Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 *
4 * Procedures for interfacing to the RTAS on CHRP machines.
5 *
6 * Peter Bergner, IBM March 2001.
7 * Copyright (C) 2001 IBM.
8 */
9
10#define pr_fmt(fmt) "rtas: " fmt
11
12#include <linux/capability.h>
13#include <linux/delay.h>
14#include <linux/export.h>
15#include <linux/init.h>
16#include <linux/kernel.h>
17#include <linux/memblock.h>
18#include <linux/of.h>
19#include <linux/of_fdt.h>
20#include <linux/reboot.h>
21#include <linux/sched.h>
22#include <linux/security.h>
23#include <linux/slab.h>
24#include <linux/spinlock.h>
25#include <linux/stdarg.h>
26#include <linux/syscalls.h>
27#include <linux/types.h>
28#include <linux/uaccess.h>
29
30#include <asm/delay.h>
31#include <asm/firmware.h>
32#include <asm/interrupt.h>
33#include <asm/machdep.h>
34#include <asm/mmu.h>
35#include <asm/page.h>
36#include <asm/rtas.h>
37#include <asm/time.h>
38#include <asm/udbg.h>
39
40/* This is here deliberately so it's only used in this file */
41void enter_rtas(unsigned long);
42
43static inline void do_enter_rtas(unsigned long args)
44{
45 unsigned long msr;
46
47 /*
48 * Make sure MSR[RI] is currently enabled as it will be forced later
49 * in enter_rtas.
50 */
51 msr = mfmsr();
52 BUG_ON(!(msr & MSR_RI));
53
54 BUG_ON(!irqs_disabled());
55
56 hard_irq_disable(); /* Ensure MSR[EE] is disabled on PPC64 */
57
58 enter_rtas(args);
59
60 srr_regs_clobbered(); /* rtas uses SRRs, invalidate */
61}
62
63struct rtas_t rtas = {
64 .lock = __ARCH_SPIN_LOCK_UNLOCKED
65};
66EXPORT_SYMBOL(rtas);
67
68DEFINE_SPINLOCK(rtas_data_buf_lock);
69EXPORT_SYMBOL(rtas_data_buf_lock);
70
71char rtas_data_buf[RTAS_DATA_BUF_SIZE] __cacheline_aligned;
72EXPORT_SYMBOL(rtas_data_buf);
73
74unsigned long rtas_rmo_buf;
75
76/*
77 * If non-NULL, this gets called when the kernel terminates.
78 * This is done like this so rtas_flash can be a module.
79 */
80void (*rtas_flash_term_hook)(int);
81EXPORT_SYMBOL(rtas_flash_term_hook);
82
83/* RTAS use home made raw locking instead of spin_lock_irqsave
84 * because those can be called from within really nasty contexts
85 * such as having the timebase stopped which would lockup with
86 * normal locks and spinlock debugging enabled
87 */
88static unsigned long lock_rtas(void)
89{
90 unsigned long flags;
91
92 local_irq_save(flags);
93 preempt_disable();
94 arch_spin_lock(&rtas.lock);
95 return flags;
96}
97
98static void unlock_rtas(unsigned long flags)
99{
100 arch_spin_unlock(&rtas.lock);
101 local_irq_restore(flags);
102 preempt_enable();
103}
104
105/*
106 * call_rtas_display_status and call_rtas_display_status_delay
107 * are designed only for very early low-level debugging, which
108 * is why the token is hard-coded to 10.
109 */
110static void call_rtas_display_status(unsigned char c)
111{
112 unsigned long s;
113
114 if (!rtas.base)
115 return;
116
117 s = lock_rtas();
118 rtas_call_unlocked(&rtas.args, 10, 1, 1, NULL, c);
119 unlock_rtas(s);
120}
121
122static void call_rtas_display_status_delay(char c)
123{
124 static int pending_newline = 0; /* did last write end with unprinted newline? */
125 static int width = 16;
126
127 if (c == '\n') {
128 while (width-- > 0)
129 call_rtas_display_status(' ');
130 width = 16;
131 mdelay(500);
132 pending_newline = 1;
133 } else {
134 if (pending_newline) {
135 call_rtas_display_status('\r');
136 call_rtas_display_status('\n');
137 }
138 pending_newline = 0;
139 if (width--) {
140 call_rtas_display_status(c);
141 udelay(10000);
142 }
143 }
144}
145
146void __init udbg_init_rtas_panel(void)
147{
148 udbg_putc = call_rtas_display_status_delay;
149}
150
151#ifdef CONFIG_UDBG_RTAS_CONSOLE
152
153/* If you think you're dying before early_init_dt_scan_rtas() does its
154 * work, you can hard code the token values for your firmware here and
155 * hardcode rtas.base/entry etc.
156 */
157static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
158static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
159
160static void udbg_rtascon_putc(char c)
161{
162 int tries;
163
164 if (!rtas.base)
165 return;
166
167 /* Add CRs before LFs */
168 if (c == '\n')
169 udbg_rtascon_putc('\r');
170
171 /* if there is more than one character to be displayed, wait a bit */
172 for (tries = 0; tries < 16; tries++) {
173 if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
174 break;
175 udelay(1000);
176 }
177}
178
179static int udbg_rtascon_getc_poll(void)
180{
181 int c;
182
183 if (!rtas.base)
184 return -1;
185
186 if (rtas_call(rtas_getchar_token, 0, 2, &c))
187 return -1;
188
189 return c;
190}
191
192static int udbg_rtascon_getc(void)
193{
194 int c;
195
196 while ((c = udbg_rtascon_getc_poll()) == -1)
197 ;
198
199 return c;
200}
201
202
203void __init udbg_init_rtas_console(void)
204{
205 udbg_putc = udbg_rtascon_putc;
206 udbg_getc = udbg_rtascon_getc;
207 udbg_getc_poll = udbg_rtascon_getc_poll;
208}
209#endif /* CONFIG_UDBG_RTAS_CONSOLE */
210
211void rtas_progress(char *s, unsigned short hex)
212{
213 struct device_node *root;
214 int width;
215 const __be32 *p;
216 char *os;
217 static int display_character, set_indicator;
218 static int display_width, display_lines, form_feed;
219 static const int *row_width;
220 static DEFINE_SPINLOCK(progress_lock);
221 static int current_line;
222 static int pending_newline = 0; /* did last write end with unprinted newline? */
223
224 if (!rtas.base)
225 return;
226
227 if (display_width == 0) {
228 display_width = 0x10;
229 if ((root = of_find_node_by_path("/rtas"))) {
230 if ((p = of_get_property(root,
231 "ibm,display-line-length", NULL)))
232 display_width = be32_to_cpu(*p);
233 if ((p = of_get_property(root,
234 "ibm,form-feed", NULL)))
235 form_feed = be32_to_cpu(*p);
236 if ((p = of_get_property(root,
237 "ibm,display-number-of-lines", NULL)))
238 display_lines = be32_to_cpu(*p);
239 row_width = of_get_property(root,
240 "ibm,display-truncation-length", NULL);
241 of_node_put(root);
242 }
243 display_character = rtas_token("display-character");
244 set_indicator = rtas_token("set-indicator");
245 }
246
247 if (display_character == RTAS_UNKNOWN_SERVICE) {
248 /* use hex display if available */
249 if (set_indicator != RTAS_UNKNOWN_SERVICE)
250 rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
251 return;
252 }
253
254 spin_lock(&progress_lock);
255
256 /*
257 * Last write ended with newline, but we didn't print it since
258 * it would just clear the bottom line of output. Print it now
259 * instead.
260 *
261 * If no newline is pending and form feed is supported, clear the
262 * display with a form feed; otherwise, print a CR to start output
263 * at the beginning of the line.
264 */
265 if (pending_newline) {
266 rtas_call(display_character, 1, 1, NULL, '\r');
267 rtas_call(display_character, 1, 1, NULL, '\n');
268 pending_newline = 0;
269 } else {
270 current_line = 0;
271 if (form_feed)
272 rtas_call(display_character, 1, 1, NULL,
273 (char)form_feed);
274 else
275 rtas_call(display_character, 1, 1, NULL, '\r');
276 }
277
278 if (row_width)
279 width = row_width[current_line];
280 else
281 width = display_width;
282 os = s;
283 while (*os) {
284 if (*os == '\n' || *os == '\r') {
285 /* If newline is the last character, save it
286 * until next call to avoid bumping up the
287 * display output.
288 */
289 if (*os == '\n' && !os[1]) {
290 pending_newline = 1;
291 current_line++;
292 if (current_line > display_lines-1)
293 current_line = display_lines-1;
294 spin_unlock(&progress_lock);
295 return;
296 }
297
298 /* RTAS wants CR-LF, not just LF */
299
300 if (*os == '\n') {
301 rtas_call(display_character, 1, 1, NULL, '\r');
302 rtas_call(display_character, 1, 1, NULL, '\n');
303 } else {
304 /* CR might be used to re-draw a line, so we'll
305 * leave it alone and not add LF.
306 */
307 rtas_call(display_character, 1, 1, NULL, *os);
308 }
309
310 if (row_width)
311 width = row_width[current_line];
312 else
313 width = display_width;
314 } else {
315 width--;
316 rtas_call(display_character, 1, 1, NULL, *os);
317 }
318
319 os++;
320
321 /* if we overwrite the screen length */
322 if (width <= 0)
323 while ((*os != 0) && (*os != '\n') && (*os != '\r'))
324 os++;
325 }
326
327 spin_unlock(&progress_lock);
328}
329EXPORT_SYMBOL(rtas_progress); /* needed by rtas_flash module */
330
331int rtas_token(const char *service)
332{
333 const __be32 *tokp;
334 if (rtas.dev == NULL)
335 return RTAS_UNKNOWN_SERVICE;
336 tokp = of_get_property(rtas.dev, service, NULL);
337 return tokp ? be32_to_cpu(*tokp) : RTAS_UNKNOWN_SERVICE;
338}
339EXPORT_SYMBOL(rtas_token);
340
341int rtas_service_present(const char *service)
342{
343 return rtas_token(service) != RTAS_UNKNOWN_SERVICE;
344}
345EXPORT_SYMBOL(rtas_service_present);
346
347#ifdef CONFIG_RTAS_ERROR_LOGGING
348
349static u32 rtas_error_log_max __ro_after_init = RTAS_ERROR_LOG_MAX;
350
351/*
352 * Return the firmware-specified size of the error log buffer
353 * for all rtas calls that require an error buffer argument.
354 * This includes 'check-exception' and 'rtas-last-error'.
