1/*
  2 * Machine check injection support.
  3 * Copyright 2008 Intel Corporation.
  4 *
  5 * This program is free software; you can redistribute it and/or
  6 * modify it under the terms of the GNU General Public License
  7 * as published by the Free Software Foundation; version 2
  8 * of the License.
  9 *
 10 * Authors:
 11 * Andi Kleen
 12 * Ying Huang
 13 */
 14#include <linux/uaccess.h>
 15#include <linux/module.h>
 16#include <linux/timer.h>
 17#include <linux/kernel.h>
 18#include <linux/string.h>
 19#include <linux/fs.h>
 20#include <linux/preempt.h>
 21#include <linux/smp.h>
 22#include <linux/notifier.h>
 23#include <linux/kdebug.h>
 24#include <linux/cpu.h>
 25#include <linux/sched.h>
 26#include <linux/gfp.h>
 27#include <asm/mce.h>
 28#include <asm/apic.h>
 29#include <asm/nmi.h>
 30
 31/* Update fake mce registers on current CPU. */
 32static void inject_mce(struct mce *m)
 33{
 34	struct mce *i = &per_cpu(injectm, m->extcpu);
 35
 36	/* Make sure no one reads partially written injectm */
 37	i->finished = 0;
 38	mb();
 39	m->finished = 0;
 40	/* First set the fields after finished */
 41	i->extcpu = m->extcpu;
 42	mb();
 43	/* Now write record in order, finished last (except above) */
 44	memcpy(i, m, sizeof(struct mce));
 45	/* Finally activate it */
 46	mb();
 47	i->finished = 1;
 48}
 49
 50static void raise_poll(struct mce *m)
 51{
 52	unsigned long flags;
 53	mce_banks_t b;
 54
 55	memset(&b, 0xff, sizeof(mce_banks_t));
 56	local_irq_save(flags);
 57	machine_check_poll(0, &b);
 58	local_irq_restore(flags);
 59	m->finished = 0;
 60}
 61
 62static void raise_exception(struct mce *m, struct pt_regs *pregs)
 63{
 64	struct pt_regs regs;
 65	unsigned long flags;
 66
 67	if (!pregs) {
 68		memset(&regs, 0, sizeof(struct pt_regs));
 69		regs.ip = m->ip;
 70		regs.cs = m->cs;
 71		pregs = &regs;
 72	}
 73	/* in mcheck exeception handler, irq will be disabled */
 74	local_irq_save(flags);
 75	do_machine_check(pregs, 0);
 76	local_irq_restore(flags);
 77	m->finished = 0;
 78}
 79
 80static cpumask_var_t mce_inject_cpumask;
 81
 82static int mce_raise_notify(unsigned int cmd, struct pt_regs *regs)
 83{
 84	int cpu = smp_processor_id();
 85	struct mce *m = &__get_cpu_var(injectm);
 86	if (!cpumask_test_cpu(cpu, mce_inject_cpumask))
 87		return NMI_DONE;
 88	cpumask_clear_cpu(cpu, mce_inject_cpumask);
 89	if (m->inject_flags & MCJ_EXCEPTION)
 90		raise_exception(m, regs);
 91	else if (m->status)
 92		raise_poll(m);
 93	return NMI_HANDLED;
 94}
 95
 96static void mce_irq_ipi(void *info)
 97{
 98	int cpu = smp_processor_id();
 99	struct mce *m = &__get_cpu_var(injectm);
100
101	if (cpumask_test_cpu(cpu, mce_inject_cpumask) &&
102			m->inject_flags & MCJ_EXCEPTION) {
103		cpumask_clear_cpu(cpu, mce_inject_cpumask);
104		raise_exception(m, NULL);
105	}
106}
107
108/* Inject mce on current CPU */
109static int raise_local(void)
110{
111	struct mce *m = &__get_cpu_var(injectm);
112	int context = MCJ_CTX(m->inject_flags);
113	int ret = 0;
114	int cpu = m->extcpu;
115
116	if (m->inject_flags & MCJ_EXCEPTION) {
117		printk(KERN_INFO "Triggering MCE exception on CPU %d\n", cpu);
118		switch (context) {
119		case MCJ_CTX_IRQ:
120			/*
121			 * Could do more to fake interrupts like
122			 * calling irq_enter, but the necessary
123			 * machinery isn't exported currently.
