio *mmio,
43 				   u32 offset, const char *name, struct drm_printer *p)
44 {
45 	u64 residency = 0;
46 	int ret;
47 
48 	ret = xe_pmt_telem_read(to_pci_dev(xe->drm.dev),
49 				xe_mmio_read32(mmio, PUNIT_TELEMETRY_GUID),
50 				&residency, offset, sizeof(residency));
51 	if (ret != sizeof(residency)) {
52 		drm_warn(&xe->drm, "%s counter failed to read, ret %d\n", name, ret);
53 		return;
54 	}
55 
56 	drm_printf(p, "%s : %llu\n", name, residency);
57 }
58 
node_to_xe(struct drm_info_node * node)59 static struct xe_device *node_to_xe(struct drm_info_node *node)
60 {
61 	return to_xe_device(node->minor->dev);
62 }
63 
info(struct seq_file * m,void * data)64 static int info(struct seq_file *m, void *data)
65 {
66 	struct xe_device *xe = node_to_xe(m->private);
67 	struct drm_printer p = drm_seq_file_printer(m);
68 	struct xe_gt *gt;
69 	u8 id;
70 
71 	xe_pm_runtime_get(xe);
72 
73 	drm_printf(&p, "graphics_verx100 %d\n", xe->info.graphics_verx100);
74 	drm_printf(&p, "media_verx100 %d\n", xe->info.media_verx100);
75 	drm_printf(&p, "stepping G:%s M:%s B:%s\n",
76 		   xe_step_name(xe->info.step.graphics),
77 		   xe_step_name(xe->info.step.media),
78 		   xe_step_name(xe->info.step.basedie));
79 	drm_printf(&p, "is_dgfx %s\n", str_yes_no(xe->info.is_dgfx));
80 	drm_printf(&p, "platform %d\n", xe->info.platform);
81 	drm_printf(&p, "subplatform %d\n",
82 		   xe->info.subplatform > XE_SUBPLATFORM_NONE ? xe->info.subplatform : 0);
83 	drm_printf(&p, "devid 0x%x\n", xe->info.devid);
84 	drm_printf(&p, "revid %d\n", xe->info.revid);
85 	drm_printf(&p, "tile_count %d\n", xe->info.tile_count);
86 	drm_printf(&p, "vm_max_level %d\n", xe->info.vm_max_level);
87 	drm_printf(&p, "force_execlist %s\n", str_yes_no(xe->info.force_execlist));
88 	drm_printf(&p, "has_flat_ccs %s\n", str_yes_no(xe->info.has_flat_ccs));
89 	drm_printf(&p, "has_usm %s\n", str_yes_no(xe->info.has_usm));
90 	drm_printf(&p, "skip_guc_pc %s\n", str_yes_no(xe->info.skip_guc_pc));
91 	for_each_gt(gt, xe, id) {
92 		drm_printf(&p, "gt%d force wake %d\n", id,
93 			   xe_force_wake_ref(gt_to_fw(gt), XE_FW_GT));
94 		drm_printf(&p, "gt%d engine_mask 0x%llx\n", id,
95 			   gt->info.engine_mask);
96 	}
97 
98 	xe_pm_runtime_put(xe);
99 	return 0;
100 }
101 
sriov_info(struct seq_file * m,void * data)102 static int sriov_info(struct seq_file *m, void *data)
103 {
104 	struct xe_device *xe = node_to_xe(m->private);
105 	struct drm_printer p = drm_seq_file_printer(m);
106 
107 	xe_sriov_print_info(xe, &p);
108 	return 0;
109 }
110 
workarounds(struct xe_device * xe,struct drm_printer * p)111 static int workarounds(struct xe_device *xe, struct drm_printer *p)
112 {
113 	xe_pm_runtime_get(xe);
114 	xe_wa_device_dump(xe, p);
115 	xe_pm_runtime_put(xe);
116 
117 	return 0;
118 }
119 
workaround_info(struct seq_file * m,void * data)120 static int workaround_info(struct seq_file *m, void *data)
121 {
122 	struct xe_device *xe = node_to_xe(m->private);
123 	struct drm_printer p = drm_seq_file_printer(m);
124 
125 	workarounds(xe, &p);
126 	return 0;
127 }
128 
dgfx_pkg_residencies_show(struct seq_file * m,void * data)129 static int dgfx_pkg_residencies_show(struct seq_file *m, void *data)
130 {
131 	struct xe_device *xe;
132 	struct xe_mmio *mmio;
133 	struct drm_printer p;
134 
135 	xe = node_to_xe(m->private);
136 	p = drm_seq_file_printer(m);
137 	xe_pm_runtime_get(xe);
138 	mmio = xe_root_tile_mmio(xe);
139 	static const struct {
140 		u32 offset;
141 		const char *name;
142 	} residencies[] = {
143 		{BMG_G2_RESIDENCY_OFFSET, "Package G2"},
144 		{BMG_G6_RESIDENCY_OFFSET, "Package G6"},