355 */
356int rtas_get_error_log_max(void)
357{
358 return rtas_error_log_max;
359}
360EXPORT_SYMBOL(rtas_get_error_log_max);
361
362static void __init init_error_log_max(void)
363{
364 static const char propname[] __initconst = "rtas-error-log-max";
365 u32 max;
366
367 if (of_property_read_u32(rtas.dev, propname, &max)) {
368 pr_warn("%s not found, using default of %u\n",
369 propname, RTAS_ERROR_LOG_MAX);
370 max = RTAS_ERROR_LOG_MAX;
371 }
372
373 if (max > RTAS_ERROR_LOG_MAX) {
374 pr_warn("%s = %u, clamping max error log size to %u\n",
375 propname, max, RTAS_ERROR_LOG_MAX);
376 max = RTAS_ERROR_LOG_MAX;
377 }
378
379 rtas_error_log_max = max;
380}
381
382
383static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
384static int rtas_last_error_token;
385
386/** Return a copy of the detailed error text associated with the
387 * most recent failed call to rtas. Because the error text
388 * might go stale if there are any other intervening rtas calls,
389 * this routine must be called atomically with whatever produced
390 * the error (i.e. with rtas.lock still held from the previous call).
391 */
392static char *__fetch_rtas_last_error(char *altbuf)
393{
394 struct rtas_args err_args, save_args;
395 u32 bufsz;
396 char *buf = NULL;
397
398 if (rtas_last_error_token == -1)
399 return NULL;
400
401 bufsz = rtas_get_error_log_max();
402
403 err_args.token = cpu_to_be32(rtas_last_error_token);
404 err_args.nargs = cpu_to_be32(2);
405 err_args.nret = cpu_to_be32(1);
406 err_args.args[0] = cpu_to_be32(__pa(rtas_err_buf));
407 err_args.args[1] = cpu_to_be32(bufsz);
408 err_args.args[2] = 0;
409
410 save_args = rtas.args;
411 rtas.args = err_args;
412
413 do_enter_rtas(__pa(&rtas.args));
414
415 err_args = rtas.args;
416 rtas.args = save_args;
417
418 /* Log the error in the unlikely case that there was one. */
419 if (unlikely(err_args.args[2] == 0)) {
420 if (altbuf) {
421 buf = altbuf;
422 } else {
423 buf = rtas_err_buf;
424 if (slab_is_available())
425 buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
426 }
427 if (buf)
428 memcpy(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
429 }
430
431 return buf;
432}
433
434#define get_errorlog_buffer() kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
435
436#else /* CONFIG_RTAS_ERROR_LOGGING */
437#define __fetch_rtas_last_error(x) NULL
438#define get_errorlog_buffer() NULL
439static void __init init_error_log_max(void) {}
440#endif
441
442
443static void
444va_rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret,
445 va_list list)
446{
447 int i;
448
449 args->token = cpu_to_be32(token);
450 args->nargs = cpu_to_be32(nargs);
451 args->nret = cpu_to_be32(nret);
452 args->rets = &(args->args[nargs]);
453
454 for (i = 0; i < nargs; ++i)
455 args->args[i] = cpu_to_be32(va_arg(list, __u32));
456
457 for (i = 0; i < nret; ++i)
458 args->rets[i] = 0;
459
460 do_enter_rtas(__pa(args));
461}
462
463void rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret, ...)
464{
465 va_list list;
466
467 va_start(list, nret);
468 va_rtas_call_unlocked(args, token, nargs, nret, list);
469 va_end(list);
470}
471
472static int ibm_open_errinjct_token;
473static int ibm_errinjct_token;
474
475/**
476 * rtas_call() - Invoke an RTAS firmware function.
477 * @token: Identifies the function being invoked.
478 * @nargs: Number of input parameters. Does not include token.
479 * @nret: Number of output parameters, including the call status.
480 * @outputs: Array of @nret output words.
481 * @....: List of @nargs input parameters.
482 *
483 * Invokes the RTAS function indicated by @token, which the caller
484 * should obtain via rtas_token().
485 *
486 * The @nargs and @nret arguments must match the number of input and
487 * output parameters specified for the RTAS function.
488 *
489 * rtas_call() returns RTAS status codes, not conventional Linux errno
490 * values. Callers must translate any failure to an appropriate errno
491 * in syscall context. Most callers of RTAS functions that can return
492 * -2 or 990x should use rtas_busy_delay() to correctly handle those
493 * statuses before calling again.
494 *
495 * The return value descriptions are adapted from 7.2.8 [RTAS] Return
496 * Codes of the PAPR and CHRP specifications.
497 *
498 * Context: Process context preferably, interrupt context if
499 * necessary. Acquires an internal spinlock and may perform
500 * GFP_ATOMIC slab allocation in error path. Unsafe for NMI
501 * context.
502 * Return:
503 * * 0 - RTAS function call succeeded.
504 * * -1 - RTAS function encountered a hardware or
505 * platform error, or the token is invalid,
506 * or the function is restricted by kernel policy.
507 * * -2 - Specs say "A necessary hardware device was busy,
508 * and the requested function could not be
509 * performed. The operation should be retried at
510 * a later time." This is misleading, at least with
511 * respect to current RTAS implementations. What it
512 * usually means in practice is that the function
513 * could not be completed while meeting RTAS's
514 * deadline for returning control to the OS (250us
515 * for PAPR/PowerVM, typically), but the call may be
516 * immediately reattempted to resume work on it.
517 * * -3 - Parameter error.
518 * * -7 - Unexpected state change.
519 * * 9000...9899 - Vendor-specific success codes.
520 * * 9900...9905 - Advisory extended delay. Caller should try
521 * again after ~10^x ms has elapsed, where x is
522 * the last digit of the status [0-5]. Again going
523 * beyond the PAPR text, 990x on PowerVM indicates
524 * contention for RTAS-internal resources. Other
525 * RTAS call sequences in progress should be
526 * allowed to complete before reattempting the
527 * call.
528 * * -9000 - Multi-level isolation error.
529 * * -9999...-9004 - Vendor-specific error codes.
530 * * Additional negative values - Function-specific error.
531 * * Additional positive values - Function-specific success.
532 */
533int rtas_call(int token, int nargs, int nret, int *outputs, ...)
534{
535 va_list list;
536 int i;
537 unsigned long s;
538 struct rtas_args *rtas_args;
539 char *buff_copy = NULL;
540 int ret;
541
542 if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
543 return -1;
544
545 if (token == ibm_open_errinjct_token || token == ibm_errinjct_token) {
546 /*
547 * It would be nicer to not discard the error value
548 * from security_locked_down(), but callers expect an
549 * RTAS status, not an errno.
550 */
551 if (security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION))
552 return -1;
553 }
554
555 if ((mfmsr() & (MSR_IR|MSR_DR)) != (MSR_IR|MSR_DR)) {
556 WARN_ON_ONCE(1);
557 return -1;
558 }
559
560 s = lock_rtas();
561
562 /* We use the global rtas args buffer */
563 rtas_args = &rtas.args;
564
565 va_start(list, outputs);
566 va_rtas_call_unlocked(rtas_args, token, nargs, nret, list);
567 va_end(list);
568
569 /* A -1 return code indicates that the last command couldn't
570 be completed due to a hardware error. */
571 if (be32_to_cpu(rtas_args->rets[0]) == -1)
572 buff_copy = __fetch_rtas_last_error(NULL);
573
574 if (nret > 1 && outputs != NULL)
575 for (i = 0; i < nret-1; ++i)
576 outputs[i] = be32_to_cpu(rtas_args->rets[i+1]);
577 ret = (nret > 0)? be32_to_cpu(rtas_args->rets[0]): 0;
578
579 unlock_rtas(s);
580
581 if (buff_copy) {
582 log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
583 if (slab_is_available())
584 kfree(buff_copy);
585 }
586 return ret;
587}
588EXPORT_SYMBOL(rtas_call);
589
590/**
591 * rtas_busy_delay_time() - From an RTAS status value, calculate the
592 * suggested delay time in milliseconds.
593 *
594 * @status: a value returned from rtas_call() or similar APIs which return
595 * the status of a RTAS function call.
596 *
597 * Context: Any context.
598 *
599 * Return:
600 * * 100000 - If @status is 9905.
601 * * 10000 - If @status is 9904.
602 * * 1000 - If @status is 9903.
603 * * 100 - If @status is 9902.
604 * * 10 - If @status is 9901.
605 * * 1 - If @status is either 9900 or -2. This is "wrong" for -2, but
606 * some callers depend on this behavior, and the worst outcome
607 * is that they will delay for longer than necessary.
608 * * 0 - If @status is not a busy or extended delay value.
609 */
610unsigned int rtas_busy_delay_time(int status)
611{
612 int order;
613 unsigned int ms = 0;
614
615 if (status == RTAS_BUSY) {
616 ms = 1;
617 } else if (status >= RTAS_EXTENDED_DELAY_MIN &&
618 status <= RTAS_EXTENDED_DELAY_MAX) {
619 order = status - RTAS_EXTENDED_DELAY_MIN;
620 for (ms = 1; order > 0; order--)
621 ms *= 10;
622 }
623
624 return ms;
625}
626EXPORT_SYMBOL(rtas_busy_delay_time);
627
628/**
629 * rtas_busy_delay() - helper for RTAS busy and extended delay statuses
630 *
631 * @status: a value returned from rtas_call() or similar APIs which return
632 * the status of a RTAS function call.
633 *
634 * Context: Process context. May sleep or schedule.
635 *
636 * Return:
637 * * true - @status is RTAS_BUSY or an extended delay hint. The
638 * caller may assume that the CPU has been yielded if necessary,
639 * and that an appropriate delay for @status has elapsed.
640 * Generally the caller should reattempt the RTAS call which
641 * yielded @status.
642 *
643 * * false - @status is not @RTAS_BUSY nor an extended delay hint. The
644 * caller is responsible for handling @status.
645 */
646bool rtas_busy_delay(int status)
647{
648 unsigned int ms;
649 bool ret;
650
651 switch (status) {
652 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
653 ret = true;
654 ms = rtas_busy_delay_time(status);
655 /*
656 * The extended delay hint can be as high as 100 seconds.
657 * Surely any function returning such a status is either
658 * buggy or isn't going to be significantly slowed by us
659 * polling at 1HZ. Clamp the sleep time to one second.
660 */
661 ms = clamp(ms, 1U, 1000U);
662 /*
663 * The delay hint is an order-of-magnitude suggestion, not
664 * a minimum. It is fine, possibly even advantageous, for
665 * us to pause for less time than hinted. For small values,
666 * use usleep_range() to ensure we don't sleep much longer
667 * than actually needed.
668 *
669 * See Documentation/timers/timers-howto.rst for
670 * explanation of the threshold used here. In effect we use
671 * usleep_range() for 9900 and 9901, msleep() for
672 * 9902-9905.
