Loading...
1/*
2 * salinfo.c
3 *
4 * Creates entries in /proc/sal for various system features.
5 *
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>
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 * 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.
30 *
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.
37 */
38
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>
47#include <linux/semaphore.h>
48
49#include <asm/sal.h>
50#include <asm/uaccess.h>
51
52MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
53MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
54MODULE_LICENSE("GPL");
55
56static int salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data);
57
58typedef struct {
59 const char *name; /* name of the proc entry */
60 unsigned long feature; /* feature bit */
61 struct proc_dir_entry *entry; /* registered entry (removal) */
62} salinfo_entry_t;
63
64/*
65 * List {name,feature} pairs for every entry in /proc/sal/<feature>
66 * that this module exports
67 */
68static 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, },
73};
74
75#define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)
76
77static char *salinfo_log_name[] = {
78 "mca",
79 "init",
80 "cmc",
81 "cpe",
82};
83
84static 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} */
88 1]; /* /proc/sal */
89
90/* Some records we get ourselves, some are accessed as saved data in buffers
91 * that are owned by mca.c.
92 */
93struct salinfo_data_saved {
94 u8* buffer;
95 u64 size;
96 u64 id;
97 int cpu;
98};
99
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.
106 *
107 * Start state is NO_DATA.
108 *
109 * 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.
115 *
116 * LOG_RECORD
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.
122 *
123 * OEMDATA
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.
129 *
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.
134 */
135enum salinfo_state {
136 STATE_NO_DATA,
137 STATE_LOG_RECORD,
138 STATE_OEMDATA,
139};
140
141struct salinfo_data {
142 cpumask_t cpu_event; /* which cpus have outstanding events */
143 struct semaphore mutex;
144 u8 *log_buffer;
145 u64 log_size;
146 u8 *oemdata; /* decoded oem data */
147 u64 oemdata_size;
148 int open; /* single-open to prevent races */
149 u8 type;
150 u8 saved_num; /* using a saved record? */
151 enum salinfo_state state :8; /* processing state */
152 u8 padding;
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 */
155};
156
157static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];
158
159static DEFINE_SPINLOCK(data_lock);
160static DEFINE_SPINLOCK(data_saved_lock);
161
162/** salinfo_platform_oemdata - optional callback to decode oemdata from an error
163 * record.
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).
167 *
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.
174 */
175int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);
176
177struct salinfo_platform_oemdata_parms {
178 const u8 *efi_guid;
179 u8 **oemdata;
180 u64 *oemdata_size;
181 int ret;
182};
183
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.
188 *
189 * This routine must be called with data_saved_lock held, to make the down/up
190 * operation atomic.
191 */
192static void
193salinfo_work_to_do(struct salinfo_data *data)
194{
195 (void)(down_trylock(&data->mutex) ?: 0);
196 up(&data->mutex);
197}
198
199static void
200salinfo_platform_oemdata_cpu(void *context)
201{
202 struct salinfo_platform_oemdata_parms *parms = context;
203 parms->ret = salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
204}
205
206static void
207shift1_data_saved (struct salinfo_data *data, int shift)
208{
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]));
213}
214
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.
220 *
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.
