4 * Creates entries in /proc/sal for various system features.
6 * Copyright (c) 2003, 2006 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.
26 * Aug 18 2005 kaos@sgi.com
27 * mca.c may not pass a buffer, a NULL buffer just indicates that a new
28 * record is available in SAL.
29 * Replace some NR_CPUS by cpus_online, for hotplug cpu.
31 * Jan 5 2006 kaos@sgi.com
32 * Handle hotplug cpus coming online.
33 * Handle hotplug cpus going offline while they still have outstanding records.
34 * Use the cpu_* macros consistently.
35 * Replace the counting semaphore with a mutex and a test if the cpumask is non-empty.
36 * Modify the locking to make the test for "work to do" an atomic operation.
39 #include <linux/capability.h>
40 #include <linux/cpu.h>
41 #include <linux/types.h>
42 #include <linux/proc_fs.h>
43 #include <linux/module.h>
44 #include <linux/smp.h>
45 #include <linux/smp_lock.h>
46 #include <linux/timer.h>
47 #include <linux/vmalloc.h>
49 #include <asm/semaphore.h>
51 #include <asm/uaccess.h>
53 MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
54 MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
55 MODULE_LICENSE("GPL");
57 static int salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data);
60 const char *name; /* name of the proc entry */
61 unsigned long feature; /* feature bit */
62 struct proc_dir_entry *entry; /* registered entry (removal) */
66 * List {name,feature} pairs for every entry in /proc/sal/<feature>
67 * that this module exports
69 static salinfo_entry_t salinfo_entries[]={
70 { "bus_lock", IA64_SAL_PLATFORM_FEATURE_BUS_LOCK, },
71 { "irq_redirection", IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT, },
72 { "ipi_redirection", IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT, },
73 { "itc_drift", IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT, },
76 #define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)
78 static char *salinfo_log_name[] = {
85 static struct proc_dir_entry *salinfo_proc_entries[
86 ARRAY_SIZE(salinfo_entries) + /* /proc/sal/bus_lock */
87 ARRAY_SIZE(salinfo_log_name) + /* /proc/sal/{mca,...} */
88 (2 * ARRAY_SIZE(salinfo_log_name)) + /* /proc/sal/mca/{event,data} */
91 /* Some records we get ourselves, some are accessed as saved data in buffers
92 * that are owned by mca.c.
94 struct salinfo_data_saved {
101 /* State transitions. Actions are :-
102 * Write "read <cpunum>" to the data file.
103 * Write "clear <cpunum>" to the data file.
104 * Write "oemdata <cpunum> <offset> to the data file.
105 * Read from the data file.
106 * Close the data file.
108 * Start state is NO_DATA.
111 * write "read <cpunum>" -> NO_DATA or LOG_RECORD.
112 * write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
113 * write "oemdata <cpunum> <offset> -> return -EINVAL.
114 * read data -> return EOF.
115 * close -> unchanged. Free record areas.
118 * write "read <cpunum>" -> NO_DATA or LOG_RECORD.
119 * write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
120 * write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
121 * read data -> return the INIT/MCA/CMC/CPE record.
122 * close -> unchanged. Keep record areas.
125 * write "read <cpunum>" -> NO_DATA or LOG_RECORD.
126 * write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
127 * write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
128 * read data -> return the formatted oemdata.
129 * close -> unchanged. Keep record areas.
131 * Closing the data file does not change the state. This allows shell scripts
132 * to manipulate salinfo data, each shell redirection opens the file, does one
133 * action then closes it again. The record areas are only freed at close when
134 * the state is NO_DATA.
142 struct salinfo_data {
143 cpumask_t cpu_event; /* which cpus have outstanding events */
144 struct semaphore mutex;
147 u8 *oemdata; /* decoded oem data */
149 int open; /* single-open to prevent races */
151 u8 saved_num; /* using a saved record? */
152 enum salinfo_state state :8; /* processing state */
154 int cpu_check; /* next CPU to check */
155 struct salinfo_data_saved data_saved[5];/* save last 5 records from mca.c, must be < 255 */
158 static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];
160 static DEFINE_SPINLOCK(data_lock);
161 static DEFINE_SPINLOCK(data_saved_lock);
163 /** salinfo_platform_oemdata - optional callback to decode oemdata from an error
165 * @sect_header: pointer to the start of the section to decode.
166 * @oemdata: returns vmalloc area containing the decded output.
167 * @oemdata_size: returns length of decoded output (strlen).