673 */
674 if (ms <= 20)
675 usleep_range(ms * 100, ms * 1000);
676 else
677 msleep(ms);
678 break;
679 case RTAS_BUSY:
680 ret = true;
681 /*
682 * We should call again immediately if there's no other
683 * work to do.
684 */
685 cond_resched();
686 break;
687 default:
688 ret = false;
689 /*
690 * Not a busy or extended delay status; the caller should
691 * handle @status itself. Ensure we warn on misuses in
692 * atomic context regardless.
693 */
694 might_sleep();
695 break;
696 }
697
698 return ret;
699}
700EXPORT_SYMBOL(rtas_busy_delay);
701
702static int rtas_error_rc(int rtas_rc)
703{
704 int rc;
705
706 switch (rtas_rc) {
707 case -1: /* Hardware Error */
708 rc = -EIO;
709 break;
710 case -3: /* Bad indicator/domain/etc */
711 rc = -EINVAL;
712 break;
713 case -9000: /* Isolation error */
714 rc = -EFAULT;
715 break;
716 case -9001: /* Outstanding TCE/PTE */
717 rc = -EEXIST;
718 break;
719 case -9002: /* No usable slot */
720 rc = -ENODEV;
721 break;
722 default:
723 pr_err("%s: unexpected error %d\n", __func__, rtas_rc);
724 rc = -ERANGE;
725 break;
726 }
727 return rc;
728}
729
730int rtas_get_power_level(int powerdomain, int *level)
731{
732 int token = rtas_token("get-power-level");
733 int rc;
734
735 if (token == RTAS_UNKNOWN_SERVICE)
736 return -ENOENT;
737
738 while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
739 udelay(1);
740
741 if (rc < 0)
742 return rtas_error_rc(rc);
743 return rc;
744}
745EXPORT_SYMBOL(rtas_get_power_level);
746
747int rtas_set_power_level(int powerdomain, int level, int *setlevel)
748{
749 int token = rtas_token("set-power-level");
750 int rc;
751
752 if (token == RTAS_UNKNOWN_SERVICE)
753 return -ENOENT;
754
755 do {
756 rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
757 } while (rtas_busy_delay(rc));
758
759 if (rc < 0)
760 return rtas_error_rc(rc);
761 return rc;
762}
763EXPORT_SYMBOL(rtas_set_power_level);
764
765int rtas_get_sensor(int sensor, int index, int *state)
766{
767 int token = rtas_token("get-sensor-state");
768 int rc;
769
770 if (token == RTAS_UNKNOWN_SERVICE)
771 return -ENOENT;
772
773 do {
774 rc = rtas_call(token, 2, 2, state, sensor, index);
775 } while (rtas_busy_delay(rc));
776
777 if (rc < 0)
778 return rtas_error_rc(rc);
779 return rc;
780}
781EXPORT_SYMBOL(rtas_get_sensor);
782
783int rtas_get_sensor_fast(int sensor, int index, int *state)
784{
785 int token = rtas_token("get-sensor-state");
786 int rc;
787
788 if (token == RTAS_UNKNOWN_SERVICE)
789 return -ENOENT;
790
791 rc = rtas_call(token, 2, 2, state, sensor, index);
792 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
793 rc <= RTAS_EXTENDED_DELAY_MAX));
794
795 if (rc < 0)
796 return rtas_error_rc(rc);
797 return rc;
798}
799
800bool rtas_indicator_present(int token, int *maxindex)
801{
802 int proplen, count, i;
803 const struct indicator_elem {
804 __be32 token;
805 __be32 maxindex;
806 } *indicators;
807
808 indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
809 if (!indicators)
810 return false;
811
812 count = proplen / sizeof(struct indicator_elem);
813
814 for (i = 0; i < count; i++) {
815 if (__be32_to_cpu(indicators[i].token) != token)
816 continue;
817 if (maxindex)
818 *maxindex = __be32_to_cpu(indicators[i].maxindex);
819 return true;
820 }
821
822 return false;
823}
824EXPORT_SYMBOL(rtas_indicator_present);
825
826int rtas_set_indicator(int indicator, int index, int new_value)
827{
828 int token = rtas_token("set-indicator");
829 int rc;
830
831 if (token == RTAS_UNKNOWN_SERVICE)
832 return -ENOENT;
833
834 do {
835 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
836 } while (rtas_busy_delay(rc));
837
838 if (rc < 0)
839 return rtas_error_rc(rc);
840 return rc;
841}
842EXPORT_SYMBOL(rtas_set_indicator);
843
844/*
845 * Ignoring RTAS extended delay
846 */
847int rtas_set_indicator_fast(int indicator, int index, int new_value)
848{
849 int rc;
850 int token = rtas_token("set-indicator");
851
852 if (token == RTAS_UNKNOWN_SERVICE)
853 return -ENOENT;
854
855 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
856
857 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
858 rc <= RTAS_EXTENDED_DELAY_MAX));
859
860 if (rc < 0)
861 return rtas_error_rc(rc);
862
863 return rc;
864}
865
866/**
867 * rtas_ibm_suspend_me() - Call ibm,suspend-me to suspend the LPAR.
868 *
869 * @fw_status: RTAS call status will be placed here if not NULL.
870 *
871 * rtas_ibm_suspend_me() should be called only on a CPU which has
872 * received H_CONTINUE from the H_JOIN hcall. All other active CPUs
873 * should be waiting to return from H_JOIN.
874 *
875 * rtas_ibm_suspend_me() may suspend execution of the OS
876 * indefinitely. Callers should take appropriate measures upon return, such as
877 * resetting watchdog facilities.
878 *
879 * Callers may choose to retry this call if @fw_status is
880 * %RTAS_THREADS_ACTIVE.
881 *
882 * Return:
883 * 0 - The partition has resumed from suspend, possibly after
884 * migration to a different host.
885 * -ECANCELED - The operation was aborted.
886 * -EAGAIN - There were other CPUs not in H_JOIN at the time of the call.
887 * -EBUSY - Some other condition prevented the suspend from succeeding.
888 * -EIO - Hardware/platform error.
889 */
890int rtas_ibm_suspend_me(int *fw_status)
891{
892 int fwrc;
893 int ret;
894
895 fwrc = rtas_call(rtas_token("ibm,suspend-me"), 0, 1, NULL);
896
897 switch (fwrc) {
898 case 0:
899 ret = 0;
900 break;
901 case RTAS_SUSPEND_ABORTED:
902 ret = -ECANCELED;
903 break;
904 case RTAS_THREADS_ACTIVE:
905 ret = -EAGAIN;
906 break;
907 case RTAS_NOT_SUSPENDABLE:
908 case RTAS_OUTSTANDING_COPROC:
909 ret = -EBUSY;
910 break;
911 case -1:
912 default:
913 ret = -EIO;
914 break;
915 }
916
917 if (fw_status)
918 *fw_status = fwrc;
919
920 return ret;
921}
922
923void __noreturn rtas_restart(char *cmd)
924{
925 if (rtas_flash_term_hook)
926 rtas_flash_term_hook(SYS_RESTART);
927 pr_emerg("system-reboot returned %d\n",
928 rtas_call(rtas_token("system-reboot"), 0, 1, NULL));
929 for (;;);
930}
931
932void rtas_power_off(void)
933{
934 if (rtas_flash_term_hook)
935 rtas_flash_term_hook(SYS_POWER_OFF);
936 /* allow power on only with power button press */
937 pr_emerg("power-off returned %d\n",
938 rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
939 for (;;);
940}
941
942void __noreturn rtas_halt(void)
943{
944 if (rtas_flash_term_hook)
945 rtas_flash_term_hook(SYS_HALT);
946 /* allow power on only with power button press */
947 pr_emerg("power-off returned %d\n",
948 rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
949 for (;;);
950}
951
952/* Must be in the RMO region, so we place it here */
953static char rtas_os_term_buf[2048];
954static s32 ibm_os_term_token = RTAS_UNKNOWN_SERVICE;
955
956void rtas_os_term(char *str)
957{
958 int status;
959
960 /*
961 * Firmware with the ibm,extended-os-term property is guaranteed
962 * to always return from an ibm,os-term call. Earlier versions without
963 * this property may terminate the partition which we want to avoid
964 * since it interferes with panic_timeout.
965 */
966 if (ibm_os_term_token == RTAS_UNKNOWN_SERVICE)
967 return;
968
969 snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
970
971 /*
972 * Keep calling as long as RTAS returns a "try again" status,
973 * but don't use rtas_busy_delay(), which potentially
974 * schedules.
975 */
976 do {
977 status = rtas_call(ibm_os_term_token, 1, 1, NULL,
978 __pa(rtas_os_term_buf));
979 } while (rtas_busy_delay_time(status));
980
981 if (status != 0)
982 pr_emerg("ibm,os-term call failed %d\n", status);
983}
984
985/**
986 * rtas_activate_firmware() - Activate a new version of firmware.
987 *
988 * Context: This function may sleep.
989 *
990 * Activate a new version of partition firmware. The OS must call this
991 * after resuming from a partition hibernation or migration in order
992 * to maintain the ability to perform live firmware updates. It's not
993 * catastrophic for this method to be absent or to fail; just log the
994 * condition in that case.
995 */
996void rtas_activate_firmware(void)
997{
998 int token;
999 int fwrc;
1000
1001 token = rtas_token("ibm,activate-firmware");
1002 if (token == RTAS_UNKNOWN_SERVICE) {
1003 pr_notice("ibm,activate-firmware method unavailable\n");
1004 return;
1005 }
1006
1007 do {
1008 fwrc = rtas_call(token, 0, 1, NULL);
1009 } while (rtas_busy_delay(fwrc));
1010
1011 if (fwrc)
1012 pr_err("ibm,activate-firmware failed (%i)\n", fwrc);
1013}
1014
1015/**
1016 * get_pseries_errorlog() - Find a specific pseries error log in an RTAS
1017 * extended event log.
1018 * @log: RTAS error/event log
1019 * @section_id: two character section identifier
1020 *
1021 * Return: A pointer to the specified errorlog or NULL if not found.
