1 // SPDX-License-Identifier: GPL-2.0-only
5 * Creates entries in /proc/sal for various system features.
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>
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
17 * Cache the record across multi-block reads from user space.
19 * Delete module_exit and MOD_INC/DEC_COUNT, salinfo cannot be a module.
21 * Jan 28 2004 kaos@sgi.com
22 * Periodically check for outstanding MCA or INIT records.
24 * Dec 5 2004 kaos@sgi.com
25 * Standardize which records are cleared automatically.
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.
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.
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>
52 #include <linux/uaccess.h>
54 MODULE_AUTHOR("Jesse Barnes <jbarnes@sgi.com>");
55 MODULE_DESCRIPTION("/proc interface to IA-64 SAL features");
56 MODULE_LICENSE("GPL");
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 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
, },
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 wait_queue_head_t read_wait
;
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 salinfo_platform_oemdata_cpu(void *context
)
186 struct salinfo_platform_oemdata_parms
*parms
= context
;
188 return salinfo_platform_oemdata(parms
->efi_guid
, parms
->oemdata
, parms
->oemdata_size
);
192 shift1_data_saved (struct salinfo_data
*data
, int shift
)
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]));
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.
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.
212 salinfo_log_wakeup(int type
, u8
*buffer
, u64 size
, int irqsafe
)
214 struct salinfo_data
*data
= salinfo_data
+ type
;
215 struct salinfo_data_saved
*data_saved
;
216 unsigned long flags
= 0;
218 int saved_size
= ARRAY_SIZE(data
->data_saved
);
220 BUG_ON(type
>= ARRAY_SIZE(salinfo_log_name
));
223 spin_lock_irqsave(&data_saved_lock
, flags
);
225 for (i
= 0, data_saved
= data
->data_saved
; i
< saved_size
; ++i
, ++data_saved
) {
226 if (!data_saved
->buffer
)
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;
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
;
243 cpumask_set_cpu(smp_processor_id(), &data
->cpu_event
);
245 wake_up_interruptible(&data
->read_wait
);
246 spin_unlock_irqrestore(&data_saved_lock
, flags
);
250 /* Check for outstanding MCA/INIT records every minute (arbitrary) */
251 #define SALINFO_TIMER_DELAY (60*HZ)
252 static struct timer_list salinfo_timer
;
253 extern void ia64_mlogbuf_dump(void);
256 salinfo_timeout_check(struct salinfo_data
*data
)
260 if (!cpumask_empty(&data
->cpu_event
))
261 wake_up_interruptible(&data
->read_wait
);
265 salinfo_timeout(struct timer_list
*unused
)
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
);
275 salinfo_event_open(struct inode
*inode
, struct file
*file
)
277 if (!capable(CAP_SYS_ADMIN
))
283 salinfo_event_read(struct file
*file
, char __user
*buffer
, size_t count
, loff_t
*ppos
)
285 struct salinfo_data
*data
= PDE_DATA(file_inode(file
));
291 if (cpumask_empty(&data
->cpu_event
)) {
292 if (file
->f_flags
& O_NONBLOCK
)
294 if (wait_event_interruptible(data
->read_wait
,
295 !cpumask_empty(&data
->cpu_event
)))
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
);
309 if (++n
== nr_cpu_ids
)
318 /* for next read, start checking at next CPU */
319 data
->cpu_check
= cpu
;
320 if (++data
->cpu_check
== nr_cpu_ids
)
323 snprintf(cmd
, sizeof(cmd
), "read %d\n", cpu
);
328 if (copy_to_user(buffer
, cmd
, size
))
334 static 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
,
341 salinfo_log_open(struct inode
*inode
, struct file
*file
)
343 struct salinfo_data
*data
= PDE_DATA(inode
);
345 if (!capable(CAP_SYS_ADMIN
))
348 spin_lock(&data_lock
);
350 spin_unlock(&data_lock
);
354 spin_unlock(&data_lock
);
356 if (data
->state
== STATE_NO_DATA
&&
357 !(data
->log_buffer
= vmalloc(ia64_sal_get_state_info_size(data
->type
)))) {
366 salinfo_log_release(struct inode
*inode
, struct file
*file
)
368 struct salinfo_data
*data
= PDE_DATA(inode
);
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
;
376 spin_lock(&data_lock
);
378 spin_unlock(&data_lock
);
383 salinfo_log_read_cpu(void *context
)
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