225 */
226void
227salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
228{
229 struct salinfo_data *data = salinfo_data + type;
230 struct salinfo_data_saved *data_saved;
231 unsigned long flags = 0;
232 int i;
233 int saved_size = ARRAY_SIZE(data->data_saved);
234
235 BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
236
237 if (irqsafe)
238 spin_lock_irqsave(&data_saved_lock, flags);
239 if (buffer) {
240 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
241 if (!data_saved->buffer)
242 break;
243 }
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;
248 } else
249 data_saved = NULL;
250 }
251 if (data_saved) {
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;
256 }
257 }
258 cpu_set(smp_processor_id(), data->cpu_event);
259 if (irqsafe) {
260 salinfo_work_to_do(data);
261 spin_unlock_irqrestore(&data_saved_lock, flags);
262 }
263}
264
265/* Check for outstanding MCA/INIT records every minute (arbitrary) */
266#define SALINFO_TIMER_DELAY (60*HZ)
267static struct timer_list salinfo_timer;
268extern void ia64_mlogbuf_dump(void);
269
270static void
271salinfo_timeout_check(struct salinfo_data *data)
272{
273 unsigned long flags;
274 if (!data->open)
275 return;
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);
280 }
281}
282
283static void
284salinfo_timeout (unsigned long arg)
285{
286 ia64_mlogbuf_dump();
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);
291}
292
293static int
294salinfo_event_open(struct inode *inode, struct file *file)
295{
296 if (!capable(CAP_SYS_ADMIN))
297 return -EPERM;
298 return 0;
299}
300
301static ssize_t
302salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
303{
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;
307 char cmd[32];
308 size_t size;
309 int i, n, cpu = -1;
310
311retry:
312 if (cpus_empty(data->cpu_event) && down_trylock(&data->mutex)) {
313 if (file->f_flags & O_NONBLOCK)
314 return -EAGAIN;
315 if (down_interruptible(&data->mutex))
316 return -EINTR;
317 }
318
319 n = data->cpu_check;
320 for (i = 0; i < nr_cpu_ids; i++) {
321 if (cpu_isset(n, data->cpu_event)) {
322 if (!cpu_online(n)) {
323 cpu_clear(n, data->cpu_event);
324 continue;
325 }
326 cpu = n;
327 break;
328 }
329 if (++n == nr_cpu_ids)
330 n = 0;
331 }
332
333 if (cpu == -1)
334 goto retry;
335
336 ia64_mlogbuf_dump();
337
338 /* for next read, start checking at next CPU */
339 data->cpu_check = cpu;
340 if (++data->cpu_check == nr_cpu_ids)
341 data->cpu_check = 0;
342
343 snprintf(cmd, sizeof(cmd), "read %d\n", cpu);
344
345 size = strlen(cmd);
346 if (size > count)
347 size = count;
348 if (copy_to_user(buffer, cmd, size))
349 return -EFAULT;
350
351 return size;
352}
353
354static const struct file_operations salinfo_event_fops = {
355 .open = salinfo_event_open,
356 .read = salinfo_event_read,
357 .llseek = noop_llseek,
358};
359
360static int
361salinfo_log_open(struct inode *inode, struct file *file)
362{
363 struct proc_dir_entry *entry = PDE(inode);
364 struct salinfo_data *data = entry->data;
365
366 if (!capable(CAP_SYS_ADMIN))
367 return -EPERM;
368
369 spin_lock(&data_lock);
370 if (data->open) {
371 spin_unlock(&data_lock);
372 return -EBUSY;
373 }
374 data->open = 1;
375 spin_unlock(&data_lock);
376
377 if (data->state == STATE_NO_DATA &&
378 !(data->log_buffer = vmalloc(ia64_sal_get_state_info_size(data->type)))) {
379 data->open = 0;
380 return -ENOMEM;
381 }
382
383 return 0;
384}
385
386static int
387salinfo_log_release(struct inode *inode, struct file *file)
388{
389 struct proc_dir_entry *entry = PDE(inode);
390 struct salinfo_data *data = entry->data;
391
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;
397 }
398 spin_lock(&data_lock);
399 data->open = 0;
400 spin_unlock(&data_lock);
401 return 0;
402}
403
404static void
405call_on_cpu(int cpu, void (*fn)(void *), void *arg)
406{
407 cpumask_t save_cpus_allowed = current->cpus_allowed;
408 set_cpus_allowed_ptr(current, cpumask_of(cpu));
409 (*fn)(arg);
410 set_cpus_allowed_ptr(current, &save_cpus_allowed);
