4 * Creates entries in /proc/sal for various system features.
6 * Copyright (c) 2003 Silicon Graphics, Inc. All rights reserved.
7 * Copyright (c) 2003 Hewlett-Packard Co
8 * Bjorn Helgaas <bjorn.helgaas@hp.com>
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
16 * Cache the record across multi-block reads from user space.
18 * Delete module_exit and MOD_INC/DEC_COUNT, salinfo cannot be a module.
20 * Jan 28 2004 kaos@sgi.com
21 * Periodically check for outstanding MCA or INIT records.
23 * Dec 5 2004 kaos@sgi.com
24 * Standardize which records are cleared automatically.
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>
35 #include <asm/semaphore.h>
37 #include <asm/uaccess.h>
39 MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
40 MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
41 MODULE_LICENSE("GPL");
43 static int salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data);
46 const char *name; /* name of the proc entry */
47 unsigned long feature; /* feature bit */
48 struct proc_dir_entry *entry; /* registered entry (removal) */
52 * List {name,feature} pairs for every entry in /proc/sal/<feature>
53 * that this module exports
55 static 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, },
62 #define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)
64 static char *salinfo_log_name[] = {
71 static 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} */
77 /* Some records we get ourselves, some are accessed as saved data in buffers
78 * that are owned by mca.c.
80 struct salinfo_data_saved {
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.
94 * Start state is 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.
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.
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.
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.
128 struct 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) */
133 u8 *oemdata; /* decoded oem data */
135 int open; /* single-open to prevent races */
137 u8 saved_num; /* using a saved record? */
138 enum salinfo_state state :8; /* processing state */
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 */
144 static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];
146 static DEFINE_SPINLOCK(data_lock);
147 static DEFINE_SPINLOCK(data_saved_lock);
149 /** salinfo_platform_oemdata - optional callback to decode oemdata from an error
151 * @sect_header: pointer to the start of the section to decode.
152 * @oemdata: returns vmalloc area containing the decded output.
153 * @oemdata_size: returns length of decoded output (strlen).
155 * Description: If user space asks for oem data to be decoded by the kernel
156 * and/or prom and the platform has set salinfo_platform_oemdata to the address
157 * of a platform specific routine then call that routine. salinfo_platform_oemdata
158 * vmalloc's and formats its output area, returning the address of the text
159 * and its strlen. Returns 0 for success, -ve for error. The callback is
160 * invoked on the cpu that generated the error record.
162 int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);
164 struct salinfo_platform_oemdata_parms {
172 salinfo_platform_oemdata_cpu(void *context)
174 struct salinfo_platform_oemdata_parms *parms = context;
175 parms->ret = salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
179 shift1_data_saved (struct salinfo_data *data, int shift)
181 memcpy(data->data_saved+shift, data->data_saved+shift+1,
182 (ARRAY_SIZE(data->data_saved) - (shift+1)) * sizeof(data->data_saved[0]));
183 memset(data->data_saved + ARRAY_SIZE(data->data_saved) - 1, 0,
184 sizeof(data->data_saved[0]));
187 /* This routine is invoked in interrupt context. Note: mca.c enables
188 * interrupts before calling this code for CMC/CPE. MCA and INIT events are
189 * not irq safe, do not call any routines that use spinlocks, they may deadlock.
190 * MCA and INIT records are recorded, a timer event will look for any
191 * outstanding events and wake up the user space code.
193 * The buffer passed from mca.c points to the output from ia64_log_get. This is
194 * a persistent buffer but its contents can change between the interrupt and
195 * when user space processes the record. Save the record id to identify
199 salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
201 struct salinfo_data *data = salinfo_data + type;
202 struct salinfo_data_saved *data_saved;
203 unsigned long flags = 0;
205 int saved_size = ARRAY_SIZE(data->data_saved);
207 BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
210 spin_lock_irqsave(&data_saved_lock, flags);
211 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
212 if (!data_saved->buffer)
215 if (i == saved_size) {
216 if (!data->saved_num) {
217 shift1_data_saved(data, 0);
218 data_saved = data->data_saved + saved_size - 1;
223 data_saved->cpu = smp_processor_id();
224 data_saved->id = ((sal_log_record_header_t *)buffer)->id;
225 data_saved->size = size;
226 data_saved->buffer = buffer;
229 spin_unlock_irqrestore(&data_saved_lock, flags);
231 if (!test_and_set_bit(smp_processor_id(), &data->cpu_event)) {
237 /* Check for outstanding MCA/INIT records every minute (arbitrary) */
238 #define SALINFO_TIMER_DELAY (60*HZ)
239 static struct timer_list salinfo_timer;
242 salinfo_timeout_check(struct salinfo_data *data)
247 for (i = 0; i < NR_CPUS; ++i) {
248 if (test_bit(i, &data->cpu_event)) {
249 /* double up() is not a problem, user space will see no
250 * records for the additional "events".
