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// 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#include <stdarg.h>
11#include <linux/kernel.h>
12#include <linux/types.h>
13#include <linux/spinlock.h>
14#include <linux/export.h>
15#include <linux/init.h>
16#include <linux/capability.h>
17#include <linux/delay.h>
18#include <linux/cpu.h>
19#include <linux/sched.h>
20#include <linux/smp.h>
21#include <linux/completion.h>
22#include <linux/cpumask.h>
23#include <linux/memblock.h>
24#include <linux/slab.h>
25#include <linux/reboot.h>
26#include <linux/syscalls.h>
27
28#include <asm/prom.h>
29#include <asm/rtas.h>
30#include <asm/hvcall.h>
31#include <asm/machdep.h>
32#include <asm/firmware.h>
33#include <asm/page.h>
34#include <asm/param.h>
35#include <asm/delay.h>
36#include <linux/uaccess.h>
37#include <asm/udbg.h>
38#include <asm/syscalls.h>
39#include <asm/smp.h>
40#include <linux/atomic.h>
41#include <asm/time.h>
42#include <asm/mmu.h>
43#include <asm/topology.h>
44
45/* This is here deliberately so it's only used in this file */
46void enter_rtas(unsigned long);
47
48struct rtas_t rtas = {
49 .lock = __ARCH_SPIN_LOCK_UNLOCKED
50};
51EXPORT_SYMBOL(rtas);
52
53DEFINE_SPINLOCK(rtas_data_buf_lock);
54EXPORT_SYMBOL(rtas_data_buf_lock);
55
56char rtas_data_buf[RTAS_DATA_BUF_SIZE] __cacheline_aligned;
57EXPORT_SYMBOL(rtas_data_buf);
58
59unsigned long rtas_rmo_buf;
60
61/*
62 * If non-NULL, this gets called when the kernel terminates.
63 * This is done like this so rtas_flash can be a module.
64 */
65void (*rtas_flash_term_hook)(int);
66EXPORT_SYMBOL(rtas_flash_term_hook);
67
68/* RTAS use home made raw locking instead of spin_lock_irqsave
69 * because those can be called from within really nasty contexts
70 * such as having the timebase stopped which would lockup with
71 * normal locks and spinlock debugging enabled
72 */
73static unsigned long lock_rtas(void)
74{
75 unsigned long flags;
76
77 local_irq_save(flags);
78 preempt_disable();
79 arch_spin_lock(&rtas.lock);
80 return flags;
81}
82
83static void unlock_rtas(unsigned long flags)
84{
85 arch_spin_unlock(&rtas.lock);
86 local_irq_restore(flags);
87 preempt_enable();
88}
89
90/*
91 * call_rtas_display_status and call_rtas_display_status_delay
92 * are designed only for very early low-level debugging, which
93 * is why the token is hard-coded to 10.
94 */
95static void call_rtas_display_status(unsigned char c)
96{
97 unsigned long s;
98
99 if (!rtas.base)
100 return;
101
102 s = lock_rtas();
103 rtas_call_unlocked(&rtas.args, 10, 1, 1, NULL, c);
104 unlock_rtas(s);
105}
106
107static void call_rtas_display_status_delay(char c)
108{
109 static int pending_newline = 0; /* did last write end with unprinted newline? */
110 static int width = 16;
111
112 if (c == '\n') {
113 while (width-- > 0)
114 call_rtas_display_status(' ');
115 width = 16;
116 mdelay(500);
117 pending_newline = 1;
118 } else {
119 if (pending_newline) {
120 call_rtas_display_status('\r');
121 call_rtas_display_status('\n');
122 }
123 pending_newline = 0;
124 if (width--) {
125 call_rtas_display_status(c);
126 udelay(10000);
127 }
128 }
129}
130
131void __init udbg_init_rtas_panel(void)
132{
133 udbg_putc = call_rtas_display_status_delay;
134}
135
136#ifdef CONFIG_UDBG_RTAS_CONSOLE
137
138/* If you think you're dying before early_init_dt_scan_rtas() does its
139 * work, you can hard code the token values for your firmware here and
140 * hardcode rtas.base/entry etc.