411}
412
413static void
414salinfo_log_read_cpu(void *context)
415{
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);
423}
424
425static void
426salinfo_log_new_read(int cpu, struct salinfo_data *data)
427{
428 struct salinfo_data_saved *data_saved;
429 unsigned long flags;
430 int i;
431 int saved_size = ARRAY_SIZE(data->data_saved);
432
433 data->saved_num = 0;
434 spin_lock_irqsave(&data_saved_lock, flags);
435retry:
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;
444 break;
445 }
446 /* saved record changed by mca.c since interrupt, discard it */
447 shift1_data_saved(data, i);
448 goto retry;
449 }
450 }
451 spin_unlock_irqrestore(&data_saved_lock, flags);
452
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);
458 } else {
459 data->state = STATE_LOG_RECORD;
460 }
461}
462
463static ssize_t
464salinfo_log_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
465{
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;
469 u8 *buf;
470 u64 bufsize;
471
472 if (data->state == STATE_LOG_RECORD) {
473 buf = data->log_buffer;
474 bufsize = data->log_size;
475 } else if (data->state == STATE_OEMDATA) {
476 buf = data->oemdata;
477 bufsize = data->oemdata_size;
478 } else {
479 buf = NULL;
480 bufsize = 0;
481 }
482 return simple_read_from_buffer(buffer, count, ppos, buf, bufsize);
483}
484
485static void
486salinfo_log_clear_cpu(void *context)
487{
488 struct salinfo_data *data = context;
489 ia64_sal_clear_state_info(data->type);
490}
491
492static int
493salinfo_log_clear(struct salinfo_data *data, int cpu)
494{
495 sal_log_record_header_t *rh;
496 unsigned long flags;
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);
501 return 0;
502 }
503 cpu_clear(cpu, data->cpu_event);
504 if (data->saved_num) {
505 shift1_data_saved(data, data->saved_num - 1);
506 data->saved_num = 0;
507 }
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);
520 }
521 return 0;
522}
523
524static ssize_t
525salinfo_log_write(struct file *file, const char __user *buffer, size_t count, loff_t *ppos)
526{
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;
530 char cmd[32];
531 size_t size;
532 u32 offset;
533 int cpu;
534
535 size = sizeof(cmd);
536 if (count < size)
537 size = count;
538 if (copy_from_user(cmd, buffer, size))
539 return -EFAULT;
540
541 if (sscanf(cmd, "read %d", &cpu) == 1) {
542 salinfo_log_new_read(cpu, data);
543 } else if (sscanf(cmd, "clear %d", &cpu) == 1) {
544 int ret;
545 if ((ret = salinfo_log_clear(data, cpu)))
546 count = ret;
547 } else if (sscanf(cmd, "oemdata %d %d", &cpu, &offset) == 2) {
548 if (data->state != STATE_LOG_RECORD && data->state != STATE_OEMDATA)
549 return -EINVAL;
550 if (offset > data->log_size - sizeof(efi_guid_t))
551 return -EINVAL;
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
558 };
559 call_on_cpu(cpu, salinfo_platform_oemdata_cpu, &parms);
560 if (parms.ret)
561 count = parms.ret;
562 } else
563 data->oemdata_size = 0;
564 } else
565 return -EINVAL;
566
567 return count;
568}
569
570static 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,
575 .llseek = default_llseek,
576};
577
578static int __cpuinit
579salinfo_cpu_callback(struct notifier_block *nb, unsigned long action, void *hcpu)
580{
581 unsigned int i, cpu = (unsigned long)hcpu;
582 unsigned long flags;
583 struct salinfo_data *data;
584 switch (action) {
585 case CPU_ONLINE:
586 case CPU_ONLINE_FROZEN:
587 spin_lock_irqsave(&data_saved_lock, flags);
588 for (i = 0, data = salinfo_data;
589 i < ARRAY_SIZE(salinfo_data);
590 ++i, ++data) {
591 cpu_set(cpu, data->cpu_event);
592 salinfo_work_to_do(data);
593 }
594 spin_unlock_irqrestore(&data_saved_lock, flags);
595 break;
596 case CPU_DEAD:
597 case CPU_DEAD_FROZEN:
598 spin_lock_irqsave(&data_saved_lock, flags);
599 for (i = 0, data = salinfo_data;
600 i < ARRAY_SIZE(salinfo_data);
601 ++i, ++data) {
602 struct salinfo_data_saved *data_saved;
603 int j;
604 for (j = ARRAY_SIZE(data->data_saved) - 1, data_saved = data->data_saved + j;