1022 */
1023noinstr struct pseries_errorlog *get_pseries_errorlog(struct rtas_error_log *log,
1024 uint16_t section_id)
1025{
1026 struct rtas_ext_event_log_v6 *ext_log =
1027 (struct rtas_ext_event_log_v6 *)log->buffer;
1028 struct pseries_errorlog *sect;
1029 unsigned char *p, *log_end;
1030 uint32_t ext_log_length = rtas_error_extended_log_length(log);
1031 uint8_t log_format = rtas_ext_event_log_format(ext_log);
1032 uint32_t company_id = rtas_ext_event_company_id(ext_log);
1033
1034 /* Check that we understand the format */
1035 if (ext_log_length < sizeof(struct rtas_ext_event_log_v6) ||
1036 log_format != RTAS_V6EXT_LOG_FORMAT_EVENT_LOG ||
1037 company_id != RTAS_V6EXT_COMPANY_ID_IBM)
1038 return NULL;
1039
1040 log_end = log->buffer + ext_log_length;
1041 p = ext_log->vendor_log;
1042
1043 while (p < log_end) {
1044 sect = (struct pseries_errorlog *)p;
1045 if (pseries_errorlog_id(sect) == section_id)
1046 return sect;
1047 p += pseries_errorlog_length(sect);
1048 }
1049
1050 return NULL;
1051}
1052
1053/*
1054 * The sys_rtas syscall, as originally designed, allows root to pass
1055 * arbitrary physical addresses to RTAS calls. A number of RTAS calls
1056 * can be abused to write to arbitrary memory and do other things that
1057 * are potentially harmful to system integrity, and thus should only
1058 * be used inside the kernel and not exposed to userspace.
1059 *
1060 * All known legitimate users of the sys_rtas syscall will only ever
1061 * pass addresses that fall within the RMO buffer, and use a known
1062 * subset of RTAS calls.
1063 *
1064 * Accordingly, we filter RTAS requests to check that the call is
1065 * permitted, and that provided pointers fall within the RMO buffer.
1066 * The rtas_filters list contains an entry for each permitted call,
1067 * with the indexes of the parameters which are expected to contain
1068 * addresses and sizes of buffers allocated inside the RMO buffer.
1069 */
1070struct rtas_filter {
1071 const char *name;
1072 int token;
1073 /* Indexes into the args buffer, -1 if not used */
1074 int buf_idx1;
1075 int size_idx1;
1076 int buf_idx2;
1077 int size_idx2;
1078
1079 int fixed_size;
1080};
1081
1082static struct rtas_filter rtas_filters[] __ro_after_init = {
1083 { "ibm,activate-firmware", -1, -1, -1, -1, -1 },
1084 { "ibm,configure-connector", -1, 0, -1, 1, -1, 4096 }, /* Special cased */
1085 { "display-character", -1, -1, -1, -1, -1 },
1086 { "ibm,display-message", -1, 0, -1, -1, -1 },
1087 { "ibm,errinjct", -1, 2, -1, -1, -1, 1024 },
1088 { "ibm,close-errinjct", -1, -1, -1, -1, -1 },
1089 { "ibm,open-errinjct", -1, -1, -1, -1, -1 },
1090 { "ibm,get-config-addr-info2", -1, -1, -1, -1, -1 },
1091 { "ibm,get-dynamic-sensor-state", -1, 1, -1, -1, -1 },
1092 { "ibm,get-indices", -1, 2, 3, -1, -1 },
1093 { "get-power-level", -1, -1, -1, -1, -1 },
1094 { "get-sensor-state", -1, -1, -1, -1, -1 },
1095 { "ibm,get-system-parameter", -1, 1, 2, -1, -1 },
1096 { "get-time-of-day", -1, -1, -1, -1, -1 },
1097 { "ibm,get-vpd", -1, 0, -1, 1, 2 },
1098 { "ibm,lpar-perftools", -1, 2, 3, -1, -1 },
1099 { "ibm,platform-dump", -1, 4, 5, -1, -1 }, /* Special cased */
1100 { "ibm,read-slot-reset-state", -1, -1, -1, -1, -1 },
1101 { "ibm,scan-log-dump", -1, 0, 1, -1, -1 },
1102 { "ibm,set-dynamic-indicator", -1, 2, -1, -1, -1 },
1103 { "ibm,set-eeh-option", -1, -1, -1, -1, -1 },
1104 { "set-indicator", -1, -1, -1, -1, -1 },
1105 { "set-power-level", -1, -1, -1, -1, -1 },
1106 { "set-time-for-power-on", -1, -1, -1, -1, -1 },
1107 { "ibm,set-system-parameter", -1, 1, -1, -1, -1 },
1108 { "set-time-of-day", -1, -1, -1, -1, -1 },
1109#ifdef CONFIG_CPU_BIG_ENDIAN
1110 { "ibm,suspend-me", -1, -1, -1, -1, -1 },
1111 { "ibm,update-nodes", -1, 0, -1, -1, -1, 4096 },
1112 { "ibm,update-properties", -1, 0, -1, -1, -1, 4096 },
1113#endif
1114 { "ibm,physical-attestation", -1, 0, 1, -1, -1 },
1115};
1116
1117static bool in_rmo_buf(u32 base, u32 end)
1118{
1119 return base >= rtas_rmo_buf &&
1120 base < (rtas_rmo_buf + RTAS_USER_REGION_SIZE) &&
1121 base <= end &&
1122 end >= rtas_rmo_buf &&
1123 end < (rtas_rmo_buf + RTAS_USER_REGION_SIZE);
1124}
1125
1126static bool block_rtas_call(int token, int nargs,
1127 struct rtas_args *args)
1128{
1129 int i;
1130
1131 for (i = 0; i < ARRAY_SIZE(rtas_filters); i++) {
1132 struct rtas_filter *f = &rtas_filters[i];
1133 u32 base, size, end;
1134
1135 if (token != f->token)
1136 continue;
1137
1138 if (f->buf_idx1 != -1) {
1139 base = be32_to_cpu(args->args[f->buf_idx1]);
1140 if (f->size_idx1 != -1)
1141 size = be32_to_cpu(args->args[f->size_idx1]);
1142 else if (f->fixed_size)
1143 size = f->fixed_size;
1144 else
1145 size = 1;
1146
1147 end = base + size - 1;
1148
1149 /*
1150 * Special case for ibm,platform-dump - NULL buffer
1151 * address is used to indicate end of dump processing
1152 */
1153 if (!strcmp(f->name, "ibm,platform-dump") &&
1154 base == 0)
1155 return false;
1156
1157 if (!in_rmo_buf(base, end))
1158 goto err;
1159 }
1160
1161 if (f->buf_idx2 != -1) {
1162 base = be32_to_cpu(args->args[f->buf_idx2]);
1163 if (f->size_idx2 != -1)
1164 size = be32_to_cpu(args->args[f->size_idx2]);
1165 else if (f->fixed_size)
1166 size = f->fixed_size;
1167 else
1168 size = 1;
1169 end = base + size - 1;
1170
1171 /*
1172 * Special case for ibm,configure-connector where the
1173 * address can be 0
1174 */
1175 if (!strcmp(f->name, "ibm,configure-connector") &&
1176 base == 0)
1177 return false;
1178
1179 if (!in_rmo_buf(base, end))
1180 goto err;
1181 }
1182
1183 return false;
1184 }
1185
1186err:
1187 pr_err_ratelimited("sys_rtas: RTAS call blocked - exploit attempt?\n");
1188 pr_err_ratelimited("sys_rtas: token=0x%x, nargs=%d (called by %s)\n",
1189 token, nargs, current->comm);
1190 return true;
1191}
1192
1193static void __init rtas_syscall_filter_init(void)
1194{
1195 unsigned int i;
1196
1197 for (i = 0; i < ARRAY_SIZE(rtas_filters); i++)
1198 rtas_filters[i].token = rtas_token(rtas_filters[i].name);
1199}
1200
1201/* We assume to be passed big endian arguments */
1202SYSCALL_DEFINE1(rtas, struct rtas_args __user *, uargs)
1203{
1204 struct rtas_args args;
1205 unsigned long flags;
1206 char *buff_copy, *errbuf = NULL;
1207 int nargs, nret, token;
1208
1209 if (!capable(CAP_SYS_ADMIN))
1210 return -EPERM;
1211
1212 if (!rtas.entry)
1213 return -EINVAL;
1214
1215 if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
1216 return -EFAULT;
1217
1218 nargs = be32_to_cpu(args.nargs);
1219 nret = be32_to_cpu(args.nret);
1220 token = be32_to_cpu(args.token);
1221
1222 if (nargs >= ARRAY_SIZE(args.args)
1223 || nret > ARRAY_SIZE(args.args)
1224 || nargs + nret > ARRAY_SIZE(args.args))
1225 return -EINVAL;
1226
1227 /* Copy in args. */
1228 if (copy_from_user(args.args, uargs->args,
1229 nargs * sizeof(rtas_arg_t)) != 0)
1230 return -EFAULT;
1231
1232 if (token == RTAS_UNKNOWN_SERVICE)
1233 return -EINVAL;
1234
1235 args.rets = &args.args[nargs];
1236 memset(args.rets, 0, nret * sizeof(rtas_arg_t));
1237
1238 if (block_rtas_call(token, nargs, &args))
1239 return -EINVAL;
1240
1241 if (token == ibm_open_errinjct_token || token == ibm_errinjct_token) {
1242 int err;
1243
1244 err = security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION);
1245 if (err)
1246 return err;
1247 }
1248
1249 /* Need to handle ibm,suspend_me call specially */
1250 if (token == rtas_token("ibm,suspend-me")) {
1251
1252 /*
1253 * rtas_ibm_suspend_me assumes the streamid handle is in cpu
1254 * endian, or at least the hcall within it requires it.
1255 */
1256 int rc = 0;
1257 u64 handle = ((u64)be32_to_cpu(args.args[0]) << 32)
1258 | be32_to_cpu(args.args[1]);
1259 rc = rtas_syscall_dispatch_ibm_suspend_me(handle);
1260 if (rc == -EAGAIN)
1261 args.rets[0] = cpu_to_be32(RTAS_NOT_SUSPENDABLE);
1262 else if (rc == -EIO)
1263 args.rets[0] = cpu_to_be32(-1);
1264 else if (rc)
1265 return rc;
1266 goto copy_return;
1267 }
1268
1269 buff_copy = get_errorlog_buffer();
1270
1271 flags = lock_rtas();
1272
1273 rtas.args = args;
1274 do_enter_rtas(__pa(&rtas.args));
1275 args = rtas.args;
1276
1277 /* A -1 return code indicates that the last command couldn't
1278 be completed due to a hardware error. */
1279 if (be32_to_cpu(args.rets[0]) == -1)
1280 errbuf = __fetch_rtas_last_error(buff_copy);
1281
1282 unlock_rtas(flags);
1283
1284 if (buff_copy) {
1285 if (errbuf)
1286 log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
1287 kfree(buff_copy);
1288 }
1289
1290 copy_return:
1291 /* Copy out args. */
1292 if (copy_to_user(uargs->args + nargs,
1293 args.args + nargs,
1294 nret * sizeof(rtas_arg_t)) != 0)
1295 return -EFAULT;
1296
1297 return 0;
1298}
1299
1300/*
1301 * Call early during boot, before mem init, to retrieve the RTAS
1302 * information from the device-tree and allocate the RMO buffer for userland
1303 * accesses.
