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/timer.h>
46 #include <linux/vmalloc.h>
48 #include <asm/semaphore.h>
50 #include <asm/uaccess.h>
52 MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
53 MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
54 MODULE_LICENSE("GPL");
56 static int salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data);
59 const char *name; /* name of the proc entry */
60 unsigned long feature; /* feature bit */
61 struct proc_dir_entry *entry; /* registered entry (removal) */
65 * List {name,feature} pairs for every entry in /proc/sal/<feature>
66 * that this module exports
68 static salinfo_entry_t salinfo_entries[]={
69 { "bus_lock", IA64_SAL_PLATFORM_FEATURE_BUS_LOCK, },
70 { "irq_redirection", IA64_SAL_PLATFORM_FEATURE_IRQ_REDIR_HINT, },
71 { "ipi_redirection", IA64_SAL_PLATFORM_FEATURE_IPI_REDIR_HINT, },
72 { "itc_drift", IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT, },
75 #define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)
77 static char *salinfo_log_name[] = {
84 static struct proc_dir_entry *salinfo_proc_entries[
85 ARRAY_SIZE(salinfo_entries) + /* /proc/sal/bus_lock */
86 ARRAY_SIZE(salinfo_log_name) + /* /proc/sal/{mca,...} */
87 (2 * ARRAY_SIZE(salinfo_log_name)) + /* /proc/sal/mca/{event,data} */
90 /* Some records we get ourselves, some are accessed as saved data in buffers
91 * that are owned by mca.c.
93 struct salinfo_data_saved {
100 /* State transitions. Actions are :-
101 * Write "read <cpunum>" to the data file.
102 * Write "clear <cpunum>" to the data file.
103 * Write "oemdata <cpunum> <offset> to the data file.
104 * Read from the data file.
105 * Close the data file.
107 * Start state is NO_DATA.
110 * write "read <cpunum>" -> NO_DATA or LOG_RECORD.
111 * write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
112 * write "oemdata <cpunum> <offset> -> return -EINVAL.
113 * read data -> return EOF.
114 * close -> unchanged. Free record areas.
117 * write "read <cpunum>" -> NO_DATA or LOG_RECORD.
118 * write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
119 * write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
120 * read data -> return the INIT/MCA/CMC/CPE record.
121 * close -> unchanged. Keep record areas.
124 * write "read <cpunum>" -> NO_DATA or LOG_RECORD.
125 * write "clear <cpunum>" -> NO_DATA or LOG_RECORD.
126 * write "oemdata <cpunum> <offset> -> format the oem data, goto OEMDATA.
127 * read data -> return the formatted oemdata.
128 * close -> unchanged. Keep record areas.
130 * Closing the data file does not change the state. This allows shell scripts
131 * to manipulate salinfo data, each shell redirection opens the file, does one
132 * action then closes it again. The record areas are only freed at close when
133 * the state is NO_DATA.
141 struct salinfo_data {
142 cpumask_t cpu_event; /* which cpus have outstanding events */
143 struct semaphore mutex;
146 u8 *oemdata; /* decoded oem data */
148 int open; /* single-open to prevent races */
150 u8 saved_num; /* using a saved record? */
151 enum salinfo_state state :8; /* processing state */
153 int cpu_check; /* next CPU to check */
154 struct salinfo_data_saved data_saved[5];/* save last 5 records from mca.c, must be < 255 */
157 static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];
159 static DEFINE_SPINLOCK(data_lock);
160 static DEFINE_SPINLOCK(data_saved_lock);
162 /** salinfo_platform_oemdata - optional callback to decode oemdata from an error
164 * @sect_header: pointer to the start of the section to decode.
165 * @oemdata: returns vmalloc area containing the decoded output.
166 * @oemdata_size: returns length of decoded output (strlen).
168 * Description: If user space asks for oem data to be decoded by the kernel
169 * and/or prom and the platform has set salinfo_platform_oemdata to the address
170 * of a platform specific routine then call that routine. salinfo_platform_oemdata
171 * vmalloc's and formats its output area, returning the address of the text
172 * and its strlen. Returns 0 for success, -ve for error. The callback is
173 * invoked on the cpu that generated the error record.
175 int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);
177 struct salinfo_platform_oemdata_parms {
184 /* Kick the mutex that tells user space that there is work to do. Instead of
185 * trying to track the state of the mutex across multiple cpus, in user
186 * context, interrupt context, non-maskable interrupt context and hotplug cpu,
187 * it is far easier just to grab the mutex if it is free then release it.