605 j >= 0;
606 --j, --data_saved) {
607 if (data_saved->buffer && data_saved->cpu == cpu) {
608 shift1_data_saved(data, j);
609 }
610 }
611 cpu_clear(cpu, data->cpu_event);
612 }
613 spin_unlock_irqrestore(&data_saved_lock, flags);
614 break;
615 }
616 return NOTIFY_OK;
617}
618
619static struct notifier_block salinfo_cpu_notifier __cpuinitdata =
620{
621 .notifier_call = salinfo_cpu_callback,
622 .priority = 0,
623};
624
625static int __init
626salinfo_init(void)
627{
628 struct proc_dir_entry *salinfo_dir; /* /proc/sal dir entry */
629 struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
630 struct proc_dir_entry *dir, *entry;
631 struct salinfo_data *data;
632 int i, j;
633
634 salinfo_dir = proc_mkdir("sal", NULL);
635 if (!salinfo_dir)
636 return 0;
637
638 for (i=0; i < NR_SALINFO_ENTRIES; i++) {
639 /* pass the feature bit in question as misc data */
640 *sdir++ = create_proc_read_entry (salinfo_entries[i].name, 0, salinfo_dir,
641 salinfo_read, (void *)salinfo_entries[i].feature);
642 }
643
644 for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
645 data = salinfo_data + i;
646 data->type = i;
647 sema_init(&data->mutex, 1);
648 dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
649 if (!dir)
650 continue;
651
652 entry = proc_create_data("event", S_IRUSR, dir,
653 &salinfo_event_fops, data);
654 if (!entry)
655 continue;
656 *sdir++ = entry;
657
658 entry = proc_create_data("data", S_IRUSR | S_IWUSR, dir,
659 &salinfo_data_fops, data);
660 if (!entry)
661 continue;
662 *sdir++ = entry;
663
664 /* we missed any events before now */
665 for_each_online_cpu(j)
666 cpu_set(j, data->cpu_event);
667
668 *sdir++ = dir;
669 }
670
671 *sdir++ = salinfo_dir;
672
673 init_timer(&salinfo_timer);
674 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
675 salinfo_timer.function = &salinfo_timeout;
676 add_timer(&salinfo_timer);
677
678 register_hotcpu_notifier(&salinfo_cpu_notifier);
679
680 return 0;
681}
682
683/*
684 * 'data' contains an integer that corresponds to the feature we're
685 * testing
686 */
687static int
688salinfo_read(char *page, char **start, off_t off, int count, int *eof, void *data)
689{
690 int len = 0;
691
692 len = sprintf(page, (sal_platform_features & (unsigned long)data) ? "1\n" : "0\n");
693
694 if (len <= off+count) *eof = 1;
695
696 *start = page + off;
697 len -= off;
698
699 if (len>count) len = count;
700 if (len<0) len = 0;
701
702 return len;
703}
704
705module_init(salinfo_init);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * salinfo.c
4 *
5 * Creates entries in /proc/sal for various system features.
6 *
7 * Copyright (c) 2003, 2006 Silicon Graphics, Inc. All rights reserved.
8 * Copyright (c) 2003 Hewlett-Packard Co
9 * Bjorn Helgaas <bjorn.helgaas@hp.com>
10 *
11 * 10/30/2001 jbarnes@sgi.com copied much of Stephane's palinfo
12 * code to create this file
13 * Oct 23 2003 kaos@sgi.com
14 * Replace IPI with set_cpus_allowed() to read a record from the required cpu.
15 * Redesign salinfo log processing to separate interrupt and user space
16 * contexts.
17 * Cache the record across multi-block reads from user space.
18 * Support > 64 cpus.
19 * Delete module_exit and MOD_INC/DEC_COUNT, salinfo cannot be a module.
20 *
21 * Jan 28 2004 kaos@sgi.com
22 * Periodically check for outstanding MCA or INIT records.
23 *
24 * Dec 5 2004 kaos@sgi.com
25 * Standardize which records are cleared automatically.
26 *
27 * Aug 18 2005 kaos@sgi.com
28 * mca.c may not pass a buffer, a NULL buffer just indicates that a new
29 * record is available in SAL.
30 * Replace some NR_CPUS by cpus_online, for hotplug cpu.
31 *
32 * Jan 5 2006 kaos@sgi.com
33 * Handle hotplug cpus coming online.
34 * Handle hotplug cpus going offline while they still have outstanding records.
35 * Use the cpu_* macros consistently.
36 * Replace the counting semaphore with a mutex and a test if the cpumask is non-empty.
37 * Modify the locking to make the test for "work to do" an atomic operation.
