1 // SPDX-License-Identifier: GPL-2.0-or-later 1
2 /*
3 *
4 * Copyright (c) 2005 Linas Vepstas <linas@lin
5 */
6
7 #include <linux/delay.h>
8 #include <linux/list.h>
9 #include <linux/sched.h>
10 #include <linux/semaphore.h>
11 #include <linux/pci.h>
12 #include <linux/slab.h>
13 #include <linux/kthread.h>
14 #include <asm/eeh_event.h>
15 #include <asm/ppc-pci.h>
16
17 /** Overview:
18 * EEH error states may be detected within ex
19 * however, the recovery processing needs to
20 * in a normal kernel context and not an inte
21 * This pair of routines creates an event and
22 * work-queue, where a worker thread can driv
23 */
24
25 static DEFINE_SPINLOCK(eeh_eventlist_lock);
26 static DECLARE_COMPLETION(eeh_eventlist_event)
27 static LIST_HEAD(eeh_eventlist);
28
29 /**
30 * eeh_event_handler - Dispatch EEH events.
31 * @dummy - unused
32 *
33 * The detection of a frozen slot can occur in
34 * where it can be hard to do anything about i
35 * routine is to pull these detection events o
36 * of the interrupt handler, and re-dispatch t
37 * at a later time in a normal context.
38 */
39 static int eeh_event_handler(void * dummy)
40 {
41 unsigned long flags;
42 struct eeh_event *event;
43
44 while (!kthread_should_stop()) {
45 if (wait_for_completion_interr
46 break;
47
48 /* Fetch EEH event from the qu
49 spin_lock_irqsave(&eeh_eventli
50 event = NULL;
51 if (!list_empty(&eeh_eventlist
52 event = list_entry(eeh
53 str
54 list_del(&event->list)
55 }
56 spin_unlock_irqrestore(&eeh_ev
57 if (!event)
58 continue;
59
60 /* We might have event without
61 if (event->pe)
62 eeh_handle_normal_even
63 else
64 eeh_handle_special_eve
65
66 kfree(event);
67 }
68
69 return 0;
70 }
71
72 /**
73 * eeh_event_init - Start kernel thread to han
74 *
75 * This routine is called to start the kernel
76 * EEH event.
77 */
78 int eeh_event_init(void)
79 {
80 struct task_struct *t;
81 int ret = 0;
82
83 t = kthread_run(eeh_event_handler, NUL
84 if (IS_ERR(t)) {
85 ret = PTR_ERR(t);
86 pr_err("%s: Failed to start EE
87 __func__, ret);
88 return ret;
89 }
90
91 return 0;
92 }
93
94 /**
95 * eeh_send_failure_event - Generate a PCI err
96 * @pe: EEH PE
97 *
98 * This routine can be called within an interr
99 * the actual event will be delivered in a nor
100 * (from a workqueue).
101 */
102 int __eeh_send_failure_event(struct eeh_pe *pe
103 {
104 unsigned long flags;
105 struct eeh_event *event;
106
107 event = kzalloc(sizeof(*event), GFP_AT
108 if (!event) {
109 pr_err("EEH: out of memory, ev
110 return -ENOMEM;
111 }
112 event->pe = pe;
113
114 /*
115 * Mark the PE as recovering before in
116 * This prevents the PE from being fre
117 * while the PE is sitting in the even
118 */
119 if (pe) {
120 #ifdef CONFIG_STACKTRACE
121 /*
122 * Save the current stack trac
123 * event handler thread.
124 */
125 pe->trace_entries = stack_trac
126 ARRAY
127 #endif /* CONFIG_STACKTRACE */
128
129 eeh_pe_state_mark(pe, EEH_PE_R
130 }
131
132 /* We may or may not be called in an i
133 spin_lock_irqsave(&eeh_eventlist_lock,
134 list_add(&event->list, &eeh_eventlist)
135 spin_unlock_irqrestore(&eeh_eventlist_
136
137 /* For EEH deamon to knick in */
138 complete(&eeh_eventlist_event);
139
140 return 0;
141 }
142
143 int eeh_send_failure_event(struct eeh_pe *pe)
144 {
145 /*
146 * If we've manually suppressed recove
147 * then just drop it on the floor.
148 */
149 if (eeh_debugfs_no_recover) {
150 pr_err("EEH: Event dropped due
151 return 0;
152 }
153
154 return __eeh_send_failure_event(pe);
155 }
156
157 /**
158 * eeh_remove_event - Remove EEH event from th
159 * @pe: Event binding to the PE
160 * @force: Event will be removed unconditional
161 *
162 * On PowerNV platform, we might have subseque
163 * is part of the former one. For that case, t
164 * coming events are totally duplicated and un
165 * they should be removed.
166 */
167 void eeh_remove_event(struct eeh_pe *pe, bool
168 {
169 unsigned long flags;
170 struct eeh_event *event, *tmp;
171
172 /*
173 * If we have NULL PE passed in, we ha
174 * or we're sure we can report all exi
175 * by the caller.
176 *
177 * With "force", the event with associ
178 * have been isolated, the event won't
179 * to avoid event lost.
180 */
181 spin_lock_irqsave(&eeh_eventlist_lock,
182 list_for_each_entry_safe(event, tmp, &
183 if (!force && event->pe &&
184 (event->pe->state & EEH_PE
185 continue;
186
187 if (!pe) {
188 list_del(&event->list)
189 kfree(event);
190 } else if (pe->type & EEH_PE_P
191 if (event->pe && event
192 list_del(&even
193 kfree(event);
194 }
195 } else if (event->pe == pe) {
196 list_del(&event->list)
197 kfree(event);
198 }
199 }
200 spin_unlock_irqrestore(&eeh_eventlist_
201 }
202
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.