1304 */
1305void __init rtas_initialize(void)
1306{
1307 unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
1308 u32 base, size, entry;
1309 int no_base, no_size, no_entry;
1310
1311 /* Get RTAS dev node and fill up our "rtas" structure with infos
1312 * about it.
1313 */
1314 rtas.dev = of_find_node_by_name(NULL, "rtas");
1315 if (!rtas.dev)
1316 return;
1317
1318 no_base = of_property_read_u32(rtas.dev, "linux,rtas-base", &base);
1319 no_size = of_property_read_u32(rtas.dev, "rtas-size", &size);
1320 if (no_base || no_size) {
1321 of_node_put(rtas.dev);
1322 rtas.dev = NULL;
1323 return;
1324 }
1325
1326 rtas.base = base;
1327 rtas.size = size;
1328 no_entry = of_property_read_u32(rtas.dev, "linux,rtas-entry", &entry);
1329 rtas.entry = no_entry ? rtas.base : entry;
1330
1331 init_error_log_max();
1332
1333 /*
1334 * Discover these now to avoid device tree lookups in the
1335 * panic path.
1336 */
1337 if (of_property_read_bool(rtas.dev, "ibm,extended-os-term"))
1338 ibm_os_term_token = rtas_token("ibm,os-term");
1339
1340 /* If RTAS was found, allocate the RMO buffer for it and look for
1341 * the stop-self token if any
1342 */
1343#ifdef CONFIG_PPC64
1344 if (firmware_has_feature(FW_FEATURE_LPAR))
1345 rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
1346#endif
1347 rtas_rmo_buf = memblock_phys_alloc_range(RTAS_USER_REGION_SIZE, PAGE_SIZE,
1348 0, rtas_region);
1349 if (!rtas_rmo_buf)
1350 panic("ERROR: RTAS: Failed to allocate %lx bytes below %pa\n",
1351 PAGE_SIZE, &rtas_region);
1352
1353#ifdef CONFIG_RTAS_ERROR_LOGGING
1354 rtas_last_error_token = rtas_token("rtas-last-error");
1355#endif
1356 ibm_open_errinjct_token = rtas_token("ibm,open-errinjct");
1357 ibm_errinjct_token = rtas_token("ibm,errinjct");
1358 rtas_syscall_filter_init();
1359}
1360
1361int __init early_init_dt_scan_rtas(unsigned long node,
1362 const char *uname, int depth, void *data)
1363{
1364 const u32 *basep, *entryp, *sizep;
1365
1366 if (depth != 1 || strcmp(uname, "rtas") != 0)
1367 return 0;
1368
1369 basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
1370 entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
1371 sizep = of_get_flat_dt_prop(node, "rtas-size", NULL);
1372
1373#ifdef CONFIG_PPC64
1374 /* need this feature to decide the crashkernel offset */
1375 if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL))
1376 powerpc_firmware_features |= FW_FEATURE_LPAR;
1377#endif
1378
1379 if (basep && entryp && sizep) {
1380 rtas.base = *basep;
1381 rtas.entry = *entryp;
1382 rtas.size = *sizep;
1383 }
1384
1385#ifdef CONFIG_UDBG_RTAS_CONSOLE
1386 basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
1387 if (basep)
1388 rtas_putchar_token = *basep;
1389
1390 basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
1391 if (basep)
1392 rtas_getchar_token = *basep;
1393
1394 if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
1395 rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
1396 udbg_init_rtas_console();
1397
1398#endif
1399
1400 /* break now */
1401 return 1;
1402}
1403
1404static arch_spinlock_t timebase_lock;
1405static u64 timebase = 0;
1406
1407void rtas_give_timebase(void)
1408{
1409 unsigned long flags;
1410
1411 local_irq_save(flags);
1412 hard_irq_disable();
1413 arch_spin_lock(&timebase_lock);
1414 rtas_call(rtas_token("freeze-time-base"), 0, 1, NULL);
1415 timebase = get_tb();
1416 arch_spin_unlock(&timebase_lock);
1417
1418 while (timebase)
1419 barrier();
1420 rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
1421 local_irq_restore(flags);
1422}
1423
1424void rtas_take_timebase(void)
1425{
1426 while (!timebase)
1427 barrier();
1428 arch_spin_lock(&timebase_lock);
1429 set_tb(timebase >> 32, timebase & 0xffffffff);
1430 timebase = 0;
1431 arch_spin_unlock(&timebase_lock);
1432}
1/*
2 *
3 * Procedures for interfacing to the RTAS on CHRP machines.
4 *
5 * Peter Bergner, IBM March 2001.
6 * Copyright (C) 2001 IBM.
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14#include <stdarg.h>
15#include <linux/kernel.h>
16#include <linux/types.h>
17#include <linux/spinlock.h>
18#include <linux/export.h>
19#include <linux/init.h>
20#include <linux/capability.h>
21#include <linux/delay.h>
22#include <linux/cpu.h>
23#include <linux/smp.h>
24#include <linux/completion.h>
25#include <linux/cpumask.h>
26#include <linux/memblock.h>
27#include <linux/slab.h>
28#include <linux/reboot.h>
29
30#include <asm/prom.h>
31#include <asm/rtas.h>
32#include <asm/hvcall.h>
33#include <asm/machdep.h>
34#include <asm/firmware.h>
35#include <asm/page.h>
36#include <asm/param.h>
37#include <asm/delay.h>
38#include <linux/uaccess.h>
39#include <asm/udbg.h>
40#include <asm/syscalls.h>
41#include <asm/smp.h>
42#include <linux/atomic.h>
43#include <asm/time.h>
44#include <asm/mmu.h>
45#include <asm/topology.h>
46
47/* This is here deliberately so it's only used in this file */
48void enter_rtas(unsigned long);
49
50struct rtas_t rtas = {
51 .lock = __ARCH_SPIN_LOCK_UNLOCKED
52};
53EXPORT_SYMBOL(rtas);
54
55DEFINE_SPINLOCK(rtas_data_buf_lock);
56EXPORT_SYMBOL(rtas_data_buf_lock);
57
58char rtas_data_buf[RTAS_DATA_BUF_SIZE] __cacheline_aligned;
59EXPORT_SYMBOL(rtas_data_buf);
60
61unsigned long rtas_rmo_buf;
62
63/*
64 * If non-NULL, this gets called when the kernel terminates.
65 * This is done like this so rtas_flash can be a module.
66 */
67void (*rtas_flash_term_hook)(int);
68EXPORT_SYMBOL(rtas_flash_term_hook);
69
70/* RTAS use home made raw locking instead of spin_lock_irqsave
71 * because those can be called from within really nasty contexts
72 * such as having the timebase stopped which would lockup with
73 * normal locks and spinlock debugging enabled
74 */
75static unsigned long lock_rtas(void)
76{
77 unsigned long flags;
78
79 local_irq_save(flags);
80 preempt_disable();
81 arch_spin_lock(&rtas.lock);
82 return flags;
83}
84
85static void unlock_rtas(unsigned long flags)
86{
87 arch_spin_unlock(&rtas.lock);
88 local_irq_restore(flags);
89 preempt_enable();
90}
91
92/*
93 * call_rtas_display_status and call_rtas_display_status_delay
94 * are designed only for very early low-level debugging, which
95 * is why the token is hard-coded to 10.
96 */
97static void call_rtas_display_status(unsigned char c)
98{
99 unsigned long s;
100
101 if (!rtas.base)
102 return;
103
104 s = lock_rtas();
105 rtas_call_unlocked(&rtas.args, 10, 1, 1, NULL, c);
106 unlock_rtas(s);
107}
108
109static void call_rtas_display_status_delay(char c)
110{
111 static int pending_newline = 0; /* did last write end with unprinted newline? */
112 static int width = 16;
113
114 if (c == '\n') {
115 while (width-- > 0)
116 call_rtas_display_status(' ');
117 width = 16;
118 mdelay(500);
119 pending_newline = 1;
120 } else {
121 if (pending_newline) {
122 call_rtas_display_status('\r');
123 call_rtas_display_status('\n');
124 }
125 pending_newline = 0;
126 if (width--) {
127 call_rtas_display_status(c);
128 udelay(10000);
129 }
130 }
131}
132
133void __init udbg_init_rtas_panel(void)
134{
135 udbg_putc = call_rtas_display_status_delay;
136}
137
138#ifdef CONFIG_UDBG_RTAS_CONSOLE
139
140/* If you think you're dying before early_init_dt_scan_rtas() does its
141 * work, you can hard code the token values for your firmware here and
142 * hardcode rtas.base/entry etc.
143 */
144static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
145static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
146
147static void udbg_rtascon_putc(char c)
148{
149 int tries;
150
151 if (!rtas.base)
152 return;
153
154 /* Add CRs before LFs */
155 if (c == '\n')
156 udbg_rtascon_putc('\r');
157
158 /* if there is more than one character to be displayed, wait a bit */
159 for (tries = 0; tries < 16; tries++) {
160 if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
161 break;
162 udelay(1000);
163 }
164}
165
166static int udbg_rtascon_getc_poll(void)
167{
168 int c;
169
170 if (!rtas.base)
171 return -1;
172
173 if (rtas_call(rtas_getchar_token, 0, 2, &c))
174 return -1;
175
176 return c;
177}
178
179static int udbg_rtascon_getc(void)
180{
181 int c;
182
183 while ((c = udbg_rtascon_getc_poll()) == -1)
184 ;
185
186 return c;
187}
188
189
190void __init udbg_init_rtas_console(void)
191{
192 udbg_putc = udbg_rtascon_putc;
193 udbg_getc = udbg_rtascon_getc;
194 udbg_getc_poll = udbg_rtascon_getc_poll;
195}
196#endif /* CONFIG_UDBG_RTAS_CONSOLE */
197
198void rtas_progress(char *s, unsigned short hex)
199{
200 struct device_node *root;
201 int width;
202 const __be32 *p;
203 char *os;
204 static int display_character, set_indicator;
205 static int display_width, display_lines, form_feed;
206 static const int *row_width;
207 static DEFINE_SPINLOCK(progress_lock);
208 static int current_line;
209 static int pending_newline = 0; /* did last write end with unprinted newline? */
210
211 if (!rtas.base)
212 return;
213
214 if (display_width == 0) {
215 display_width = 0x10;
216 if ((root = of_find_node_by_path("/rtas"))) {
217 if ((p = of_get_property(root,
218 "ibm,display-line-length", NULL)))
219 display_width = be32_to_cpu(*p);
220 if ((p = of_get_property(root,
221 "ibm,form-feed", NULL)))
222 form_feed = be32_to_cpu(*p);
223 if ((p = of_get_property(root,
224 "ibm,display-number-of-lines", NULL)))
225 display_lines = be32_to_cpu(*p);
226 row_width = of_get_property(root,
227 "ibm,display-truncation-length", NULL);
228 of_node_put(root);
229 }
230 display_character = rtas_token("display-character");
231 set_indicator = rtas_token("set-indicator");
232 }
233
234 if (display_character == RTAS_UNKNOWN_SERVICE) {
235 /* use hex display if available */
236 if (set_indicator != RTAS_UNKNOWN_SERVICE)
237 rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
238 return;
239 }
240
241 spin_lock(&progress_lock);
242
243 /*
244 * Last write ended with newline, but we didn't print it since
245 * it would just clear the bottom line of output. Print it now
246 * instead.
