xref: /linux/drivers/gpu/drm/xe/xe_devcoredump.c (revision 24b10e5f8e0d2bee1a10fc67011ea5d936c1a389)
1 // SPDX-License-Identifier: MIT
2 /*
3  * Copyright © 2023 Intel Corporation
4  */
5 
6 #include "xe_devcoredump.h"
7 #include "xe_devcoredump_types.h"
8 
9 #include <linux/devcoredump.h>
10 #include <generated/utsrelease.h>
11 
12 #include "xe_device.h"
13 #include "xe_exec_queue.h"
14 #include "xe_force_wake.h"
15 #include "xe_gt.h"
16 #include "xe_guc_ct.h"
17 #include "xe_guc_submit.h"
18 #include "xe_hw_engine.h"
19 
20 /**
21  * DOC: Xe device coredump
22  *
23  * Devices overview:
24  * Xe uses dev_coredump infrastructure for exposing the crash errors in a
25  * standardized way.
26  * devcoredump exposes a temporary device under /sys/class/devcoredump/
27  * which is linked with our card device directly.
28  * The core dump can be accessed either from
29  * /sys/class/drm/card<n>/device/devcoredump/ or from
30  * /sys/class/devcoredump/devcd<m> where
31  * /sys/class/devcoredump/devcd<m>/failing_device is a link to
32  * /sys/class/drm/card<n>/device/.
33  *
34  * Snapshot at hang:
35  * The 'data' file is printed with a drm_printer pointer at devcoredump read
36  * time. For this reason, we need to take snapshots from when the hang has
37  * happened, and not only when the user is reading the file. Otherwise the
38  * information is outdated since the resets might have happened in between.
39  *
40  * 'First' failure snapshot:
41  * In general, the first hang is the most critical one since the following hangs
42  * can be a consequence of the initial hang. For this reason we only take the
43  * snapshot of the 'first' failure and ignore subsequent calls of this function,
44  * at least while the coredump device is alive. Dev_coredump has a delayed work
45  * queue that will eventually delete the device and free all the dump
46  * information.
47  */
48 
49 #ifdef CONFIG_DEV_COREDUMP
50 
51 static struct xe_device *coredump_to_xe(const struct xe_devcoredump *coredump)
52 {
53 	return container_of(coredump, struct xe_device, devcoredump);
54 }
55 
56 static struct xe_guc *exec_queue_to_guc(struct xe_exec_queue *q)
57 {
58 	return &q->gt->uc.guc;
59 }
60 
61 static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,
62 				   size_t count, void *data, size_t datalen)
63 {
64 	struct xe_devcoredump *coredump = data;
65 	struct xe_devcoredump_snapshot *ss;
66 	struct drm_printer p;
67 	struct drm_print_iterator iter;
68 	struct timespec64 ts;
69 	int i;
70 
71 	/* Our device is gone already... */
72 	if (!data || !coredump_to_xe(coredump))
73 		return -ENODEV;
74 
75 	iter.data = buffer;
76 	iter.offset = 0;
77 	iter.start = offset;
78 	iter.remain = count;
79 
80 	ss = &coredump->snapshot;
81 	p = drm_coredump_printer(&iter);
82 
83 	drm_printf(&p, "**** Xe Device Coredump ****\n");
84 	drm_printf(&p, "kernel: " UTS_RELEASE "\n");
85 	drm_printf(&p, "module: " KBUILD_MODNAME "\n");
86 
87 	ts = ktime_to_timespec64(ss->snapshot_time);
88 	drm_printf(&p, "Snapshot time: %lld.%09ld\n", ts.tv_sec, ts.tv_nsec);
89 	ts = ktime_to_timespec64(ss->boot_time);
90 	drm_printf(&p, "Uptime: %lld.%09ld\n", ts.tv_sec, ts.tv_nsec);
91 
92 	drm_printf(&p, "\n**** GuC CT ****\n");
93 	xe_guc_ct_snapshot_print(coredump->snapshot.ct, &p);
94 	xe_guc_exec_queue_snapshot_print(coredump->snapshot.ge, &p);
95 