38 */
39
40#include <linux/capability.h>
41#include <linux/cpu.h>
42#include <linux/types.h>
43#include <linux/proc_fs.h>
44#include <linux/seq_file.h>
45#include <linux/module.h>
46#include <linux/smp.h>
47#include <linux/timer.h>
48#include <linux/vmalloc.h>
49#include <linux/semaphore.h>
50
51#include <asm/sal.h>
52#include <linux/uaccess.h>
53
54MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
55MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
56MODULE_LICENSE("GPL");
57
58typedef struct {
59 const char *name; /* name of the proc entry */
60 unsigned long feature; /* feature bit */
61 struct proc_dir_entry *entry; /* registered entry (removal) */
62} salinfo_entry_t;
63
64/*
65 * List {name,feature} pairs for every entry in /proc/sal/<feature>
66 * that this module exports
67 */
68static const 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, },
73};
74
75#define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)
76
77static char *salinfo_log_name[] = {
78 "mca",
79 "init",
80 "cmc",
81 "cpe",
82};
83
84static 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} */
88 1]; /* /proc/sal */
89
90/* Some records we get ourselves, some are accessed as saved data in buffers
91 * that are owned by mca.c.
92 */
93struct salinfo_data_saved {
94 u8* buffer;
95 u64 size;
96 u64 id;
97 int cpu;
98};
99
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.
106 *
107 * Start state is NO_DATA.
108 *
109 * 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.
115 *
116 * LOG_RECORD
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.
122 *
123 * OEMDATA
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.
129 *
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.
134 */
135enum salinfo_state {
136 STATE_NO_DATA,
137 STATE_LOG_RECORD,
138 STATE_OEMDATA,
139};
140
141struct salinfo_data {
142 cpumask_t cpu_event; /* which cpus have outstanding events */
143 wait_queue_head_t read_wait;
144 u8 *log_buffer;
145 u64 log_size;
146 u8 *oemdata; /* decoded oem data */
147 u64 oemdata_size;
148 int open; /* single-open to prevent races */
149 u8 type;
150 u8 saved_num; /* using a saved record? */
151 enum salinfo_state state :8; /* processing state */
152 u8 padding;
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 */
155};
156
157static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];
158
159static DEFINE_SPINLOCK(data_lock);
160static DEFINE_SPINLOCK(data_saved_lock);
161
162/** salinfo_platform_oemdata - optional callback to decode oemdata from an error
163 * record.
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).
167 *
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.
174 */
175int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);
176
177struct salinfo_platform_oemdata_parms {
178 const u8 *efi_guid;
179 u8 **oemdata;
180 u64 *oemdata_size;
181};
182
183static long
184salinfo_platform_oemdata_cpu(void *context)
185{
186 struct salinfo_platform_oemdata_parms *parms = context;
187
188 return salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
189}
190
191static void
192shift1_data_saved (struct salinfo_data *data, int shift)
193{
194 memcpy(data->data_saved+shift, data->data_saved+shift+1,
195 (ARRAY_SIZE(data->data_saved) - (shift+1)) * sizeof(data->data_saved[0]));
196 memset(data->data_saved + ARRAY_SIZE(data->data_saved) - 1, 0,
197 sizeof(data->data_saved[0]));
198}
199
200/* This routine is invoked in interrupt context. Note: mca.c enables
201 * interrupts before calling this code for CMC/CPE. MCA and INIT events are
202 * not irq safe, do not call any routines that use spinlocks, they may deadlock.
203 * MCA and INIT records are recorded, a timer event will look for any
204 * outstanding events and wake up the user space code.
205 *
206 * The buffer passed from mca.c points to the output from ia64_log_get. This is
207 * a persistent buffer but its contents can change between the interrupt and
208 * when user space processes the record. Save the record id to identify
209 * changes. If the buffer is NULL then just update the bitmap.
