PageRenderTime 58ms CodeModel.GetById 17ms app.highlight 35ms RepoModel.GetById 1ms app.codeStats 0ms

/arch/ia64/kernel/salinfo.c

https://bitbucket.org/evzijst/gittest
C | 629 lines | 447 code | 75 blank | 107 comment | 85 complexity | eac085d045f71b56e9a51964fe4628dc MD5 | raw file
  1/*
  2 * salinfo.c
  3 *
  4 * Creates entries in /proc/sal for various system features.
  5 *
  6 * Copyright (c) 2003 Silicon Graphics, Inc.  All rights reserved.
  7 * Copyright (c) 2003 Hewlett-Packard Co
  8 *	Bjorn Helgaas <bjorn.helgaas@hp.com>
  9 *
 10 * 10/30/2001	jbarnes@sgi.com		copied much of Stephane's palinfo
 11 *					code to create this file
 12 * Oct 23 2003	kaos@sgi.com
 13 *   Replace IPI with set_cpus_allowed() to read a record from the required cpu.
 14 *   Redesign salinfo log processing to separate interrupt and user space
 15 *   contexts.
 16 *   Cache the record across multi-block reads from user space.
 17 *   Support > 64 cpus.
 18 *   Delete module_exit and MOD_INC/DEC_COUNT, salinfo cannot be a module.
 19 *
 20 * Jan 28 2004	kaos@sgi.com
 21 *   Periodically check for outstanding MCA or INIT records.
 22 *
 23 * Dec  5 2004	kaos@sgi.com
 24 *   Standardize which records are cleared automatically.
 25 */
 26
 27#include <linux/types.h>
 28#include <linux/proc_fs.h>
 29#include <linux/module.h>
 30#include <linux/smp.h>
 31#include <linux/smp_lock.h>
 32#include <linux/timer.h>
 33#include <linux/vmalloc.h>
 34
 35#include <asm/semaphore.h>
 36#include <asm/sal.h>
 37#include <asm/uaccess.h>
 38
 39MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
 40MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
 41MODULE_LICENSE("GPL");
 42
 43static int salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data);
 44
 45typedef struct {
 46	const char		*name;		/* name of the proc entry */
 47	unsigned long           feature;        /* feature bit */
 48	struct proc_dir_entry	*entry;		/* registered entry (removal) */
 49} salinfo_entry_t;
 50
 51/*
 52 * List {name,feature} pairs for every entry in /proc/sal/<feature>
 53 * that this module exports
 54 */
 55static salinfo_entry_t salinfo_entries[]={
 56	{ "bus_lock",           IA64_SAL_PLATFORM_FEATURE_BUS_LOCK, },
 57	{ "irq_redirection",	IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT, },
 58	{ "ipi_redirection",	IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT, },
 59	{ "itc_drift",		IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT, },
 60};
 61
 62#define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)
 63
 64static char *salinfo_log_name[] = {
 65	"mca",
 66	"init",
 67	"cmc",
 68	"cpe",
 69};
 70
 71static struct proc_dir_entry *salinfo_proc_entries[
 72	ARRAY_SIZE(salinfo_entries) +			/* /proc/sal/bus_lock */
 73	ARRAY_SIZE(salinfo_log_name) +			/* /proc/sal/{mca,...} */
 74	(2 * ARRAY_SIZE(salinfo_log_name)) +		/* /proc/sal/mca/{event,data} */
 75	1];						/* /proc/sal */
 76
 77/* Some records we get ourselves, some are accessed as saved data in buffers
 78 * that are owned by mca.c.
 79 */
 80struct salinfo_data_saved {
 81	u8*			buffer;
 82	u64			size;
 83	u64			id;
 84	int			cpu;
 85};
 86
 87/* State transitions.  Actions are :-
 88 *   Write "read <cpunum>" to the data file.
 89 *   Write "clear <cpunum>" to the data file.
 90 *   Write "oemdata <cpunum> <offset> to the data file.
 91 *   Read from the data file.
 92 *   Close the data file.
 93 *
 94 * Start state is NO_DATA.
 95 *
 96 * NO_DATA
 97 *    write "read <cpunum>" -> NO_DATA or LOG_RECORD.
 98 *    write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
 99 *    write "oemdata <cpunum> <offset> -> return -EINVAL.
100 *    read data -> return EOF.
101 *    close -> unchanged.  Free record areas.