258 salinfo_timeout (unsigned long arg)
260 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
261 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_INIT);
262 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
263 add_timer(&salinfo_timer);
267 salinfo_event_open(struct inode *inode, struct file *file)
269 if (!capable(CAP_SYS_ADMIN))
275 salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
277 struct inode *inode = file->f_dentry->d_inode;
278 struct proc_dir_entry *entry = PDE(inode);
279 struct salinfo_data *data = entry->data;
285 if (down_trylock(&data->sem)) {
286 if (file->f_flags & O_NONBLOCK)
288 if (down_interruptible(&data->sem))
293 for (i = 0; i < NR_CPUS; i++) {
294 if (test_bit(n, &data->cpu_event)) {
305 /* events are sticky until the user says "clear" */
308 /* for next read, start checking at next CPU */
309 data->cpu_check = cpu;
310 if (++data->cpu_check == NR_CPUS)
313 snprintf(cmd, sizeof(cmd), "read %d\n", cpu);
318 if (copy_to_user(buffer, cmd, size))
324 static struct file_operations salinfo_event_fops = {
325 .open = salinfo_event_open,
326 .read = salinfo_event_read,
330 salinfo_log_open(struct inode *inode, struct file *file)
332 struct proc_dir_entry *entry = PDE(inode);
333 struct salinfo_data *data = entry->data;
335 if (!capable(CAP_SYS_ADMIN))
338 spin_lock(&data_lock);
340 spin_unlock(&data_lock);
344 spin_unlock(&data_lock);
346 if (data->state == STATE_NO_DATA &&
347 !(data->log_buffer = vmalloc(ia64_sal_get_state_info_size(data->type)))) {
356 salinfo_log_release(struct inode *inode, struct file *file)
358 struct proc_dir_entry *entry = PDE(inode);
359 struct salinfo_data *data = entry->data;
361 if (data->state == STATE_NO_DATA) {
362 vfree(data->log_buffer);
363 vfree(data->oemdata);
364 data->log_buffer = NULL;
365 data->oemdata = NULL;
367 spin_lock(&data_lock);
369 spin_unlock(&data_lock);
374 call_on_cpu(int cpu, void (*fn)(void *), void *arg)
376 cpumask_t save_cpus_allowed, new_cpus_allowed;
377 memcpy(&save_cpus_allowed, ¤t->cpus_allowed, sizeof(save_cpus_allowed));
378 memset(&new_cpus_allowed, 0, sizeof(new_cpus_allowed));
379 set_bit(cpu, &new_cpus_allowed);
380 set_cpus_allowed(current, new_cpus_allowed);
382 set_cpus_allowed(current, save_cpus_allowed);
386 salinfo_log_read_cpu(void *context)
388 struct salinfo_data *data = context;
389 sal_log_record_header_t *rh;
390 data->log_size = ia64_sal_get_state_info(data->type, (u64 *) data->log_buffer);
391 rh = (sal_log_record_header_t *)(data->log_buffer);
392 /* Clear corrected errors as they are read from SAL */
393 if (rh->severity == sal_log_severity_corrected)
394 ia64_sal_clear_state_info(data->type);
398 salinfo_log_new_read(int cpu, struct salinfo_data *data)
400 struct salinfo_data_saved *data_saved;
403 int saved_size = ARRAY_SIZE(data->data_saved);
406 spin_lock_irqsave(&data_saved_lock, flags);
408 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
409 if (data_saved->buffer && data_saved->cpu == cpu) {
410 sal_log_record_header_t *rh = (sal_log_record_header_t *)(data_saved->buffer);
411 data->log_size = data_saved->size;
412 memcpy(data->log_buffer, rh, data->log_size);
413 barrier(); /* id check must not be moved */
414 if (rh->id == data_saved->id) {
415 data->saved_num = i+1;
418 /* saved record changed by mca.c since interrupt, discard it */
419 shift1_data_saved(data, i);
423 spin_unlock_irqrestore(&data_saved_lock, flags);
425 if (!data->saved_num)
426 call_on_cpu(cpu, salinfo_log_read_cpu, data);
427 if (!data->log_size) {
428 data->state = STATE_NO_DATA;
429 clear_bit(cpu, &data->cpu_event);
431 data->state = STATE_LOG_RECORD;
436 salinfo_log_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
438 struct inode *inode = file->f_dentry->d_inode;
439 struct proc_dir_entry *entry = PDE(inode);
440 struct salinfo_data *data = entry->data;