210 */
211void
212salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
213{
214 struct salinfo_data *data = salinfo_data + type;
215 struct salinfo_data_saved *data_saved;
216 unsigned long flags = 0;
217 int i;
218 int saved_size = ARRAY_SIZE(data->data_saved);
219
220 BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
221
222 if (irqsafe)
223 spin_lock_irqsave(&data_saved_lock, flags);
224 if (buffer) {
225 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
226 if (!data_saved->buffer)
227 break;
228 }
229 if (i == saved_size) {
230 if (!data->saved_num) {
231 shift1_data_saved(data, 0);
232 data_saved = data->data_saved + saved_size - 1;
233 } else
234 data_saved = NULL;
235 }
236 if (data_saved) {
237 data_saved->cpu = smp_processor_id();
238 data_saved->id = ((sal_log_record_header_t *)buffer)->id;
239 data_saved->size = size;
240 data_saved->buffer = buffer;
241 }
242 }
243 cpumask_set_cpu(smp_processor_id(), &data->cpu_event);
244 if (irqsafe) {
245 wake_up_interruptible(&data->read_wait);
246 spin_unlock_irqrestore(&data_saved_lock, flags);
247 }
248}
249
250/* Check for outstanding MCA/INIT records every minute (arbitrary) */
251#define SALINFO_TIMER_DELAY (60*HZ)
252static struct timer_list salinfo_timer;
253extern void ia64_mlogbuf_dump(void);
254
255static void
256salinfo_timeout_check(struct salinfo_data *data)
257{
258 if (!data->open)
259 return;
260 if (!cpumask_empty(&data->cpu_event))
261 wake_up_interruptible(&data->read_wait);
262}
263
264static void
265salinfo_timeout(struct timer_list *unused)
266{
267 ia64_mlogbuf_dump();
268 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
269 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_INIT);
270 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
271 add_timer(&salinfo_timer);
272}
273
274static int
275salinfo_event_open(struct inode *inode, struct file *file)
276{
277 if (!capable(CAP_SYS_ADMIN))
278 return -EPERM;
279 return 0;
280}
281
282static ssize_t
283salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
284{
285 struct salinfo_data *data = PDE_DATA(file_inode(file));
286 char cmd[32];
287 size_t size;
288 int i, n, cpu = -1;
289
290retry:
291 if (cpumask_empty(&data->cpu_event)) {
292 if (file->f_flags & O_NONBLOCK)
293 return -EAGAIN;
294 if (wait_event_interruptible(data->read_wait,
295 !cpumask_empty(&data->cpu_event)))
296 return -EINTR;
297 }
298
299 n = data->cpu_check;
300 for (i = 0; i < nr_cpu_ids; i++) {
301 if (cpumask_test_cpu(n, &data->cpu_event)) {
302 if (!cpu_online(n)) {
303 cpumask_clear_cpu(n, &data->cpu_event);
304 continue;
305 }
306 cpu = n;
307 break;
308 }
309 if (++n == nr_cpu_ids)
310 n = 0;
311 }
312
313 if (cpu == -1)
314 goto retry;
315
316 ia64_mlogbuf_dump();
317
318 /* for next read, start checking at next CPU */
319 data->cpu_check = cpu;
320 if (++data->cpu_check == nr_cpu_ids)
321 data->cpu_check = 0;
322
323 snprintf(cmd, sizeof(cmd), "read %d\n", cpu);
324
325 size = strlen(cmd);
326 if (size > count)
327 size = count;
328 if (copy_to_user(buffer, cmd, size))
329 return -EFAULT;
330
331 return size;
332}
333
334static const struct proc_ops salinfo_event_proc_ops = {
335 .proc_open = salinfo_event_open,
336 .proc_read = salinfo_event_read,
337 .proc_lseek = noop_llseek,
338};
339
340static int
341salinfo_log_open(struct inode *inode, struct file *file)
342{
343 struct salinfo_data *data = PDE_DATA(inode);
344
345 if (!capable(CAP_SYS_ADMIN))
346 return -EPERM;
347
348 spin_lock(&data_lock);
349 if (data->open) {
350 spin_unlock(&data_lock);
351 return -EBUSY;
352 }
353 data->open = 1;
354 spin_unlock(&data_lock);
355
356 if (data->state == STATE_NO_DATA &&
357 !(data->log_buffer = vmalloc(ia64_sal_get_state_info_size(data->type)))) {
358 data->open = 0;
359 return -ENOMEM;
360 }
361
362 return 0;
363}
364
365static int
