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#define pr_fmt(fmt) "rtas: " fmt
11
12#include <linux/bsearch.h>
13#include <linux/capability.h>
14#include <linux/delay.h>
15#include <linux/export.h>
16#include <linux/init.h>
17#include <linux/kconfig.h>
18#include <linux/kernel.h>
19#include <linux/lockdep.h>
20#include <linux/memblock.h>
21#include <linux/mutex.h>
22#include <linux/nospec.h>
23#include <linux/of.h>
24#include <linux/of_fdt.h>
25#include <linux/reboot.h>
26#include <linux/sched.h>
27#include <linux/security.h>
28#include <linux/slab.h>
29#include <linux/spinlock.h>
30#include <linux/stdarg.h>
31#include <linux/syscalls.h>
32#include <linux/types.h>
33#include <linux/uaccess.h>
34#include <linux/xarray.h>
35
36#include <asm/delay.h>
37#include <asm/firmware.h>
38#include <asm/interrupt.h>
39#include <asm/machdep.h>
40#include <asm/mmu.h>
41#include <asm/page.h>
42#include <asm/rtas-work-area.h>
43#include <asm/rtas.h>
44#include <asm/time.h>
45#include <asm/trace.h>
46#include <asm/udbg.h>
47
48struct rtas_filter {
49 /* Indexes into the args buffer, -1 if not used */
50 const int buf_idx1;
51 const int size_idx1;
52 const int buf_idx2;
53 const int size_idx2;
54 /*
55 * Assumed buffer size per the spec if the function does not
56 * have a size parameter, e.g. ibm,errinjct. 0 if unused.
57 */
58 const int fixed_size;
59};
60
61/**
62 * struct rtas_function - Descriptor for RTAS functions.
63 *
64 * @token: Value of @name if it exists under the /rtas node.
65 * @name: Function name.
66 * @filter: If non-NULL, invoking this function via the rtas syscall is
67 * generally allowed, and @filter describes constraints on the
68 * arguments. See also @banned_for_syscall_on_le.
69 * @banned_for_syscall_on_le: Set when call via sys_rtas is generally allowed
70 * but specifically restricted on ppc64le. Such
71 * functions are believed to have no users on
72 * ppc64le, and we want to keep it that way. It does
73 * not make sense for this to be set when @filter
74 * is NULL.
75 * @lock: Pointer to an optional dedicated per-function mutex. This
76 * should be set for functions that require multiple calls in
77 * sequence to complete a single operation, and such sequences
78 * will disrupt each other if allowed to interleave. Users of
79 * this function are required to hold the associated lock for
80 * the duration of the call sequence. Add an explanatory
81 * comment to the function table entry if setting this member.
82 */
83struct rtas_function {
84 s32 token;
85 const bool banned_for_syscall_on_le:1;
86 const char * const name;
87 const struct rtas_filter *filter;
88 struct mutex *lock;
89};
90
91/*
92 * Per-function locks for sequence-based RTAS functions.
93 */
94static DEFINE_MUTEX(rtas_ibm_activate_firmware_lock);
95static DEFINE_MUTEX(rtas_ibm_get_dynamic_sensor_state_lock);
96static DEFINE_MUTEX(rtas_ibm_get_indices_lock);
97static DEFINE_MUTEX(rtas_ibm_lpar_perftools_lock);
98static DEFINE_MUTEX(rtas_ibm_physical_attestation_lock);
99static DEFINE_MUTEX(rtas_ibm_set_dynamic_indicator_lock);
100DEFINE_MUTEX(rtas_ibm_get_vpd_lock);
101
102static struct rtas_function rtas_function_table[] __ro_after_init = {
103 [RTAS_FNIDX__CHECK_EXCEPTION] = {
104 .name = "check-exception",
105 },
106 [RTAS_FNIDX__DISPLAY_CHARACTER] = {
107 .name = "display-character",
108 .filter = &(const struct rtas_filter) {
109 .buf_idx1 = -1, .size_idx1 = -1,
110 .buf_idx2 = -1, .size_idx2 = -1,
111 },
112 },
113 [RTAS_FNIDX__EVENT_SCAN] = {
114 .name = "event-scan",
115 },
116 [RTAS_FNIDX__FREEZE_TIME_BASE] = {
117 .name = "freeze-time-base",
118 },
119 [RTAS_FNIDX__GET_POWER_LEVEL] = {
120 .name = "get-power-level",
121 .filter = &(const struct rtas_filter) {
122 .buf_idx1 = -1, .size_idx1 = -1,
123 .buf_idx2 = -1, .size_idx2 = -1,
124 },
125 },
126 [RTAS_FNIDX__GET_SENSOR_STATE] = {
127 .name = "get-sensor-state",
128 .filter = &(const struct rtas_filter) {
129 .buf_idx1 = -1, .size_idx1 = -1,
130 .buf_idx2 = -1, .size_idx2 = -1,
131 },
132 },
133 [RTAS_FNIDX__GET_TERM_CHAR] = {
134 .name = "get-term-char",
135 },
136 [RTAS_FNIDX__GET_TIME_OF_DAY] = {
137 .name = "get-time-of-day",
138 .filter = &(const struct rtas_filter) {
139 .buf_idx1 = -1, .size_idx1 = -1,
140 .buf_idx2 = -1, .size_idx2 = -1,
141 },
142 },
143 [RTAS_FNIDX__IBM_ACTIVATE_FIRMWARE] = {
144 .name = "ibm,activate-firmware",
145 .filter = &(const struct rtas_filter) {
146 .buf_idx1 = -1, .size_idx1 = -1,
147 .buf_idx2 = -1, .size_idx2 = -1,
148 },
149 /*
150 * PAPR+ as of v2.13 doesn't explicitly impose any
151 * restriction, but this typically requires multiple
152 * calls before success, and there's no reason to
153 * allow sequences to interleave.
154 */
155 .lock = &rtas_ibm_activate_firmware_lock,
156 },
157 [RTAS_FNIDX__IBM_CBE_START_PTCAL] = {
158 .name = "ibm,cbe-start-ptcal",
159 },
160 [RTAS_FNIDX__IBM_CBE_STOP_PTCAL] = {
161 .name = "ibm,cbe-stop-ptcal",
162 },
163 [RTAS_FNIDX__IBM_CHANGE_MSI] = {
164 .name = "ibm,change-msi",
165 },
166 [RTAS_FNIDX__IBM_CLOSE_ERRINJCT] = {
167 .name = "ibm,close-errinjct",
168 .filter = &(const struct rtas_filter) {
169 .buf_idx1 = -1, .size_idx1 = -1,
170 .buf_idx2 = -1, .size_idx2 = -1,
171 },
172 },
173 [RTAS_FNIDX__IBM_CONFIGURE_BRIDGE] = {
174 .name = "ibm,configure-bridge",
175 },
176 [RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR] = {
177 .name = "ibm,configure-connector",
178 .filter = &(const struct rtas_filter) {
179 .buf_idx1 = 0, .size_idx1 = -1,
180 .buf_idx2 = 1, .size_idx2 = -1,
181 .fixed_size = 4096,
182 },
183 },
184 [RTAS_FNIDX__IBM_CONFIGURE_KERNEL_DUMP] = {
185 .name = "ibm,configure-kernel-dump",
186 },
187 [RTAS_FNIDX__IBM_CONFIGURE_PE] = {
188 .name = "ibm,configure-pe",
189 },
190 [RTAS_FNIDX__IBM_CREATE_PE_DMA_WINDOW] = {
191 .name = "ibm,create-pe-dma-window",
192 },
193 [RTAS_FNIDX__IBM_DISPLAY_MESSAGE] = {
194 .name = "ibm,display-message",
195 .filter = &(const struct rtas_filter) {
196 .buf_idx1 = 0, .size_idx1 = -1,
197 .buf_idx2 = -1, .size_idx2 = -1,
198 },
199 },
200 [RTAS_FNIDX__IBM_ERRINJCT] = {
201 .name = "ibm,errinjct",
202 .filter = &(const struct rtas_filter) {
203 .buf_idx1 = 2, .size_idx1 = -1,
204 .buf_idx2 = -1, .size_idx2 = -1,
205 .fixed_size = 1024,
206 },
207 },
208 [RTAS_FNIDX__IBM_EXTI2C] = {
209 .name = "ibm,exti2c",
210 },
211 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO] = {
212 .name = "ibm,get-config-addr-info",
213 },
214 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO2] = {
215 .name = "ibm,get-config-addr-info2",
216 .filter = &(const struct rtas_filter) {
217 .buf_idx1 = -1, .size_idx1 = -1,
218 .buf_idx2 = -1, .size_idx2 = -1,
219 },
220 },
221 [RTAS_FNIDX__IBM_GET_DYNAMIC_SENSOR_STATE] = {
222 .name = "ibm,get-dynamic-sensor-state",
223 .filter = &(const struct rtas_filter) {
224 .buf_idx1 = 1, .size_idx1 = -1,
225 .buf_idx2 = -1, .size_idx2 = -1,
226 },
227 /*
228 * PAPR+ v2.13 R1–7.3.19–3 is explicit that the OS
229 * must not call ibm,get-dynamic-sensor-state with
230 * different inputs until a non-retry status has been
231 * returned.
232 */
233 .lock = &rtas_ibm_get_dynamic_sensor_state_lock,
234 },
235 [RTAS_FNIDX__IBM_GET_INDICES] = {
236 .name = "ibm,get-indices",
237 .filter = &(const struct rtas_filter) {
238 .buf_idx1 = 2, .size_idx1 = 3,
239 .buf_idx2 = -1, .size_idx2 = -1,
240 },
241 /*
242 * PAPR+ v2.13 R1–7.3.17–2 says that the OS must not
243 * interleave ibm,get-indices call sequences with
244 * different inputs.