102 *
103 * LOG_RECORD
104 *    write "read <cpunum>" -> NO_DATA or LOG_RECORD.
105 *    write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
106 *    write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
107 *    read data -> return the INIT/MCA/CMC/CPE record.
108 *    close -> unchanged.  Keep record areas.
109 *
110 * OEMDATA
111 *    write "read <cpunum>" -> NO_DATA or LOG_RECORD.
112 *    write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
113 *    write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
114 *    read data -> return the formatted oemdata.
115 *    close -> unchanged.  Keep record areas.
116 *
117 * Closing the data file does not change the state.  This allows shell scripts
118 * to manipulate salinfo data, each shell redirection opens the file, does one
119 * action then closes it again.  The record areas are only freed at close when
120 * the state is NO_DATA.
121 */
122enum salinfo_state {
123	STATE_NO_DATA,
124	STATE_LOG_RECORD,
125	STATE_OEMDATA,
126};
127
128struct salinfo_data {
129	volatile cpumask_t	cpu_event;	/* which cpus have outstanding events */
130	struct semaphore	sem;		/* count of cpus with outstanding events (bits set in cpu_event) */
131	u8			*log_buffer;
132	u64			log_size;
133	u8			*oemdata;	/* decoded oem data */
134	u64			oemdata_size;
135	int			open;		/* single-open to prevent races */
136	u8			type;
137	u8			saved_num;	/* using a saved record? */
138	enum salinfo_state	state :8;	/* processing state */
139	u8			padding;
140	int			cpu_check;	/* next CPU to check */
141	struct salinfo_data_saved data_saved[5];/* save last 5 records from mca.c, must be < 255 */
142};
143
144static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];
145
146static spinlock_t data_lock, data_saved_lock;
147
148/** salinfo_platform_oemdata - optional callback to decode oemdata from an error
149 * record.
150 * @sect_header: pointer to the start of the section to decode.
151 * @oemdata: returns vmalloc area containing the decded output.
152 * @oemdata_size: returns length of decoded output (strlen).
153 *
154 * Description: If user space asks for oem data to be decoded by the kernel
155 * and/or prom and the platform has set salinfo_platform_oemdata to the address
156 * of a platform specific routine then call that routine.  salinfo_platform_oemdata
157 * vmalloc's and formats its output area, returning the address of the text
158 * and its strlen.  Returns 0 for success, -ve for error.  The callback is
159 * invoked on the cpu that generated the error record.
160 */
161int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);
162
163struct salinfo_platform_oemdata_parms {
164	const u8 *efi_guid;
165	u8 **oemdata;
166	u64 *oemdata_size;
167	int ret;
168};
169
170static void
171salinfo_platform_oemdata_cpu(void *context)
172{
173	struct salinfo_platform_oemdata_parms *parms = context;
174	parms->ret = salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
175}
176
177static void
178shift1_data_saved (struct salinfo_data *data, int shift)
179{
180	memcpy(data->data_saved+shift, data->data_saved+shift+1,
181	       (ARRAY_SIZE(data->data_saved) - (shift+1)) * sizeof(data->data_saved[0]));
182	memset(data->data_saved + ARRAY_SIZE(data->data_saved) - 1, 0,
183	       sizeof(data->data_saved[0]));
184}
185
186/* This routine is invoked in interrupt context.  Note: mca.c enables
187 * interrupts before calling this code for CMC/CPE.  MCA and INIT events are
188 * not irq safe, do not call any routines that use spinlocks, they may deadlock.
189 * MCA and INIT records are recorded, a timer event will look for any
190 * outstanding events and wake up the user space code.
191 *
192 * The buffer passed from mca.c points to the output from ia64_log_get. This is
193 * a persistent buffer but its contents can change between the interrupt and
194 * when user space processes the record.  Save the record id to identify
195 * changes.