247 *
248 * If no newline is pending and form feed is supported, clear the
249 * display with a form feed; otherwise, print a CR to start output
250 * at the beginning of the line.
251 */
252 if (pending_newline) {
253 rtas_call(display_character, 1, 1, NULL, '\r');
254 rtas_call(display_character, 1, 1, NULL, '\n');
255 pending_newline = 0;
256 } else {
257 current_line = 0;
258 if (form_feed)
259 rtas_call(display_character, 1, 1, NULL,
260 (char)form_feed);
261 else
262 rtas_call(display_character, 1, 1, NULL, '\r');
263 }
264
265 if (row_width)
266 width = row_width[current_line];
267 else
268 width = display_width;
269 os = s;
270 while (*os) {
271 if (*os == '\n' || *os == '\r') {
272 /* If newline is the last character, save it
273 * until next call to avoid bumping up the
274 * display output.
275 */
276 if (*os == '\n' && !os[1]) {
277 pending_newline = 1;
278 current_line++;
279 if (current_line > display_lines-1)
280 current_line = display_lines-1;
281 spin_unlock(&progress_lock);
282 return;
283 }
284
285 /* RTAS wants CR-LF, not just LF */
286
287 if (*os == '\n') {
288 rtas_call(display_character, 1, 1, NULL, '\r');
289 rtas_call(display_character, 1, 1, NULL, '\n');
290 } else {
291 /* CR might be used to re-draw a line, so we'll
292 * leave it alone and not add LF.
293 */
294 rtas_call(display_character, 1, 1, NULL, *os);
295 }
296
297 if (row_width)
298 width = row_width[current_line];
299 else
300 width = display_width;
301 } else {
302 width--;
303 rtas_call(display_character, 1, 1, NULL, *os);
304 }
305
306 os++;
307
308 /* if we overwrite the screen length */
309 if (width <= 0)
310 while ((*os != 0) && (*os != '\n') && (*os != '\r'))
311 os++;
312 }
313
314 spin_unlock(&progress_lock);
315}
316EXPORT_SYMBOL(rtas_progress); /* needed by rtas_flash module */
317
318int rtas_token(const char *service)
319{
320 const __be32 *tokp;
321 if (rtas.dev == NULL)
322 return RTAS_UNKNOWN_SERVICE;
323 tokp = of_get_property(rtas.dev, service, NULL);
324 return tokp ? be32_to_cpu(*tokp) : RTAS_UNKNOWN_SERVICE;
325}
326EXPORT_SYMBOL(rtas_token);
327
328int rtas_service_present(const char *service)
329{
330 return rtas_token(service) != RTAS_UNKNOWN_SERVICE;
331}
332EXPORT_SYMBOL(rtas_service_present);
333
334#ifdef CONFIG_RTAS_ERROR_LOGGING
335/*
336 * Return the firmware-specified size of the error log buffer
337 * for all rtas calls that require an error buffer argument.
338 * This includes 'check-exception' and 'rtas-last-error'.
339 */
340int rtas_get_error_log_max(void)
341{
342 static int rtas_error_log_max;
343 if (rtas_error_log_max)
344 return rtas_error_log_max;
345
346 rtas_error_log_max = rtas_token ("rtas-error-log-max");
347 if ((rtas_error_log_max == RTAS_UNKNOWN_SERVICE) ||
348 (rtas_error_log_max > RTAS_ERROR_LOG_MAX)) {
349 printk (KERN_WARNING "RTAS: bad log buffer size %d\n",
350 rtas_error_log_max);
351 rtas_error_log_max = RTAS_ERROR_LOG_MAX;
352 }
353 return rtas_error_log_max;
354}
355EXPORT_SYMBOL(rtas_get_error_log_max);
356
357
358static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
359static int rtas_last_error_token;
360
361/** Return a copy of the detailed error text associated with the
362 * most recent failed call to rtas. Because the error text
363 * might go stale if there are any other intervening rtas calls,
364 * this routine must be called atomically with whatever produced
365 * the error (i.e. with rtas.lock still held from the previous call).
366 */
367static char *__fetch_rtas_last_error(char *altbuf)
368{
369 struct rtas_args err_args, save_args;
370 u32 bufsz;
371 char *buf = NULL;
372
373 if (rtas_last_error_token == -1)
374 return NULL;
375
376 bufsz = rtas_get_error_log_max();
377
378 err_args.token = cpu_to_be32(rtas_last_error_token);
379 err_args.nargs = cpu_to_be32(2);
380 err_args.nret = cpu_to_be32(1);
381 err_args.args[0] = cpu_to_be32(__pa(rtas_err_buf));
382 err_args.args[1] = cpu_to_be32(bufsz);
383 err_args.args[2] = 0;
384
385 save_args = rtas.args;
386 rtas.args = err_args;
387
388 enter_rtas(__pa(&rtas.args));
389
390 err_args = rtas.args;
391 rtas.args = save_args;
392
393 /* Log the error in the unlikely case that there was one. */
394 if (unlikely(err_args.args[2] == 0)) {
395 if (altbuf) {
396 buf = altbuf;
397 } else {
398 buf = rtas_err_buf;
399 if (slab_is_available())
400 buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
401 }
402 if (buf)
403 memcpy(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
404 }
405
406 return buf;
407}
408
409#define get_errorlog_buffer() kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
410
411#else /* CONFIG_RTAS_ERROR_LOGGING */
412#define __fetch_rtas_last_error(x) NULL
413#define get_errorlog_buffer() NULL
414#endif
415
416
417static void
418va_rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret,
419 va_list list)
420{
421 int i;
422
423 args->token = cpu_to_be32(token);
424 args->nargs = cpu_to_be32(nargs);
425 args->nret = cpu_to_be32(nret);
426 args->rets = &(args->args[nargs]);
427
428 for (i = 0; i < nargs; ++i)
429 args->args[i] = cpu_to_be32(va_arg(list, __u32));
430
431 for (i = 0; i < nret; ++i)
432 args->rets[i] = 0;
433
434 enter_rtas(__pa(args));
435}
436
437void rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret, ...)
438{
439 va_list list;
440
441 va_start(list, nret);
442 va_rtas_call_unlocked(args, token, nargs, nret, list);
443 va_end(list);
444}
445
446int rtas_call(int token, int nargs, int nret, int *outputs, ...)
447{
448 va_list list;
449 int i;
450 unsigned long s;
451 struct rtas_args *rtas_args;
452 char *buff_copy = NULL;
453 int ret;
454
455 if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
456 return -1;
457
458 s = lock_rtas();
459
460 /* We use the global rtas args buffer */
461 rtas_args = &rtas.args;
462
463 va_start(list, outputs);
464 va_rtas_call_unlocked(rtas_args, token, nargs, nret, list);
465 va_end(list);
466
467 /* A -1 return code indicates that the last command couldn't
468 be completed due to a hardware error. */
469 if (be32_to_cpu(rtas_args->rets[0]) == -1)
470 buff_copy = __fetch_rtas_last_error(NULL);
471
472 if (nret > 1 && outputs != NULL)
473 for (i = 0; i < nret-1; ++i)
474 outputs[i] = be32_to_cpu(rtas_args->rets[i+1]);
475 ret = (nret > 0)? be32_to_cpu(rtas_args->rets[0]): 0;
476
477 unlock_rtas(s);
478
479 if (buff_copy) {
480 log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
481 if (slab_is_available())
482 kfree(buff_copy);
483 }
484 return ret;
485}
486EXPORT_SYMBOL(rtas_call);
487
488/* For RTAS_BUSY (-2), delay for 1 millisecond. For an extended busy status
489 * code of 990n, perform the hinted delay of 10^n (last digit) milliseconds.