245 */
246 .lock = &rtas_ibm_get_indices_lock,
247 },
248 [RTAS_FNIDX__IBM_GET_RIO_TOPOLOGY] = {
249 .name = "ibm,get-rio-topology",
250 },
251 [RTAS_FNIDX__IBM_GET_SYSTEM_PARAMETER] = {
252 .name = "ibm,get-system-parameter",
253 .filter = &(const struct rtas_filter) {
254 .buf_idx1 = 1, .size_idx1 = 2,
255 .buf_idx2 = -1, .size_idx2 = -1,
256 },
257 },
258 [RTAS_FNIDX__IBM_GET_VPD] = {
259 .name = "ibm,get-vpd",
260 .filter = &(const struct rtas_filter) {
261 .buf_idx1 = 0, .size_idx1 = -1,
262 .buf_idx2 = 1, .size_idx2 = 2,
263 },
264 /*
265 * PAPR+ v2.13 R1–7.3.20–4 indicates that sequences
266 * should not be allowed to interleave.
267 */
268 .lock = &rtas_ibm_get_vpd_lock,
269 },
270 [RTAS_FNIDX__IBM_GET_XIVE] = {
271 .name = "ibm,get-xive",
272 },
273 [RTAS_FNIDX__IBM_INT_OFF] = {
274 .name = "ibm,int-off",
275 },
276 [RTAS_FNIDX__IBM_INT_ON] = {
277 .name = "ibm,int-on",
278 },
279 [RTAS_FNIDX__IBM_IO_QUIESCE_ACK] = {
280 .name = "ibm,io-quiesce-ack",
281 },
282 [RTAS_FNIDX__IBM_LPAR_PERFTOOLS] = {
283 .name = "ibm,lpar-perftools",
284 .filter = &(const struct rtas_filter) {
285 .buf_idx1 = 2, .size_idx1 = 3,
286 .buf_idx2 = -1, .size_idx2 = -1,
287 },
288 /*
289 * PAPR+ v2.13 R1–7.3.26–6 says the OS should allow
290 * only one call sequence in progress at a time.
291 */
292 .lock = &rtas_ibm_lpar_perftools_lock,
293 },
294 [RTAS_FNIDX__IBM_MANAGE_FLASH_IMAGE] = {
295 .name = "ibm,manage-flash-image",
296 },
297 [RTAS_FNIDX__IBM_MANAGE_STORAGE_PRESERVATION] = {
298 .name = "ibm,manage-storage-preservation",
299 },
300 [RTAS_FNIDX__IBM_NMI_INTERLOCK] = {
301 .name = "ibm,nmi-interlock",
302 },
303 [RTAS_FNIDX__IBM_NMI_REGISTER] = {
304 .name = "ibm,nmi-register",
305 },
306 [RTAS_FNIDX__IBM_OPEN_ERRINJCT] = {
307 .name = "ibm,open-errinjct",
308 .filter = &(const struct rtas_filter) {
309 .buf_idx1 = -1, .size_idx1 = -1,
310 .buf_idx2 = -1, .size_idx2 = -1,
311 },
312 },
313 [RTAS_FNIDX__IBM_OPEN_SRIOV_ALLOW_UNFREEZE] = {
314 .name = "ibm,open-sriov-allow-unfreeze",
315 },
316 [RTAS_FNIDX__IBM_OPEN_SRIOV_MAP_PE_NUMBER] = {
317 .name = "ibm,open-sriov-map-pe-number",
318 },
319 [RTAS_FNIDX__IBM_OS_TERM] = {
320 .name = "ibm,os-term",
321 },
322 [RTAS_FNIDX__IBM_PARTNER_CONTROL] = {
323 .name = "ibm,partner-control",
324 },
325 [RTAS_FNIDX__IBM_PHYSICAL_ATTESTATION] = {
326 .name = "ibm,physical-attestation",
327 .filter = &(const struct rtas_filter) {
328 .buf_idx1 = 0, .size_idx1 = 1,
329 .buf_idx2 = -1, .size_idx2 = -1,
330 },
331 /*
332 * This follows a sequence-based pattern similar to
333 * ibm,get-vpd et al. Since PAPR+ restricts
334 * interleaving call sequences for other functions of
335 * this style, assume the restriction applies here,
336 * even though it's not explicit in the spec.
337 */
338 .lock = &rtas_ibm_physical_attestation_lock,
339 },
340 [RTAS_FNIDX__IBM_PLATFORM_DUMP] = {
341 .name = "ibm,platform-dump",
342 .filter = &(const struct rtas_filter) {
343 .buf_idx1 = 4, .size_idx1 = 5,
344 .buf_idx2 = -1, .size_idx2 = -1,
345 },
346 /*
347 * PAPR+ v2.13 7.3.3.4.1 indicates that concurrent
348 * sequences of ibm,platform-dump are allowed if they
349 * are operating on different dump tags. So leave the
350 * lock pointer unset for now. This may need
351 * reconsideration if kernel-internal users appear.
352 */
353 },
354 [RTAS_FNIDX__IBM_POWER_OFF_UPS] = {
355 .name = "ibm,power-off-ups",
356 },
357 [RTAS_FNIDX__IBM_QUERY_INTERRUPT_SOURCE_NUMBER] = {
358 .name = "ibm,query-interrupt-source-number",
359 },
360 [RTAS_FNIDX__IBM_QUERY_PE_DMA_WINDOW] = {
361 .name = "ibm,query-pe-dma-window",
362 },
363 [RTAS_FNIDX__IBM_READ_PCI_CONFIG] = {
364 .name = "ibm,read-pci-config",
365 },
366 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE] = {
367 .name = "ibm,read-slot-reset-state",
368 .filter = &(const struct rtas_filter) {
369 .buf_idx1 = -1, .size_idx1 = -1,
370 .buf_idx2 = -1, .size_idx2 = -1,
371 },
372 },
373 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE2] = {
374 .name = "ibm,read-slot-reset-state2",
375 },
376 [RTAS_FNIDX__IBM_REMOVE_PE_DMA_WINDOW] = {
377 .name = "ibm,remove-pe-dma-window",
378 },
379 [RTAS_FNIDX__IBM_RESET_PE_DMA_WINDOW] = {
380 /*
381 * Note: PAPR+ v2.13 7.3.31.4.1 spells this as
382 * "ibm,reset-pe-dma-windows" (plural), but RTAS
383 * implementations use the singular form in practice.
384 */
385 .name = "ibm,reset-pe-dma-window",
386 },
387 [RTAS_FNIDX__IBM_SCAN_LOG_DUMP] = {
388 .name = "ibm,scan-log-dump",
389 .filter = &(const struct rtas_filter) {
390 .buf_idx1 = 0, .size_idx1 = 1,
391 .buf_idx2 = -1, .size_idx2 = -1,
392 },
393 },
394 [RTAS_FNIDX__IBM_SET_DYNAMIC_INDICATOR] = {
395 .name = "ibm,set-dynamic-indicator",
396 .filter = &(const struct rtas_filter) {
397 .buf_idx1 = 2, .size_idx1 = -1,
398 .buf_idx2 = -1, .size_idx2 = -1,
399 },
400 /*
401 * PAPR+ v2.13 R1–7.3.18–3 says the OS must not call
402 * this function with different inputs until a
403 * non-retry status has been returned.