366salinfo_log_release(struct inode *inode, struct file *file)
367{
368 struct salinfo_data *data = PDE_DATA(inode);
369
370 if (data->state == STATE_NO_DATA) {
371 vfree(data->log_buffer);
372 vfree(data->oemdata);
373 data->log_buffer = NULL;
374 data->oemdata = NULL;
375 }
376 spin_lock(&data_lock);
377 data->open = 0;
378 spin_unlock(&data_lock);
379 return 0;
380}
381
382static long
383salinfo_log_read_cpu(void *context)
384{
385 struct salinfo_data *data = context;
386 sal_log_record_header_t *rh;
387 data->log_size = ia64_sal_get_state_info(data->type, (u64 *) data->log_buffer);
388 rh = (sal_log_record_header_t *)(data->log_buffer);
389 /* Clear corrected errors as they are read from SAL */
390 if (rh->severity == sal_log_severity_corrected)
391 ia64_sal_clear_state_info(data->type);
392 return 0;
393}
394
395static void
396salinfo_log_new_read(int cpu, struct salinfo_data *data)
397{
398 struct salinfo_data_saved *data_saved;
399 unsigned long flags;
400 int i;
401 int saved_size = ARRAY_SIZE(data->data_saved);
402
403 data->saved_num = 0;
404 spin_lock_irqsave(&data_saved_lock, flags);
405retry:
406 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
407 if (data_saved->buffer && data_saved->cpu == cpu) {
408 sal_log_record_header_t *rh = (sal_log_record_header_t *)(data_saved->buffer);
409 data->log_size = data_saved->size;
410 memcpy(data->log_buffer, rh, data->log_size);
411 barrier(); /* id check must not be moved */
412 if (rh->id == data_saved->id) {
413 data->saved_num = i+1;
414 break;
415 }
416 /* saved record changed by mca.c since interrupt, discard it */
417 shift1_data_saved(data, i);
418 goto retry;
419 }
420 }
421 spin_unlock_irqrestore(&data_saved_lock, flags);
422
423 if (!data->saved_num)
424 work_on_cpu_safe(cpu, salinfo_log_read_cpu, data);
425 if (!data->log_size) {
426 data->state = STATE_NO_DATA;
427 cpumask_clear_cpu(cpu, &data->cpu_event);
428 } else {
429 data->state = STATE_LOG_RECORD;
430 }
431}
432
433static ssize_t
434salinfo_log_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
435{
436 struct salinfo_data *data = PDE_DATA(file_inode(file));
437 u8 *buf;
438 u64 bufsize;
439
440 if (data->state == STATE_LOG_RECORD) {
441 buf = data->log_buffer;
442 bufsize = data->log_size;
443 } else if (data->state == STATE_OEMDATA) {
444 buf = data->oemdata;
445 bufsize = data->oemdata_size;
446 } else {
447 buf = NULL;
448 bufsize = 0;
449 }
450 return simple_read_from_buffer(buffer, count, ppos, buf, bufsize);
451}
452
453static long
454salinfo_log_clear_cpu(void *context)
455{
456 struct salinfo_data *data = context;
457
458 ia64_sal_clear_state_info(data->type);
459 return 0;
460}
461
462static int
463salinfo_log_clear(struct salinfo_data *data, int cpu)
464{
465 sal_log_record_header_t *rh;
466 unsigned long flags;
467 spin_lock_irqsave(&data_saved_lock, flags);
468 data->state = STATE_NO_DATA;
469 if (!cpumask_test_cpu(cpu, &data->cpu_event)) {
470 spin_unlock_irqrestore(&data_saved_lock, flags);
471 return 0;
472 }
473 cpumask_clear_cpu(cpu, &data->cpu_event);
474 if (data->saved_num) {
475 shift1_data_saved(data, data->saved_num - 1);
476 data->saved_num = 0;
477 }
478 spin_unlock_irqrestore(&data_saved_lock, flags);
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 work_on_cpu_safe(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 spin_lock_irqsave(&data_saved_lock, flags);
487 cpumask_set_cpu(cpu, &data->cpu_event);
488 wake_up_interruptible(&data->read_wait);
489 spin_unlock_irqrestore(&data_saved_lock, flags);
490 }
491 return 0;
492}
493
494static ssize_t
495salinfo_log_write(struct file *file, const char __user *buffer, size_t count, loff_t *ppos)
496{
497 struct salinfo_data *data = PDE_DATA(file_inode(file));
498 char cmd[32];
499 size_t size;
500 u32 offset;
501 int cpu;
502
503 size = sizeof(cmd);
504 if (count < size)
505 size = count;