169 * Description: If user space asks for oem data to be decoded by the kernel
170 * and/or prom and the platform has set salinfo_platform_oemdata to the address
171 * of a platform specific routine then call that routine. salinfo_platform_oemdata
172 * vmalloc's and formats its output area, returning the address of the text
173 * and its strlen. Returns 0 for success, -ve for error. The callback is
174 * invoked on the cpu that generated the error record.
176 int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);
178 struct salinfo_platform_oemdata_parms {
185 /* Kick the mutex that tells user space that there is work to do. Instead of
186 * trying to track the state of the mutex across multiple cpus, in user
187 * context, interrupt context, non-maskable interrupt context and hotplug cpu,
188 * it is far easier just to grab the mutex if it is free then release it.
190 * This routine must be called with data_saved_lock held, to make the down/up
194 salinfo_work_to_do(struct salinfo_data *data)
196 down_trylock(&data->mutex);
201 salinfo_platform_oemdata_cpu(void *context)
203 struct salinfo_platform_oemdata_parms *parms = context;
204 parms->ret = salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
208 shift1_data_saved (struct salinfo_data *data, int shift)
210 memcpy(data->data_saved+shift, data->data_saved+shift+1,
211 (ARRAY_SIZE(data->data_saved) - (shift+1)) * sizeof(data->data_saved[0]));
212 memset(data->data_saved + ARRAY_SIZE(data->data_saved) - 1, 0,
213 sizeof(data->data_saved[0]));
216 /* This routine is invoked in interrupt context. Note: mca.c enables
217 * interrupts before calling this code for CMC/CPE. MCA and INIT events are
218 * not irq safe, do not call any routines that use spinlocks, they may deadlock.
219 * MCA and INIT records are recorded, a timer event will look for any
220 * outstanding events and wake up the user space code.
222 * The buffer passed from mca.c points to the output from ia64_log_get. This is
223 * a persistent buffer but its contents can change between the interrupt and
224 * when user space processes the record. Save the record id to identify
225 * changes. If the buffer is NULL then just update the bitmap.
228 salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
230 struct salinfo_data *data = salinfo_data + type;
231 struct salinfo_data_saved *data_saved;
232 unsigned long flags = 0;
234 int saved_size = ARRAY_SIZE(data->data_saved);
236 BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
239 spin_lock_irqsave(&data_saved_lock, flags);
241 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
242 if (!data_saved->buffer)
245 if (i == saved_size) {
246 if (!data->saved_num) {
247 shift1_data_saved(data, 0);
248 data_saved = data->data_saved + saved_size - 1;
253 data_saved->cpu = smp_processor_id();
254 data_saved->id = ((sal_log_record_header_t *)buffer)->id;
255 data_saved->size = size;
256 data_saved->buffer = buffer;
259 cpu_set(smp_processor_id(), data->cpu_event);
261 salinfo_work_to_do(data);
262 spin_unlock_irqrestore(&data_saved_lock, flags);
266 /* Check for outstanding MCA/INIT records every minute (arbitrary) */
267 #define SALINFO_TIMER_DELAY (60*HZ)
268 static struct timer_list salinfo_timer;
269 extern void ia64_mlogbuf_dump(void);
272 salinfo_timeout_check(struct salinfo_data *data)
277 if (!cpus_empty(data->cpu_event)) {
278 spin_lock_irqsave(&data_saved_lock, flags);
279 salinfo_work_to_do(data);
280 spin_unlock_irqrestore(&data_saved_lock, flags);
285 salinfo_timeout (unsigned long arg)
288 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
289 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_INIT);
290 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
291 add_timer(&salinfo_timer);
295 salinfo_event_open(struct inode *inode, struct file *file)
297 if (!capable(CAP_SYS_ADMIN))
303 salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
305 struct inode *inode = file->f_path.dentry->d_inode;
306 struct proc_dir_entry *entry = PDE(inode);
307 struct salinfo_data *data = entry->data;
