Loading...
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 file_operations salinfo_event_fops = {
335 .open = salinfo_event_open,
336 .read = salinfo_event_read,
337 .llseek = 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 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 .llseek = 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_fops, data);
621 if (!entry)
622 continue;
623 *sdir++ = entry;
624
625 entry = proc_create_data("data", S_IRUSR | S_IWUSR, dir,
626 &salinfo_data_fops, 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);
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/seq_file.h>
44#include <linux/module.h>
45#include <linux/smp.h>
46#include <linux/timer.h>
47#include <linux/vmalloc.h>
48#include <linux/semaphore.h>
49
50#include <asm/sal.h>
51#include <linux/uaccess.h>
52
53MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
54MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
55MODULE_LICENSE("GPL");
56
57static const struct file_operations proc_salinfo_fops;
58
59typedef struct {
60 const char *name; /* name of the proc entry */
61 unsigned long feature; /* feature bit */
62 struct proc_dir_entry *entry; /* registered entry (removal) */
63} salinfo_entry_t;
64
65/*
66 * List {name,feature} pairs for every entry in /proc/sal/<feature>
67 * that this module exports
68 */
69static const 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, },
74};
75
76#define NR_SALINFO_ENTRIES ARRAY_SIZE(salinfo_entries)
77
78static char *salinfo_log_name[] = {
79 "mca",
80 "init",
81 "cmc",
82 "cpe",
83};
84
85static 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} */
89 1]; /* /proc/sal */
90
91/* Some records we get ourselves, some are accessed as saved data in buffers
92 * that are owned by mca.c.
93 */
94struct salinfo_data_saved {
95 u8* buffer;
96 u64 size;
97 u64 id;
98 int cpu;
99};
100
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.
107 *
108 * Start state is NO_DATA.
109 *
110 * 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.
116 *
117 * LOG_RECORD
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.
123 *
124 * OEMDATA
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.
130 *
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.
135 */
136enum salinfo_state {
137 STATE_NO_DATA,
138 STATE_LOG_RECORD,
139 STATE_OEMDATA,
140};
141
142struct salinfo_data {
143 cpumask_t cpu_event; /* which cpus have outstanding events */
144 wait_queue_head_t read_wait;
145 u8 *log_buffer;
146 u64 log_size;
147 u8 *oemdata; /* decoded oem data */
148 u64 oemdata_size;
149 int open; /* single-open to prevent races */
150 u8 type;
151 u8 saved_num; /* using a saved record? */
152 enum salinfo_state state :8; /* processing state */
153 u8 padding;
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 */
156};
157
158static struct salinfo_data salinfo_data[ARRAY_SIZE(salinfo_log_name)];
159
160static DEFINE_SPINLOCK(data_lock);
161static DEFINE_SPINLOCK(data_saved_lock);
162
163/** salinfo_platform_oemdata - optional callback to decode oemdata from an error
164 * record.
165 * @sect_header: pointer to the start of the section to decode.
166 * @oemdata: returns vmalloc area containing the decoded output.
167 * @oemdata_size: returns length of decoded output (strlen).
168 *
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.
175 */
176int (*salinfo_platform_oemdata)(const u8 *sect_header, u8 **oemdata, u64 *oemdata_size);
177
178struct salinfo_platform_oemdata_parms {
179 const u8 *efi_guid;
180 u8 **oemdata;
181 u64 *oemdata_size;
182 int ret;
183};
184
185static void
186salinfo_platform_oemdata_cpu(void *context)
187{
188 struct salinfo_platform_oemdata_parms *parms = context;
189 parms->ret = salinfo_platform_oemdata(parms->efi_guid, parms->oemdata, parms->oemdata_size);
190}
191
192static void
193shift1_data_saved (struct salinfo_data *data, int shift)
194{
195 memcpy(data->data_saved+shift, data->data_saved+shift+1,
196 (ARRAY_SIZE(data->data_saved) - (shift+1)) * sizeof(data->data_saved[0]));
197 memset(data->data_saved + ARRAY_SIZE(data->data_saved) - 1, 0,
198 sizeof(data->data_saved[0]));
199}
200
201/* This routine is invoked in interrupt context. Note: mca.c enables
202 * interrupts before calling this code for CMC/CPE. MCA and INIT events are
203 * not irq safe, do not call any routines that use spinlocks, they may deadlock.