124			 */
125			/*FALL THROUGH*/
126		case MCJ_CTX_PROCESS:
127			raise_exception(m, NULL);
128			break;
129		default:
130			printk(KERN_INFO "Invalid MCE context\n");
131			ret = -EINVAL;
132		}
133		printk(KERN_INFO "MCE exception done on CPU %d\n", cpu);
134	} else if (m->status) {
135		printk(KERN_INFO "Starting machine check poll CPU %d\n", cpu);
136		raise_poll(m);
137		mce_notify_irq();
138		printk(KERN_INFO "Machine check poll done on CPU %d\n", cpu);
139	} else
140		m->finished = 0;
141
142	return ret;
143}
144
145static void raise_mce(struct mce *m)
146{
147	int context = MCJ_CTX(m->inject_flags);
148
149	inject_mce(m);
150
151	if (context == MCJ_CTX_RANDOM)
152		return;
153
154#ifdef CONFIG_X86_LOCAL_APIC
155	if (m->inject_flags & (MCJ_IRQ_BRAODCAST | MCJ_NMI_BROADCAST)) {
156		unsigned long start;
157		int cpu;
158
159		get_online_cpus();
160		cpumask_copy(mce_inject_cpumask, cpu_online_mask);
161		cpumask_clear_cpu(get_cpu(), mce_inject_cpumask);
162		for_each_online_cpu(cpu) {
163			struct mce *mcpu = &per_cpu(injectm, cpu);
164			if (!mcpu->finished ||
165			    MCJ_CTX(mcpu->inject_flags) != MCJ_CTX_RANDOM)
166				cpumask_clear_cpu(cpu, mce_inject_cpumask);
167		}
168		if (!cpumask_empty(mce_inject_cpumask)) {
169			if (m->inject_flags & MCJ_IRQ_BRAODCAST) {
170				/*
171				 * don't wait because mce_irq_ipi is necessary
172				 * to be sync with following raise_local
173				 */
174				preempt_disable();
175				smp_call_function_many(mce_inject_cpumask,
176					mce_irq_ipi, NULL, 0);
177				preempt_enable();
178			} else if (m->inject_flags & MCJ_NMI_BROADCAST)
179				apic->send_IPI_mask(mce_inject_cpumask,
180						NMI_VECTOR);
181		}
182		start = jiffies;
183		while (!cpumask_empty(mce_inject_cpumask)) {
184			if (!time_before(jiffies, start + 2*HZ)) {
185				printk(KERN_ERR
186				"Timeout waiting for mce inject %lx\n",
187					*cpumask_bits(mce_inject_cpumask));
188				break;
189			}
190			cpu_relax();
191		}
192		raise_local();
193		put_cpu();
194		put_online_cpus();
195	} else
196#endif
197		raise_local();
198}
199
200/* Error injection interface */
201static ssize_t mce_write(struct file *filp, const char __user *ubuf,
202			 size_t usize, loff_t *off)
203{
204	struct mce m;
205
206	if (!capable(CAP_SYS_ADMIN))
207		return -EPERM;
208	/*
209	 * There are some cases where real MSR reads could slip
210	 * through.
211	 */
212	if (!boot_cpu_has(X86_FEATURE_MCE) || !boot_cpu_has(X86_FEATURE_MCA))
213		return -EIO;
214
215	if ((unsigned long)usize > sizeof(struct mce))
216		usize = sizeof(struct mce);
217	if (copy_from_user(&m, ubuf, usize))
218		return -EFAULT;
219
220	if (m.extcpu >= num_possible_cpus() || !cpu_online(m.extcpu))
221		return -EINVAL;
222
223	/*
224	 * Need to give user space some time to set everything up,
225	 * so do it a jiffie or two later everywhere.
226	 */
227	schedule_timeout(2);
228	raise_mce(&m);
229	return usize;
230}
231
232static int inject_init(void)
233{
234	if (!alloc_cpumask_var(&mce_inject_cpumask, GFP_KERNEL))
235		return -ENOMEM;
236	printk(KERN_INFO "Machine check injector initialized\n");
237	register_mce_write_callback(mce_write);
238	register_nmi_handler(NMI_LOCAL, mce_raise_notify, 0,
239				"mce_notify");
240	return 0;
241}
242
243module_init(inject_init);
244/*
245 * Cannot tolerate unloading currently because we cannot
246 * guarantee all openers of mce_chrdev will get a reference to us.
247 */
248MODULE_LICENSE("GPL");