444 if (data->state == STATE_LOG_RECORD) {
445 buf = data->log_buffer;
446 bufsize = data->log_size;
447 } else if (data->state == STATE_OEMDATA) {
449 bufsize = data->oemdata_size;
454 return simple_read_from_buffer(buffer, count, ppos, buf, bufsize);
458 salinfo_log_clear_cpu(void *context)
460 struct salinfo_data *data = context;
461 ia64_sal_clear_state_info(data->type);
465 salinfo_log_clear(struct salinfo_data *data, int cpu)
467 sal_log_record_header_t *rh;
468 data->state = STATE_NO_DATA;
469 if (!test_bit(cpu, &data->cpu_event))
472 clear_bit(cpu, &data->cpu_event);
473 if (data->saved_num) {
475 spin_lock_irqsave(&data_saved_lock, flags);
476 shift1_data_saved(data, data->saved_num - 1 );
478 spin_unlock_irqrestore(&data_saved_lock, flags);
480 rh = (sal_log_record_header_t *)(data->log_buffer);
481 /* Corrected errors have already been cleared from SAL */
482 if (rh->severity != sal_log_severity_corrected)
483 call_on_cpu(cpu, salinfo_log_clear_cpu, data);
484 /* clearing a record may make a new record visible */
485 salinfo_log_new_read(cpu, data);
486 if (data->state == STATE_LOG_RECORD &&
487 !test_and_set_bit(cpu, &data->cpu_event))
493 salinfo_log_write(struct file *file, const char __user *buffer, size_t count, loff_t *ppos)
495 struct inode *inode = file->f_dentry->d_inode;
496 struct proc_dir_entry *entry = PDE(inode);
497 struct salinfo_data *data = entry->data;
506 if (copy_from_user(cmd, buffer, size))
509 if (sscanf(cmd, "read %d", &cpu) == 1) {
510 salinfo_log_new_read(cpu, data);
511 } else if (sscanf(cmd, "clear %d", &cpu) == 1) {
513 if ((ret = salinfo_log_clear(data, cpu)))
515 } else if (sscanf(cmd, "oemdata %d %d", &cpu, &offset) == 2) {
516 if (data->state != STATE_LOG_RECORD && data->state != STATE_OEMDATA)
518 if (offset > data->log_size - sizeof(efi_guid_t))
520 data->state = STATE_OEMDATA;
521 if (salinfo_platform_oemdata) {
522 struct salinfo_platform_oemdata_parms parms = {
523 .efi_guid = data->log_buffer + offset,
524 .oemdata = &data->oemdata,
525 .oemdata_size = &data->oemdata_size
527 call_on_cpu(cpu, salinfo_platform_oemdata_cpu, &parms);
531 data->oemdata_size = 0;
538 static struct file_operations salinfo_data_fops = {
539 .open = salinfo_log_open,
540 .release = salinfo_log_release,
541 .read = salinfo_log_read,
542 .write = salinfo_log_write,
548 struct proc_dir_entry *salinfo_dir; /* /proc/sal dir entry */
549 struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
550 struct proc_dir_entry *dir, *entry;
551 struct salinfo_data *data;
554 salinfo_dir = proc_mkdir("sal", NULL);
558 for (i=0; i < NR_SALINFO_ENTRIES; i++) {
559 /* pass the feature bit in question as misc data */
560 *sdir++ = create_proc_read_entry (salinfo_entries[i].name, 0, salinfo_dir,
561 salinfo_read, (void *)salinfo_entries[i].feature);
564 for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
565 data = salinfo_data + i;
567 sema_init(&data->sem, 0);
568 dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
572 entry = create_proc_entry("event", S_IRUSR, dir);
576 entry->proc_fops = &salinfo_event_fops;
579 entry = create_proc_entry("data", S_IRUSR | S_IWUSR, dir);
583 entry->proc_fops = &salinfo_data_fops;
586 /* we missed any events before now */
588 for (j = 0; j < NR_CPUS; j++)
590 set_bit(j, &data->cpu_event);
593 sema_init(&data->sem, online);
598 *sdir++ = salinfo_dir;
600 init_timer(&salinfo_timer);
601 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
602 salinfo_timer.function = &salinfo_timeout;
603 add_timer(&salinfo_timer);
609 * 'data' contains an integer that corresponds to the feature we're
613 salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data)
617 len = sprintf(page, (sal_platform_features & (unsigned long)data) ? "1\n" : "0\n");
619 if (len <= off+count) *eof = 1;
624 if (len>count) len = count;
630 module_init(salinfo_init);