313 if (cpus_empty(data->cpu_event) && down_trylock(&data->mutex)) {
314 if (file->f_flags & O_NONBLOCK)
316 if (down_interruptible(&data->mutex))
321 for (i = 0; i < NR_CPUS; i++) {
322 if (cpu_isset(n, data->cpu_event)) {
323 if (!cpu_online(n)) {
324 cpu_clear(n, data->cpu_event);
339 /* for next read, start checking at next CPU */
340 data->cpu_check = cpu;
341 if (++data->cpu_check == NR_CPUS)
344 snprintf(cmd, sizeof(cmd), "read %d\n", cpu);
349 if (copy_to_user(buffer, cmd, size))
355 static const struct file_operations salinfo_event_fops = {
356 .open = salinfo_event_open,
357 .read = salinfo_event_read,
361 salinfo_log_open(struct inode *inode, struct file *file)
363 struct proc_dir_entry *entry = PDE(inode);
364 struct salinfo_data *data = entry->data;
366 if (!capable(CAP_SYS_ADMIN))
369 spin_lock(&data_lock);
371 spin_unlock(&data_lock);
375 spin_unlock(&data_lock);
377 if (data->state == STATE_NO_DATA &&
378 !(data->log_buffer = vmalloc(ia64_sal_get_state_info_size(data->type)))) {
387 salinfo_log_release(struct inode *inode, struct file *file)
389 struct proc_dir_entry *entry = PDE(inode);
390 struct salinfo_data *data = entry->data;
392 if (data->state == STATE_NO_DATA) {
393 vfree(data->log_buffer);
394 vfree(data->oemdata);
395 data->log_buffer = NULL;
396 data->oemdata = NULL;
398 spin_lock(&data_lock);
400 spin_unlock(&data_lock);
405 call_on_cpu(int cpu, void (*fn)(void *), void *arg)
407 cpumask_t save_cpus_allowed = current->cpus_allowed;
408 cpumask_t new_cpus_allowed = cpumask_of_cpu(cpu);
409 set_cpus_allowed(current, new_cpus_allowed);
411 set_cpus_allowed(current, save_cpus_allowed);
415 salinfo_log_read_cpu(void *context)
417 struct salinfo_data *data = context;
418 sal_log_record_header_t *rh;
419 data->log_size = ia64_sal_get_state_info(data->type, (u64 *) data->log_buffer);
420 rh = (sal_log_record_header_t *)(data->log_buffer);
421 /* Clear corrected errors as they are read from SAL */
422 if (rh->severity == sal_log_severity_corrected)
423 ia64_sal_clear_state_info(data->type);
427 salinfo_log_new_read(int cpu, struct salinfo_data *data)
429 struct salinfo_data_saved *data_saved;
432 int saved_size = ARRAY_SIZE(data->data_saved);
435 spin_lock_irqsave(&data_saved_lock, flags);
437 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
438 if (data_saved->buffer && data_saved->cpu == cpu) {
439 sal_log_record_header_t *rh = (sal_log_record_header_t *)(data_saved->buffer);
440 data->log_size = data_saved->size;
441 memcpy(data->log_buffer, rh, data->log_size);
442 barrier(); /* id check must not be moved */
443 if (rh->id == data_saved->id) {
444 data->saved_num = i+1;
447 /* saved record changed by mca.c since interrupt, discard it */
448 shift1_data_saved(data, i);
452 spin_unlock_irqrestore(&data_saved_lock, flags);
454 if (!data->saved_num)
455 call_on_cpu(cpu, salinfo_log_read_cpu, data);
456 if (!data->log_size) {
457 data->state = STATE_NO_DATA;
458 cpu_clear(cpu, data->cpu_event);
460 data->state = STATE_LOG_RECORD;
465 salinfo_log_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
467 struct inode *inode = file->f_path.dentry->d_inode;
468 struct proc_dir_entry *entry = PDE(inode);
469 struct salinfo_data *data = entry->data;
473 if (data->state == STATE_LOG_RECORD) {
474 buf = data->log_buffer;
475 bufsize = data->log_size;
476 } else if (data->state == STATE_OEMDATA) {
478 bufsize = data->oemdata_size;
483 return simple_read_from_buffer(buffer, count, ppos, buf, bufsize);
487 salinfo_log_clear_cpu(void *context)
489 struct salinfo_data *data = context;
490 ia64_sal_clear_state_info(data->type);
494 salinfo_log_clear(struct salinfo_data *data, int cpu)
496 sal_log_record_header_t *rh;
498 spin_lock_irqsave(&data_saved_lock, flags);
499 data->state = STATE_NO_DATA;
500 if (!cpu_isset(cpu, data->cpu_event)) {
501 spin_unlock_irqrestore(&data_saved_lock, flags);
504 cpu_clear(cpu, data->cpu_event);
505 if (data->saved_num) {
506 shift1_data_saved(data, data->saved_num - 1);
509 spin_unlock_irqrestore(&data_saved_lock, flags);
510 rh = (sal_log_record_header_t *)(data->log_buffer);