141 */
142static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
143static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
144
145static void udbg_rtascon_putc(char c)
146{
147 int tries;
148
149 if (!rtas.base)
150 return;
151
152 /* Add CRs before LFs */
153 if (c == '\n')
154 udbg_rtascon_putc('\r');
155
156 /* if there is more than one character to be displayed, wait a bit */
157 for (tries = 0; tries < 16; tries++) {
158 if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
159 break;
160 udelay(1000);
161 }
162}
163
164static int udbg_rtascon_getc_poll(void)
165{
166 int c;
167
168 if (!rtas.base)
169 return -1;
170
171 if (rtas_call(rtas_getchar_token, 0, 2, &c))
172 return -1;
173
174 return c;
175}
176
177static int udbg_rtascon_getc(void)
178{
179 int c;
180
181 while ((c = udbg_rtascon_getc_poll()) == -1)
182 ;
183
184 return c;
185}
186
187
188void __init udbg_init_rtas_console(void)
189{
190 udbg_putc = udbg_rtascon_putc;
191 udbg_getc = udbg_rtascon_getc;
192 udbg_getc_poll = udbg_rtascon_getc_poll;
193}
194#endif /* CONFIG_UDBG_RTAS_CONSOLE */
195
196void rtas_progress(char *s, unsigned short hex)
197{
198 struct device_node *root;
199 int width;
200 const __be32 *p;
201 char *os;
202 static int display_character, set_indicator;
203 static int display_width, display_lines, form_feed;
204 static const int *row_width;
205 static DEFINE_SPINLOCK(progress_lock);
206 static int current_line;
207 static int pending_newline = 0; /* did last write end with unprinted newline? */
208
209 if (!rtas.base)
210 return;
211
212 if (display_width == 0) {
213 display_width = 0x10;
214 if ((root = of_find_node_by_path("/rtas"))) {
215 if ((p = of_get_property(root,
216 "ibm,display-line-length", NULL)))
217 display_width = be32_to_cpu(*p);
218 if ((p = of_get_property(root,
219 "ibm,form-feed", NULL)))
220 form_feed = be32_to_cpu(*p);
221 if ((p = of_get_property(root,
222 "ibm,display-number-of-lines", NULL)))
223 display_lines = be32_to_cpu(*p);
224 row_width = of_get_property(root,
225 "ibm,display-truncation-length", NULL);
226 of_node_put(root);
227 }
228 display_character = rtas_token("display-character");
229 set_indicator = rtas_token("set-indicator");
230 }
231
232 if (display_character == RTAS_UNKNOWN_SERVICE) {
233 /* use hex display if available */
234 if (set_indicator != RTAS_UNKNOWN_SERVICE)
235 rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
236 return;
237 }
238
239 spin_lock(&progress_lock);
240
241 /*
242 * Last write ended with newline, but we didn't print it since
243 * it would just clear the bottom line of output. Print it now
244 * instead.
245 *
246 * If no newline is pending and form feed is supported, clear the
247 * display with a form feed; otherwise, print a CR to start output
248 * at the beginning of the line.
249 */
250 if (pending_newline) {
251 rtas_call(display_character, 1, 1, NULL, '\r');
252 rtas_call(display_character, 1, 1, NULL, '\n');
253 pending_newline = 0;
254 } else {
255 current_line = 0;
256 if (form_feed)
257 rtas_call(display_character, 1, 1, NULL,
258 (char)form_feed);
259 else
260 rtas_call(display_character, 1, 1, NULL, '\r');
261 }
262
263 if (row_width)
264 width = row_width[current_line];
265 else
266 width = display_width;
267 os = s;
268 while (*os) {
269 if (*os == '\n' || *os == '\r') {
270 /* If newline is the last character, save it
271 * until next call to avoid bumping up the
272 * display output.
273 */
274 if (*os == '\n' && !os[1]) {
275 pending_newline = 1;
276 current_line++;
277 if (current_line > display_lines-1)
278 current_line = display_lines-1;
279 spin_unlock(&progress_lock);
280 return;
281 }
282
283 /* RTAS wants CR-LF, not just LF */
284
285 if (*os == '\n') {
286 rtas_call(display_character, 1, 1, NULL, '\r');
287 rtas_call(display_character, 1, 1, NULL, '\n');
288 } else {
289 /* CR might be used to re-draw a line, so we'll
290 * leave it alone and not add LF.
291 */
292 rtas_call(display_character, 1, 1, NULL, *os);
293 }
294
295 if (row_width)
296 width = row_width[current_line];
297 else
298 width = display_width;
299 } else {
300 width--;
301 rtas_call(display_character, 1, 1, NULL, *os);
302 }
303
304 os++;
305
306 /* if we overwrite the screen length */
307 if (width <= 0)
308 while ((*os != 0) && (*os != '\n') && (*os != '\r'))
309 os++;
310 }
311
312 spin_unlock(&progress_lock);
313}
314EXPORT_SYMBOL(rtas_progress); /* needed by rtas_flash module */
315
316int rtas_token(const char *service)
317{
318 const __be32 *tokp;
319 if (rtas.dev == NULL)
320 return RTAS_UNKNOWN_SERVICE;
321 tokp = of_get_property(rtas.dev, service, NULL);
322 return tokp ? be32_to_cpu(*tokp) : RTAS_UNKNOWN_SERVICE;
323}
324EXPORT_SYMBOL(rtas_token);
325
326int rtas_service_present(const char *service)
327{
328 return rtas_token(service) != RTAS_UNKNOWN_SERVICE;
329}
330EXPORT_SYMBOL(rtas_service_present);
331
332#ifdef CONFIG_RTAS_ERROR_LOGGING
333/*
334 * Return the firmware-specified size of the error log buffer
335 * for all rtas calls that require an error buffer argument.