204 * MCA and INIT records are recorded, a timer event will look for any
205 * outstanding events and wake up the user space code.
206 *
207 * The buffer passed from mca.c points to the output from ia64_log_get. This is
208 * a persistent buffer but its contents can change between the interrupt and
209 * when user space processes the record. Save the record id to identify
210 * changes. If the buffer is NULL then just update the bitmap.
211 */
212void
213salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
214{
215 struct salinfo_data *data = salinfo_data + type;
216 struct salinfo_data_saved *data_saved;
217 unsigned long flags = 0;
218 int i;
219 int saved_size = ARRAY_SIZE(data->data_saved);
220
221 BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
222
223 if (irqsafe)
224 spin_lock_irqsave(&data_saved_lock, flags);
225 if (buffer) {
226 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
227 if (!data_saved->buffer)
228 break;
229 }
230 if (i == saved_size) {
231 if (!data->saved_num) {
232 shift1_data_saved(data, 0);
233 data_saved = data->data_saved + saved_size - 1;
234 } else
235 data_saved = NULL;
236 }
237 if (data_saved) {
238 data_saved->cpu = smp_processor_id();
239 data_saved->id = ((sal_log_record_header_t *)buffer)->id;
240 data_saved->size = size;
241 data_saved->buffer = buffer;
242 }
243 }
244 cpumask_set_cpu(smp_processor_id(), &data->cpu_event);
245 if (irqsafe) {
246 wake_up_interruptible(&data->read_wait);
247 spin_unlock_irqrestore(&data_saved_lock, flags);
248 }
249}
250
251/* Check for outstanding MCA/INIT records every minute (arbitrary) */
252#define SALINFO_TIMER_DELAY (60*HZ)
253static struct timer_list salinfo_timer;
254extern void ia64_mlogbuf_dump(void);
255
256static void
257salinfo_timeout_check(struct salinfo_data *data)
258{
259 if (!data->open)
260 return;
261 if (!cpumask_empty(&data->cpu_event))
262 wake_up_interruptible(&data->read_wait);
263}
264
265static void
266salinfo_timeout (unsigned long arg)
267{
268 ia64_mlogbuf_dump();
269 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_MCA);
270 salinfo_timeout_check(salinfo_data + SAL_INFO_TYPE_INIT);
271 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
272 add_timer(&salinfo_timer);
273}
274
275static int
276salinfo_event_open(struct inode *inode, struct file *file)
277{
278 if (!capable(CAP_SYS_ADMIN))
279 return -EPERM;
280 return 0;
281}
282
283static ssize_t
284salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
285{
286 struct salinfo_data *data = PDE_DATA(file_inode(file));
287 char cmd[32];
288 size_t size;
289 int i, n, cpu = -1;
290
291retry:
292 if (cpumask_empty(&data->cpu_event)) {
293 if (file->f_flags & O_NONBLOCK)
294 return -EAGAIN;
295 if (wait_event_interruptible(data->read_wait,
296 !cpumask_empty(&data->cpu_event)))
297 return -EINTR;
298 }
299
300 n = data->cpu_check;
301 for (i = 0; i < nr_cpu_ids; i++) {
302 if (cpumask_test_cpu(n, &data->cpu_event)) {
303 if (!cpu_online(n)) {
304 cpumask_clear_cpu(n, &data->cpu_event);
305 continue;
306 }
307 cpu = n;
308 break;
309 }
310 if (++n == nr_cpu_ids)
311 n = 0;
312 }
313
314 if (cpu == -1)
315 goto retry;
316
317 ia64_mlogbuf_dump();
318
319 /* for next read, start checking at next CPU */
320 data->cpu_check = cpu;
