1/******************************************************************************
2 * mcelog.c
3 * Driver for receiving and transferring machine check error infomation
4 *
5 * Copyright (c) 2012 Intel Corporation
6 * Author: Liu, Jinsong <jinsong.liu@intel.com>
7 * Author: Jiang, Yunhong <yunhong.jiang@intel.com>
8 * Author: Ke, Liping <liping.ke@intel.com>
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License version 2
12 * as published by the Free Software Foundation; or, when distributed
13 * separately from the Linux kernel or incorporated into other
14 * software packages, subject to the following license:
15 *
16 * Permission is hereby granted, free of charge, to any person obtaining a copy
17 * of this source file (the "Software"), to deal in the Software without
18 * restriction, including without limitation the rights to use, copy, modify,
19 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
20 * and to permit persons to whom the Software is furnished to do so, subject to
21 * the following conditions:
22 *
23 * The above copyright notice and this permission notice shall be included in
24 * all copies or substantial portions of the Software.
25 *
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
27 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
28 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
29 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
30 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
31 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 * IN THE SOFTWARE.
33 */
34
35#define pr_fmt(fmt) "xen_mcelog: " fmt
36
37#include <linux/init.h>
38#include <linux/types.h>
39#include <linux/kernel.h>
40#include <linux/slab.h>
41#include <linux/fs.h>
42#include <linux/device.h>
43#include <linux/miscdevice.h>
44#include <linux/uaccess.h>
45#include <linux/capability.h>
46#include <linux/poll.h>
47#include <linux/sched.h>
48
49#include <xen/interface/xen.h>
50#include <xen/events.h>
51#include <xen/interface/vcpu.h>
52#include <xen/xen.h>
53#include <asm/xen/hypercall.h>
54#include <asm/xen/hypervisor.h>
55
56static struct mc_info g_mi;
57static struct mcinfo_logical_cpu *g_physinfo;
58static uint32_t ncpus;
59
60static DEFINE_MUTEX(mcelog_lock);
61
62static struct xen_mce_log xen_mcelog = {
63	.signature	= XEN_MCE_LOG_SIGNATURE,
64	.len		= XEN_MCE_LOG_LEN,
65	.recordlen	= sizeof(struct xen_mce),
66};
67
68static DEFINE_SPINLOCK(xen_mce_chrdev_state_lock);
69static int xen_mce_chrdev_open_count;	/* #times opened */
70static int xen_mce_chrdev_open_exclu;	/* already open exclusive? */
71
72static DECLARE_WAIT_QUEUE_HEAD(xen_mce_chrdev_wait);
73
74static int xen_mce_chrdev_open(struct inode *inode, struct file *file)
75{
76	spin_lock(&xen_mce_chrdev_state_lock);
77
78	if (xen_mce_chrdev_open_exclu ||
79	    (xen_mce_chrdev_open_count && (file->f_flags & O_EXCL))) {
80		spin_unlock(&xen_mce_chrdev_state_lock);
81
82		return -EBUSY;
83	}
84
85	if (file->f_flags & O_EXCL)
86		xen_mce_chrdev_open_exclu = 1;
87	xen_mce_chrdev_open_count++;
88
89	spin_unlock(&xen_mce_chrdev_state_lock);
90
91	return nonseekable_open(inode, file);
92}
93
94static int xen_mce_chrdev_release(struct inode *inode, struct file *file)
95{
96	spin_lock(&xen_mce_chrdev_state_lock);
97
98	xen_mce_chrdev_open_count--;
99	xen_mce_chrdev_open_exclu = 0;
100
101	spin_unlock(&xen_mce_chrdev_state_lock);
102
103	return 0;
104}
105
106static ssize_t xen_mce_chrdev_read(struct file *filp, char __user *ubuf,
107				size_t usize, loff_t *off)
108{
109	char __user *buf = ubuf;
110	unsigned num;
111	int i, err;
112
113	mutex_lock(&mcelog_lock);
114
115	num = xen_mcelog.next;
116
117	/* Only supports full reads right now */
118	err = -EINVAL;
119	if (*off != 0 || usize < XEN_MCE_LOG_LEN*sizeof(struct xen_mce))
120		goto out;
121