196 */
197void
198salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
199{
200	struct salinfo_data *data = salinfo_data + type;
201	struct salinfo_data_saved *data_saved;
202	unsigned long flags = 0;
203	int i;
204	int saved_size = ARRAY_SIZE(data->data_saved);
205
206	BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
207
208	if (irqsafe)
209		spin_lock_irqsave(&data_saved_lock, flags);
210	for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
211		if (!data_saved->buffer)
212			break;
213	}
214	if (i == saved_size) {
215		if (!data->saved_num) {
216			shift1_data_saved(data, 0);
217			data_saved = data->data_saved + saved_size - 1;
218		} else
219			data_saved = NULL;
220	}
221	if (data_saved) {
222		data_saved->cpu = smp_processor_id();
223		data_saved->id = ((sal_log_record_header_t *)buffer)->id;
224		data_saved->size = size;
225		data_saved->buffer = buffer;
226	}
227	if (irqsafe)
228		spin_unlock_irqrestore(&data_saved_lock, flags);
229
230	if (!test_and_set_bit(smp_processor_id(), &data->cpu_event)) {
231		if (irqsafe)
232			up(&data->sem);
233	}
234}
235
236/* Check for outstanding MCA/INIT records every minute (arbitrary) */
237#define SALINFO_TIMER_DELAY (60*HZ)
238static struct timer_list salinfo_timer;
239
240static void
241salinfo_timeout_check(struct salinfo_data *data)
242{
243	int i;
244	if (!data->open)
245		return;
246	for (i = 0; i < NR_CPUS; ++i) {
247		if (test_bit(i, &data->cpu_event)) {
248			/* double up() is not a problem, user space will see no
249			 * records for the additional "events".
250			 */
251			up(&data->sem);
252		}
253	}
254}
255
256static void 
257salinfo_timeout (unsigned long arg)
258{
259	salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
260	salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_INIT);
261	salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
262	add_timer(&salinfo_timer);
263}
264
265static int
266salinfo_event_open(struct inode *inode, struct file *file)
267{
268	if (!capable(CAP_SYS_ADMIN))
269		return -EPERM;
270	return 0;
271}
272
273static ssize_t
274salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
275{
276	struct inode *inode = file->f_dentry->d_inode;
277	struct proc_dir_entry *entry = PDE(inode);
278	struct salinfo_data *data = entry->data;
279	char cmd[32];
280	size_t size;
281	int i, n, cpu = -1;
282
283retry:
284	if (down_trylock(&data->sem)) {
285		if (file->f_flags & O_NONBLOCK)
286			return -EAGAIN;
287		if (down_interruptible(&data->sem))
288			return -ERESTARTSYS;
289	}
290
291	n = data->cpu_check;
292	for (i = 0; i < NR_CPUS; i++) {
293		if (test_bit(n, &data->cpu_event)) {
294			cpu = n;
295			break;
296		}
297		if (++n == NR_CPUS)
298			n = 0;
299	}
300
301	if (cpu == -1)
302		goto retry;
303
304	/* events are sticky until the user says "clear" */
305	up(&data->sem);
306
307	/* for next read, start checking at next CPU */
308	data->cpu_check = cpu;
309	if (++data->cpu_check == NR_CPUS)
310		data->cpu_check = 0;
311
312	snprintf(cmd, sizeof(cmd), "read %d\n", cpu);
313
314	size = strlen(cmd);
315	if (size > count)
316		size = count;
317	if (copy_to_user(buffer, cmd, size))
318		return -EFAULT;
319
320	return size;
321}
322
323static struct file_operations salinfo_event_fops = {
324	.open  = salinfo_event_open,
325	.read  = salinfo_event_read,
326};
327
328static int
329salinfo_log_open(struct inode *inode, struct file *file)
330{
331	struct proc_dir_entry *entry = PDE(inode);
332	struct salinfo_data *data = entry->data;
333
334	if (!capable(CAP_SYS_ADMIN))
335		return -EPERM;
336
337	spin_lock(&data_lock);
338	if (data->open) {
339		spin_unlock(&data_lock);
340		return -EBUSY;
341	}
342	data->open = 1;
343	spin_unlock(&data_lock);
344
345	if (data->state == STATE_NO_DATA &&
346	    !(data->log_buffer = vmalloc(ia64_sal_get_state_info_size(data->type)))) {
347		data->open = 0;
348		return -ENOMEM;
349	}
350
351	return 0;
352}
353
354static int
355salinfo_log_release(struct inode *inode, struct file *file)