404 */
405 .lock = &rtas_ibm_set_dynamic_indicator_lock,
406 },
407 [RTAS_FNIDX__IBM_SET_EEH_OPTION] = {
408 .name = "ibm,set-eeh-option",
409 .filter = &(const struct rtas_filter) {
410 .buf_idx1 = -1, .size_idx1 = -1,
411 .buf_idx2 = -1, .size_idx2 = -1,
412 },
413 },
414 [RTAS_FNIDX__IBM_SET_SLOT_RESET] = {
415 .name = "ibm,set-slot-reset",
416 },
417 [RTAS_FNIDX__IBM_SET_SYSTEM_PARAMETER] = {
418 .name = "ibm,set-system-parameter",
419 .filter = &(const struct rtas_filter) {
420 .buf_idx1 = 1, .size_idx1 = -1,
421 .buf_idx2 = -1, .size_idx2 = -1,
422 },
423 },
424 [RTAS_FNIDX__IBM_SET_XIVE] = {
425 .name = "ibm,set-xive",
426 },
427 [RTAS_FNIDX__IBM_SLOT_ERROR_DETAIL] = {
428 .name = "ibm,slot-error-detail",
429 },
430 [RTAS_FNIDX__IBM_SUSPEND_ME] = {
431 .name = "ibm,suspend-me",
432 .banned_for_syscall_on_le = true,
433 .filter = &(const struct rtas_filter) {
434 .buf_idx1 = -1, .size_idx1 = -1,
435 .buf_idx2 = -1, .size_idx2 = -1,
436 },
437 },
438 [RTAS_FNIDX__IBM_TUNE_DMA_PARMS] = {
439 .name = "ibm,tune-dma-parms",
440 },
441 [RTAS_FNIDX__IBM_UPDATE_FLASH_64_AND_REBOOT] = {
442 .name = "ibm,update-flash-64-and-reboot",
443 },
444 [RTAS_FNIDX__IBM_UPDATE_NODES] = {
445 .name = "ibm,update-nodes",
446 .banned_for_syscall_on_le = true,
447 .filter = &(const struct rtas_filter) {
448 .buf_idx1 = 0, .size_idx1 = -1,
449 .buf_idx2 = -1, .size_idx2 = -1,
450 .fixed_size = 4096,
451 },
452 },
453 [RTAS_FNIDX__IBM_UPDATE_PROPERTIES] = {
454 .name = "ibm,update-properties",
455 .banned_for_syscall_on_le = true,
456 .filter = &(const struct rtas_filter) {
457 .buf_idx1 = 0, .size_idx1 = -1,
458 .buf_idx2 = -1, .size_idx2 = -1,
459 .fixed_size = 4096,
460 },
461 },
462 [RTAS_FNIDX__IBM_VALIDATE_FLASH_IMAGE] = {
463 .name = "ibm,validate-flash-image",
464 },
465 [RTAS_FNIDX__IBM_WRITE_PCI_CONFIG] = {
466 .name = "ibm,write-pci-config",
467 },
468 [RTAS_FNIDX__NVRAM_FETCH] = {
469 .name = "nvram-fetch",
470 },
471 [RTAS_FNIDX__NVRAM_STORE] = {
472 .name = "nvram-store",
473 },
474 [RTAS_FNIDX__POWER_OFF] = {
475 .name = "power-off",
476 },
477 [RTAS_FNIDX__PUT_TERM_CHAR] = {
478 .name = "put-term-char",
479 },
480 [RTAS_FNIDX__QUERY_CPU_STOPPED_STATE] = {
481 .name = "query-cpu-stopped-state",
482 },
483 [RTAS_FNIDX__READ_PCI_CONFIG] = {
484 .name = "read-pci-config",
485 },
486 [RTAS_FNIDX__RTAS_LAST_ERROR] = {
487 .name = "rtas-last-error",
488 },
489 [RTAS_FNIDX__SET_INDICATOR] = {
490 .name = "set-indicator",
491 .filter = &(const struct rtas_filter) {
492 .buf_idx1 = -1, .size_idx1 = -1,
493 .buf_idx2 = -1, .size_idx2 = -1,
494 },
495 },
496 [RTAS_FNIDX__SET_POWER_LEVEL] = {
497 .name = "set-power-level",
498 .filter = &(const struct rtas_filter) {
499 .buf_idx1 = -1, .size_idx1 = -1,
500 .buf_idx2 = -1, .size_idx2 = -1,
501 },
502 },
503 [RTAS_FNIDX__SET_TIME_FOR_POWER_ON] = {
504 .name = "set-time-for-power-on",
505 .filter = &(const struct rtas_filter) {
506 .buf_idx1 = -1, .size_idx1 = -1,
507 .buf_idx2 = -1, .size_idx2 = -1,
508 },
509 },
510 [RTAS_FNIDX__SET_TIME_OF_DAY] = {
511 .name = "set-time-of-day",
512 .filter = &(const struct rtas_filter) {
513 .buf_idx1 = -1, .size_idx1 = -1,
514 .buf_idx2 = -1, .size_idx2 = -1,
515 },
516 },
517 [RTAS_FNIDX__START_CPU] = {
518 .name = "start-cpu",
519 },
520 [RTAS_FNIDX__STOP_SELF] = {
521 .name = "stop-self",
522 },
523 [RTAS_FNIDX__SYSTEM_REBOOT] = {
524 .name = "system-reboot",
525 },
526 [RTAS_FNIDX__THAW_TIME_BASE] = {
527 .name = "thaw-time-base",
528 },
529 [RTAS_FNIDX__WRITE_PCI_CONFIG] = {
530 .name = "write-pci-config",
531 },
532};
533
534#define for_each_rtas_function(funcp) \
535 for (funcp = &rtas_function_table[0]; \
536 funcp < &rtas_function_table[ARRAY_SIZE(rtas_function_table)]; \
537 ++funcp)
538
539/*
540 * Nearly all RTAS calls need to be serialized. All uses of the
541 * default rtas_args block must hold rtas_lock.
542 *
543 * Exceptions to the RTAS serialization requirement (e.g. stop-self)
544 * must use a separate rtas_args structure.
545 */
546static DEFINE_RAW_SPINLOCK(rtas_lock);
547static struct rtas_args rtas_args;
548
549/**
550 * rtas_function_token() - RTAS function token lookup.
551 * @handle: Function handle, e.g. RTAS_FN_EVENT_SCAN.
552 *
553 * Context: Any context.
554 * Return: the token value for the function if implemented by this platform,
555 * otherwise RTAS_UNKNOWN_SERVICE.
556 */
557s32 rtas_function_token(const rtas_fn_handle_t handle)
558{
559 const size_t index = handle.index;
560 const bool out_of_bounds = index >= ARRAY_SIZE(rtas_function_table);
561
562 if (WARN_ONCE(out_of_bounds, "invalid function index %zu", index))
563 return RTAS_UNKNOWN_SERVICE;
564 /*
565 * Various drivers attempt token lookups on non-RTAS
566 * platforms.
567 */
568 if (!rtas.dev)
569 return RTAS_UNKNOWN_SERVICE;
570
571 return rtas_function_table[index].token;
572}
573EXPORT_SYMBOL_GPL(rtas_function_token);
574
575static int rtas_function_cmp(const void *a, const void *b)
576{
577 const struct rtas_function *f1 = a;
578 const struct rtas_function *f2 = b;
579
580 return strcmp(f1->name, f2->name);
581}
582
583/*
584 * Boot-time initialization of the function table needs the lookup to
585 * return a non-const-qualified object. Use rtas_name_to_function()
586 * in all other contexts.
587 */
588static struct rtas_function *__rtas_name_to_function(const char *name)
589{
590 const struct rtas_function key = {
591 .name = name,
592 };
593 struct rtas_function *found;
594
595 found = bsearch(&key, rtas_function_table, ARRAY_SIZE(rtas_function_table),
596 sizeof(rtas_function_table[0]), rtas_function_cmp);
597
598 return found;
599}
600
601static const struct rtas_function *rtas_name_to_function(const char *name)
602{
603 return __rtas_name_to_function(name);
604}
605
606static DEFINE_XARRAY(rtas_token_to_function_xarray);
607
608static int __init rtas_token_to_function_xarray_init(void)
609{
610 const struct rtas_function *func;
611 int err = 0;
612
613 for_each_rtas_function(func) {
614 const s32 token = func->token;
615
616 if (token == RTAS_UNKNOWN_SERVICE)
617 continue;
618
619 err = xa_err(xa_store(&rtas_token_to_function_xarray,
620 token, (void *)func, GFP_KERNEL));
621 if (err)
622 break;
623 }
624
625 return err;
626}
627arch_initcall(rtas_token_to_function_xarray_init);
628
629/*
630 * For use by sys_rtas(), where the token value is provided by user
631 * space and we don't want to warn on failed lookups.
632 */
633static const struct rtas_function *rtas_token_to_function_untrusted(s32 token)
634{
635 return xa_load(&rtas_token_to_function_xarray, token);
636}
637
638/*
639 * Reverse lookup for deriving the function descriptor from a
640 * known-good token value in contexts where the former is not already
641 * available. @token must be valid, e.g. derived from the result of a
642 * prior lookup against the function table.
643 */
644static const struct rtas_function *rtas_token_to_function(s32 token)
645{
646 const struct rtas_function *func;
647
648 if (WARN_ONCE(token < 0, "invalid token %d", token))
649 return NULL;
650
651 func = rtas_token_to_function_untrusted(token);
652 if (func)
653 return func;
654 /*
655 * Fall back to linear scan in case the reverse mapping hasn't
656 * been initialized yet.
657 */
658 if (xa_empty(&rtas_token_to_function_xarray)) {
659 for_each_rtas_function(func) {
660 if (func->token == token)
661 return func;
662 }
663 }
664
665 WARN_ONCE(true, "unexpected failed lookup for token %d", token);
666 return NULL;
667}
668
669/* This is here deliberately so it's only used in this file */
670void enter_rtas(unsigned long);
671
672static void __do_enter_rtas(struct rtas_args *args)
673{
674 enter_rtas(__pa(args));
675 srr_regs_clobbered(); /* rtas uses SRRs, invalidate */
676}
677
678static void __do_enter_rtas_trace(struct rtas_args *args)
679{
680 const struct rtas_function *func = rtas_token_to_function(be32_to_cpu(args->token));
681
682 /*
683 * If there is a per-function lock, it must be held by the
684 * caller.
685 */
686 if (func->lock)
687 lockdep_assert_held(func->lock);
688
689 if (args == &rtas_args)
690 lockdep_assert_held(&rtas_lock);
691
692 trace_rtas_input(args, func->name);
693 trace_rtas_ll_entry(args);
694
695 __do_enter_rtas(args);
696
697 trace_rtas_ll_exit(args);
698 trace_rtas_output(args, func->name);
699}
700
701static void do_enter_rtas(struct rtas_args *args)
702{
703 const unsigned long msr = mfmsr();
704 /*
705 * Situations where we want to skip any active tracepoints for
706 * safety reasons:
707 *
708 * 1. The last code executed on an offline CPU as it stops,
709 * i.e. we're about to call stop-self. The tracepoints'
710 * function name lookup uses xarray, which uses RCU, which
711 * isn't valid to call on an offline CPU. Any events
712 * emitted on an offline CPU will be discarded anyway.
713 *
714 * 2. In real mode, as when invoking ibm,nmi-interlock from
715 * the pseries MCE handler. We cannot count on trace
716 * buffers or the entries in rtas_token_to_function_xarray
717 * to be contained in the RMO.
718 */
719 const unsigned long mask = MSR_IR | MSR_DR;
720 const bool can_trace = likely(cpu_online(raw_smp_processor_id()) &&
721 (msr & mask) == mask);
722 /*
723 * Make sure MSR[RI] is currently enabled as it will be forced later
724 * in enter_rtas.
725 */
726 BUG_ON(!(msr & MSR_RI));
727
728 BUG_ON(!irqs_disabled());
729
730 hard_irq_disable(); /* Ensure MSR[EE] is disabled on PPC64 */
731
732 if (can_trace)
733 __do_enter_rtas_trace(args);
734 else
735 __do_enter_rtas(args);
736}
737
738struct rtas_t rtas;
739
740DEFINE_SPINLOCK(rtas_data_buf_lock);
741EXPORT_SYMBOL_GPL(rtas_data_buf_lock);
742
743char rtas_data_buf[RTAS_DATA_BUF_SIZE] __aligned(SZ_4K);
744EXPORT_SYMBOL_GPL(rtas_data_buf);
745
746unsigned long rtas_rmo_buf;
747
748/*
749 * If non-NULL, this gets called when the kernel terminates.