336 * This includes 'check-exception' and 'rtas-last-error'.
337 */
338int rtas_get_error_log_max(void)
339{
340 static int rtas_error_log_max;
341 if (rtas_error_log_max)
342 return rtas_error_log_max;
343
344 rtas_error_log_max = rtas_token ("rtas-error-log-max");
345 if ((rtas_error_log_max == RTAS_UNKNOWN_SERVICE) ||
346 (rtas_error_log_max > RTAS_ERROR_LOG_MAX)) {
347 printk (KERN_WARNING "RTAS: bad log buffer size %d\n",
348 rtas_error_log_max);
349 rtas_error_log_max = RTAS_ERROR_LOG_MAX;
350 }
351 return rtas_error_log_max;
352}
353EXPORT_SYMBOL(rtas_get_error_log_max);
354
355
356static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
357static int rtas_last_error_token;
358
359/** Return a copy of the detailed error text associated with the
360 * most recent failed call to rtas. Because the error text
361 * might go stale if there are any other intervening rtas calls,
362 * this routine must be called atomically with whatever produced
363 * the error (i.e. with rtas.lock still held from the previous call).
364 */
365static char *__fetch_rtas_last_error(char *altbuf)
366{
367 struct rtas_args err_args, save_args;
368 u32 bufsz;
369 char *buf = NULL;
370
371 if (rtas_last_error_token == -1)
372 return NULL;
373
374 bufsz = rtas_get_error_log_max();
375
376 err_args.token = cpu_to_be32(rtas_last_error_token);
377 err_args.nargs = cpu_to_be32(2);
378 err_args.nret = cpu_to_be32(1);
379 err_args.args[0] = cpu_to_be32(__pa(rtas_err_buf));
380 err_args.args[1] = cpu_to_be32(bufsz);
381 err_args.args[2] = 0;
382
383 save_args = rtas.args;
384 rtas.args = err_args;
385
386 enter_rtas(__pa(&rtas.args));
387
388 err_args = rtas.args;
389 rtas.args = save_args;
390
391 /* Log the error in the unlikely case that there was one. */
392 if (unlikely(err_args.args[2] == 0)) {
393 if (altbuf) {
394 buf = altbuf;
395 } else {
396 buf = rtas_err_buf;
397 if (slab_is_available())
398 buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
399 }
400 if (buf)
401 memcpy(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
402 }
403
404 return buf;
405}
406
407#define get_errorlog_buffer() kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
408
409#else /* CONFIG_RTAS_ERROR_LOGGING */
410#define __fetch_rtas_last_error(x) NULL
411#define get_errorlog_buffer() NULL
412#endif
413
414
415static void
416va_rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret,
417 va_list list)
418{
419 int i;
420
421 args->token = cpu_to_be32(token);
422 args->nargs = cpu_to_be32(nargs);
423 args->nret = cpu_to_be32(nret);
424 args->rets = &(args->args[nargs]);
425
426 for (i = 0; i < nargs; ++i)
427 args->args[i] = cpu_to_be32(va_arg(list, __u32));
428
429 for (i = 0; i < nret; ++i)
430 args->rets[i] = 0;
431
432 enter_rtas(__pa(args));
433}
434
435void rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret, ...)
436{
437 va_list list;
438
439 va_start(list, nret);
440 va_rtas_call_unlocked(args, token, nargs, nret, list);
441 va_end(list);
442}
443
444int rtas_call(int token, int nargs, int nret, int *outputs, ...)
445{
446 va_list list;
447 int i;
448 unsigned long s;
449 struct rtas_args *rtas_args;
450 char *buff_copy = NULL;
451 int ret;
452
453 if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
454 return -1;
455
456 s = lock_rtas();
457
458 /* We use the global rtas args buffer */
459 rtas_args = &rtas.args;
460
461 va_start(list, outputs);
462 va_rtas_call_unlocked(rtas_args, token, nargs, nret, list);
463 va_end(list);
464
465 /* A -1 return code indicates that the last command couldn't
466 be completed due to a hardware error. */
467 if (be32_to_cpu(rtas_args->rets[0]) == -1)
468 buff_copy = __fetch_rtas_last_error(NULL);
469
470 if (nret > 1 && outputs != NULL)
471 for (i = 0; i < nret-1; ++i)
472 outputs[i] = be32_to_cpu(rtas_args->rets[i+1]);
473 ret = (nret > 0)? be32_to_cpu(rtas_args->rets[0]): 0;
474
475 unlock_rtas(s);
476
477 if (buff_copy) {
478 log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
479 if (slab_is_available())
480 kfree(buff_copy);
481 }
482 return ret;
483}
484EXPORT_SYMBOL(rtas_call);
485
486/* For RTAS_BUSY (-2), delay for 1 millisecond. For an extended busy status
487 * code of 990n, perform the hinted delay of 10^n (last digit) milliseconds.