145 		{BMG_G7_RESIDENCY_OFFSET, "Package G7"},
146 		{BMG_G8_RESIDENCY_OFFSET, "Package G8"},
147 		{BMG_G10_RESIDENCY_OFFSET, "Package G10"},
148 		{BMG_MODS_RESIDENCY_OFFSET, "Package ModS"}
149 	};
150 
151 	for (int i = 0; i < ARRAY_SIZE(residencies); i++)
152 		read_residency_counter(xe, mmio, residencies[i].offset, residencies[i].name, &p);
153 
154 	xe_pm_runtime_put(xe);
155 	return 0;
156 }
157 
dgfx_pcie_link_residencies_show(struct seq_file * m,void * data)158 static int dgfx_pcie_link_residencies_show(struct seq_file *m, void *data)
159 {
160 	struct xe_device *xe;
161 	struct xe_mmio *mmio;
162 	struct drm_printer p;
163 
164 	xe = node_to_xe(m->private);
165 	p = drm_seq_file_printer(m);
166 	xe_pm_runtime_get(xe);
167 	mmio = xe_root_tile_mmio(xe);
168 
169 	static const struct {
170 		u32 offset;
171 		const char *name;
172 	} residencies[] = {
173 		{BMG_PCIE_LINK_L0_RESIDENCY_OFFSET, "PCIE LINK L0 RESIDENCY"},
174 		{BMG_PCIE_LINK_L1_RESIDENCY_OFFSET, "PCIE LINK L1 RESIDENCY"},
175 		{BMG_PCIE_LINK_L1_2_RESIDENCY_OFFSET, "PCIE LINK L1.2 RESIDENCY"}
176 	};
177 
178 	for (int i = 0; i < ARRAY_SIZE(residencies); i++)
179 		read_residency_counter(xe, mmio, residencies[i].offset, residencies[i].name, &p);
180 
181 	xe_pm_runtime_put(xe);
182 	return 0;
183 }
184 
185 static const struct drm_info_list debugfs_list[] = {
186 	{"info", info, 0},
187 	{ .name = "sriov_info", .show = sriov_info, },
188 	{ .name = "workarounds", .show = workaround_info, },
189 };
190 
191 static const struct drm_info_list debugfs_residencies[] = {
192 	{ .name = "dgfx_pkg_residencies", .show = dgfx_pkg_residencies_show, },
193 	{ .name = "dgfx_pcie_link_residencies", .show = dgfx_pcie_link_residencies_show, },
194 };
195 
forcewake_open(struct inode * inode,struct file * file)196 static int forcewake_open(struct inode *inode, struct file *file)
197 {
198 	struct xe_device *xe = inode->i_private;
199 	struct xe_gt *gt;
200 	u8 id, last_gt;
201 	unsigned int fw_ref;
202 
203 	xe_pm_runtime_get(xe);
204 	for_each_gt(gt, xe, id) {
205 		last_gt = id;
206 
207 		fw_ref = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
208 		if (!xe_force_wake_ref_has_domain(fw_ref, XE_FORCEWAKE_ALL))
209 			goto err_fw_get;
210 	}
211 
212 	return 0;
213 
214 err_fw_get:
215 	for_each_gt(gt, xe, id) {
216 		if (id < last_gt)
217 			xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
218 		else if (id == last_gt)
219 			xe_force_wake_put(gt_to_fw(gt), fw_ref);
220 		else
221 			break;
222 	}
223 
224 	xe_pm_runtime_put(xe);
225 	return -ETIMEDOUT;
226 }
227 
forcewake_release(struct inode * inode,struct file * file)228 static int forcewake_release(struct inode *inode, struct file *file)
229 {
230 	struct xe_device *xe = inode->i_private;
231 	struct xe_gt *gt;
232 	u8 id;
233 
234 	for_each_gt(gt, xe, id)
235 		xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
236 	xe_pm_runtime_put(xe);
237 
238 	return 0;
239 }
240 
241 static const struct file_operations forcewake_all_fops = {
242 	.owner = THIS_MODULE,
243 	.open = forcewake_open,
244 	.release = forcewake_release,
245 };
246 
wedged_mode_show(struct file * f,char __user * ubuf,size_t size,loff_t * pos)247 static ssize_t wedged_mode_show(struct file *f, char __user *ubuf,
248 				size_t size, loff_t *pos)
249 {
250 	struct xe_device *xe = file_inode(f)->i_private;
251 	char buf[32];
252 	int len = 0;
253 
254 	len = scnprintf(buf, sizeof(buf), "%d\n", xe->wedged.mode);
255 
256 	return simple_read_from_buffer(ubuf, size, pos, buf, len);
257 }
258 