750 * This is done like this so rtas_flash can be a module.
751 */
752void (*rtas_flash_term_hook)(int);
753EXPORT_SYMBOL_GPL(rtas_flash_term_hook);
754
755/*
756 * call_rtas_display_status and call_rtas_display_status_delay
757 * are designed only for very early low-level debugging, which
758 * is why the token is hard-coded to 10.
759 */
760static void call_rtas_display_status(unsigned char c)
761{
762 unsigned long flags;
763
764 if (!rtas.base)
765 return;
766
767 raw_spin_lock_irqsave(&rtas_lock, flags);
768 rtas_call_unlocked(&rtas_args, 10, 1, 1, NULL, c);
769 raw_spin_unlock_irqrestore(&rtas_lock, flags);
770}
771
772static void call_rtas_display_status_delay(char c)
773{
774 static int pending_newline = 0; /* did last write end with unprinted newline? */
775 static int width = 16;
776
777 if (c == '\n') {
778 while (width-- > 0)
779 call_rtas_display_status(' ');
780 width = 16;
781 mdelay(500);
782 pending_newline = 1;
783 } else {
784 if (pending_newline) {
785 call_rtas_display_status('\r');
786 call_rtas_display_status('\n');
787 }
788 pending_newline = 0;
789 if (width--) {
790 call_rtas_display_status(c);
791 udelay(10000);
792 }
793 }
794}
795
796void __init udbg_init_rtas_panel(void)
797{
798 udbg_putc = call_rtas_display_status_delay;
799}
800
801#ifdef CONFIG_UDBG_RTAS_CONSOLE
802
803/* If you think you're dying before early_init_dt_scan_rtas() does its
804 * work, you can hard code the token values for your firmware here and
805 * hardcode rtas.base/entry etc.
806 */
807static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
808static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
809
810static void udbg_rtascon_putc(char c)
811{
812 int tries;
813
814 if (!rtas.base)
815 return;
816
817 /* Add CRs before LFs */
818 if (c == '\n')
819 udbg_rtascon_putc('\r');
820
821 /* if there is more than one character to be displayed, wait a bit */
822 for (tries = 0; tries < 16; tries++) {
823 if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
824 break;
825 udelay(1000);
826 }
827}
828
829static int udbg_rtascon_getc_poll(void)
830{
831 int c;
832
833 if (!rtas.base)
834 return -1;
835
836 if (rtas_call(rtas_getchar_token, 0, 2, &c))
837 return -1;
838
839 return c;
840}
841
842static int udbg_rtascon_getc(void)
843{
844 int c;
845
846 while ((c = udbg_rtascon_getc_poll()) == -1)
847 ;
848
849 return c;
850}
851
852
853void __init udbg_init_rtas_console(void)
854{
855 udbg_putc = udbg_rtascon_putc;
856 udbg_getc = udbg_rtascon_getc;
857 udbg_getc_poll = udbg_rtascon_getc_poll;
858}
859#endif /* CONFIG_UDBG_RTAS_CONSOLE */
860
861void rtas_progress(char *s, unsigned short hex)
862{
863 struct device_node *root;
864 int width;
865 const __be32 *p;
866 char *os;
867 static int display_character, set_indicator;
868 static int display_width, display_lines, form_feed;
869 static const int *row_width;
870 static DEFINE_SPINLOCK(progress_lock);
871 static int current_line;
872 static int pending_newline = 0; /* did last write end with unprinted newline? */
873
874 if (!rtas.base)
875 return;
876
877 if (display_width == 0) {
878 display_width = 0x10;
879 if ((root = of_find_node_by_path("/rtas"))) {
880 if ((p = of_get_property(root,
881 "ibm,display-line-length", NULL)))
882 display_width = be32_to_cpu(*p);
883 if ((p = of_get_property(root,
884 "ibm,form-feed", NULL)))
885 form_feed = be32_to_cpu(*p);
886 if ((p = of_get_property(root,
887 "ibm,display-number-of-lines", NULL)))
888 display_lines = be32_to_cpu(*p);
889 row_width = of_get_property(root,
890 "ibm,display-truncation-length", NULL);
891 of_node_put(root);
892 }
893 display_character = rtas_function_token(RTAS_FN_DISPLAY_CHARACTER);
894 set_indicator = rtas_function_token(RTAS_FN_SET_INDICATOR);
895 }
896
897 if (display_character == RTAS_UNKNOWN_SERVICE) {
898 /* use hex display if available */
899 if (set_indicator != RTAS_UNKNOWN_SERVICE)
900 rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
901 return;
902 }
903
904 spin_lock(&progress_lock);
905
906 /*
907 * Last write ended with newline, but we didn't print it since
908 * it would just clear the bottom line of output. Print it now
909 * instead.
910 *
911 * If no newline is pending and form feed is supported, clear the
912 * display with a form feed; otherwise, print a CR to start output
913 * at the beginning of the line.
914 */
915 if (pending_newline) {
916 rtas_call(display_character, 1, 1, NULL, '\r');
917 rtas_call(display_character, 1, 1, NULL, '\n');
918 pending_newline = 0;
919 } else {
920 current_line = 0;
921 if (form_feed)
922 rtas_call(display_character, 1, 1, NULL,
923 (char)form_feed);
924 else
925 rtas_call(display_character, 1, 1, NULL, '\r');
926 }
927
928 if (row_width)
929 width = row_width[current_line];
930 else
931 width = display_width;
932 os = s;
933 while (*os) {
934 if (*os == '\n' || *os == '\r') {
935 /* If newline is the last character, save it
936 * until next call to avoid bumping up the
937 * display output.
938 */
939 if (*os == '\n' && !os[1]) {
940 pending_newline = 1;
941 current_line++;
942 if (current_line > display_lines-1)
943 current_line = display_lines-1;
944 spin_unlock(&progress_lock);
945 return;
946 }
947
948 /* RTAS wants CR-LF, not just LF */
949
950 if (*os == '\n') {
951 rtas_call(display_character, 1, 1, NULL, '\r');
952 rtas_call(display_character, 1, 1, NULL, '\n');
953 } else {
954 /* CR might be used to re-draw a line, so we'll
955 * leave it alone and not add LF.
956 */
957 rtas_call(display_character, 1, 1, NULL, *os);
958 }
959
960 if (row_width)
961 width = row_width[current_line];
962 else
963 width = display_width;
964 } else {
965 width--;
966 rtas_call(display_character, 1, 1, NULL, *os);
967 }
968
969 os++;
970
971 /* if we overwrite the screen length */
972 if (width <= 0)
973 while ((*os != 0) && (*os != '\n') && (*os != '\r'))
974 os++;
975 }
976
977 spin_unlock(&progress_lock);
978}
979EXPORT_SYMBOL_GPL(rtas_progress); /* needed by rtas_flash module */
980
981int rtas_token(const char *service)
982{
983 const struct rtas_function *func;
984 const __be32 *tokp;
985
986 if (rtas.dev == NULL)
987 return RTAS_UNKNOWN_SERVICE;
988
989 func = rtas_name_to_function(service);
990 if (func)
991 return func->token;
992 /*
993 * The caller is looking up a name that is not known to be an
994 * RTAS function. Either it's a function that needs to be
995 * added to the table, or they're misusing rtas_token() to
996 * access non-function properties of the /rtas node. Warn and
997 * fall back to the legacy behavior.
998 */
999 WARN_ONCE(1, "unknown function `%s`, should it be added to rtas_function_table?\n",
1000 service);
1001
1002 tokp = of_get_property(rtas.dev, service, NULL);
1003 return tokp ? be32_to_cpu(*tokp) : RTAS_UNKNOWN_SERVICE;
1004}
1005EXPORT_SYMBOL_GPL(rtas_token);
1006
1007#ifdef CONFIG_RTAS_ERROR_LOGGING
1008
1009static u32 rtas_error_log_max __ro_after_init = RTAS_ERROR_LOG_MAX;
1010
1011/*
1012 * Return the firmware-specified size of the error log buffer
1013 * for all rtas calls that require an error buffer argument.
1014 * This includes 'check-exception' and 'rtas-last-error'.
1015 */
1016int rtas_get_error_log_max(void)
1017{
1018 return rtas_error_log_max;
1019}
1020
1021static void __init init_error_log_max(void)
1022{
1023 static const char propname[] __initconst = "rtas-error-log-max";
1024 u32 max;
1025
1026 if (of_property_read_u32(rtas.dev, propname, &max)) {
1027 pr_warn("%s not found, using default of %u\n",
1028 propname, RTAS_ERROR_LOG_MAX);
1029 max = RTAS_ERROR_LOG_MAX;
1030 }
1031
1032 if (max > RTAS_ERROR_LOG_MAX) {
1033 pr_warn("%s = %u, clamping max error log size to %u\n",
1034 propname, max, RTAS_ERROR_LOG_MAX);
1035 max = RTAS_ERROR_LOG_MAX;
1036 }
1037
1038 rtas_error_log_max = max;
1039}
1040
1041
1042static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
1043
1044/** Return a copy of the detailed error text associated with the
1045 * most recent failed call to rtas. Because the error text
1046 * might go stale if there are any other intervening rtas calls,
1047 * this routine must be called atomically with whatever produced
1048 * the error (i.e. with rtas_lock still held from the previous call).