122	err = 0;
123	for (i = 0; i < num; i++) {
124		struct xen_mce *m = &xen_mcelog.entry[i];
125
126		err |= copy_to_user(buf, m, sizeof(*m));
127		buf += sizeof(*m);
128	}
129
130	memset(xen_mcelog.entry, 0, num * sizeof(struct xen_mce));
131	xen_mcelog.next = 0;
132
133	if (err)
134		err = -EFAULT;
135
136out:
137	mutex_unlock(&mcelog_lock);
138
139	return err ? err : buf - ubuf;
140}
141
142static __poll_t xen_mce_chrdev_poll(struct file *file, poll_table *wait)
143{
144	poll_wait(file, &xen_mce_chrdev_wait, wait);
145
146	if (xen_mcelog.next)
147		return EPOLLIN | EPOLLRDNORM;
148
149	return 0;
150}
151
152static long xen_mce_chrdev_ioctl(struct file *f, unsigned int cmd,
153				unsigned long arg)
154{
155	int __user *p = (int __user *)arg;
156
157	if (!capable(CAP_SYS_ADMIN))
158		return -EPERM;
159
160	switch (cmd) {
161	case MCE_GET_RECORD_LEN:
162		return put_user(sizeof(struct xen_mce), p);
163	case MCE_GET_LOG_LEN:
164		return put_user(XEN_MCE_LOG_LEN, p);
165	case MCE_GETCLEAR_FLAGS: {
166		unsigned flags;
167
168		do {
169			flags = xen_mcelog.flags;
170		} while (cmpxchg(&xen_mcelog.flags, flags, 0) != flags);
171
172		return put_user(flags, p);
173	}
174	default:
175		return -ENOTTY;
176	}
177}
178
179static const struct file_operations xen_mce_chrdev_ops = {
180	.open			= xen_mce_chrdev_open,
181	.release		= xen_mce_chrdev_release,
182	.read			= xen_mce_chrdev_read,
183	.poll			= xen_mce_chrdev_poll,
184	.unlocked_ioctl		= xen_mce_chrdev_ioctl,
185	.llseek			= no_llseek,
186};
187
188static struct miscdevice xen_mce_chrdev_device = {
189	MISC_MCELOG_MINOR,
190	"mcelog",
191	&xen_mce_chrdev_ops,
192};
193
194/*
195 * Caller should hold the mcelog_lock
196 */
197static void xen_mce_log(struct xen_mce *mce)
198{
199	unsigned entry;
200
201	entry = xen_mcelog.next;
202
203	/*
204	 * When the buffer fills up discard new entries.
205	 * Assume that the earlier errors are the more
206	 * interesting ones:
207	 */
208	if (entry >= XEN_MCE_LOG_LEN) {
209		set_bit(XEN_MCE_OVERFLOW,
210			(unsigned long *)&xen_mcelog.flags);
211		return;
212	}
213
214	memcpy(xen_mcelog.entry + entry, mce, sizeof(struct xen_mce));
215
216	xen_mcelog.next++;
217}
218
219static int convert_log(struct mc_info *mi)
220{
221	struct mcinfo_common *mic;
222	struct mcinfo_global *mc_global;
223	struct mcinfo_bank *mc_bank;
224	struct xen_mce m;
225	unsigned int i, j;
226
227	mic = NULL;
228	x86_mcinfo_lookup(&mic, mi, MC_TYPE_GLOBAL);
229	if (unlikely(!mic)) {
230		pr_warn("Failed to find global error info\n");
231		return -ENODEV;
232	}
233
234	memset(&m, 0, sizeof(struct xen_mce));
235
236	mc_global = (struct mcinfo_global *)mic;
237	m.mcgstatus = mc_global->mc_gstatus;
238	m.apicid = mc_global->mc_apicid;
239
240	for (i = 0; i < ncpus; i++)
241		if (g_physinfo[i].mc_apicid == m.apicid)
242			break;
243	if (unlikely(i == ncpus)) {
244		pr_warn("Failed to match cpu with apicid %d\n", m.apicid);
245		return -ENODEV;
246	}
247
248	m.socketid = g_physinfo[i].mc_chipid;
249	m.cpu = m.extcpu = g_physinfo[i].mc_cpunr;
250	m.cpuvendor = (__u8)g_physinfo[i].mc_vendor;
251	for (j = 0; j < g_physinfo[i].mc_nmsrvals; ++j)
252		switch (g_physinfo[i].mc_msrvalues[j].reg) {
253		case MSR_IA32_MCG_CAP:
254			m.mcgcap = g_physinfo[i].mc_msrvalues[j].value;
255			break;
256
257		case MSR_PPIN:
258		case MSR_AMD_PPIN:
259			m.ppin = g_physinfo[i].mc_msrvalues[j].value;
260			break;
261		}
262
263	mic = NULL;
264	x86_mcinfo_lookup(&mic, mi, MC_TYPE_BANK);
265	if (unlikely(!mic)) {
266		pr_warn("Fail to find bank error info\n");
267		return -ENODEV;
268	}
269
270	do {
271		if ((!mic) || (mic->size == 0) ||
272		    (mic->type != MC_TYPE_GLOBAL   &&
273		     mic->type != MC_TYPE_BANK     &&
274		     mic->type != MC_TYPE_EXTENDED &&
275		     mic->type != MC_TYPE_RECOVERY))