__wedged_mode_set_reset_policy(struct xe_gt * gt,enum xe_wedged_mode mode)259 static int __wedged_mode_set_reset_policy(struct xe_gt *gt, enum xe_wedged_mode mode)
260 {
261 	bool enable_engine_reset;
262 	int ret;
263 
264 	enable_engine_reset = (mode != XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET);
265 	ret = xe_guc_ads_scheduler_policy_toggle_reset(&gt->uc.guc.ads,
266 						       enable_engine_reset);
267 	if (ret)
268 		xe_gt_err(gt, "Failed to update GuC ADS scheduler policy (%pe)\n", ERR_PTR(ret));
269 
270 	return ret;
271 }
272 
wedged_mode_set_reset_policy(struct xe_device * xe,enum xe_wedged_mode mode)273 static int wedged_mode_set_reset_policy(struct xe_device *xe, enum xe_wedged_mode mode)
274 {
275 	struct xe_gt *gt;
276 	int ret;
277 	u8 id;
278 
279 	guard(xe_pm_runtime)(xe);
280 	for_each_gt(gt, xe, id) {
281 		ret = __wedged_mode_set_reset_policy(gt, mode);
282 		if (ret) {
283 			if (id > 0) {
284 				xe->wedged.inconsistent_reset = true;
285 				drm_err(&xe->drm, "Inconsistent reset policy state between GTs\n");
286 			}
287 			return ret;
288 		}
289 	}
290 
291 	xe->wedged.inconsistent_reset = false;
292 
293 	return 0;
294 }
295 
wedged_mode_needs_policy_update(struct xe_device * xe,enum xe_wedged_mode mode)296 static bool wedged_mode_needs_policy_update(struct xe_device *xe, enum xe_wedged_mode mode)
297 {
298 	if (xe->wedged.inconsistent_reset)
299 		return true;
300 
301 	if (xe->wedged.mode == mode)
302 		return false;
303 
304 	if (xe->wedged.mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET ||
305 	    mode == XE_WEDGED_MODE_UPON_ANY_HANG_NO_RESET)
306 		return true;
307 
308 	return false;
309 }
310 
wedged_mode_set(struct file * f,const char __user * ubuf,size_t size,loff_t * pos)311 static ssize_t wedged_mode_set(struct file *f, const char __user *ubuf,
312 			       size_t size, loff_t *pos)
313 {
314 	struct xe_device *xe = file_inode(f)->i_private;
315 	u32 wedged_mode;
316 	ssize_t ret;
317 
318 	ret = kstrtouint_from_user(ubuf, size, 0, &wedged_mode);
319 	if (ret)
320 		return ret;
321 
322 	if (wedged_mode > 2)
323 		return -EINVAL;
324 
325 	if (wedged_mode_needs_policy_update(xe, wedged_mode)) {
326 		ret = wedged_mode_set_reset_policy(xe, wedged_mode);
327 		if (ret)
328 			return ret;
329 	}
330 
331 	xe->wedged.mode = wedged_mode;
332 
333 	return size;
334 }
335 
336 static const struct file_operations wedged_mode_fops = {
337 	.owner = THIS_MODULE,
338 	.read = wedged_mode_show,
339 	.write = wedged_mode_set,
340 };
341 
atomic_svm_timeslice_ms_show(struct file * f,char __user * ubuf,size_t size,loff_t * pos)342 static ssize_t atomic_svm_timeslice_ms_show(struct file *f, char __user *ubuf,
343 					    size_t size, loff_t *pos)
344 {
345 	struct xe_device *xe = file_inode(f)->i_private;
346 	char buf[32];
347 	int len = 0;
348 
349 	len = scnprintf(buf, sizeof(buf), "%d\n", xe->atomic_svm_timeslice_ms);
350 
351 	return simple_read_from_buffer(ubuf, size, pos, buf, len);
352 }
353 
atomic_svm_timeslice_ms_set(struct file * f,const char __user * ubuf,size_t size,loff_t * pos)354 static ssize_t atomic_svm_timeslice_ms_set(struct file *f,
355 					   const char __user *ubuf,
356 					   size_t size, loff_t *pos)
357 {
358 	struct xe_device *xe = file_inode(f)->i_private;
359 	u32 atomic_svm_timeslice_ms;
360 	ssize_t ret;
361 
362 	ret = kstrtouint_from_user(ubuf, size, 0, &atomic_svm_timeslice_ms);
363 	if (ret)
364 		return ret;
365 
366 	xe->atomic_svm_timeslice_ms = atomic_svm_timeslice_ms;
367 
368 	return size;
369 }
370 
371 static const struct file_operations atomic_svm_timeslice_ms_fops = {