356{
357	struct proc_dir_entry *entry = PDE(inode);
358	struct salinfo_data *data = entry->data;
359
360	if (data->state == STATE_NO_DATA) {
361		vfree(data->log_buffer);
362		vfree(data->oemdata);
363		data->log_buffer = NULL;
364		data->oemdata = NULL;
365	}
366	spin_lock(&data_lock);
367	data->open = 0;
368	spin_unlock(&data_lock);
369	return 0;
370}
371
372static void
373call_on_cpu(int cpu, void (*fn)(void *), void *arg)
374{
375	cpumask_t save_cpus_allowed, new_cpus_allowed;
376	memcpy(&save_cpus_allowed, &current->cpus_allowed, sizeof(save_cpus_allowed));
377	memset(&new_cpus_allowed, 0, sizeof(new_cpus_allowed));
378	set_bit(cpu, &new_cpus_allowed);
379	set_cpus_allowed(current, new_cpus_allowed);
380	(*fn)(arg);
381	set_cpus_allowed(current, save_cpus_allowed);
382}
383
384static void
385salinfo_log_read_cpu(void *context)
386{
387	struct salinfo_data *data = context;
388	sal_log_record_header_t *rh;
389	data->log_size = ia64_sal_get_state_info(data->type, (u64 *) data->log_buffer);
390	rh = (sal_log_record_header_t *)(data->log_buffer);
391	/* Clear corrected errors as they are read from SAL */
392	if (rh->severity == sal_log_severity_corrected)
393		ia64_sal_clear_state_info(data->type);
394}
395
396static void
397salinfo_log_new_read(int cpu, struct salinfo_data *data)
398{
399	struct salinfo_data_saved *data_saved;
400	unsigned long flags;
401	int i;
402	int saved_size = ARRAY_SIZE(data->data_saved);
403
404	data->saved_num = 0;
405	spin_lock_irqsave(&data_saved_lock, flags);
406retry:
407	for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
408		if (data_saved->buffer && data_saved->cpu == cpu) {
409			sal_log_record_header_t *rh = (sal_log_record_header_t *)(data_saved->buffer);
410			data->log_size = data_saved->size;
411			memcpy(data->log_buffer, rh, data->log_size);
412			barrier();	/* id check must not be moved */
413			if (rh->id == data_saved->id) {
414				data->saved_num = i+1;
415				break;
416			}
417			/* saved record changed by mca.c since interrupt, discard it */
418			shift1_data_saved(data, i);
419			goto retry;
420		}
421	}
422	spin_unlock_irqrestore(&data_saved_lock, flags);
423
424	if (!data->saved_num)
425		call_on_cpu(cpu, salinfo_log_read_cpu, data);
426	if (!data->log_size) {
427	        data->state = STATE_NO_DATA;
428	        clear_bit(cpu, &data->cpu_event);
429	} else {
430	        data->state = STATE_LOG_RECORD;
431	}
432}
433
434static ssize_t
435salinfo_log_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
436{
437	struct inode *inode = file->f_dentry->d_inode;
438	struct proc_dir_entry *entry = PDE(inode);
439	struct salinfo_data *data = entry->data;
440	u8 *buf;
441	u64 bufsize;
442
443	if (data->state == STATE_LOG_RECORD) {
444		buf = data->log_buffer;
445		bufsize = data->log_size;
446	} else if (data->state == STATE_OEMDATA) {
447		buf = data->oemdata;
448		bufsize = data->oemdata_size;
449	} else {
450		buf = NULL;
451		bufsize = 0;
452	}
453	return simple_read_from_buffer(buffer, count, ppos, buf, bufsize);
454}
455
456static void
457salinfo_log_clear_cpu(void *context)
458{
459	struct salinfo_data *data = context;
460	ia64_sal_clear_state_info(data->type);
461}
462
463static int
464salinfo_log_clear(struct salinfo_data *data, int cpu)
465{
466	sal_log_record_header_t *rh;
467	data->state = STATE_NO_DATA;
468	if (!test_bit(cpu, &data->cpu_event))
469		return 0;
470	down(&data->sem);
471	clear_bit(cpu, &data->cpu_event);
472	if (data->saved_num) {
473		unsigned long flags;
474		spin_lock_irqsave(&data_saved_lock, flags);
475		shift1_data_saved(data, data->saved_num - 1 );
476		data->saved_num = 0;
477		spin_unlock_irqrestore(&data_saved_lock, flags);
478	}
479	rh = (sal_log_record_header_t *)(data->log_buffer);
480	/* Corrected errors have already been cleared from SAL */
481	if (rh->severity != sal_log_severity_corrected)
482		call_on_cpu(cpu, salinfo_log_clear_cpu, data);
483	/* clearing a record may make a new record visible */