189 * This routine must be called with data_saved_lock held, to make the down/up
193 salinfo_work_to_do(struct salinfo_data *data)
195 down_trylock(&data->mutex);
200 salinfo_platform_oemdata_cpu(void *context)
202 struct salinfo_platform_oemdata_parms *parms = context;
203 parms->ret = salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
207 shift1_data_saved (struct salinfo_data *data, int shift)
209 memcpy(data->data_saved+shift, data->data_saved+shift+1,
210 (ARRAY_SIZE(data->data_saved) - (shift+1)) * sizeof(data->data_saved[0]));
211 memset(data->data_saved + ARRAY_SIZE(data->data_saved) - 1, 0,
212 sizeof(data->data_saved[0]));
215 /* This routine is invoked in interrupt context. Note: mca.c enables
216 * interrupts before calling this code for CMC/CPE. MCA and INIT events are
217 * not irq safe, do not call any routines that use spinlocks, they may deadlock.
218 * MCA and INIT records are recorded, a timer event will look for any
219 * outstanding events and wake up the user space code.
221 * The buffer passed from mca.c points to the output from ia64_log_get. This is
222 * a persistent buffer but its contents can change between the interrupt and
223 * when user space processes the record. Save the record id to identify
224 * changes. If the buffer is NULL then just update the bitmap.
227 salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
229 struct salinfo_data *data = salinfo_data + type;
230 struct salinfo_data_saved *data_saved;
231 unsigned long flags = 0;
233 int saved_size = ARRAY_SIZE(data->data_saved);
235 BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
238 spin_lock_irqsave(&data_saved_lock, flags);
240 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
241 if (!data_saved->buffer)
244 if (i == saved_size) {
245 if (!data->saved_num) {
246 shift1_data_saved(data, 0);
247 data_saved = data->data_saved + saved_size - 1;
252 data_saved->cpu = smp_processor_id();
253 data_saved->id = ((sal_log_record_header_t *)buffer)->id;
254 data_saved->size = size;
255 data_saved->buffer = buffer;
258 cpu_set(smp_processor_id(), data->cpu_event);
260 salinfo_work_to_do(data);
261 spin_unlock_irqrestore(&data_saved_lock, flags);
265 /* Check for outstanding MCA/INIT records every minute (arbitrary) */
266 #define SALINFO_TIMER_DELAY (60*HZ)
267 static struct timer_list salinfo_timer;
268 extern void ia64_mlogbuf_dump(void);
271 salinfo_timeout_check(struct salinfo_data *data)
276 if (!cpus_empty(data->cpu_event)) {
277 spin_lock_irqsave(&data_saved_lock, flags);
278 salinfo_work_to_do(data);
279 spin_unlock_irqrestore(&data_saved_lock, flags);
284 salinfo_timeout (unsigned long arg)
287 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
288 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_INIT);
289 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
290 add_timer(&salinfo_timer);
294 salinfo_event_open(struct inode *inode, struct file *file)
296 if (!capable(CAP_SYS_ADMIN))
302 salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
304 struct inode *inode = file->f_path.dentry->d_inode;
305 struct proc_dir_entry *entry = PDE(inode);
306 struct salinfo_data *data = entry->data;
312 if (cpus_empty(data->cpu_event) && down_trylock(&data->mutex)) {
313 if (file->f_flags & O_NONBLOCK)
315 if (down_interruptible(&data->mutex))
320 for (i = 0; i < NR_CPUS; i++) {
321 if (cpu_isset(n, data->cpu_event)) {
322 if (!cpu_online(n)) {
323 cpu_clear(n, data->cpu_event);
338 /* for next read, start checking at next CPU */
339 data->cpu_check = cpu;
340 if (++data->cpu_check == NR_CPUS)
343 snprintf(cmd, sizeof(cmd), "read %d\n", cpu);
348 if (copy_to_user(buffer, cmd, size))
354 static const struct file_operations salinfo_event_fops = {
355 .open = salinfo_event_open,