1049 */
1050static char *__fetch_rtas_last_error(char *altbuf)
1051{
1052 const s32 token = rtas_function_token(RTAS_FN_RTAS_LAST_ERROR);
1053 struct rtas_args err_args, save_args;
1054 u32 bufsz;
1055 char *buf = NULL;
1056
1057 lockdep_assert_held(&rtas_lock);
1058
1059 if (token == -1)
1060 return NULL;
1061
1062 bufsz = rtas_get_error_log_max();
1063
1064 err_args.token = cpu_to_be32(token);
1065 err_args.nargs = cpu_to_be32(2);
1066 err_args.nret = cpu_to_be32(1);
1067 err_args.args[0] = cpu_to_be32(__pa(rtas_err_buf));
1068 err_args.args[1] = cpu_to_be32(bufsz);
1069 err_args.args[2] = 0;
1070
1071 save_args = rtas_args;
1072 rtas_args = err_args;
1073
1074 do_enter_rtas(&rtas_args);
1075
1076 err_args = rtas_args;
1077 rtas_args = save_args;
1078
1079 /* Log the error in the unlikely case that there was one. */
1080 if (unlikely(err_args.args[2] == 0)) {
1081 if (altbuf) {
1082 buf = altbuf;
1083 } else {
1084 buf = rtas_err_buf;
1085 if (slab_is_available())
1086 buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
1087 }
1088 if (buf)
1089 memmove(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
1090 }
1091
1092 return buf;
1093}
1094
1095#define get_errorlog_buffer() kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
1096
1097#else /* CONFIG_RTAS_ERROR_LOGGING */
1098#define __fetch_rtas_last_error(x) NULL
1099#define get_errorlog_buffer() NULL
1100static void __init init_error_log_max(void) {}
1101#endif
1102
1103
1104static void
1105va_rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret,
1106 va_list list)
1107{
1108 int i;
1109
1110 args->token = cpu_to_be32(token);
1111 args->nargs = cpu_to_be32(nargs);
1112 args->nret = cpu_to_be32(nret);
1113 args->rets = &(args->args[nargs]);
1114
1115 for (i = 0; i < nargs; ++i)
1116 args->args[i] = cpu_to_be32(va_arg(list, __u32));
1117
1118 for (i = 0; i < nret; ++i)
1119 args->rets[i] = 0;
1120
1121 do_enter_rtas(args);
1122}
1123
1124/**
1125 * rtas_call_unlocked() - Invoke an RTAS firmware function without synchronization.
1126 * @args: RTAS parameter block to be used for the call, must obey RTAS addressing
1127 * constraints.
1128 * @token: Identifies the function being invoked.
1129 * @nargs: Number of input parameters. Does not include token.
1130 * @nret: Number of output parameters, including the call status.
1131 * @....: List of @nargs input parameters.
1132 *
1133 * Invokes the RTAS function indicated by @token, which the caller
1134 * should obtain via rtas_function_token().
1135 *
1136 * This function is similar to rtas_call(), but must be used with a
1137 * limited set of RTAS calls specifically exempted from the general
1138 * requirement that only one RTAS call may be in progress at any
1139 * time. Examples include stop-self and ibm,nmi-interlock.
1140 */
1141void rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret, ...)
1142{
1143 va_list list;
1144
1145 va_start(list, nret);
1146 va_rtas_call_unlocked(args, token, nargs, nret, list);
1147 va_end(list);
1148}
1149
1150static bool token_is_restricted_errinjct(s32 token)
1151{
1152 return token == rtas_function_token(RTAS_FN_IBM_OPEN_ERRINJCT) ||
1153 token == rtas_function_token(RTAS_FN_IBM_ERRINJCT);
1154}
1155
1156/**
1157 * rtas_call() - Invoke an RTAS firmware function.
1158 * @token: Identifies the function being invoked.
1159 * @nargs: Number of input parameters. Does not include token.
1160 * @nret: Number of output parameters, including the call status.
1161 * @outputs: Array of @nret output words.
1162 * @....: List of @nargs input parameters.
1163 *
1164 * Invokes the RTAS function indicated by @token, which the caller
1165 * should obtain via rtas_function_token().
1166 *
1167 * The @nargs and @nret arguments must match the number of input and
1168 * output parameters specified for the RTAS function.
1169 *
1170 * rtas_call() returns RTAS status codes, not conventional Linux errno
1171 * values. Callers must translate any failure to an appropriate errno
1172 * in syscall context. Most callers of RTAS functions that can return
1173 * -2 or 990x should use rtas_busy_delay() to correctly handle those
1174 * statuses before calling again.
1175 *
1176 * The return value descriptions are adapted from 7.2.8 [RTAS] Return
1177 * Codes of the PAPR and CHRP specifications.
1178 *
1179 * Context: Process context preferably, interrupt context if
1180 * necessary. Acquires an internal spinlock and may perform
1181 * GFP_ATOMIC slab allocation in error path. Unsafe for NMI
1182 * context.
1183 * Return:
1184 * * 0 - RTAS function call succeeded.
1185 * * -1 - RTAS function encountered a hardware or
1186 * platform error, or the token is invalid,
1187 * or the function is restricted by kernel policy.
1188 * * -2 - Specs say "A necessary hardware device was busy,
1189 * and the requested function could not be
1190 * performed. The operation should be retried at
1191 * a later time." This is misleading, at least with
1192 * respect to current RTAS implementations. What it
1193 * usually means in practice is that the function
1194 * could not be completed while meeting RTAS's
1195 * deadline for returning control to the OS (250us
1196 * for PAPR/PowerVM, typically), but the call may be
1197 * immediately reattempted to resume work on it.
1198 * * -3 - Parameter error.
1199 * * -7 - Unexpected state change.
1200 * * 9000...9899 - Vendor-specific success codes.
1201 * * 9900...9905 - Advisory extended delay. Caller should try
1202 * again after ~10^x ms has elapsed, where x is
1203 * the last digit of the status [0-5]. Again going
1204 * beyond the PAPR text, 990x on PowerVM indicates
1205 * contention for RTAS-internal resources. Other
1206 * RTAS call sequences in progress should be
1207 * allowed to complete before reattempting the
1208 * call.
1209 * * -9000 - Multi-level isolation error.
1210 * * -9999...-9004 - Vendor-specific error codes.
1211 * * Additional negative values - Function-specific error.
1212 * * Additional positive values - Function-specific success.
1213 */
1214int rtas_call(int token, int nargs, int nret, int *outputs, ...)
1215{
1216 struct pin_cookie cookie;
1217 va_list list;
1218 int i;
1219 unsigned long flags;
1220 struct rtas_args *args;
1221 char *buff_copy = NULL;
1222 int ret;
1223
1224 if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
1225 return -1;
1226
1227 if (token_is_restricted_errinjct(token)) {
1228 /*
1229 * It would be nicer to not discard the error value
1230 * from security_locked_down(), but callers expect an
1231 * RTAS status, not an errno.
1232 */
1233 if (security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION))
1234 return -1;
1235 }
1236
1237 if ((mfmsr() & (MSR_IR|MSR_DR)) != (MSR_IR|MSR_DR)) {
1238 WARN_ON_ONCE(1);
1239 return -1;
1240 }
1241
1242 raw_spin_lock_irqsave(&rtas_lock, flags);
1243 cookie = lockdep_pin_lock(&rtas_lock);
1244
1245 /* We use the global rtas args buffer */
1246 args = &rtas_args;
1247
1248 va_start(list, outputs);
1249 va_rtas_call_unlocked(args, token, nargs, nret, list);
1250 va_end(list);
1251
1252 /* A -1 return code indicates that the last command couldn't
1253 be completed due to a hardware error. */
1254 if (be32_to_cpu(args->rets[0]) == -1)
1255 buff_copy = __fetch_rtas_last_error(NULL);
1256
1257 if (nret > 1 && outputs != NULL)
1258 for (i = 0; i < nret-1; ++i)
1259 outputs[i] = be32_to_cpu(args->rets[i + 1]);
1260 ret = (nret > 0) ? be32_to_cpu(args->rets[0]) : 0;
1261
1262 lockdep_unpin_lock(&rtas_lock, cookie);
1263 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1264
1265 if (buff_copy) {
1266 log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
1267 if (slab_is_available())
1268 kfree(buff_copy);
1269 }
1270 return ret;
1271}
1272EXPORT_SYMBOL_GPL(rtas_call);
1273
1274/**
1275 * rtas_busy_delay_time() - From an RTAS status value, calculate the
1276 * suggested delay time in milliseconds.
1277 *
1278 * @status: a value returned from rtas_call() or similar APIs which return
1279 * the status of a RTAS function call.
1280 *
1281 * Context: Any context.
1282 *
1283 * Return:
1284 * * 100000 - If @status is 9905.
1285 * * 10000 - If @status is 9904.
1286 * * 1000 - If @status is 9903.
1287 * * 100 - If @status is 9902.
1288 * * 10 - If @status is 9901.
1289 * * 1 - If @status is either 9900 or -2. This is "wrong" for -2, but
1290 * some callers depend on this behavior, and the worst outcome
1291 * is that they will delay for longer than necessary.
1292 * * 0 - If @status is not a busy or extended delay value.
1293 */
1294unsigned int rtas_busy_delay_time(int status)
1295{
1296 int order;
1297 unsigned int ms = 0;
1298
1299 if (status == RTAS_BUSY) {
1300 ms = 1;
1301 } else if (status >= RTAS_EXTENDED_DELAY_MIN &&
1302 status <= RTAS_EXTENDED_DELAY_MAX) {
1303 order = status - RTAS_EXTENDED_DELAY_MIN;
1304 for (ms = 1; order > 0; order--)
1305 ms *= 10;
1306 }
1307
1308 return ms;
1309}
1310
1311/*
1312 * Early boot fallback for rtas_busy_delay().