276			break;
277
278		if (mic->type == MC_TYPE_BANK) {
279			mc_bank = (struct mcinfo_bank *)mic;
280			m.misc = mc_bank->mc_misc;
281			m.status = mc_bank->mc_status;
282			m.addr = mc_bank->mc_addr;
283			m.tsc = mc_bank->mc_tsc;
284			m.bank = mc_bank->mc_bank;
285			m.finished = 1;
286			/*log this record*/
287			xen_mce_log(&m);
288		}
289		mic = x86_mcinfo_next(mic);
290	} while (1);
291
292	return 0;
293}
294
295static int mc_queue_handle(uint32_t flags)
296{
297	struct xen_mc mc_op;
298	int ret = 0;
299
300	mc_op.cmd = XEN_MC_fetch;
301	set_xen_guest_handle(mc_op.u.mc_fetch.data, &g_mi);
302	do {
303		mc_op.u.mc_fetch.flags = flags;
304		ret = HYPERVISOR_mca(&mc_op);
305		if (ret) {
306			pr_err("Failed to fetch %surgent error log\n",
307			       flags == XEN_MC_URGENT ? "" : "non");
308			break;
309		}
310
311		if (mc_op.u.mc_fetch.flags & XEN_MC_NODATA ||
312		    mc_op.u.mc_fetch.flags & XEN_MC_FETCHFAILED)
313			break;
314		else {
315			ret = convert_log(&g_mi);
316			if (ret)
317				pr_warn("Failed to convert this error log, continue acking it anyway\n");
318
319			mc_op.u.mc_fetch.flags = flags | XEN_MC_ACK;
320			ret = HYPERVISOR_mca(&mc_op);
321			if (ret) {
322				pr_err("Failed to ack previous error log\n");
323				break;
324			}
325		}
326	} while (1);
327
328	return ret;
329}
330
331/* virq handler for machine check error info*/
332static void xen_mce_work_fn(struct work_struct *work)
333{
334	int err;
335
336	mutex_lock(&mcelog_lock);
337
338	/* urgent mc_info */
339	err = mc_queue_handle(XEN_MC_URGENT);
340	if (err)
341		pr_err("Failed to handle urgent mc_info queue, continue handling nonurgent mc_info queue anyway\n");
342
343	/* nonurgent mc_info */
344	err = mc_queue_handle(XEN_MC_NONURGENT);
345	if (err)
346		pr_err("Failed to handle nonurgent mc_info queue\n");
347
348	/* wake processes polling /dev/mcelog */
349	wake_up_interruptible(&xen_mce_chrdev_wait);
350
351	mutex_unlock(&mcelog_lock);
352}
353static DECLARE_WORK(xen_mce_work, xen_mce_work_fn);
354
355static irqreturn_t xen_mce_interrupt(int irq, void *dev_id)
356{
357	schedule_work(&xen_mce_work);
358	return IRQ_HANDLED;
359}
360
361static int bind_virq_for_mce(void)
362{
363	int ret;
364	struct xen_mc mc_op;
365
366	memset(&mc_op, 0, sizeof(struct xen_mc));
367
368	/* Fetch physical CPU Numbers */
369	mc_op.cmd = XEN_MC_physcpuinfo;
370	set_xen_guest_handle(mc_op.u.mc_physcpuinfo.info, g_physinfo);
371	ret = HYPERVISOR_mca(&mc_op);
372	if (ret) {
373		pr_err("Failed to get CPU numbers\n");
374		return ret;
375	}
376
377	/* Fetch each CPU Physical Info for later reference*/
378	ncpus = mc_op.u.mc_physcpuinfo.ncpus;
379	g_physinfo = kcalloc(ncpus, sizeof(struct mcinfo_logical_cpu),
380			     GFP_KERNEL);
381	if (!g_physinfo)
382		return -ENOMEM;
383	set_xen_guest_handle(mc_op.u.mc_physcpuinfo.info, g_physinfo);
384	ret = HYPERVISOR_mca(&mc_op);
385	if (ret) {
386		pr_err("Failed to get CPU info\n");
387		kfree(g_physinfo);
388		return ret;
389	}
390
391	ret  = bind_virq_to_irqhandler(VIRQ_MCA, 0,
392				       xen_mce_interrupt, 0, "mce", NULL);
393	if (ret < 0) {
394		pr_err("Failed to bind virq\n");
395		kfree(g_physinfo);
396		return ret;
397	}
398
399	return 0;
400}
401
402static int __init xen_late_init_mcelog(void)
403{
404	int ret;
405
406	/* Only DOM0 is responsible for MCE logging */
407	if (!xen_initial_domain())
408		return -ENODEV;
409
410	/* register character device /dev/mcelog for xen mcelog */
411	ret = misc_register(&xen_mce_chrdev_device);
412	if (ret)
413		return ret;
414
415	ret = bind_virq_for_mce();
416	if (ret)
417		goto deregister;
418
419	pr_info("/dev/mcelog registered by Xen\n");
420
421	return 0;
422
423deregister:
424	misc_deregister(&xen_mce_chrdev_device);
425	return ret;
426}
427device_initcall(xen_late_init_mcelog);
428