321 if (++data->cpu_check == nr_cpu_ids)
322 data->cpu_check = 0;
323
324 snprintf(cmd, sizeof(cmd), "read %d\n", cpu);
325
326 size = strlen(cmd);
327 if (size > count)
328 size = count;
329 if (copy_to_user(buffer, cmd, size))
330 return -EFAULT;
331
332 return size;
333}
334
335static const struct file_operations salinfo_event_fops = {
336 .open = salinfo_event_open,
337 .read = salinfo_event_read,
338 .llseek = noop_llseek,
339};
340
341static int
342salinfo_log_open(struct inode *inode, struct file *file)
343{
344 struct salinfo_data *data = PDE_DATA(inode);
345
346 if (!capable(CAP_SYS_ADMIN))
347 return -EPERM;
348
349 spin_lock(&data_lock);
350 if (data->open) {
351 spin_unlock(&data_lock);
352 return -EBUSY;
353 }
354 data->open = 1;
355 spin_unlock(&data_lock);
356
357 if (data->state == STATE_NO_DATA &&
358 !(data->log_buffer = vmalloc(ia64_sal_get_state_info_size(data->type)))) {
359 data->open = 0;
360 return -ENOMEM;
361 }
362
363 return 0;
364}
365
366static int
367salinfo_log_release(struct inode *inode, struct file *file)
368{
369 struct salinfo_data *data = PDE_DATA(inode);
370
371 if (data->state == STATE_NO_DATA) {
372 vfree(data->log_buffer);
373 vfree(data->oemdata);
374 data->log_buffer = NULL;
375 data->oemdata = NULL;
376 }
377 spin_lock(&data_lock);
378 data->open = 0;
379 spin_unlock(&data_lock);
380 return 0;
381}
382
383static void
384call_on_cpu(int cpu, void (*fn)(void *), void *arg)
385{
386 cpumask_t save_cpus_allowed = current->cpus_allowed;
387 set_cpus_allowed_ptr(current, cpumask_of(cpu));
388 (*fn)(arg);
389 set_cpus_allowed_ptr(current, &save_cpus_allowed);
390}
391
392static void
393salinfo_log_read_cpu(void *context)
394{
395 struct salinfo_data *data = context;
396 sal_log_record_header_t *rh;
397 data->log_size = ia64_sal_get_state_info(data->type, (u64 *) data->log_buffer);
398 rh = (sal_log_record_header_t *)(data->log_buffer);
399 /* Clear corrected errors as they are read from SAL */
400 if (rh->severity == sal_log_severity_corrected)
401 ia64_sal_clear_state_info(data->type);
402}
403
404static void
405salinfo_log_new_read(int cpu, struct salinfo_data *data)
406{
407 struct salinfo_data_saved *data_saved;
408 unsigned long flags;
409 int i;
410 int saved_size = ARRAY_SIZE(data->data_saved);
411
412 data->saved_num = 0;
413 spin_lock_irqsave(&data_saved_lock, flags);
414retry:
415 for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
416 if (data_saved->buffer && data_saved->cpu == cpu) {
417 sal_log_record_header_t *rh = (sal_log_record_header_t *)(data_saved->buffer);
418 data->log_size = data_saved->size;
419 memcpy(data->log_buffer, rh, data->log_size);
420 barrier(); /* id check must not be moved */
421 if (rh->id == data_saved->id) {
422 data->saved_num = i+1;
423 break;
424 }
425 /* saved record changed by mca.c since interrupt, discard it */
426 shift1_data_saved(data, i);
427 goto retry;
428 }
429 }
430 spin_unlock_irqrestore(&data_saved_lock, flags);
431
432 if (!data->saved_num)
433 call_on_cpu(cpu, salinfo_log_read_cpu, data);
434 if (!data->log_size) {
435 data->state = STATE_NO_DATA;
436 cpumask_clear_cpu(cpu, &data->cpu_event);
437 } else {
438 data->state = STATE_LOG_RECORD;
439 }
440}
441
442static ssize_t