1313 */
1314static bool __init rtas_busy_delay_early(int status)
1315{
1316 static size_t successive_ext_delays __initdata;
1317 bool retry;
1318
1319 switch (status) {
1320 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1321 /*
1322 * In the unlikely case that we receive an extended
1323 * delay status in early boot, the OS is probably not
1324 * the cause, and there's nothing we can do to clear
1325 * the condition. Best we can do is delay for a bit
1326 * and hope it's transient. Lie to the caller if it
1327 * seems like we're stuck in a retry loop.
1328 */
1329 mdelay(1);
1330 retry = true;
1331 successive_ext_delays += 1;
1332 if (successive_ext_delays > 1000) {
1333 pr_err("too many extended delays, giving up\n");
1334 dump_stack();
1335 retry = false;
1336 successive_ext_delays = 0;
1337 }
1338 break;
1339 case RTAS_BUSY:
1340 retry = true;
1341 successive_ext_delays = 0;
1342 break;
1343 default:
1344 retry = false;
1345 successive_ext_delays = 0;
1346 break;
1347 }
1348
1349 return retry;
1350}
1351
1352/**
1353 * rtas_busy_delay() - helper for RTAS busy and extended delay statuses
1354 *
1355 * @status: a value returned from rtas_call() or similar APIs which return
1356 * the status of a RTAS function call.
1357 *
1358 * Context: Process context. May sleep or schedule.
1359 *
1360 * Return:
1361 * * true - @status is RTAS_BUSY or an extended delay hint. The
1362 * caller may assume that the CPU has been yielded if necessary,
1363 * and that an appropriate delay for @status has elapsed.
1364 * Generally the caller should reattempt the RTAS call which
1365 * yielded @status.
1366 *
1367 * * false - @status is not @RTAS_BUSY nor an extended delay hint. The
1368 * caller is responsible for handling @status.
1369 */
1370bool __ref rtas_busy_delay(int status)
1371{
1372 unsigned int ms;
1373 bool ret;
1374
1375 /*
1376 * Can't do timed sleeps before timekeeping is up.
1377 */
1378 if (system_state < SYSTEM_SCHEDULING)
1379 return rtas_busy_delay_early(status);
1380
1381 switch (status) {
1382 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1383 ret = true;
1384 ms = rtas_busy_delay_time(status);
1385 /*
1386 * The extended delay hint can be as high as 100 seconds.
1387 * Surely any function returning such a status is either
1388 * buggy or isn't going to be significantly slowed by us
1389 * polling at 1HZ. Clamp the sleep time to one second.
1390 */
1391 ms = clamp(ms, 1U, 1000U);
1392 /*
1393 * The delay hint is an order-of-magnitude suggestion, not a
1394 * minimum. It is fine, possibly even advantageous, for us to
1395 * pause for less time than hinted. To make sure pause time will
1396 * not be way longer than requested independent of HZ
1397 * configuration, use fsleep(). See fsleep() for details of
1398 * used sleeping functions.
1399 */
1400 fsleep(ms * 1000);
1401 break;
1402 case RTAS_BUSY:
1403 ret = true;
1404 /*
1405 * We should call again immediately if there's no other
1406 * work to do.
1407 */
1408 cond_resched();
1409 break;
1410 default:
1411 ret = false;
1412 /*
1413 * Not a busy or extended delay status; the caller should
1414 * handle @status itself. Ensure we warn on misuses in
1415 * atomic context regardless.
1416 */
1417 might_sleep();
1418 break;
1419 }
1420
1421 return ret;
1422}
1423EXPORT_SYMBOL_GPL(rtas_busy_delay);
1424
1425int rtas_error_rc(int rtas_rc)
1426{
1427 int rc;
1428
1429 switch (rtas_rc) {
1430 case RTAS_HARDWARE_ERROR: /* Hardware Error */
1431 rc = -EIO;
1432 break;
1433 case RTAS_INVALID_PARAMETER: /* Bad indicator/domain/etc */
1434 rc = -EINVAL;
1435 break;
1436 case -9000: /* Isolation error */
1437 rc = -EFAULT;
1438 break;
1439 case -9001: /* Outstanding TCE/PTE */
1440 rc = -EEXIST;
1441 break;
1442 case -9002: /* No usable slot */
1443 rc = -ENODEV;
1444 break;
1445 default:
1446 pr_err("%s: unexpected error %d\n", __func__, rtas_rc);
1447 rc = -ERANGE;
1448 break;
1449 }
1450 return rc;
1451}
1452EXPORT_SYMBOL_GPL(rtas_error_rc);
1453
1454int rtas_get_power_level(int powerdomain, int *level)
1455{
1456 int token = rtas_function_token(RTAS_FN_GET_POWER_LEVEL);
1457 int rc;
1458
1459 if (token == RTAS_UNKNOWN_SERVICE)
1460 return -ENOENT;
1461
1462 while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
1463 udelay(1);
1464
1465 if (rc < 0)
1466 return rtas_error_rc(rc);
1467 return rc;
1468}
1469EXPORT_SYMBOL_GPL(rtas_get_power_level);
1470
1471int rtas_set_power_level(int powerdomain, int level, int *setlevel)
1472{
1473 int token = rtas_function_token(RTAS_FN_SET_POWER_LEVEL);
1474 int rc;
1475
1476 if (token == RTAS_UNKNOWN_SERVICE)
1477 return -ENOENT;
1478
1479 do {
1480 rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
1481 } while (rtas_busy_delay(rc));
1482
1483 if (rc < 0)
1484 return rtas_error_rc(rc);
1485 return rc;
1486}
1487EXPORT_SYMBOL_GPL(rtas_set_power_level);
1488
1489int rtas_get_sensor(int sensor, int index, int *state)
1490{
1491 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1492 int rc;
1493
1494 if (token == RTAS_UNKNOWN_SERVICE)
1495 return -ENOENT;
1496
1497 do {
1498 rc = rtas_call(token, 2, 2, state, sensor, index);
1499 } while (rtas_busy_delay(rc));
1500
1501 if (rc < 0)
1502 return rtas_error_rc(rc);
1503 return rc;
1504}
1505EXPORT_SYMBOL_GPL(rtas_get_sensor);
1506
1507int rtas_get_sensor_fast(int sensor, int index, int *state)
1508{
1509 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1510 int rc;
1511
1512 if (token == RTAS_UNKNOWN_SERVICE)
1513 return -ENOENT;
1514
1515 rc = rtas_call(token, 2, 2, state, sensor, index);
1516 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1517 rc <= RTAS_EXTENDED_DELAY_MAX));
1518
1519 if (rc < 0)
1520 return rtas_error_rc(rc);
1521 return rc;
1522}
1523
1524bool rtas_indicator_present(int token, int *maxindex)
1525{
1526 int proplen, count, i;
1527 const struct indicator_elem {
1528 __be32 token;
1529 __be32 maxindex;
1530 } *indicators;
1531
1532 indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
1533 if (!indicators)
1534 return false;
1535
1536 count = proplen / sizeof(struct indicator_elem);
1537
1538 for (i = 0; i < count; i++) {
1539 if (__be32_to_cpu(indicators[i].token) != token)
1540 continue;
1541 if (maxindex)
1542 *maxindex = __be32_to_cpu(indicators[i].maxindex);
1543 return true;
1544 }
1545
1546 return false;
1547}
1548
1549int rtas_set_indicator(int indicator, int index, int new_value)
1550{
1551 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1552 int rc;
1553
1554 if (token == RTAS_UNKNOWN_SERVICE)
1555 return -ENOENT;
1556
1557 do {
1558 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1559 } while (rtas_busy_delay(rc));
1560
1561 if (rc < 0)
1562 return rtas_error_rc(rc);
1563 return rc;
1564}
1565EXPORT_SYMBOL_GPL(rtas_set_indicator);
1566
1567/*
1568 * Ignoring RTAS extended delay
1569 */
1570int rtas_set_indicator_fast(int indicator, int index, int new_value)
1571{
1572 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1573 int rc;
1574
1575 if (token == RTAS_UNKNOWN_SERVICE)
1576 return -ENOENT;
1577
1578 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1579
1580 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1581 rc <= RTAS_EXTENDED_DELAY_MAX));
1582
1583 if (rc < 0)
1584 return rtas_error_rc(rc);
1585
1586 return rc;
1587}
1588
1589/**
1590 * rtas_ibm_suspend_me() - Call ibm,suspend-me to suspend the LPAR.
1591 *
1592 * @fw_status: RTAS call status will be placed here if not NULL.
1593 *
1594 * rtas_ibm_suspend_me() should be called only on a CPU which has
1595 * received H_CONTINUE from the H_JOIN hcall. All other active CPUs
1596 * should be waiting to return from H_JOIN.
1597 *
1598 * rtas_ibm_suspend_me() may suspend execution of the OS
1599 * indefinitely. Callers should take appropriate measures upon return, such as
1600 * resetting watchdog facilities.
1601 *
1602 * Callers may choose to retry this call if @fw_status is
1603 * %RTAS_THREADS_ACTIVE.
1604 *
1605 * Return:
1606 * 0 - The partition has resumed from suspend, possibly after
1607 * migration to a different host.
1608 * -ECANCELED - The operation was aborted.
1609 * -EAGAIN - There were other CPUs not in H_JOIN at the time of the call.
1610 * -EBUSY - Some other condition prevented the suspend from succeeding.
1611 * -EIO - Hardware/platform error.