488 */
489unsigned int rtas_busy_delay_time(int status)
490{
491 int order;
492 unsigned int ms = 0;
493
494 if (status == RTAS_BUSY) {
495 ms = 1;
496 } else if (status >= RTAS_EXTENDED_DELAY_MIN &&
497 status <= RTAS_EXTENDED_DELAY_MAX) {
498 order = status - RTAS_EXTENDED_DELAY_MIN;
499 for (ms = 1; order > 0; order--)
500 ms *= 10;
501 }
502
503 return ms;
504}
505EXPORT_SYMBOL(rtas_busy_delay_time);
506
507/* For an RTAS busy status code, perform the hinted delay. */
508unsigned int rtas_busy_delay(int status)
509{
510 unsigned int ms;
511
512 might_sleep();
513 ms = rtas_busy_delay_time(status);
514 if (ms && need_resched())
515 msleep(ms);
516
517 return ms;
518}
519EXPORT_SYMBOL(rtas_busy_delay);
520
521static int rtas_error_rc(int rtas_rc)
522{
523 int rc;
524
525 switch (rtas_rc) {
526 case -1: /* Hardware Error */
527 rc = -EIO;
528 break;
529 case -3: /* Bad indicator/domain/etc */
530 rc = -EINVAL;
531 break;
532 case -9000: /* Isolation error */
533 rc = -EFAULT;
534 break;
535 case -9001: /* Outstanding TCE/PTE */
536 rc = -EEXIST;
537 break;
538 case -9002: /* No usable slot */
539 rc = -ENODEV;
540 break;
541 default:
542 printk(KERN_ERR "%s: unexpected RTAS error %d\n",
543 __func__, rtas_rc);
544 rc = -ERANGE;
545 break;
546 }
547 return rc;
548}
549
550int rtas_get_power_level(int powerdomain, int *level)
551{
552 int token = rtas_token("get-power-level");
553 int rc;
554
555 if (token == RTAS_UNKNOWN_SERVICE)
556 return -ENOENT;
557
558 while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
559 udelay(1);
560
561 if (rc < 0)
562 return rtas_error_rc(rc);
563 return rc;
564}
565EXPORT_SYMBOL(rtas_get_power_level);
566
567int rtas_set_power_level(int powerdomain, int level, int *setlevel)
568{
569 int token = rtas_token("set-power-level");
570 int rc;
571
572 if (token == RTAS_UNKNOWN_SERVICE)
573 return -ENOENT;
574
575 do {
576 rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
577 } while (rtas_busy_delay(rc));
578
579 if (rc < 0)
580 return rtas_error_rc(rc);
581 return rc;
582}
583EXPORT_SYMBOL(rtas_set_power_level);
584
585int rtas_get_sensor(int sensor, int index, int *state)
586{
587 int token = rtas_token("get-sensor-state");
588 int rc;
589
590 if (token == RTAS_UNKNOWN_SERVICE)
591 return -ENOENT;
592
593 do {
594 rc = rtas_call(token, 2, 2, state, sensor, index);
595 } while (rtas_busy_delay(rc));
596
597 if (rc < 0)
598 return rtas_error_rc(rc);
599 return rc;
600}
601EXPORT_SYMBOL(rtas_get_sensor);
602
603int rtas_get_sensor_fast(int sensor, int index, int *state)
604{
605 int token = rtas_token("get-sensor-state");
606 int rc;
607
608 if (token == RTAS_UNKNOWN_SERVICE)
609 return -ENOENT;
610
611 rc = rtas_call(token, 2, 2, state, sensor, index);
612 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
613 rc <= RTAS_EXTENDED_DELAY_MAX));
614
615 if (rc < 0)
616 return rtas_error_rc(rc);
617 return rc;
618}
619
620bool rtas_indicator_present(int token, int *maxindex)
621{
622 int proplen, count, i;
623 const struct indicator_elem {
624 __be32 token;
625 __be32 maxindex;
626 } *indicators;
627
628 indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
629 if (!indicators)
630 return false;
631
632 count = proplen / sizeof(struct indicator_elem);
633
634 for (i = 0; i < count; i++) {
635 if (__be32_to_cpu(indicators[i].token) != token)
636 continue;
637 if (maxindex)
638 *maxindex = __be32_to_cpu(indicators[i].maxindex);
639 return true;
640 }
641
642 return false;
643}
644EXPORT_SYMBOL(rtas_indicator_present);
645
646int rtas_set_indicator(int indicator, int index, int new_value)
647{
648 int token = rtas_token("set-indicator");
649 int rc;
650
651 if (token == RTAS_UNKNOWN_SERVICE)
652 return -ENOENT;
653
654 do {
655 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
656 } while (rtas_busy_delay(rc));
657
658 if (rc < 0)
659 return rtas_error_rc(rc);
660 return rc;
661}
662EXPORT_SYMBOL(rtas_set_indicator);
663
664/*
665 * Ignoring RTAS extended delay
666 */