443salinfo_log_read(struct file *file, char __user *buffer, size_t count, loff_t *ppos)
444{
445 struct salinfo_data *data = PDE_DATA(file_inode(file));
446 u8 *buf;
447 u64 bufsize;
448
449 if (data->state == STATE_LOG_RECORD) {
450 buf = data->log_buffer;
451 bufsize = data->log_size;
452 } else if (data->state == STATE_OEMDATA) {
453 buf = data->oemdata;
454 bufsize = data->oemdata_size;
455 } else {
456 buf = NULL;
457 bufsize = 0;
458 }
459 return simple_read_from_buffer(buffer, count, ppos, buf, bufsize);
460}
461
462static void
463salinfo_log_clear_cpu(void *context)
464{
465 struct salinfo_data *data = context;
466 ia64_sal_clear_state_info(data->type);
467}
468
469static int
470salinfo_log_clear(struct salinfo_data *data, int cpu)
471{
472 sal_log_record_header_t *rh;
473 unsigned long flags;
474 spin_lock_irqsave(&data_saved_lock, flags);
475 data->state = STATE_NO_DATA;
476 if (!cpumask_test_cpu(cpu, &data->cpu_event)) {
477 spin_unlock_irqrestore(&data_saved_lock, flags);
478 return 0;
479 }
480 cpumask_clear_cpu(cpu, &data->cpu_event);
481 if (data->saved_num) {
482 shift1_data_saved(data, data->saved_num - 1);
483 data->saved_num = 0;
484 }
485 spin_unlock_irqrestore(&data_saved_lock, flags);
486 rh = (sal_log_record_header_t *)(data->log_buffer);
487 /* Corrected errors have already been cleared from SAL */
488 if (rh->severity != sal_log_severity_corrected)
489 call_on_cpu(cpu, salinfo_log_clear_cpu, data);
490 /* clearing a record may make a new record visible */
491 salinfo_log_new_read(cpu, data);
492 if (data->state == STATE_LOG_RECORD) {
493 spin_lock_irqsave(&data_saved_lock, flags);
494 cpumask_set_cpu(cpu, &data->cpu_event);
495 wake_up_interruptible(&data->read_wait);
496 spin_unlock_irqrestore(&data_saved_lock, flags);
497 }
498 return 0;
499}
500
501static ssize_t
502salinfo_log_write(struct file *file, const char __user *buffer, size_t count, loff_t *ppos)
503{
504 struct salinfo_data *data = PDE_DATA(file_inode(file));
505 char cmd[32];
506 size_t size;
507 u32 offset;
508 int cpu;
509
510 size = sizeof(cmd);
511 if (count < size)
512 size = count;
513 if (copy_from_user(cmd, buffer, size))
514 return -EFAULT;
515
516 if (sscanf(cmd, "read %d", &cpu) == 1) {
517 salinfo_log_new_read(cpu, data);
518 } else if (sscanf(cmd, "clear %d", &cpu) == 1) {
519 int ret;
520 if ((ret = salinfo_log_clear(data, cpu)))
521 count = ret;
522 } else if (sscanf(cmd, "oemdata %d %d", &cpu, &offset) == 2) {
523 if (data->state != STATE_LOG_RECORD && data->state != STATE_OEMDATA)
524 return -EINVAL;
525 if (offset > data->log_size - sizeof(efi_guid_t))
526 return -EINVAL;
527 data->state = STATE_OEMDATA;
528 if (salinfo_platform_oemdata) {
529 struct salinfo_platform_oemdata_parms parms = {
530 .efi_guid = data->log_buffer + offset,
531 .oemdata = &data->oemdata,
532 .oemdata_size = &data->oemdata_size
533 };
534 call_on_cpu(cpu, salinfo_platform_oemdata_cpu, &parms);
535 if (parms.ret)
536 count = parms.ret;
537 } else
538 data->oemdata_size = 0;
539 } else
540 return -EINVAL;
541
542 return count;
543}
544
545static const struct file_operations salinfo_data_fops = {
546 .open = salinfo_log_open,
547 .release = salinfo_log_release,
548 .read = salinfo_log_read,
549 .write = salinfo_log_write,
550 .llseek = default_llseek,