1612 */
1613int rtas_ibm_suspend_me(int *fw_status)
1614{
1615 int token = rtas_function_token(RTAS_FN_IBM_SUSPEND_ME);
1616 int fwrc;
1617 int ret;
1618
1619 fwrc = rtas_call(token, 0, 1, NULL);
1620
1621 switch (fwrc) {
1622 case 0:
1623 ret = 0;
1624 break;
1625 case RTAS_SUSPEND_ABORTED:
1626 ret = -ECANCELED;
1627 break;
1628 case RTAS_THREADS_ACTIVE:
1629 ret = -EAGAIN;
1630 break;
1631 case RTAS_NOT_SUSPENDABLE:
1632 case RTAS_OUTSTANDING_COPROC:
1633 ret = -EBUSY;
1634 break;
1635 case -1:
1636 default:
1637 ret = -EIO;
1638 break;
1639 }
1640
1641 if (fw_status)
1642 *fw_status = fwrc;
1643
1644 return ret;
1645}
1646
1647void __noreturn rtas_restart(char *cmd)
1648{
1649 if (rtas_flash_term_hook)
1650 rtas_flash_term_hook(SYS_RESTART);
1651 pr_emerg("system-reboot returned %d\n",
1652 rtas_call(rtas_function_token(RTAS_FN_SYSTEM_REBOOT), 0, 1, NULL));
1653 for (;;);
1654}
1655
1656void rtas_power_off(void)
1657{
1658 if (rtas_flash_term_hook)
1659 rtas_flash_term_hook(SYS_POWER_OFF);
1660 /* allow power on only with power button press */
1661 pr_emerg("power-off returned %d\n",
1662 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1663 for (;;);
1664}
1665
1666void __noreturn rtas_halt(void)
1667{
1668 if (rtas_flash_term_hook)
1669 rtas_flash_term_hook(SYS_HALT);
1670 /* allow power on only with power button press */
1671 pr_emerg("power-off returned %d\n",
1672 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1673 for (;;);
1674}
1675
1676/* Must be in the RMO region, so we place it here */
1677static char rtas_os_term_buf[2048];
1678static bool ibm_extended_os_term;
1679
1680void rtas_os_term(char *str)
1681{
1682 s32 token = rtas_function_token(RTAS_FN_IBM_OS_TERM);
1683 static struct rtas_args args;
1684 int status;
1685
1686 /*
1687 * Firmware with the ibm,extended-os-term property is guaranteed
1688 * to always return from an ibm,os-term call. Earlier versions without
1689 * this property may terminate the partition which we want to avoid
1690 * since it interferes with panic_timeout.
1691 */
1692
1693 if (token == RTAS_UNKNOWN_SERVICE || !ibm_extended_os_term)
1694 return;
1695
1696 snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
1697
1698 /*
1699 * Keep calling as long as RTAS returns a "try again" status,
1700 * but don't use rtas_busy_delay(), which potentially
1701 * schedules.
1702 */
1703 do {
1704 rtas_call_unlocked(&args, token, 1, 1, NULL, __pa(rtas_os_term_buf));
1705 status = be32_to_cpu(args.rets[0]);
1706 } while (rtas_busy_delay_time(status));
1707
1708 if (status != 0)
1709 pr_emerg("ibm,os-term call failed %d\n", status);
1710}
1711
1712/**
1713 * rtas_activate_firmware() - Activate a new version of firmware.
1714 *
1715 * Context: This function may sleep.
1716 *
1717 * Activate a new version of partition firmware. The OS must call this
1718 * after resuming from a partition hibernation or migration in order
1719 * to maintain the ability to perform live firmware updates. It's not
1720 * catastrophic for this method to be absent or to fail; just log the
1721 * condition in that case.
1722 */
1723void rtas_activate_firmware(void)
1724{
1725 int token = rtas_function_token(RTAS_FN_IBM_ACTIVATE_FIRMWARE);
1726 int fwrc;
1727
1728 if (token == RTAS_UNKNOWN_SERVICE) {
1729 pr_notice("ibm,activate-firmware method unavailable\n");
1730 return;
1731 }
1732
1733 mutex_lock(&rtas_ibm_activate_firmware_lock);
1734
1735 do {
1736 fwrc = rtas_call(token, 0, 1, NULL);
1737 } while (rtas_busy_delay(fwrc));
1738
1739 mutex_unlock(&rtas_ibm_activate_firmware_lock);
1740
1741 if (fwrc)
1742 pr_err("ibm,activate-firmware failed (%i)\n", fwrc);
1743}
1744
1745/**
1746 * get_pseries_errorlog() - Find a specific pseries error log in an RTAS
1747 * extended event log.
1748 * @log: RTAS error/event log
1749 * @section_id: two character section identifier
1750 *
1751 * Return: A pointer to the specified errorlog or NULL if not found.
1752 */
1753noinstr struct pseries_errorlog *get_pseries_errorlog(struct rtas_error_log *log,
1754 uint16_t section_id)
1755{
1756 struct rtas_ext_event_log_v6 *ext_log =
1757 (struct rtas_ext_event_log_v6 *)log->buffer;
1758 struct pseries_errorlog *sect;
1759 unsigned char *p, *log_end;
1760 uint32_t ext_log_length = rtas_error_extended_log_length(log);
1761 uint8_t log_format = rtas_ext_event_log_format(ext_log);
1762 uint32_t company_id = rtas_ext_event_company_id(ext_log);
1763
1764 /* Check that we understand the format */
1765 if (ext_log_length < sizeof(struct rtas_ext_event_log_v6) ||
1766 log_format != RTAS_V6EXT_LOG_FORMAT_EVENT_LOG ||
1767 company_id != RTAS_V6EXT_COMPANY_ID_IBM)
1768 return NULL;
1769
1770 log_end = log->buffer + ext_log_length;
1771 p = ext_log->vendor_log;
1772
1773 while (p < log_end) {
1774 sect = (struct pseries_errorlog *)p;
1775 if (pseries_errorlog_id(sect) == section_id)
1776 return sect;
1777 p += pseries_errorlog_length(sect);
1778 }
1779
1780 return NULL;
1781}
1782
1783/*
1784 * The sys_rtas syscall, as originally designed, allows root to pass
1785 * arbitrary physical addresses to RTAS calls. A number of RTAS calls
1786 * can be abused to write to arbitrary memory and do other things that
1787 * are potentially harmful to system integrity, and thus should only
1788 * be used inside the kernel and not exposed to userspace.
1789 *
1790 * All known legitimate users of the sys_rtas syscall will only ever
1791 * pass addresses that fall within the RMO buffer, and use a known
1792 * subset of RTAS calls.
1793 *
1794 * Accordingly, we filter RTAS requests to check that the call is
1795 * permitted, and that provided pointers fall within the RMO buffer.
1796 * If a function is allowed to be invoked via the syscall, then its
1797 * entry in the rtas_functions table points to a rtas_filter that
1798 * describes its constraints, with the indexes of the parameters which
1799 * are expected to contain addresses and sizes of buffers allocated
1800 * inside the RMO buffer.
1801 */
1802
1803static bool in_rmo_buf(u32 base, u32 end)
1804{
1805 return base >= rtas_rmo_buf &&
1806 base < (rtas_rmo_buf + RTAS_USER_REGION_SIZE) &&
1807 base <= end &&
1808 end >= rtas_rmo_buf &&
1809 end < (rtas_rmo_buf + RTAS_USER_REGION_SIZE);
1810}
1811
1812static bool block_rtas_call(const struct rtas_function *func, int nargs,
1813 struct rtas_args *args)
1814{
1815 const struct rtas_filter *f;
1816 const bool is_platform_dump =
1817 func == &rtas_function_table[RTAS_FNIDX__IBM_PLATFORM_DUMP];
1818 const bool is_config_conn =
1819 func == &rtas_function_table[RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR];
1820 u32 base, size, end;
1821
1822 /*
1823 * Only functions with filters attached are allowed.
1824 */
1825 f = func->filter;
1826 if (!f)
1827 goto err;
1828 /*
1829 * And some functions aren't allowed on LE.
1830 */
1831 if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) && func->banned_for_syscall_on_le)
1832 goto err;
1833
1834 if (f->buf_idx1 != -1) {
1835 base = be32_to_cpu(args->args[f->buf_idx1]);
1836 if (f->size_idx1 != -1)
1837 size = be32_to_cpu(args->args[f->size_idx1]);
1838 else if (f->fixed_size)
1839 size = f->fixed_size;
1840 else
1841 size = 1;
1842
1843 end = base + size - 1;
1844
1845 /*
1846 * Special case for ibm,platform-dump - NULL buffer
1847 * address is used to indicate end of dump processing
1848 */
1849 if (is_platform_dump && base == 0)
1850 return false;
1851
1852 if (!in_rmo_buf(base, end))
1853 goto err;
1854 }
1855
1856 if (f->buf_idx2 != -1) {
1857 base = be32_to_cpu(args->args[f->buf_idx2]);
1858 if (f->size_idx2 != -1)
1859 size = be32_to_cpu(args->args[f->size_idx2]);
1860 else if (f->fixed_size)
1861 size = f->fixed_size;
1862 else
1863 size = 1;
1864 end = base + size - 1;
1865
1866 /*
1867 * Special case for ibm,configure-connector where the
1868 * address can be 0
1869 */
1870 if (is_config_conn && base == 0)
1871 return false;
1872
1873 if (!in_rmo_buf(base, end))
1874 goto err;
1875 }
1876
1877 return false;
1878err:
1879 pr_err_ratelimited("sys_rtas: RTAS call blocked - exploit attempt?\n");
1880 pr_err_ratelimited("sys_rtas: %s nargs=%d (called by %s)\n",
1881 func->name, nargs, current->comm);
1882 return true;
1883}
1884
1885/* We assume to be passed big endian arguments */
1886SYSCALL_DEFINE1(rtas, struct rtas_args __user *, uargs)
1887{
1888 const struct rtas_function *func;
1889 struct pin_cookie cookie;
1890 struct rtas_args args;
1891 unsigned long flags;
1892 char *buff_copy, *errbuf = NULL;
1893 int nargs, nret, token;
1894
1895 if (!capable(CAP_SYS_ADMIN))
1896 return -EPERM;
1897
1898 if (!rtas.entry)
1899 return -EINVAL;
1900
1901 if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
1902 return -EFAULT;
1903
1904 nargs = be32_to_cpu(args.nargs);
1905 nret = be32_to_cpu(args.nret);
1906 token = be32_to_cpu(args.token);
1907
1908 if (nargs >= ARRAY_SIZE(args.args)
1909 || nret > ARRAY_SIZE(args.args)
1910 || nargs + nret > ARRAY_SIZE(args.args))
1911 return -EINVAL;
1912
1913 nargs = array_index_nospec(nargs, ARRAY_SIZE(args.args));
1914 nret = array_index_nospec(nret, ARRAY_SIZE(args.args) - nargs);
1915
1916 /* Copy in args. */
1917 if (copy_from_user(args.args, uargs->args,
1918 nargs * sizeof(rtas_arg_t)) != 0)
1919 return -EFAULT;
1920
1921 /*
1922 * If this token doesn't correspond to a function the kernel
1923 * understands, you're not allowed to call it.