490 */
491unsigned int rtas_busy_delay_time(int status)
492{
493 int order;
494 unsigned int ms = 0;
495
496 if (status == RTAS_BUSY) {
497 ms = 1;
498 } else if (status >= RTAS_EXTENDED_DELAY_MIN &&
499 status <= RTAS_EXTENDED_DELAY_MAX) {
500 order = status - RTAS_EXTENDED_DELAY_MIN;
501 for (ms = 1; order > 0; order--)
502 ms *= 10;
503 }
504
505 return ms;
506}
507EXPORT_SYMBOL(rtas_busy_delay_time);
508
509/* For an RTAS busy status code, perform the hinted delay. */
510unsigned int rtas_busy_delay(int status)
511{
512 unsigned int ms;
513
514 might_sleep();
515 ms = rtas_busy_delay_time(status);
516 if (ms && need_resched())
517 msleep(ms);
518
519 return ms;
520}
521EXPORT_SYMBOL(rtas_busy_delay);
522
523static int rtas_error_rc(int rtas_rc)
524{
525 int rc;
526
527 switch (rtas_rc) {
528 case -1: /* Hardware Error */
529 rc = -EIO;
530 break;
531 case -3: /* Bad indicator/domain/etc */
532 rc = -EINVAL;
533 break;
534 case -9000: /* Isolation error */
535 rc = -EFAULT;
536 break;
537 case -9001: /* Outstanding TCE/PTE */
538 rc = -EEXIST;
539 break;
540 case -9002: /* No usable slot */
541 rc = -ENODEV;
542 break;
543 default:
544 printk(KERN_ERR "%s: unexpected RTAS error %d\n",
545 __func__, rtas_rc);
546 rc = -ERANGE;
547 break;
548 }
549 return rc;
550}
551
552int rtas_get_power_level(int powerdomain, int *level)
553{
554 int token = rtas_token("get-power-level");
555 int rc;
556
557 if (token == RTAS_UNKNOWN_SERVICE)
558 return -ENOENT;
559
560 while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
561 udelay(1);
562
563 if (rc < 0)
564 return rtas_error_rc(rc);
565 return rc;
566}
567EXPORT_SYMBOL(rtas_get_power_level);
568
569int rtas_set_power_level(int powerdomain, int level, int *setlevel)
570{
571 int token = rtas_token("set-power-level");
572 int rc;
573
574 if (token == RTAS_UNKNOWN_SERVICE)
575 return -ENOENT;
576
577 do {
578 rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
579 } while (rtas_busy_delay(rc));
580
581 if (rc < 0)
582 return rtas_error_rc(rc);
583 return rc;
584}
585EXPORT_SYMBOL(rtas_set_power_level);
586
587int rtas_get_sensor(int sensor, int index, int *state)
588{
589 int token = rtas_token("get-sensor-state");
590 int rc;
591
592 if (token == RTAS_UNKNOWN_SERVICE)
593 return -ENOENT;
594
595 do {
596 rc = rtas_call(token, 2, 2, state, sensor, index);
597 } while (rtas_busy_delay(rc));
598
599 if (rc < 0)
600 return rtas_error_rc(rc);
601 return rc;
602}
603EXPORT_SYMBOL(rtas_get_sensor);
604
605int rtas_get_sensor_fast(int sensor, int index, int *state)
606{
607 int token = rtas_token("get-sensor-state");
608 int rc;
609
610 if (token == RTAS_UNKNOWN_SERVICE)
611 return -ENOENT;
612
613 rc = rtas_call(token, 2, 2, state, sensor, index);
614 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
615 rc <= RTAS_EXTENDED_DELAY_MAX));
616
617 if (rc < 0)
618 return rtas_error_rc(rc);
619 return rc;
620}
621
622bool rtas_indicator_present(int token, int *maxindex)
623{
624 int proplen, count, i;
625 const struct indicator_elem {
626 __be32 token;
627 __be32 maxindex;
628 } *indicators;
629
630 indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
631 if (!indicators)
632 return false;
633
634 count = proplen / sizeof(struct indicator_elem);
635
636 for (i = 0; i < count; i++) {
637 if (__be32_to_cpu(indicators[i].token) != token)
638 continue;
639 if (maxindex)
640 *maxindex = __be32_to_cpu(indicators[i].maxindex);
641 return true;
642 }
643
644 return false;
645}
646EXPORT_SYMBOL(rtas_indicator_present);
647
648int rtas_set_indicator(int indicator, int index, int new_value)
649{
650 int token = rtas_token("set-indicator");
651 int rc;
652
653 if (token == RTAS_UNKNOWN_SERVICE)
654 return -ENOENT;
655
656 do {
657 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
658 } while (rtas_busy_delay(rc));
659
660 if (rc < 0)
661 return rtas_error_rc(rc);
662 return rc;
663}
664EXPORT_SYMBOL(rtas_set_indicator);
665
666/*
667 * Ignoring RTAS extended delay
668 */
669int rtas_set_indicator_fast(int indicator, int index, int new_value)
670{
671 int rc;
672 int token = rtas_token("set-indicator");
673
674 if (token == RTAS_UNKNOWN_SERVICE)
675 return -ENOENT;
676
677 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
678
679 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
680 rc <= RTAS_EXTENDED_DELAY_MAX));
681
682 if (rc < 0)
683 return rtas_error_rc(rc);
684
685 return rc;
686}
687
688void __noreturn rtas_restart(char *cmd)
689{
690 if (rtas_flash_term_hook)
691 rtas_flash_term_hook(SYS_RESTART);
692 printk("RTAS system-reboot returned %d\n",
693 rtas_call(rtas_token("system-reboot"), 0, 1, NULL));
694 for (;;);
695}
696
697void rtas_power_off(void)
698{
699 if (rtas_flash_term_hook)
700 rtas_flash_term_hook(SYS_POWER_OFF);
701 /* allow power on only with power button press */
702 printk("RTAS power-off returned %d\n",
703 rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
704 for (;;);
705}
706
707void __noreturn rtas_halt(void)
708{
709 if (rtas_flash_term_hook)
710 rtas_flash_term_hook(SYS_HALT);
711 /* allow power on only with power button press */
712 printk("RTAS power-off returned %d\n",
713 rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
714 for (;;);
715}
716
717/* Must be in the RMO region, so we place it here */
718static char rtas_os_term_buf[2048];
719
720void rtas_os_term(char *str)
721{
722 int status;
723
724 /*
725 * Firmware with the ibm,extended-os-term property is guaranteed
726 * to always return from an ibm,os-term call. Earlier versions without
727 * this property may terminate the partition which we want to avoid
728 * since it interferes with panic_timeout.
729 */
730 if (RTAS_UNKNOWN_SERVICE == rtas_token("ibm,os-term") ||
731 RTAS_UNKNOWN_SERVICE == rtas_token("ibm,extended-os-term"))
732 return;
733
734 snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
735
736 do {
737 status = rtas_call(rtas_token("ibm,os-term"), 1, 1, NULL,
738 __pa(rtas_os_term_buf));
739 } while (rtas_busy_delay(status));
740
741 if (status != 0)
742 printk(KERN_EMERG "ibm,os-term call failed %d\n", status);
743}
744
745static int ibm_suspend_me_token = RTAS_UNKNOWN_SERVICE;
746#ifdef CONFIG_PPC_PSERIES
747static int __rtas_suspend_last_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
748{
749 u16 slb_size = mmu_slb_size;
750 int rc = H_MULTI_THREADS_ACTIVE;
751 int cpu;
752
753 slb_set_size(SLB_MIN_SIZE);
754 printk(KERN_DEBUG "calling ibm,suspend-me on cpu %i\n", smp_processor_id());
755
756 while (rc == H_MULTI_THREADS_ACTIVE && !atomic_read(&data->done) &&
757 !atomic_read(&data->error))
758 rc = rtas_call(data->token, 0, 1, NULL);
759
760 if (rc || atomic_read(&data->error)) {
761 printk(KERN_DEBUG "ibm,suspend-me returned %d\n", rc);
762 slb_set_size(slb_size);
763 }
764
765 if (atomic_read(&data->error))
766 rc = atomic_read(&data->error);
767
768 atomic_set(&data->error, rc);
769 pSeries_coalesce_init();
770
771 if (wake_when_done) {
772 atomic_set(&data->done, 1);
773
774 for_each_online_cpu(cpu)
775 plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
776 }
777
778 if (atomic_dec_return(&data->working) == 0)
779 complete(data->complete);
780
781 return rc;
782}
783
784int rtas_suspend_last_cpu(struct rtas_suspend_me_data *data)
785{
786 atomic_inc(&data->working);
787 return __rtas_suspend_last_cpu(data, 0);
788}
789
790static int __rtas_suspend_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
791{
792 long rc = H_SUCCESS;
793 unsigned long msr_save;
794 int cpu;
795
796 atomic_inc(&data->working);
797
798 /* really need to ensure MSR.EE is off for H_JOIN */
799 msr_save = mfmsr();
800 mtmsr(msr_save & ~(MSR_EE));
801
802 while (rc == H_SUCCESS && !atomic_read(&data->done) && !atomic_read(&data->error))
803 rc = plpar_hcall_norets(H_JOIN);
804
805 mtmsr(msr_save);
806
807 if (rc == H_SUCCESS) {
808 /* This cpu was prodded and the suspend is complete. */
809 goto out;
810 } else if (rc == H_CONTINUE) {
811 /* All other cpus are in H_JOIN, this cpu does
812 * the suspend.
813 */
814 return __rtas_suspend_last_cpu(data, wake_when_done);
815 } else {
816 printk(KERN_ERR "H_JOIN on cpu %i failed with rc = %ld\n",
817 smp_processor_id(), rc);
818 atomic_set(&data->error, rc);
819 }
820
821 if (wake_when_done) {
822 atomic_set(&data->done, 1);
823
824 /* This cpu did the suspend or got an error; in either case,
825 * we need to prod all other other cpus out of join state.
826 * Extra prods are harmless.
827 */
828 for_each_online_cpu(cpu)
829 plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
830 }
831out:
832 if (atomic_dec_return(&data->working) == 0)
833 complete(data->complete);
834 return rc;
835}
836
837int rtas_suspend_cpu(struct rtas_suspend_me_data *data)
838{
839 return __rtas_suspend_cpu(data, 0);
840}
841
842static void rtas_percpu_suspend_me(void *info)
843{
844 __rtas_suspend_cpu((struct rtas_suspend_me_data *)info, 1);
845}
846
847enum rtas_cpu_state {
848 DOWN,
849 UP,
850};
851
852#ifndef CONFIG_SMP
853static int rtas_cpu_state_change_mask(enum rtas_cpu_state state,
854 cpumask_var_t cpus)
855{
856 if (!cpumask_empty(cpus)) {
857 cpumask_clear(cpus);
858 return -EINVAL;
859 } else
860 return 0;
861}
862#else
863/* On return cpumask will be altered to indicate CPUs changed.