484	salinfo_log_new_read(cpu, data);
485	if (data->state == STATE_LOG_RECORD &&
486	    !test_and_set_bit(cpu,  &data->cpu_event))
487		up(&data->sem);
488	return 0;
489}
490
491static ssize_t
492salinfo_log_write(struct file *file, const char __user *buffer, size_t count, loff_t *ppos)
493{
494	struct inode *inode = file->f_dentry->d_inode;
495	struct proc_dir_entry *entry = PDE(inode);
496	struct salinfo_data *data = entry->data;
497	char cmd[32];
498	size_t size;
499	u32 offset;
500	int cpu;
501
502	size = sizeof(cmd);
503	if (count < size)
504		size = count;
505	if (copy_from_user(cmd, buffer, size))
506		return -EFAULT;
507
508	if (sscanf(cmd, "read %d", &cpu) == 1) {
509		salinfo_log_new_read(cpu, data);
510	} else if (sscanf(cmd, "clear %d", &cpu) == 1) {
511		int ret;
512		if ((ret = salinfo_log_clear(data, cpu)))
513			count = ret;
514	} else if (sscanf(cmd, "oemdata %d %d", &cpu, &offset) == 2) {
515		if (data->state != STATE_LOG_RECORD && data->state != STATE_OEMDATA)
516			return -EINVAL;
517		if (offset > data->log_size - sizeof(efi_guid_t))
518			return -EINVAL;
519		data->state = STATE_OEMDATA;
520		if (salinfo_platform_oemdata) {
521			struct salinfo_platform_oemdata_parms parms = {
522				.efi_guid = data->log_buffer + offset,
523				.oemdata = &data->oemdata,
524				.oemdata_size = &data->oemdata_size
525			};
526			call_on_cpu(cpu, salinfo_platform_oemdata_cpu, &parms);
527			if (parms.ret)
528				count = parms.ret;
529		} else
530			data->oemdata_size = 0;
531	} else
532		return -EINVAL;
533
534	return count;
535}
536
537static struct file_operations salinfo_data_fops = {
538	.open    = salinfo_log_open,
539	.release = salinfo_log_release,
540	.read    = salinfo_log_read,
541	.write   = salinfo_log_write,
542};
543
544static int __init
545salinfo_init(void)
546{
547	struct proc_dir_entry *salinfo_dir; /* /proc/sal dir entry */
548	struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
549	struct proc_dir_entry *dir, *entry;
550	struct salinfo_data *data;
551	int i, j, online;
552
553	salinfo_dir = proc_mkdir("sal", NULL);
554	if (!salinfo_dir)
555		return 0;
556
557	for (i=0; i < NR_SALINFO_ENTRIES; i++) {
558		/* pass the feature bit in question as misc data */
559		*sdir++ = create_proc_read_entry (salinfo_entries[i].name, 0, salinfo_dir,
560						  salinfo_read, (void *)salinfo_entries[i].feature);
561	}
562
563	for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
564		data = salinfo_data + i;
565		data->type = i;
566		sema_init(&data->sem, 0);
567		dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
568		if (!dir)
569			continue;
570
571		entry = create_proc_entry("event", S_IRUSR, dir);
572		if (!entry)
573			continue;
574		entry->data = data;
575		entry->proc_fops = &salinfo_event_fops;
576		*sdir++ = entry;
577
578		entry = create_proc_entry("data", S_IRUSR | S_IWUSR, dir);
579		if (!entry)
580			continue;
581		entry->data = data;
582		entry->proc_fops = &salinfo_data_fops;
583		*sdir++ = entry;
584
585		/* we missed any events before now */
586		online = 0;
587		for (j = 0; j < NR_CPUS; j++)
588			if (cpu_online(j)) {
589				set_bit(j, &data->cpu_event);
590				++online;
591			}
592		sema_init(&data->sem, online);
593
594		*sdir++ = dir;
595	}
596
597	*sdir++ = salinfo_dir;
598
599	init_timer(&salinfo_timer);
600	salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
601	salinfo_timer.function = &salinfo_timeout;
602	add_timer(&salinfo_timer);
603
604	return 0;
605}
606
607/*
608 * 'data' contains an integer that corresponds to the feature we're
609 * testing
610 */
611static int
612salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data)
613{
614	int len = 0;
615
616	len = sprintf(page, (sal_platform_features & (unsigned long)data) ? "1\n" : "0\n");
617
618	if (len <= off+count) *eof = 1;
619
620	*start = page + off;
621	len   -= off;
622
623	if (len>count) len = count;
624	if (len<0) len = 0;
625
626	return len;
627}
628
629module_init(salinfo_init);