1924 */
1925 func = rtas_token_to_function_untrusted(token);
1926 if (!func)
1927 return -EINVAL;
1928
1929 args.rets = &args.args[nargs];
1930 memset(args.rets, 0, nret * sizeof(rtas_arg_t));
1931
1932 if (block_rtas_call(func, nargs, &args))
1933 return -EINVAL;
1934
1935 if (token_is_restricted_errinjct(token)) {
1936 int err;
1937
1938 err = security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION);
1939 if (err)
1940 return err;
1941 }
1942
1943 /* Need to handle ibm,suspend_me call specially */
1944 if (token == rtas_function_token(RTAS_FN_IBM_SUSPEND_ME)) {
1945
1946 /*
1947 * rtas_ibm_suspend_me assumes the streamid handle is in cpu
1948 * endian, or at least the hcall within it requires it.
1949 */
1950 int rc = 0;
1951 u64 handle = ((u64)be32_to_cpu(args.args[0]) << 32)
1952 | be32_to_cpu(args.args[1]);
1953 rc = rtas_syscall_dispatch_ibm_suspend_me(handle);
1954 if (rc == -EAGAIN)
1955 args.rets[0] = cpu_to_be32(RTAS_NOT_SUSPENDABLE);
1956 else if (rc == -EIO)
1957 args.rets[0] = cpu_to_be32(-1);
1958 else if (rc)
1959 return rc;
1960 goto copy_return;
1961 }
1962
1963 buff_copy = get_errorlog_buffer();
1964
1965 /*
1966 * If this function has a mutex assigned to it, we must
1967 * acquire it to avoid interleaving with any kernel-based uses
1968 * of the same function. Kernel-based sequences acquire the
1969 * appropriate mutex explicitly.
1970 */
1971 if (func->lock)
1972 mutex_lock(func->lock);
1973
1974 raw_spin_lock_irqsave(&rtas_lock, flags);
1975 cookie = lockdep_pin_lock(&rtas_lock);
1976
1977 rtas_args = args;
1978 do_enter_rtas(&rtas_args);
1979 args = rtas_args;
1980
1981 /* A -1 return code indicates that the last command couldn't
1982 be completed due to a hardware error. */
1983 if (be32_to_cpu(args.rets[0]) == -1)
1984 errbuf = __fetch_rtas_last_error(buff_copy);
1985
1986 lockdep_unpin_lock(&rtas_lock, cookie);
1987 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1988
1989 if (func->lock)
1990 mutex_unlock(func->lock);
1991
1992 if (buff_copy) {
1993 if (errbuf)
1994 log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
1995 kfree(buff_copy);
1996 }
1997
1998 copy_return:
1999 /* Copy out args. */
2000 if (copy_to_user(uargs->args + nargs,
2001 args.args + nargs,
2002 nret * sizeof(rtas_arg_t)) != 0)
2003 return -EFAULT;
2004
2005 return 0;
2006}
2007
2008static void __init rtas_function_table_init(void)
2009{
2010 struct property *prop;
2011
2012 for (size_t i = 0; i < ARRAY_SIZE(rtas_function_table); ++i) {
2013 struct rtas_function *curr = &rtas_function_table[i];
2014 struct rtas_function *prior;
2015 int cmp;
2016
2017 curr->token = RTAS_UNKNOWN_SERVICE;
2018
2019 if (i == 0)
2020 continue;
2021 /*
2022 * Ensure table is sorted correctly for binary search
2023 * on function names.
2024 */
2025 prior = &rtas_function_table[i - 1];
2026
2027 cmp = strcmp(prior->name, curr->name);
2028 if (cmp < 0)
2029 continue;
2030
2031 if (cmp == 0) {
2032 pr_err("'%s' has duplicate function table entries\n",
2033 curr->name);
2034 } else {
2035 pr_err("function table unsorted: '%s' wrongly precedes '%s'\n",
2036 prior->name, curr->name);
2037 }
2038 }
2039
2040 for_each_property_of_node(rtas.dev, prop) {
2041 struct rtas_function *func;
2042
2043 if (prop->length != sizeof(u32))
2044 continue;
2045
2046 func = __rtas_name_to_function(prop->name);
2047 if (!func)
2048 continue;
2049
2050 func->token = be32_to_cpup((__be32 *)prop->value);
2051
2052 pr_debug("function %s has token %u\n", func->name, func->token);
2053 }
2054}
2055
2056/*
2057 * Call early during boot, before mem init, to retrieve the RTAS
2058 * information from the device-tree and allocate the RMO buffer for userland
2059 * accesses.
2060 */
2061void __init rtas_initialize(void)
2062{
2063 unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
2064 u32 base, size, entry;
2065 int no_base, no_size, no_entry;
2066
2067 /* Get RTAS dev node and fill up our "rtas" structure with infos
2068 * about it.
2069 */
2070 rtas.dev = of_find_node_by_name(NULL, "rtas");
2071 if (!rtas.dev)
2072 return;
2073
2074 no_base = of_property_read_u32(rtas.dev, "linux,rtas-base", &base);
2075 no_size = of_property_read_u32(rtas.dev, "rtas-size", &size);
2076 if (no_base || no_size) {
2077 of_node_put(rtas.dev);
2078 rtas.dev = NULL;
2079 return;
2080 }
2081
2082 rtas.base = base;
2083 rtas.size = size;
2084 no_entry = of_property_read_u32(rtas.dev, "linux,rtas-entry", &entry);
2085 rtas.entry = no_entry ? rtas.base : entry;
2086
2087 init_error_log_max();
2088
2089 /* Must be called before any function token lookups */
2090 rtas_function_table_init();
2091
2092 /*
2093 * Discover this now to avoid a device tree lookup in the
2094 * panic path.
2095 */
2096 ibm_extended_os_term = of_property_read_bool(rtas.dev, "ibm,extended-os-term");
2097
2098 /* If RTAS was found, allocate the RMO buffer for it and look for
2099 * the stop-self token if any
2100 */
2101#ifdef CONFIG_PPC64
2102 if (firmware_has_feature(FW_FEATURE_LPAR))
2103 rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
2104#endif
2105 rtas_rmo_buf = memblock_phys_alloc_range(RTAS_USER_REGION_SIZE, PAGE_SIZE,
2106 0, rtas_region);
2107 if (!rtas_rmo_buf)
2108 panic("ERROR: RTAS: Failed to allocate %lx bytes below %pa\n",
2109 PAGE_SIZE, &rtas_region);
2110
2111 rtas_work_area_reserve_arena(rtas_region);
2112}
2113
2114int __init early_init_dt_scan_rtas(unsigned long node,
2115 const char *uname, int depth, void *data)
2116{
2117 const u32 *basep, *entryp, *sizep;
2118
2119 if (depth != 1 || strcmp(uname, "rtas") != 0)
2120 return 0;
2121
2122 basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
2123 entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
2124 sizep = of_get_flat_dt_prop(node, "rtas-size", NULL);
2125
2126#ifdef CONFIG_PPC64
2127 /* need this feature to decide the crashkernel offset */
2128 if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL))
2129 powerpc_firmware_features |= FW_FEATURE_LPAR;
2130#endif
2131
2132 if (basep && entryp && sizep) {
2133 rtas.base = *basep;
2134 rtas.entry = *entryp;
2135 rtas.size = *sizep;
2136 }
2137
2138#ifdef CONFIG_UDBG_RTAS_CONSOLE
2139 basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
2140 if (basep)
2141 rtas_putchar_token = *basep;
2142
2143 basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
2144 if (basep)
2145 rtas_getchar_token = *basep;
2146
2147 if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
2148 rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
2149 udbg_init_rtas_console();
2150
2151#endif
2152
2153 /* break now */
2154 return 1;
2155}
2156
2157static DEFINE_RAW_SPINLOCK(timebase_lock);
2158static u64 timebase = 0;
2159
2160void rtas_give_timebase(void)
2161{
2162 unsigned long flags;
2163
2164 raw_spin_lock_irqsave(&timebase_lock, flags);
2165 hard_irq_disable();
2166 rtas_call(rtas_function_token(RTAS_FN_FREEZE_TIME_BASE), 0, 1, NULL);
2167 timebase = get_tb();
2168 raw_spin_unlock(&timebase_lock);
2169
2170 while (timebase)
2171 barrier();
2172 rtas_call(rtas_function_token(RTAS_FN_THAW_TIME_BASE), 0, 1, NULL);
2173 local_irq_restore(flags);
2174}
2175
2176void rtas_take_timebase(void)
2177{
2178 while (!timebase)
2179 barrier();
2180 raw_spin_lock(&timebase_lock);
2181 set_tb(timebase >> 32, timebase & 0xffffffff);
2182 timebase = 0;
2183 raw_spin_unlock(&timebase_lock);
2184}