511 /* Corrected errors have already been cleared from SAL */
512 if (rh->severity != sal_log_severity_corrected)
513 call_on_cpu(cpu, salinfo_log_clear_cpu, data);
514 /* clearing a record may make a new record visible */
515 salinfo_log_new_read(cpu, data);
516 if (data->state == STATE_LOG_RECORD) {
517 spin_lock_irqsave(&data_saved_lock, flags);
518 cpu_set(cpu, data->cpu_event);
519 salinfo_work_to_do(data);
520 spin_unlock_irqrestore(&data_saved_lock, flags);
526 salinfo_log_write(struct file *file, const char __user *buffer, size_t count, loff_t *ppos)
528 struct inode *inode = file->f_path.dentry->d_inode;
529 struct proc_dir_entry *entry = PDE(inode);
530 struct salinfo_data *data = entry->data;
539 if (copy_from_user(cmd, buffer, size))
542 if (sscanf(cmd, "read %d", &cpu) == 1) {
543 salinfo_log_new_read(cpu, data);
544 } else if (sscanf(cmd, "clear %d", &cpu) == 1) {
546 if ((ret = salinfo_log_clear(data, cpu)))
548 } else if (sscanf(cmd, "oemdata %d %d", &cpu, &offset) == 2) {
549 if (data->state != STATE_LOG_RECORD && data->state != STATE_OEMDATA)
551 if (offset > data->log_size - sizeof(efi_guid_t))
553 data->state = STATE_OEMDATA;
554 if (salinfo_platform_oemdata) {
555 struct salinfo_platform_oemdata_parms parms = {
556 .efi_guid = data->log_buffer + offset,
557 .oemdata = &data->oemdata,
558 .oemdata_size = &data->oemdata_size
560 call_on_cpu(cpu, salinfo_platform_oemdata_cpu, &parms);
564 data->oemdata_size = 0;
571 static const struct file_operations salinfo_data_fops = {
572 .open = salinfo_log_open,
573 .release = salinfo_log_release,
574 .read = salinfo_log_read,
575 .write = salinfo_log_write,
579 salinfo_cpu_callback(struct notifier_block *nb, unsigned long action, void *hcpu)
581 unsigned int i, cpu = (unsigned long)hcpu;
583 struct salinfo_data *data;
586 spin_lock_irqsave(&data_saved_lock, flags);
587 for (i = 0, data = salinfo_data;
588 i < ARRAY_SIZE(salinfo_data);
590 cpu_set(cpu, data->cpu_event);
591 salinfo_work_to_do(data);
593 spin_unlock_irqrestore(&data_saved_lock, flags);
596 spin_lock_irqsave(&data_saved_lock, flags);
597 for (i = 0, data = salinfo_data;
598 i < ARRAY_SIZE(salinfo_data);
600 struct salinfo_data_saved *data_saved;
602 for (j = ARRAY_SIZE(data->data_saved) - 1, data_saved = data->data_saved + j;
605 if (data_saved->buffer && data_saved->cpu == cpu) {
606 shift1_data_saved(data, j);
609 cpu_clear(cpu, data->cpu_event);
611 spin_unlock_irqrestore(&data_saved_lock, flags);
617 static struct notifier_block salinfo_cpu_notifier =
619 .notifier_call = salinfo_cpu_callback,
626 struct proc_dir_entry *salinfo_dir; /* /proc/sal dir entry */
627 struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
628 struct proc_dir_entry *dir, *entry;
629 struct salinfo_data *data;
632 salinfo_dir = proc_mkdir("sal", NULL);
636 for (i=0; i < NR_SALINFO_ENTRIES; i++) {
637 /* pass the feature bit in question as misc data */
638 *sdir++ = create_proc_read_entry (salinfo_entries[i].name, 0, salinfo_dir,
639 salinfo_read, (void *)salinfo_entries[i].feature);
642 for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
643 data = salinfo_data + i;
645 init_MUTEX(&data->mutex);
646 dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
650 entry = create_proc_entry("event", S_IRUSR, dir);
654 entry->proc_fops = &salinfo_event_fops;
657 entry = create_proc_entry("data", S_IRUSR | S_IWUSR, dir);
661 entry->proc_fops = &salinfo_data_fops;
664 /* we missed any events before now */
665 for_each_online_cpu(j)
666 cpu_set(j, data->cpu_event);
671 *sdir++ = salinfo_dir;
673 init_timer(&salinfo_timer);
674 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
675 salinfo_timer.function = &salinfo_timeout;
676 add_timer(&salinfo_timer);
678 register_hotcpu_notifier(&salinfo_cpu_notifier);
684 * 'data' contains an integer that corresponds to the feature we're
688 salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data)
692 len = sprintf(page, (sal_platform_features & (unsigned long)data) ? "1\n" : "0\n");
694 if (len <= off+count) *eof = 1;
699 if (len>count) len = count;
705 module_init(salinfo_init);