356 .read = salinfo_event_read,
360 salinfo_log_open(struct inode *inode, struct file *file)
362 struct proc_dir_entry *entry = PDE(inode);
363 struct salinfo_data *data = entry->data;
365 if (!capable(CAP_SYS_ADMIN))
368 spin_lock(&data_lock);
370 spin_unlock(&data_lock);
374 spin_unlock(&data_lock);
376 if (data->state == STATE_NO_DATA &&
377 !(data->log_buffer = vmalloc(ia64_sal_get_state_info_size(data->type)))) {
386 salinfo_log_release(struct inode *inode, struct file *file)
388 struct proc_dir_entry *entry = PDE(inode);
389 struct salinfo_data *data = entry->data;
391 if (data->state == STATE_NO_DATA) {
392 vfree(data->log_buffer);
393 vfree(data->oemdata);
394 data->log_buffer = NULL;
395 data->oemdata = NULL;
397 spin_lock(&data_lock);
399 spin_unlock(&data_lock);
404 call_on_cpu(int cpu, void (*fn)(void *), void *arg)
406 cpumask_t save_cpus_allowed = current->cpus_allowed;
407 cpumask_t new_cpus_allowed = cpumask_of_cpu(cpu);
408 set_cpus_allowed(current, new_cpus_allowed);
410 set_cpus_allowed(current, save_cpus_allowed);
414 salinfo_log_read_cpu(void *context)
416 struct salinfo_data *data = context;
417 sal_log_record_header_t *rh;
418 data->log_size = ia64_sal_get_state_info(data->type, (u64 *) data->log_buffer);
419 rh = (sal_log_record_header_t *)(data->log_buffer);
420 /* Clear corrected errors as they are read from SAL */
421 if (rh->severity == sal_log_severity_corrected)
422 ia64_sal_clear_state_info(data->type);
426 salinfo_log_new_read(int cpu, struct salinfo_data *data)
428 struct salinfo_data_saved *data_saved;
431 int saved_size = ARRAY_SIZE(data->data_saved);
434 spin_lock_irqsave(&data_saved_lock, flags);
436 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
437 if (data_saved->buffer && data_saved->cpu == cpu) {
438 sal_log_record_header_t *rh = (sal_log_record_header_t *)(data_saved->buffer);
439 data->log_size = data_saved->size;
440 memcpy(data->log_buffer, rh, data->log_size);
441 barrier(); /* id check must not be moved */
442 if (rh->id == data_saved->id) {
443 data->saved_num = i+1;
446 /* saved record changed by mca.c since interrupt, discard it */
447 shift1_data_saved(data, i);
451 spin_unlock_irqrestore(&data_saved_lock, flags);
453 if (!data->saved_num)
454 call_on_cpu(cpu, salinfo_log_read_cpu, data);
455 if (!data->log_size) {
456 data->state = STATE_NO_DATA;
457 cpu_clear(cpu, data->cpu_event);
459 data->state = STATE_LOG_RECORD;
464 salinfo_log_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
466 struct inode *inode = file->f_path.dentry->d_inode;
467 struct proc_dir_entry *entry = PDE(inode);
468 struct salinfo_data *data = entry->data;
472 if (data->state == STATE_LOG_RECORD) {
473 buf = data->log_buffer;
474 bufsize = data->log_size;
475 } else if (data->state == STATE_OEMDATA) {
477 bufsize = data->oemdata_size;
482 return simple_read_from_buffer(buffer, count, ppos, buf, bufsize);
486 salinfo_log_clear_cpu(void *context)
488 struct salinfo_data *data = context;
489 ia64_sal_clear_state_info(data->type);
493 salinfo_log_clear(struct salinfo_data *data, int cpu)
495 sal_log_record_header_t *rh;
497 spin_lock_irqsave(&data_saved_lock, flags);
498 data->state = STATE_NO_DATA;
499 if (!cpu_isset(cpu, data->cpu_event)) {
500 spin_unlock_irqrestore(&data_saved_lock, flags);
503 cpu_clear(cpu, data->cpu_event);
504 if (data->saved_num) {
505 shift1_data_saved(data, data->saved_num - 1);
508 spin_unlock_irqrestore(&data_saved_lock, flags);
509 rh = (sal_log_record_header_t *)(data->log_buffer);
510 /* Corrected errors have already been cleared from SAL */
511 if (rh->severity != sal_log_severity_corrected)
512 call_on_cpu(cpu, salinfo_log_clear_cpu, data);
513 /* clearing a record may make a new record visible */