96 	drm_printf(&p, "\n**** HW Engines ****\n");
97 	for (i = 0; i < XE_NUM_HW_ENGINES; i++)
98 		if (coredump->snapshot.hwe[i])
99 			xe_hw_engine_snapshot_print(coredump->snapshot.hwe[i],
100 						    &p);
101 
102 	return count - iter.remain;
103 }
104 
105 static void xe_devcoredump_free(void *data)
106 {
107 	struct xe_devcoredump *coredump = data;
108 	int i;
109 
110 	/* Our device is gone. Nothing to do... */
111 	if (!data || !coredump_to_xe(coredump))
112 		return;
113 
114 	xe_guc_ct_snapshot_free(coredump->snapshot.ct);
115 	xe_guc_exec_queue_snapshot_free(coredump->snapshot.ge);
116 	for (i = 0; i < XE_NUM_HW_ENGINES; i++)
117 		if (coredump->snapshot.hwe[i])
118 			xe_hw_engine_snapshot_free(coredump->snapshot.hwe[i]);
119 
120 	coredump->captured = false;
121 	drm_info(&coredump_to_xe(coredump)->drm,
122 		 "Xe device coredump has been deleted.\n");
123 }
124 
125 static void devcoredump_snapshot(struct xe_devcoredump *coredump,
126 				 struct xe_exec_queue *q)
127 {
128 	struct xe_devcoredump_snapshot *ss = &coredump->snapshot;
129 	struct xe_guc *guc = exec_queue_to_guc(q);
130 	struct xe_hw_engine *hwe;
131 	enum xe_hw_engine_id id;
132 	u32 adj_logical_mask = q->logical_mask;
133 	u32 width_mask = (0x1 << q->width) - 1;
134 	int i;
135 	bool cookie;
136 
137 	ss->snapshot_time = ktime_get_real();
138 	ss->boot_time = ktime_get_boottime();
139 
140 	cookie = dma_fence_begin_signalling();
141 	for (i = 0; q->width > 1 && i < XE_HW_ENGINE_MAX_INSTANCE;) {
142 		if (adj_logical_mask & BIT(i)) {
143 			adj_logical_mask |= width_mask << i;
144 			i += q->width;
145 		} else {
146 			++i;
147 		}
148 	}
149 
150 	xe_force_wake_get(gt_to_fw(q->gt), XE_FORCEWAKE_ALL);
151 
152 	coredump->snapshot.ct = xe_guc_ct_snapshot_capture(&guc->ct, true);
153 	coredump->snapshot.ge = xe_guc_exec_queue_snapshot_capture(q);
154 
155 	for_each_hw_engine(hwe, q->gt, id) {
156 		if (hwe->class != q->hwe->class ||
157 		    !(BIT(hwe->logical_instance) & adj_logical_mask)) {
158 			coredump->snapshot.hwe[id] = NULL;
159 			continue;
160 		}
161 		coredump->snapshot.hwe[id] = xe_hw_engine_snapshot_capture(hwe);
162 	}
163 
164 	xe_force_wake_put(gt_to_fw(q->gt), XE_FORCEWAKE_ALL);
165 	dma_fence_end_signalling(cookie);
166 }
167 
168 /**
169  * xe_devcoredump - Take the required snapshots and initialize coredump device.
170  * @q: The faulty xe_exec_queue, where the issue was detected.
171  *
172  * This function should be called at the crash time within the serialized
173  * gt_reset. It is skipped if we still have the core dump device available
174  * with the information of the 'first' snapshot.
175  */
176 void xe_devcoredump(struct xe_exec_queue *q)
177 {
178 	struct xe_device *xe = gt_to_xe(q->gt);
179 	struct xe_devcoredump *coredump = &xe->devcoredump;
180 
181 	if (coredump->captured) {
182 		drm_dbg(&xe->drm, "Multiple hangs are occurring, but only the first snapshot was taken\n");
183 		return;
184 	}
185 
186 	coredump->captured = true;
187 	devcoredump_snapshot(coredump, q);
188 
189 	drm_info(&xe->drm, "Xe device coredump has been created\n");
190 	drm_info(&xe->drm, "Check your /sys/class/drm/card%d/device/devcoredump/data\n",
191 		 xe->drm.primary->index);
192 
193 	dev_coredumpm(xe->drm.dev, THIS_MODULE, coredump, 0, GFP_KERNEL,
194 		      xe_devcoredump_read, xe_devcoredump_free);
195 }
196 #endif
197