506 if (copy_from_user(cmd, buffer, size))
507 return -EFAULT;
508
509 if (sscanf(cmd, "read %d", &cpu) == 1) {
510 salinfo_log_new_read(cpu, data);
511 } else if (sscanf(cmd, "clear %d", &cpu) == 1) {
512 int ret;
513 if ((ret = salinfo_log_clear(data, cpu)))
514 count = ret;
515 } else if (sscanf(cmd, "oemdata %d %d", &cpu, &offset) == 2) {
516 if (data->state != STATE_LOG_RECORD && data->state != STATE_OEMDATA)
517 return -EINVAL;
518 if (offset > data->log_size - sizeof(efi_guid_t))
519 return -EINVAL;
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
526 };
527 count = work_on_cpu_safe(cpu, salinfo_platform_oemdata_cpu,
528 &parms);
529 } else
530 data->oemdata_size = 0;
531 } else
532 return -EINVAL;
533
534 return count;
535}
536
537static const struct proc_ops salinfo_data_proc_ops = {
538 .proc_open = salinfo_log_open,
539 .proc_release = salinfo_log_release,
540 .proc_read = salinfo_log_read,
541 .proc_write = salinfo_log_write,
542 .proc_lseek = default_llseek,
543};
544
545static int salinfo_cpu_online(unsigned int cpu)
546{
547 unsigned int i, end = ARRAY_SIZE(salinfo_data);
548 struct salinfo_data *data;
549
550 spin_lock_irq(&data_saved_lock);
551 for (i = 0, data = salinfo_data; i < end; ++i, ++data) {
552 cpumask_set_cpu(cpu, &data->cpu_event);
553 wake_up_interruptible(&data->read_wait);
554 }
555 spin_unlock_irq(&data_saved_lock);
556 return 0;
557}
558
559static int salinfo_cpu_pre_down(unsigned int cpu)
560{
561 unsigned int i, end = ARRAY_SIZE(salinfo_data);
562 struct salinfo_data *data;
563
564 spin_lock_irq(&data_saved_lock);
565 for (i = 0, data = salinfo_data; i < end; ++i, ++data) {
566 struct salinfo_data_saved *data_saved;
567 int j = ARRAY_SIZE(data->data_saved) - 1;
568
569 for (data_saved = data->data_saved + j; j >= 0;
570 --j, --data_saved) {
571 if (data_saved->buffer && data_saved->cpu == cpu)
572 shift1_data_saved(data, j);
573 }
574 cpumask_clear_cpu(cpu, &data->cpu_event);
575 }
576 spin_unlock_irq(&data_saved_lock);
577 return 0;
578}
579
580/*
581 * 'data' contains an integer that corresponds to the feature we're
582 * testing
583 */
584static int proc_salinfo_show(struct seq_file *m, void *v)
585{
586 unsigned long data = (unsigned long)v;
587 seq_puts(m, (sal_platform_features & data) ? "1\n" : "0\n");
588 return 0;
589}
590
591static int __init
592salinfo_init(void)
593{
594 struct proc_dir_entry *salinfo_dir; /* /proc/sal dir entry */
595 struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
596 struct proc_dir_entry *dir, *entry;
597 struct salinfo_data *data;
598 int i;
599
600 salinfo_dir = proc_mkdir("sal", NULL);
601 if (!salinfo_dir)
602 return 0;
603
604 for (i=0; i < NR_SALINFO_ENTRIES; i++) {
605 /* pass the feature bit in question as misc data */
606 *sdir++ = proc_create_single_data(salinfo_entries[i].name, 0,
607 salinfo_dir, proc_salinfo_show,
608 (void *)salinfo_entries[i].feature);
609 }
610
611 for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
612 data = salinfo_data + i;
613 data->type = i;
614 init_waitqueue_head(&data->read_wait);
615 dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
616 if (!dir)
617 continue;
618
619 entry = proc_create_data("event", S_IRUSR, dir,
620 &salinfo_event_proc_ops, data);
621 if (!entry)
622 continue;
623 *sdir++ = entry;
624
625 entry = proc_create_data("data", S_IRUSR | S_IWUSR, dir,
626 &salinfo_data_proc_ops, data);
627 if (!entry)
628 continue;
629 *sdir++ = entry;
630
631 *sdir++ = dir;
632 }
633
634 *sdir++ = salinfo_dir;
635
636 timer_setup(&salinfo_timer, salinfo_timeout, 0);
637 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
638 add_timer(&salinfo_timer);
639
640 i = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "ia64/salinfo:online",
641 salinfo_cpu_online, salinfo_cpu_pre_down);
642 WARN_ON(i < 0);
643 return 0;
644}
645
646module_init(salinfo_init);