514 salinfo_log_new_read(cpu, data);
515 if (data->state == STATE_LOG_RECORD) {
516 spin_lock_irqsave(&data_saved_lock, flags);
517 cpu_set(cpu, data->cpu_event);
518 salinfo_work_to_do(data);
519 spin_unlock_irqrestore(&data_saved_lock, flags);
525 salinfo_log_write(struct file *file, const char __user *buffer, size_t count, loff_t *ppos)
527 struct inode *inode = file->f_path.dentry->d_inode;
528 struct proc_dir_entry *entry = PDE(inode);
529 struct salinfo_data *data = entry->data;
538 if (copy_from_user(cmd, buffer, size))
541 if (sscanf(cmd, "read %d", &cpu) == 1) {
542 salinfo_log_new_read(cpu, data);
543 } else if (sscanf(cmd, "clear %d", &cpu) == 1) {
545 if ((ret = salinfo_log_clear(data, cpu)))
547 } else if (sscanf(cmd, "oemdata %d %d", &cpu, &offset) == 2) {
548 if (data->state != STATE_LOG_RECORD && data->state != STATE_OEMDATA)
550 if (offset > data->log_size - sizeof(efi_guid_t))
552 data->state = STATE_OEMDATA;
553 if (salinfo_platform_oemdata) {
554 struct salinfo_platform_oemdata_parms parms = {
555 .efi_guid = data->log_buffer + offset,
556 .oemdata = &data->oemdata,
557 .oemdata_size = &data->oemdata_size
559 call_on_cpu(cpu, salinfo_platform_oemdata_cpu, &parms);
563 data->oemdata_size = 0;
570 static const struct file_operations salinfo_data_fops = {
571 .open = salinfo_log_open,
572 .release = salinfo_log_release,
573 .read = salinfo_log_read,
574 .write = salinfo_log_write,
578 salinfo_cpu_callback(struct notifier_block *nb, unsigned long action, void *hcpu)
580 unsigned int i, cpu = (unsigned long)hcpu;
582 struct salinfo_data *data;
585 case CPU_ONLINE_FROZEN:
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 case CPU_DEAD_FROZEN:
597 spin_lock_irqsave(&data_saved_lock, flags);
598 for (i = 0, data = salinfo_data;
599 i < ARRAY_SIZE(salinfo_data);
601 struct salinfo_data_saved *data_saved;
603 for (j = ARRAY_SIZE(data->data_saved) - 1, data_saved = data->data_saved + j;
606 if (data_saved->buffer && data_saved->cpu == cpu) {
607 shift1_data_saved(data, j);
610 cpu_clear(cpu, data->cpu_event);
612 spin_unlock_irqrestore(&data_saved_lock, flags);
618 static struct notifier_block salinfo_cpu_notifier =
620 .notifier_call = salinfo_cpu_callback,
627 struct proc_dir_entry *salinfo_dir; /* /proc/sal dir entry */
628 struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
629 struct proc_dir_entry *dir, *entry;
630 struct salinfo_data *data;
633 salinfo_dir = proc_mkdir("sal", NULL);
637 for (i=0; i < NR_SALINFO_ENTRIES; i++) {
638 /* pass the feature bit in question as misc data */
639 *sdir++ = create_proc_read_entry (salinfo_entries[i].name, 0, salinfo_dir,
640 salinfo_read, (void *)salinfo_entries[i].feature);
643 for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
644 data = salinfo_data + i;
646 init_MUTEX(&data->mutex);
647 dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
651 entry = create_proc_entry("event", S_IRUSR, dir);
655 entry->proc_fops = &salinfo_event_fops;
658 entry = create_proc_entry("data", S_IRUSR | S_IWUSR, dir);
662 entry->proc_fops = &salinfo_data_fops;
665 /* we missed any events before now */
666 for_each_online_cpu(j)
667 cpu_set(j, data->cpu_event);
672 *sdir++ = salinfo_dir;
674 init_timer(&salinfo_timer);
675 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
676 salinfo_timer.function = &salinfo_timeout;
677 add_timer(&salinfo_timer);
679 register_hotcpu_notifier(&salinfo_cpu_notifier);
685 * 'data' contains an integer that corresponds to the feature we're
689 salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data)
693 len = sprintf(page, (sal_platform_features & (unsigned long)data) ? "1\n" : "0\n");
695 if (len <= off+count) *eof = 1;
700 if (len>count) len = count;
706 module_init(salinfo_init);