xref: /openbmc/linux/arch/powerpc/kernel/eeh_event.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *
4  * Copyright (c) 2005 Linas Vepstas <linas@linas.org>
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 exception handlers;
19  *  however, the recovery processing needs to occur asynchronously
20  *  in a normal kernel context and not an interrupt context.
21  *  This pair of routines creates an event and queues it onto a
22  *  work-queue, where a worker thread can drive recovery.
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 inside an interrupt,
34  * where it can be hard to do anything about it.  The goal of this
35  * routine is to pull these detection events out of the context
36  * of the interrupt handler, and re-dispatch them for processing
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 	struct eeh_pe *pe;
44 
45 	while (!kthread_should_stop()) {
46 		if (wait_for_completion_interruptible(&eeh_eventlist_event))
47 			break;
48 
49 		/* Fetch EEH event from the queue */
50 		spin_lock_irqsave(&eeh_eventlist_lock, flags);
51 		event = NULL;
52 		if (!list_empty(&eeh_eventlist)) {
53 			event = list_entry(eeh_eventlist.next,
54 					   struct eeh_event, list);
55 			list_del(&event->list);
56 		}
57 		spin_unlock_irqrestore(&eeh_eventlist_lock, flags);
58 		if (!event)
59 			continue;
60 
61 		/* We might have event without binding PE */
62 		pe = event->pe;
63 		if (pe) {
64 			if (pe->type & EEH_PE_PHB)
65 				pr_info("EEH: Detected error on PHB#%x\n",
66 					 pe->phb->global_number);
67 			else
68 				pr_info("EEH: Detected PCI bus error on "
69 					"PHB#%x-PE#%x\n",
70 					pe->phb->global_number, pe->addr);
71 			eeh_handle_normal_event(pe);
72 		} else {
73 			eeh_handle_special_event();
74 		}
75 
76 		kfree(event);
77 	}
78 
79 	return 0;
80 }
81 
82 /**
83  * eeh_event_init - Start kernel thread to handle EEH events
84  *
85  * This routine is called to start the kernel thread for processing
86  * EEH event.
87  */
88 int eeh_event_init(void)
89 {
90 	struct task_struct *t;
91 	int ret = 0;
92 
93 	t = kthread_run(eeh_event_handler, NULL, "eehd");
94 	if (IS_ERR(t)) {
95 		ret = PTR_ERR(t);
96 		pr_err("%s: Failed to start EEH daemon (%d)\n",
97 			__func__, ret);
98 		return ret;
99 	}
100 
101 	return 0;
102 }
103 
104 /**
105  * eeh_send_failure_event - Generate a PCI error event
106  * @pe: EEH PE
107  *
108  * This routine can be called within an interrupt context;
109  * the actual event will be delivered in a normal context
110  * (from a workqueue).
111  */
112 int __eeh_send_failure_event(struct eeh_pe *pe)
113 {
114 	unsigned long flags;
115 	struct eeh_event *event;
116 
117 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
118 	if (!event) {
119 		pr_err("EEH: out of memory, event not handled\n");
120 		return -ENOMEM;
121 	}
122 	event->pe = pe;
123 
124 	/* We may or may not be called in an interrupt context */
125 	spin_lock_irqsave(&eeh_eventlist_lock, flags);
126 	list_add(&event->list, &eeh_eventlist);
127 	spin_unlock_irqrestore(&eeh_eventlist_lock, flags);
128 
129 	/* For EEH deamon to knick in */
130 	complete(&eeh_eventlist_event);
131 
132 	return 0;
133 }
134 
135 int eeh_send_failure_event(struct eeh_pe *pe)
136 {
137 	/*
138 	 * If we've manually supressed recovery events via debugfs
139 	 * then just drop it on the floor.
140 	 */
141 	if (eeh_debugfs_no_recover) {
142 		pr_err("EEH: Event dropped due to no_recover setting\n");
143 		return 0;
144 	}
145 
146 	return __eeh_send_failure_event(pe);
147 }
148 
149 /**
150  * eeh_remove_event - Remove EEH event from the queue
151  * @pe: Event binding to the PE
152  * @force: Event will be removed unconditionally
153  *
154  * On PowerNV platform, we might have subsequent coming events
155  * is part of the former one. For that case, those subsequent
156  * coming events are totally duplicated and unnecessary, thus
157  * they should be removed.
158  */
159 void eeh_remove_event(struct eeh_pe *pe, bool force)
160 {
161 	unsigned long flags;
162 	struct eeh_event *event, *tmp;
163 
164 	/*
165 	 * If we have NULL PE passed in, we have dead IOC
166 	 * or we're sure we can report all existing errors
167 	 * by the caller.
168 	 *
169 	 * With "force", the event with associated PE that
170 	 * have been isolated, the event won't be removed
171 	 * to avoid event lost.
172 	 */
173 	spin_lock_irqsave(&eeh_eventlist_lock, flags);
174 	list_for_each_entry_safe(event, tmp, &eeh_eventlist, list) {
175 		if (!force && event->pe &&
176 		    (event->pe->state & EEH_PE_ISOLATED))
177 			continue;
178 
179 		if (!pe) {
180 			list_del(&event->list);
181 			kfree(event);
182 		} else if (pe->type & EEH_PE_PHB) {
183 			if (event->pe && event->pe->phb == pe->phb) {
184 				list_del(&event->list);
185 				kfree(event);
186 			}
187 		} else if (event->pe == pe) {
188 			list_del(&event->list);
189 			kfree(event);
190 		}
191 	}
192 	spin_unlock_irqrestore(&eeh_eventlist_lock, flags);
193 }
194