);
396 salinfo_log_new_read(int cpu
, struct salinfo_data
*data
)
398 struct salinfo_data_saved
*data_saved
;
401 int saved_size
= ARRAY_SIZE(data
->data_saved
);
404 spin_lock_irqsave(&data_saved_lock
, flags
);
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;
416 /* saved record changed by mca.c since interrupt, discard it */
417 shift1_data_saved(data
, i
);
421 spin_unlock_irqrestore(&data_saved_lock
, flags
);
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
);
429 data
->state
= STATE_LOG_RECORD
;
434 salinfo_log_read(struct file
*file
, char __user
*buffer
, size_t count
, loff_t
*ppos
)
436 struct salinfo_data
*data
= PDE_DATA(file_inode(file
));
440 if (data
->state
== STATE_LOG_RECORD
) {
441 buf
= data
->log_buffer
;
442 bufsize
= data
->log_size
;
443 } else if (data
->state
== STATE_OEMDATA
) {
445 bufsize
= data
->oemdata_size
;
450 return simple_read_from_buffer(buffer
, count
, ppos
, buf
, bufsize
);
454 salinfo_log_clear_cpu(void *context
)
456 struct salinfo_data
*data
= context
;
458 ia64_sal_clear_state_info(data
->type
);
463 salinfo_log_clear(struct salinfo_data
*data
, int cpu
)
465 sal_log_record_header_t
*rh
;
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
);
473 cpumask_clear_cpu(cpu
, &data
->cpu_event
);
474 if (data
->saved_num
) {
475 shift1_data_saved(data
, data
->saved_num
- 1);
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
);
495 salinfo_log_write(struct file
*file
, const char __user
*buffer
, size_t count
, loff_t
*ppos
)
497 struct salinfo_data
*data
= PDE_DATA(file_inode(file
));
506 if (copy_from_user(cmd
, buffer
, size
))
509 if (sscanf(cmd
, "read %d", &cpu
) == 1) {
510 salinfo_log_new_read(cpu
, data
);
511 } else if (sscanf(cmd
, "clear %d", &cpu
) == 1) {
513 if ((ret
= salinfo_log_clear(data
, cpu
)))
515 } else if (sscanf(cmd
, "oemdata %d %d", &cpu
, &offset
) == 2) {
516 if (data
->state
!= STATE_LOG_RECORD
&& data
->state
!= STATE_OEMDATA
)
518 if (offset
> data
->log_size
- sizeof(efi_guid_t
))
520 data
->state
= STATE_OEMDATA
;
521 if (salinfo_platform_oemdata
) {
522 struct salinfo_platform_oemdata_parms parms
= {
523 .efi_guid
= data
->log_buffer
+ offset
,
524 .oemdata
= &data
->oemdata
,
525 .oemdata_size
= &data
->oemdata_size
527 count
= work_on_cpu_safe(cpu
, salinfo_platform_oemdata_cpu
,
530 data
->oemdata_size
= 0;
537 static 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
,
545 static int salinfo_cpu_online(unsigned int cpu
)
547 unsigned int i
, end
= ARRAY_SIZE(salinfo_data
);
548 struct salinfo_data
*data
;
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
);
555 spin_unlock_irq(&data_saved_lock
);
559 static int salinfo_cpu_pre_down(unsigned int cpu
)
561 unsigned int i
, end
= ARRAY_SIZE(salinfo_data
);
562 struct salinfo_data
*data
;
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;
569 for (data_saved
= data
->data_saved
+ j
; j
>= 0;
571 if (data_saved
->buffer
&& data_saved
->cpu
== cpu
)
572 shift1_data_saved(data
, j
);
574 cpumask_clear_cpu(cpu
, &data
->cpu_event
);
576 spin_unlock_irq(&data_saved_lock
);
581 * 'data' contains an integer that corresponds to the feature we're
584 static int proc_salinfo_show(struct seq_file
*m
, void *v
)
586 unsigned long data
= (unsigned long)v
;
587 seq_puts(m
, (sal_platform_features
& data
) ? "1\n" : "0\n");
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
;
600 salinfo_dir
= proc_mkdir("sal", NULL
);
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
);
611 for (i
= 0; i
< ARRAY_SIZE(salinfo_log_name
); i
++) {
612 data
= salinfo_data
+ i
;
614 init_waitqueue_head(&data
->read_wait
);
615 dir
= proc_mkdir(salinfo_log_name
[i
], salinfo_dir
);
619 entry
= proc_create_data("event", S_IRUSR
, dir
,
620 &salinfo_event_proc_ops
, data
);
625 entry
= proc_create_data("data", S_IRUSR
| S_IWUSR
, dir
,
626 &salinfo_data_proc_ops
, data
);
634 *sdir
++ = salinfo_dir
;
636 timer_setup(&salinfo_timer
, salinfo_timeout
, 0);
637 salinfo_timer
.expires
= jiffies
+ SALINFO_TIMER_DELAY
;
638 add_timer(&salinfo_timer
);
640 i
= cpuhp_setup_state(CPUHP_AP_ONLINE_DYN
, "ia64/salinfo:online",
641 salinfo_cpu_online
, salinfo_cpu_pre_down
);
646 module_init(salinfo_init
);