372 	.owner = THIS_MODULE,
373 	.read = atomic_svm_timeslice_ms_show,
374 	.write = atomic_svm_timeslice_ms_set,
375 };
376 
disable_late_binding_show(struct file * f,char __user * ubuf,size_t size,loff_t * pos)377 static ssize_t disable_late_binding_show(struct file *f, char __user *ubuf,
378 					 size_t size, loff_t *pos)
379 {
380 	struct xe_device *xe = file_inode(f)->i_private;
381 	struct xe_late_bind *late_bind = &xe->late_bind;
382 	char buf[32];
383 	int len;
384 
385 	len = scnprintf(buf, sizeof(buf), "%d\n", late_bind->disable);
386 
387 	return simple_read_from_buffer(ubuf, size, pos, buf, len);
388 }
389 
disable_late_binding_set(struct file * f,const char __user * ubuf,size_t size,loff_t * pos)390 static ssize_t disable_late_binding_set(struct file *f, const char __user *ubuf,
391 					size_t size, loff_t *pos)
392 {
393 	struct xe_device *xe = file_inode(f)->i_private;
394 	struct xe_late_bind *late_bind = &xe->late_bind;
395 	bool val;
396 	int ret;
397 
398 	ret = kstrtobool_from_user(ubuf, size, &val);
399 	if (ret)
400 		return ret;
401 
402 	late_bind->disable = val;
403 	return size;
404 }
405 
406 static const struct file_operations disable_late_binding_fops = {
407 	.owner = THIS_MODULE,
408 	.read = disable_late_binding_show,
409 	.write = disable_late_binding_set,
410 };
411 
xe_debugfs_register(struct xe_device * xe)412 void xe_debugfs_register(struct xe_device *xe)
413 {
414 	struct ttm_device *bdev = &xe->ttm;
415 	struct drm_minor *minor = xe->drm.primary;
416 	struct dentry *root = minor->debugfs_root;
417 	struct ttm_resource_manager *man;
418 	struct xe_tile *tile;
419 	struct xe_gt *gt;
420 	u32 mem_type;
421 	u8 tile_id;
422 	u8 id;
423 
424 	drm_debugfs_create_files(debugfs_list,
425 				 ARRAY_SIZE(debugfs_list),
426 				 root, minor);
427 
428 	if (xe->info.platform == XE_BATTLEMAGE && !IS_SRIOV_VF(xe)) {
429 		drm_debugfs_create_files(debugfs_residencies,
430 					 ARRAY_SIZE(debugfs_residencies),
431 					 root, minor);
432 		fault_create_debugfs_attr("inject_csc_hw_error", root,
433 					  &inject_csc_hw_error);
434 	}
435 
436 	debugfs_create_file("forcewake_all", 0400, root, xe,
437 			    &forcewake_all_fops);
438 
439 	debugfs_create_file("wedged_mode", 0600, root, xe,
440 			    &wedged_mode_fops);
441 
442 	debugfs_create_file("atomic_svm_timeslice_ms", 0600, root, xe,
443 			    &atomic_svm_timeslice_ms_fops);
444 
445 	debugfs_create_file("disable_late_binding", 0600, root, xe,
446 			    &disable_late_binding_fops);
447 
448 	for (mem_type = XE_PL_VRAM0; mem_type <= XE_PL_VRAM1; ++mem_type) {
449 		man = ttm_manager_type(bdev, mem_type);
450 
451 		if (man) {
452 			char name[16];
453 
454 			snprintf(name, sizeof(name), "vram%d_mm", mem_type - XE_PL_VRAM0);
455 			ttm_resource_manager_create_debugfs(man, root, name);
456 		}
457 	}
458 
459 	man = ttm_manager_type(bdev, XE_PL_TT);
460 	ttm_resource_manager_create_debugfs(man, root, "gtt_mm");
461 
462 	man = ttm_manager_type(bdev, XE_PL_STOLEN);
463 	if (man)
464 		ttm_resource_manager_create_debugfs(man, root, "stolen_mm");
465 
466 	for_each_tile(tile, xe, tile_id)
467 		xe_tile_debugfs_register(tile);
468 
469 	for_each_gt(gt, xe, id)
470 		xe_gt_debugfs_register(gt);
471 
472 	xe_pxp_debugfs_register(xe->pxp);
473 
474 	xe_psmi_debugfs_register(xe);
475 
476 	fault_create_debugfs_attr("fail_gt_reset", root, &gt_reset_failure);
477 
478 	if (IS_SRIOV_PF(xe))
479 		xe_sriov_pf_debugfs_register(xe, root);
480 	else if (IS_SRIOV_VF(xe))
481 		xe_sriov_vf_debugfs_register(xe, root);
482 }
483