667int rtas_set_indicator_fast(int indicator, int index, int new_value)
668{
669 int rc;
670 int token = rtas_token("set-indicator");
671
672 if (token == RTAS_UNKNOWN_SERVICE)
673 return -ENOENT;
674
675 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
676
677 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
678 rc <= RTAS_EXTENDED_DELAY_MAX));
679
680 if (rc < 0)
681 return rtas_error_rc(rc);
682
683 return rc;
684}
685
686void __noreturn rtas_restart(char *cmd)
687{
688 if (rtas_flash_term_hook)
689 rtas_flash_term_hook(SYS_RESTART);
690 printk("RTAS system-reboot returned %d\n",
691 rtas_call(rtas_token("system-reboot"), 0, 1, NULL));
692 for (;;);
693}
694
695void rtas_power_off(void)
696{
697 if (rtas_flash_term_hook)
698 rtas_flash_term_hook(SYS_POWER_OFF);
699 /* allow power on only with power button press */
700 printk("RTAS power-off returned %d\n",
701 rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
702 for (;;);
703}
704
705void __noreturn rtas_halt(void)
706{
707 if (rtas_flash_term_hook)
708 rtas_flash_term_hook(SYS_HALT);
709 /* allow power on only with power button press */
710 printk("RTAS power-off returned %d\n",
711 rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
712 for (;;);
713}
714
715/* Must be in the RMO region, so we place it here */
716static char rtas_os_term_buf[2048];
717
718void rtas_os_term(char *str)
719{
720 int status;
721
722 /*
723 * Firmware with the ibm,extended-os-term property is guaranteed
724 * to always return from an ibm,os-term call. Earlier versions without
725 * this property may terminate the partition which we want to avoid
726 * since it interferes with panic_timeout.
727 */
728 if (RTAS_UNKNOWN_SERVICE == rtas_token("ibm,os-term") ||
729 RTAS_UNKNOWN_SERVICE == rtas_token("ibm,extended-os-term"))
730 return;
731
732 snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
733
734 do {
735 status = rtas_call(rtas_token("ibm,os-term"), 1, 1, NULL,
736 __pa(rtas_os_term_buf));
737 } while (rtas_busy_delay(status));
738
739 if (status != 0)
740 printk(KERN_EMERG "ibm,os-term call failed %d\n", status);
741}
742
743static int ibm_suspend_me_token = RTAS_UNKNOWN_SERVICE;
744#ifdef CONFIG_PPC_PSERIES
745static int __rtas_suspend_last_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
746{
747 u16 slb_size = mmu_slb_size;
748 int rc = H_MULTI_THREADS_ACTIVE;
749 int cpu;
750
751 slb_set_size(SLB_MIN_SIZE);
752 printk(KERN_DEBUG "calling ibm,suspend-me on cpu %i\n", smp_processor_id());
753
754 while (rc == H_MULTI_THREADS_ACTIVE && !atomic_read(&data->done) &&
755 !atomic_read(&data->error))
756 rc = rtas_call(data->token, 0, 1, NULL);
757
758 if (rc || atomic_read(&data->error)) {
759 printk(KERN_DEBUG "ibm,suspend-me returned %d\n", rc);
760 slb_set_size(slb_size);
761 }
762
763 if (atomic_read(&data->error))
764 rc = atomic_read(&data->error);
765
766 atomic_set(&data->error, rc);
767 pSeries_coalesce_init();
768
769 if (wake_when_done) {
770 atomic_set(&data->done, 1);
771
772 for_each_online_cpu(cpu)
773 plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
774 }
775
776 if (atomic_dec_return(&data->working) == 0)
777 complete(data->complete);
778
779 return rc;
780}
781
782int rtas_suspend_last_cpu(struct rtas_suspend_me_data *data)
783{
784 atomic_inc(&data->working);
785 return __rtas_suspend_last_cpu(data, 0);
786}
787
788static int __rtas_suspend_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
789{
790 long rc = H_SUCCESS;
791 unsigned long msr_save;
792 int cpu;
793
794 atomic_inc(&data->working);
795
796 /* really need to ensure MSR.EE is off for H_JOIN */
797 msr_save = mfmsr();
798 mtmsr(msr_save & ~(MSR_EE));
799
800 while (rc == H_SUCCESS && !atomic_read(&data->done) && !atomic_read(&data->error))
801 rc = plpar_hcall_norets(H_JOIN);
802
803 mtmsr(msr_save);
804
805 if (rc == H_SUCCESS) {
806 /* This cpu was prodded and the suspend is complete. */
807 goto out;
808 } else if (rc == H_CONTINUE) {
809 /* All other cpus are in H_JOIN, this cpu does
810 * the suspend.