864 * CPUs with states changed will be set in the mask,
865 * CPUs with status unchanged will be unset in the mask. */
866static int rtas_cpu_state_change_mask(enum rtas_cpu_state state,
867 cpumask_var_t cpus)
868{
869 int cpu;
870 int cpuret = 0;
871 int ret = 0;
872
873 if (cpumask_empty(cpus))
874 return 0;
875
876 for_each_cpu(cpu, cpus) {
877 switch (state) {
878 case DOWN:
879 cpuret = cpu_down(cpu);
880 break;
881 case UP:
882 cpuret = cpu_up(cpu);
883 break;
884 }
885 if (cpuret) {
886 pr_debug("%s: cpu_%s for cpu#%d returned %d.\n",
887 __func__,
888 ((state == UP) ? "up" : "down"),
889 cpu, cpuret);
890 if (!ret)
891 ret = cpuret;
892 if (state == UP) {
893 /* clear bits for unchanged cpus, return */
894 cpumask_shift_right(cpus, cpus, cpu);
895 cpumask_shift_left(cpus, cpus, cpu);
896 break;
897 } else {
898 /* clear bit for unchanged cpu, continue */
899 cpumask_clear_cpu(cpu, cpus);
900 }
901 }
902 }
903
904 return ret;
905}
906#endif
907
908int rtas_online_cpus_mask(cpumask_var_t cpus)
909{
910 int ret;
911
912 ret = rtas_cpu_state_change_mask(UP, cpus);
913
914 if (ret) {
915 cpumask_var_t tmp_mask;
916
917 if (!alloc_cpumask_var(&tmp_mask, GFP_KERNEL))
918 return ret;
919
920 /* Use tmp_mask to preserve cpus mask from first failure */
921 cpumask_copy(tmp_mask, cpus);
922 rtas_offline_cpus_mask(tmp_mask);
923 free_cpumask_var(tmp_mask);
924 }
925
926 return ret;
927}
928EXPORT_SYMBOL(rtas_online_cpus_mask);
929
930int rtas_offline_cpus_mask(cpumask_var_t cpus)
931{
932 return rtas_cpu_state_change_mask(DOWN, cpus);
933}
934EXPORT_SYMBOL(rtas_offline_cpus_mask);
935
936int rtas_ibm_suspend_me(u64 handle)
937{
938 long state;
939 long rc;
940 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
941 struct rtas_suspend_me_data data;
942 DECLARE_COMPLETION_ONSTACK(done);
943 cpumask_var_t offline_mask;
944 int cpuret;
945
946 if (!rtas_service_present("ibm,suspend-me"))
947 return -ENOSYS;
948
949 /* Make sure the state is valid */
950 rc = plpar_hcall(H_VASI_STATE, retbuf, handle);
951
952 state = retbuf[0];
953
954 if (rc) {
955 printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned %ld\n",rc);
956 return rc;
957 } else if (state == H_VASI_ENABLED) {
958 return -EAGAIN;
959 } else if (state != H_VASI_SUSPENDING) {
960 printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned state %ld\n",
961 state);
962 return -EIO;
963 }
964
965 if (!alloc_cpumask_var(&offline_mask, GFP_KERNEL))
966 return -ENOMEM;
967
968 atomic_set(&data.working, 0);
969 atomic_set(&data.done, 0);
970 atomic_set(&data.error, 0);
971 data.token = rtas_token("ibm,suspend-me");
972 data.complete = &done;
973
974 /* All present CPUs must be online */
975 cpumask_andnot(offline_mask, cpu_present_mask, cpu_online_mask);
976 cpuret = rtas_online_cpus_mask(offline_mask);
977 if (cpuret) {
978 pr_err("%s: Could not bring present CPUs online.\n", __func__);
979 atomic_set(&data.error, cpuret);
980 goto out;
981 }
982
983 stop_topology_update();
984
985 /* Call function on all CPUs. One of us will make the
986 * rtas call
987 */
988 if (on_each_cpu(rtas_percpu_suspend_me, &data, 0))
989 atomic_set(&data.error, -EINVAL);
990
991 wait_for_completion(&done);
992
993 if (atomic_read(&data.error) != 0)
994 printk(KERN_ERR "Error doing global join\n");
995
996 start_topology_update();
997
998 /* Take down CPUs not online prior to suspend */
999 cpuret = rtas_offline_cpus_mask(offline_mask);
1000 if (cpuret)
1001 pr_warn("%s: Could not restore CPUs to offline state.\n",
1002 __func__);
1003
1004out:
1005 free_cpumask_var(offline_mask);
1006 return atomic_read(&data.error);
1007}
1008#else /* CONFIG_PPC_PSERIES */
1009int rtas_ibm_suspend_me(u64 handle)
1010{
1011 return -ENOSYS;
1012}
1013#endif
1014
1015/**
1016 * Find a specific pseries error log in an RTAS extended event log.
1017 * @log: RTAS error/event log
1018 * @section_id: two character section identifier
1019 *
1020 * Returns a pointer to the specified errorlog or NULL if not found.
1021 */
1022struct pseries_errorlog *get_pseries_errorlog(struct rtas_error_log *log,
1023 uint16_t section_id)
1024{
1025 struct rtas_ext_event_log_v6 *ext_log =
1026 (struct rtas_ext_event_log_v6 *)log->buffer;
1027 struct pseries_errorlog *sect;
1028 unsigned char *p, *log_end;
1029 uint32_t ext_log_length = rtas_error_extended_log_length(log);
1030 uint8_t log_format = rtas_ext_event_log_format(ext_log);
1031 uint32_t company_id = rtas_ext_event_company_id(ext_log);
1032
1033 /* Check that we understand the format */
1034 if (ext_log_length < sizeof(struct rtas_ext_event_log_v6) ||
1035 log_format != RTAS_V6EXT_LOG_FORMAT_EVENT_LOG ||
1036 company_id != RTAS_V6EXT_COMPANY_ID_IBM)
1037 return NULL;
1038
1039 log_end = log->buffer + ext_log_length;
1040 p = ext_log->vendor_log;
1041
1042 while (p < log_end) {
1043 sect = (struct pseries_errorlog *)p;
1044 if (pseries_errorlog_id(sect) == section_id)
1045 return sect;
1046 p += pseries_errorlog_length(sect);
1047 }
1048
1049 return NULL;
1050}
1051
1052/* We assume to be passed big endian arguments */
1053asmlinkage int ppc_rtas(struct rtas_args __user *uargs)
1054{
1055 struct rtas_args args;
1056 unsigned long flags;
1057 char *buff_copy, *errbuf = NULL;
1058 int nargs, nret, token;
1059
1060 if (!capable(CAP_SYS_ADMIN))
1061 return -EPERM;
1062
1063 if (!rtas.entry)
1064 return -EINVAL;
1065
1066 if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
1067 return -EFAULT;
1068
1069 nargs = be32_to_cpu(args.nargs);
1070 nret = be32_to_cpu(args.nret);
1071 token = be32_to_cpu(args.token);
1072
1073 if (nargs >= ARRAY_SIZE(args.args)
1074 || nret > ARRAY_SIZE(args.args)
1075 || nargs + nret > ARRAY_SIZE(args.args))
1076 return -EINVAL;
1077
1078 /* Copy in args. */
1079 if (copy_from_user(args.args, uargs->args,
1080 nargs * sizeof(rtas_arg_t)) != 0)
1081 return -EFAULT;
1082
1083 if (token == RTAS_UNKNOWN_SERVICE)
1084 return -EINVAL;
1085
1086 args.rets = &args.args[nargs];
1087 memset(args.rets, 0, nret * sizeof(rtas_arg_t));
1088
1089 /* Need to handle ibm,suspend_me call specially */
1090 if (token == ibm_suspend_me_token) {
1091
1092 /*
1093 * rtas_ibm_suspend_me assumes the streamid handle is in cpu
1094 * endian, or at least the hcall within it requires it.
1095 */
1096 int rc = 0;
1097 u64 handle = ((u64)be32_to_cpu(args.args[0]) << 32)
1098 | be32_to_cpu(args.args[1]);
1099 rc = rtas_ibm_suspend_me(handle);
1100 if (rc == -EAGAIN)
1101 args.rets[0] = cpu_to_be32(RTAS_NOT_SUSPENDABLE);
1102 else if (rc == -EIO)
1103 args.rets[0] = cpu_to_be32(-1);
1104 else if (rc)
1105 return rc;
1106 goto copy_return;
1107 }
1108
1109 buff_copy = get_errorlog_buffer();
1110
1111 flags = lock_rtas();
1112
1113 rtas.args = args;
1114 enter_rtas(__pa(&rtas.args));
1115 args = rtas.args;
1116
1117 /* A -1 return code indicates that the last command couldn't
1118 be completed due to a hardware error. */
1119 if (be32_to_cpu(args.rets[0]) == -1)
1120 errbuf = __fetch_rtas_last_error(buff_copy);
1121
1122 unlock_rtas(flags);
1123
1124 if (buff_copy) {
1125 if (errbuf)
1126 log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
1127 kfree(buff_copy);
1128 }
1129
1130 copy_return:
1131 /* Copy out args. */
1132 if (copy_to_user(uargs->args + nargs,
1133 args.args + nargs,
1134 nret * sizeof(rtas_arg_t)) != 0)
1135 return -EFAULT;
1136
1137 return 0;
1138}
1139
1140/*
1141 * Call early during boot, before mem init, to retrieve the RTAS
1142 * information from the device-tree and allocate the RMO buffer for userland
1143 * accesses.
1144 */
1145void __init rtas_initialize(void)
1146{
1147 unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
1148 u32 base, size, entry;
1149 int no_base, no_size, no_entry;
1150
1151 /* Get RTAS dev node and fill up our "rtas" structure with infos
1152 * about it.
1153 */
1154 rtas.dev = of_find_node_by_name(NULL, "rtas");
1155 if (!rtas.dev)
1156 return;
1157
1158 no_base = of_property_read_u32(rtas.dev, "linux,rtas-base", &base);
1159 no_size = of_property_read_u32(rtas.dev, "rtas-size", &size);
1160 if (no_base || no_size) {
1161 of_node_put(rtas.dev);
1162 rtas.dev = NULL;
1163 return;
1164 }
1165
1166 rtas.base = base;
1167 rtas.size = size;
1168 no_entry = of_property_read_u32(rtas.dev, "linux,rtas-entry", &entry);
1169 rtas.entry = no_entry ? rtas.base : entry;
1170
1171 /* If RTAS was found, allocate the RMO buffer for it and look for
1172 * the stop-self token if any
1173 */
1174#ifdef CONFIG_PPC64
1175 if (firmware_has_feature(FW_FEATURE_LPAR)) {
1176 rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
1177 ibm_suspend_me_token = rtas_token("ibm,suspend-me");
1178 }
1179#endif
1180 rtas_rmo_buf = memblock_alloc_base(RTAS_RMOBUF_MAX, PAGE_SIZE, rtas_region);
1181
1182#ifdef CONFIG_RTAS_ERROR_LOGGING
1183 rtas_last_error_token = rtas_token("rtas-last-error");
1184#endif
1185}
1186
1187int __init early_init_dt_scan_rtas(unsigned long node,
1188 const char *uname, int depth, void *data)
1189{
1190 const u32 *basep, *entryp, *sizep;
1191
1192 if (depth != 1 || strcmp(uname, "rtas") != 0)
1193 return 0;
1194
1195 basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
1196 entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
1197 sizep = of_get_flat_dt_prop(node, "rtas-size", NULL);
1198
1199 if (basep && entryp && sizep) {
1200 rtas.base = *basep;
1201 rtas.entry = *entryp;
1202 rtas.size = *sizep;
1203 }
1204
1205#ifdef CONFIG_UDBG_RTAS_CONSOLE
1206 basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
1207 if (basep)
1208 rtas_putchar_token = *basep;
1209
1210 basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
1211 if (basep)
1212 rtas_getchar_token = *basep;
1213
1214 if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
1215 rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
1216 udbg_init_rtas_console();
1217
1218#endif
1219
1220 /* break now */
1221 return 1;
1222}
1223
1224static arch_spinlock_t timebase_lock;
1225static u64 timebase = 0;
1226
1227void rtas_give_timebase(void)
1228{
1229 unsigned long flags;
1230
1231 local_irq_save(flags);
1232 hard_irq_disable();
1233 arch_spin_lock(&timebase_lock);
1234 rtas_call(rtas_token("freeze-time-base"), 0, 1, NULL);
1235 timebase = get_tb();
1236 arch_spin_unlock(&timebase_lock);
1237
1238 while (timebase)
1239 barrier();
1240 rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
1241 local_irq_restore(flags);
1242}
1243
1244void rtas_take_timebase(void)
1245{
1246 while (!timebase)
1247 barrier();
1248 arch_spin_lock(&timebase_lock);
1249 set_tb(timebase >> 32, timebase & 0xffffffff);
1250 timebase = 0;
1251 arch_spin_unlock(&timebase_lock);
1252}