551};
552
553static int salinfo_cpu_online(unsigned int cpu)
554{
555 unsigned int i, end = ARRAY_SIZE(salinfo_data);
556 struct salinfo_data *data;
557
558 spin_lock_irq(&data_saved_lock);
559 for (i = 0, data = salinfo_data; i < end; ++i, ++data) {
560 cpumask_set_cpu(cpu, &data->cpu_event);
561 wake_up_interruptible(&data->read_wait);
562 }
563 spin_unlock_irq(&data_saved_lock);
564 return 0;
565}
566
567static int salinfo_cpu_pre_down(unsigned int cpu)
568{
569 unsigned int i, end = ARRAY_SIZE(salinfo_data);
570 struct salinfo_data *data;
571
572 spin_lock_irq(&data_saved_lock);
573 for (i = 0, data = salinfo_data; i < end; ++i, ++data) {
574 struct salinfo_data_saved *data_saved;
575 int j = ARRAY_SIZE(data->data_saved) - 1;
576
577 for (data_saved = data->data_saved + j; j >= 0;
578 --j, --data_saved) {
579 if (data_saved->buffer && data_saved->cpu == cpu)
580 shift1_data_saved(data, j);
581 }
582 cpumask_clear_cpu(cpu, &data->cpu_event);
583 }
584 spin_unlock_irq(&data_saved_lock);
585 return 0;
586}
587
588static int __init
589salinfo_init(void)
590{
591 struct proc_dir_entry *salinfo_dir; /* /proc/sal dir entry */
592 struct proc_dir_entry **sdir = salinfo_proc_entries; /* keeps track of every entry */
593 struct proc_dir_entry *dir, *entry;
594 struct salinfo_data *data;
595 int i;
596
597 salinfo_dir = proc_mkdir("sal", NULL);
598 if (!salinfo_dir)
599 return 0;
600
601 for (i=0; i < NR_SALINFO_ENTRIES; i++) {
602 /* pass the feature bit in question as misc data */
603 *sdir++ = proc_create_data(salinfo_entries[i].name, 0, salinfo_dir,
604 &proc_salinfo_fops,
605 (void *)salinfo_entries[i].feature);
606 }
607
608 for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
609 data = salinfo_data + i;
610 data->type = i;
611 init_waitqueue_head(&data->read_wait);
612 dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
613 if (!dir)
614 continue;
615
616 entry = proc_create_data("event", S_IRUSR, dir,
617 &salinfo_event_fops, data);
618 if (!entry)
619 continue;
620 *sdir++ = entry;
621
622 entry = proc_create_data("data", S_IRUSR | S_IWUSR, dir,
623 &salinfo_data_fops, data);
624 if (!entry)
625 continue;
626 *sdir++ = entry;
627
628 *sdir++ = dir;
629 }
630
631 *sdir++ = salinfo_dir;
632
633 init_timer(&salinfo_timer);
634 salinfo_timer.expires = jiffies + SALINFO_TIMER_DELAY;
635 salinfo_timer.function = &salinfo_timeout;
636 add_timer(&salinfo_timer);
637
638 i = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "ia64/salinfo:online",
639 salinfo_cpu_online, salinfo_cpu_pre_down);
640 WARN_ON(i < 0);
641 return 0;
642}
643
644/*
645 * 'data' contains an integer that corresponds to the feature we're
646 * testing
647 */
648static int proc_salinfo_show(struct seq_file *m, void *v)
649{
650 unsigned long data = (unsigned long)v;
651 seq_puts(m, (sal_platform_features & data) ? "1\n" : "0\n");
652 return 0;
653}
654
655static int proc_salinfo_open(struct inode *inode, struct file *file)
656{
657 return single_open(file, proc_salinfo_show, PDE_DATA(inode));
658}
659
660static const struct file_operations proc_salinfo_fops = {
661 .open = proc_salinfo_open,
662 .read = seq_read,
663 .llseek = seq_lseek,
664 .release = single_release,
665};
666
667module_init(salinfo_init);