811 */
812 return __rtas_suspend_last_cpu(data, wake_when_done);
813 } else {
814 printk(KERN_ERR "H_JOIN on cpu %i failed with rc = %ld\n",
815 smp_processor_id(), rc);
816 atomic_set(&data->error, rc);
817 }
818
819 if (wake_when_done) {
820 atomic_set(&data->done, 1);
821
822 /* This cpu did the suspend or got an error; in either case,
823 * we need to prod all other other cpus out of join state.
824 * Extra prods are harmless.
825 */
826 for_each_online_cpu(cpu)
827 plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
828 }
829out:
830 if (atomic_dec_return(&data->working) == 0)
831 complete(data->complete);
832 return rc;
833}
834
835int rtas_suspend_cpu(struct rtas_suspend_me_data *data)
836{
837 return __rtas_suspend_cpu(data, 0);
838}
839
840static void rtas_percpu_suspend_me(void *info)
841{
842 __rtas_suspend_cpu((struct rtas_suspend_me_data *)info, 1);
843}
844
845enum rtas_cpu_state {
846 DOWN,
847 UP,
848};
849
850#ifndef CONFIG_SMP
851static int rtas_cpu_state_change_mask(enum rtas_cpu_state state,
852 cpumask_var_t cpus)
853{
854 if (!cpumask_empty(cpus)) {
855 cpumask_clear(cpus);
856 return -EINVAL;
857 } else
858 return 0;
859}
860#else
861/* On return cpumask will be altered to indicate CPUs changed.
862 * CPUs with states changed will be set in the mask,
863 * CPUs with status unchanged will be unset in the mask. */
864static int rtas_cpu_state_change_mask(enum rtas_cpu_state state,
865 cpumask_var_t cpus)
866{
867 int cpu;
868 int cpuret = 0;
869 int ret = 0;
870
871 if (cpumask_empty(cpus))
872 return 0;
873
874 for_each_cpu(cpu, cpus) {
875 struct device *dev = get_cpu_device(cpu);
876
877 switch (state) {
878 case DOWN:
879 cpuret = device_offline(dev);
880 break;
881 case UP:
882 cpuret = device_online(dev);
883 break;
884 }
885 if (cpuret < 0) {
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 cond_resched();
903 }
904
905 return ret;
906}
907#endif
908
909int rtas_online_cpus_mask(cpumask_var_t cpus)
910{
911 int ret;
912
913 ret = rtas_cpu_state_change_mask(UP, cpus);
914
915 if (ret) {
916 cpumask_var_t tmp_mask;
917
918 if (!alloc_cpumask_var(&tmp_mask, GFP_KERNEL))
919 return ret;
920
921 /* Use tmp_mask to preserve cpus mask from first failure */
922 cpumask_copy(tmp_mask, cpus);
923 rtas_offline_cpus_mask(tmp_mask);
924 free_cpumask_var(tmp_mask);
925 }
926
927 return ret;
928}
929
930int rtas_offline_cpus_mask(cpumask_var_t cpus)
931{
932 return rtas_cpu_state_change_mask(DOWN, cpus);
933}
934
935int rtas_ibm_suspend_me(u64 handle)
936{
937 long state;
938 long rc;
939 unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
940 struct rtas_suspend_me_data data;
941 DECLARE_COMPLETION_ONSTACK(done);
942 cpumask_var_t offline_mask;
943 int cpuret;
944
945 if (!rtas_service_present("ibm,suspend-me"))
946 return -ENOSYS;
947
948 /* Make sure the state is valid */
949 rc = plpar_hcall(H_VASI_STATE, retbuf, handle);
950
951 state = retbuf[0];
952
953 if (rc) {
954 printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned %ld\n",rc);
955 return rc;
956 } else if (state == H_VASI_ENABLED) {
957 return -EAGAIN;
958 } else if (state != H_VASI_SUSPENDING) {
959 printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned state %ld\n",
960 state);
961 return -EIO;
962 }
963
964 if (!alloc_cpumask_var(&offline_mask, GFP_KERNEL))
965 return -ENOMEM;
966
967 atomic_set(&data.working, 0);
968 atomic_set(&data.done, 0);
969 atomic_set(&data.error, 0);
970 data.token = rtas_token("ibm,suspend-me");
971 data.complete = &done;
972
973 lock_device_hotplug();
974
975 /* All present CPUs must be online */
976 cpumask_andnot(offline_mask, cpu_present_mask, cpu_online_mask);
977 cpuret = rtas_online_cpus_mask(offline_mask);
978 if (cpuret) {
979 pr_err("%s: Could not bring present CPUs online.\n", __func__);
980 atomic_set(&data.error, cpuret);
981 goto out;
982 }
983
984 cpu_hotplug_disable();
985
986 /* Check if we raced with a CPU-Offline Operation */
987 if (!cpumask_equal(cpu_present_mask, cpu_online_mask)) {
988 pr_info("%s: Raced against a concurrent CPU-Offline\n", __func__);
989 atomic_set(&data.error, -EAGAIN);
990 goto out_hotplug_enable;
991 }
992
993 /* Call function on all CPUs. One of us will make the
994 * rtas call
995 */
996 on_each_cpu(rtas_percpu_suspend_me, &data, 0);
997
998 wait_for_completion(&done);
999
1000 if (atomic_read(&data.error) != 0)
1001 printk(KERN_ERR "Error doing global join\n");
1002
1003out_hotplug_enable:
1004 cpu_hotplug_enable();
1005
1006 /* Take down CPUs not online prior to suspend */
1007 cpuret = rtas_offline_cpus_mask(offline_mask);
1008 if (cpuret)
1009 pr_warn("%s: Could not restore CPUs to offline state.\n",
1010 __func__);
1011
1012out:
1013 unlock_device_hotplug();
1014 free_cpumask_var(offline_mask);
1015 return atomic_read(&data.error);
1016}
1017#else /* CONFIG_PPC_PSERIES */
1018int rtas_ibm_suspend_me(u64 handle)
1019{
1020 return -ENOSYS;
1021}
1022#endif
1023
1024/**
1025 * Find a specific pseries error log in an RTAS extended event log.
1026 * @log: RTAS error/event log
1027 * @section_id: two character section identifier
1028 *
1029 * Returns a pointer to the specified errorlog or NULL if not found.
1030 */
1031struct pseries_errorlog *get_pseries_errorlog(struct rtas_error_log *log,
1032 uint16_t section_id)
1033{
1034 struct rtas_ext_event_log_v6 *ext_log =
1035 (struct rtas_ext_event_log_v6 *)log->buffer;
1036 struct pseries_errorlog *sect;
1037 unsigned char *p, *log_end;
1038 uint32_t ext_log_length = rtas_error_extended_log_length(log);
1039 uint8_t log_format = rtas_ext_event_log_format(ext_log);
1040 uint32_t company_id = rtas_ext_event_company_id(ext_log);
1041
1042 /* Check that we understand the format */
1043 if (ext_log_length < sizeof(struct rtas_ext_event_log_v6) ||
1044 log_format != RTAS_V6EXT_LOG_FORMAT_EVENT_LOG ||
1045 company_id != RTAS_V6EXT_COMPANY_ID_IBM)
1046 return NULL;
1047
1048 log_end = log->buffer + ext_log_length;
1049 p = ext_log->vendor_log;
1050
1051 while (p < log_end) {
1052 sect = (struct pseries_errorlog *)p;
1053 if (pseries_errorlog_id(sect) == section_id)
1054 return sect;
1055 p += pseries_errorlog_length(sect);
1056 }
1057
1058 return NULL;
1059}
1060
1061/* We assume to be passed big endian arguments */
1062SYSCALL_DEFINE1(rtas, struct rtas_args __user *, uargs)
1063{
1064 struct rtas_args args;
1065 unsigned long flags;
1066 char *buff_copy, *errbuf = NULL;
1067 int nargs, nret, token;
1068
1069 if (!capable(CAP_SYS_ADMIN))
1070 return -EPERM;
1071
1072 if (!rtas.entry)
1073 return -EINVAL;
1074
1075 if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
1076 return -EFAULT;
1077
1078 nargs = be32_to_cpu(args.nargs);
1079 nret = be32_to_cpu(args.nret);
1080 token = be32_to_cpu(args.token);
1081
1082 if (nargs >= ARRAY_SIZE(args.args)
1083 || nret > ARRAY_SIZE(args.args)
1084 || nargs + nret > ARRAY_SIZE(args.args))
1085 return -EINVAL;
1086
1087 /* Copy in args. */
1088 if (copy_from_user(args.args, uargs->args,
1089 nargs * sizeof(rtas_arg_t)) != 0)
1090 return -EFAULT;
1091
1092 if (token == RTAS_UNKNOWN_SERVICE)
1093 return -EINVAL;
1094
1095 args.rets = &args.args[nargs];
1096 memset(args.rets, 0, nret * sizeof(rtas_arg_t));
1097
1098 /* Need to handle ibm,suspend_me call specially */
1099 if (token == ibm_suspend_me_token) {
1100
1101 /*
1102 * rtas_ibm_suspend_me assumes the streamid handle is in cpu
1103 * endian, or at least the hcall within it requires it.
1104 */
1105 int rc = 0;
1106 u64 handle = ((u64)be32_to_cpu(args.args[0]) << 32)
1107 | be32_to_cpu(args.args[1]);
1108 rc = rtas_ibm_suspend_me(handle);
1109 if (rc == -EAGAIN)
1110 args.rets[0] = cpu_to_be32(RTAS_NOT_SUSPENDABLE);
1111 else if (rc == -EIO)
1112 args.rets[0] = cpu_to_be32(-1);
1113 else if (rc)
1114 return rc;
1115 goto copy_return;
1116 }
1117
1118 buff_copy = get_errorlog_buffer();
1119
1120 flags = lock_rtas();
1121
1122 rtas.args = args;
1123 enter_rtas(__pa(&rtas.args));
1124 args = rtas.args;
1125
1126 /* A -1 return code indicates that the last command couldn't
1127 be completed due to a hardware error. */
1128 if (be32_to_cpu(args.rets[0]) == -1)
1129 errbuf = __fetch_rtas_last_error(buff_copy);
1130
1131 unlock_rtas(flags);
1132
1133 if (buff_copy) {
1134 if (errbuf)
1135 log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
1136 kfree(buff_copy);
1137 }
1138
1139 copy_return:
1140 /* Copy out args. */
1141 if (copy_to_user(uargs->args + nargs,
1142 args.args + nargs,
1143 nret * sizeof(rtas_arg_t)) != 0)
1144 return -EFAULT;
1145
1146 return 0;
1147}
1148
1149/*
1150 * Call early during boot, before mem init, to retrieve the RTAS
1151 * information from the device-tree and allocate the RMO buffer for userland
1152 * accesses.
1153 */
1154void __init rtas_initialize(void)
1155{
1156 unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
1157 u32 base, size, entry;
1158 int no_base, no_size, no_entry;
1159
1160 /* Get RTAS dev node and fill up our "rtas" structure with infos
1161 * about it.
1162 */
1163 rtas.dev = of_find_node_by_name(NULL, "rtas");
1164 if (!rtas.dev)
1165 return;
1166
1167 no_base = of_property_read_u32(rtas.dev, "linux,rtas-base", &base);
1168 no_size = of_property_read_u32(rtas.dev, "rtas-size", &size);
1169 if (no_base || no_size) {
1170 of_node_put(rtas.dev);
1171 rtas.dev = NULL;
1172 return;
1173 }
1174
1175 rtas.base = base;
1176 rtas.size = size;
1177 no_entry = of_property_read_u32(rtas.dev, "linux,rtas-entry", &entry);
1178 rtas.entry = no_entry ? rtas.base : entry;
1179
1180 /* If RTAS was found, allocate the RMO buffer for it and look for
1181 * the stop-self token if any
1182 */
1183#ifdef CONFIG_PPC64
1184 if (firmware_has_feature(FW_FEATURE_LPAR)) {
1185 rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
1186 ibm_suspend_me_token = rtas_token("ibm,suspend-me");
1187 }
1188#endif
1189 rtas_rmo_buf = memblock_phys_alloc_range(RTAS_RMOBUF_MAX, PAGE_SIZE,
1190 0, rtas_region);
1191 if (!rtas_rmo_buf)
1192 panic("ERROR: RTAS: Failed to allocate %lx bytes below %pa\n",
1193 PAGE_SIZE, &rtas_region);
1194
1195#ifdef CONFIG_RTAS_ERROR_LOGGING
1196 rtas_last_error_token = rtas_token("rtas-last-error");
1197#endif
1198}
1199
1200int __init early_init_dt_scan_rtas(unsigned long node,
1201 const char *uname, int depth, void *data)
1202{
1203 const u32 *basep, *entryp, *sizep;
1204
1205 if (depth != 1 || strcmp(uname, "rtas") != 0)
1206 return 0;
1207
1208 basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
1209 entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
1210 sizep = of_get_flat_dt_prop(node, "rtas-size", NULL);
1211
1212 if (basep && entryp && sizep) {
1213 rtas.base = *basep;
1214 rtas.entry = *entryp;
1215 rtas.size = *sizep;
1216 }
1217
1218#ifdef CONFIG_UDBG_RTAS_CONSOLE
1219 basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
1220 if (basep)
1221 rtas_putchar_token = *basep;
1222
1223 basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
1224 if (basep)
1225 rtas_getchar_token = *basep;
1226
1227 if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
1228 rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
1229 udbg_init_rtas_console();
1230
1231#endif
1232
1233 /* break now */
1234 return 1;
1235}
1236
1237static arch_spinlock_t timebase_lock;
1238static u64 timebase = 0;
1239
1240void rtas_give_timebase(void)
1241{
1242 unsigned long flags;
1243
1244 local_irq_save(flags);
1245 hard_irq_disable();
1246 arch_spin_lock(&timebase_lock);
1247 rtas_call(rtas_token("freeze-time-base"), 0, 1, NULL);
1248 timebase = get_tb();
1249 arch_spin_unlock(&timebase_lock);
1250
1251 while (timebase)
1252 barrier();
1253 rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
1254 local_irq_restore(flags);
1255}
1256
1257void rtas_take_timebase(void)
1258{
1259 while (!timebase)
1260 barrier();
1261 arch_spin_lock(&timebase_lock);
1262 set_tb(timebase >> 32, timebase & 0xffffffff);
1263 timebase = 0;
1264 arch_spin_unlock(&timebase_lock);
1265}