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/of.h>
23#include <linux/of_fdt.h>
24#include <linux/reboot.h>
25#include <linux/sched.h>
26#include <linux/security.h>
27#include <linux/slab.h>
28#include <linux/spinlock.h>
29#include <linux/stdarg.h>
30#include <linux/syscalls.h>
31#include <linux/types.h>
32#include <linux/uaccess.h>
33#include <linux/xarray.h>
34
35#include <asm/delay.h>
36#include <asm/firmware.h>
37#include <asm/interrupt.h>
38#include <asm/machdep.h>
39#include <asm/mmu.h>
40#include <asm/page.h>
41#include <asm/rtas-work-area.h>
42#include <asm/rtas.h>
43#include <asm/time.h>
44#include <asm/trace.h>
45#include <asm/udbg.h>
46
47struct rtas_filter {
48 /* Indexes into the args buffer, -1 if not used */
49 const int buf_idx1;
50 const int size_idx1;
51 const int buf_idx2;
52 const int size_idx2;
53 /*
54 * Assumed buffer size per the spec if the function does not
55 * have a size parameter, e.g. ibm,errinjct. 0 if unused.
56 */
57 const int fixed_size;
58};
59
60/**
61 * struct rtas_function - Descriptor for RTAS functions.
62 *
63 * @token: Value of @name if it exists under the /rtas node.
64 * @name: Function name.
65 * @filter: If non-NULL, invoking this function via the rtas syscall is
66 * generally allowed, and @filter describes constraints on the
67 * arguments. See also @banned_for_syscall_on_le.
68 * @banned_for_syscall_on_le: Set when call via sys_rtas is generally allowed
69 * but specifically restricted on ppc64le. Such
70 * functions are believed to have no users on
71 * ppc64le, and we want to keep it that way. It does
72 * not make sense for this to be set when @filter
73 * is NULL.
74 * @lock: Pointer to an optional dedicated per-function mutex. This
75 * should be set for functions that require multiple calls in
76 * sequence to complete a single operation, and such sequences
77 * will disrupt each other if allowed to interleave. Users of
78 * this function are required to hold the associated lock for
79 * the duration of the call sequence. Add an explanatory
80 * comment to the function table entry if setting this member.
81 */
82struct rtas_function {
83 s32 token;
84 const bool banned_for_syscall_on_le:1;
85 const char * const name;
86 const struct rtas_filter *filter;
87 struct mutex *lock;
88};
89
90/*
91 * Per-function locks for sequence-based RTAS functions.
92 */
93static DEFINE_MUTEX(rtas_ibm_activate_firmware_lock);
94static DEFINE_MUTEX(rtas_ibm_get_dynamic_sensor_state_lock);
95static DEFINE_MUTEX(rtas_ibm_get_indices_lock);
96static DEFINE_MUTEX(rtas_ibm_lpar_perftools_lock);
97static DEFINE_MUTEX(rtas_ibm_physical_attestation_lock);
98static DEFINE_MUTEX(rtas_ibm_set_dynamic_indicator_lock);
99DEFINE_MUTEX(rtas_ibm_get_vpd_lock);
100
101static struct rtas_function rtas_function_table[] __ro_after_init = {
102 [RTAS_FNIDX__CHECK_EXCEPTION] = {
103 .name = "check-exception",
104 },
105 [RTAS_FNIDX__DISPLAY_CHARACTER] = {
106 .name = "display-character",
107 .filter = &(const struct rtas_filter) {
108 .buf_idx1 = -1, .size_idx1 = -1,
109 .buf_idx2 = -1, .size_idx2 = -1,
110 },
111 },
112 [RTAS_FNIDX__EVENT_SCAN] = {
113 .name = "event-scan",
114 },
115 [RTAS_FNIDX__FREEZE_TIME_BASE] = {
116 .name = "freeze-time-base",
117 },
118 [RTAS_FNIDX__GET_POWER_LEVEL] = {
119 .name = "get-power-level",
120 .filter = &(const struct rtas_filter) {
121 .buf_idx1 = -1, .size_idx1 = -1,
122 .buf_idx2 = -1, .size_idx2 = -1,
123 },
124 },
125 [RTAS_FNIDX__GET_SENSOR_STATE] = {
126 .name = "get-sensor-state",
127 .filter = &(const struct rtas_filter) {
128 .buf_idx1 = -1, .size_idx1 = -1,
129 .buf_idx2 = -1, .size_idx2 = -1,
130 },
131 },
132 [RTAS_FNIDX__GET_TERM_CHAR] = {
133 .name = "get-term-char",
134 },
135 [RTAS_FNIDX__GET_TIME_OF_DAY] = {
136 .name = "get-time-of-day",
137 .filter = &(const struct rtas_filter) {
138 .buf_idx1 = -1, .size_idx1 = -1,
139 .buf_idx2 = -1, .size_idx2 = -1,
140 },
141 },
142 [RTAS_FNIDX__IBM_ACTIVATE_FIRMWARE] = {
143 .name = "ibm,activate-firmware",
144 .filter = &(const struct rtas_filter) {
145 .buf_idx1 = -1, .size_idx1 = -1,
146 .buf_idx2 = -1, .size_idx2 = -1,
147 },
148 /*
149 * PAPR+ as of v2.13 doesn't explicitly impose any
150 * restriction, but this typically requires multiple
151 * calls before success, and there's no reason to
152 * allow sequences to interleave.
153 */
154 .lock = &rtas_ibm_activate_firmware_lock,
155 },
156 [RTAS_FNIDX__IBM_CBE_START_PTCAL] = {
157 .name = "ibm,cbe-start-ptcal",
158 },
159 [RTAS_FNIDX__IBM_CBE_STOP_PTCAL] = {
160 .name = "ibm,cbe-stop-ptcal",
161 },
162 [RTAS_FNIDX__IBM_CHANGE_MSI] = {
163 .name = "ibm,change-msi",
164 },
165 [RTAS_FNIDX__IBM_CLOSE_ERRINJCT] = {
166 .name = "ibm,close-errinjct",
167 .filter = &(const struct rtas_filter) {
168 .buf_idx1 = -1, .size_idx1 = -1,
169 .buf_idx2 = -1, .size_idx2 = -1,
170 },
171 },
172 [RTAS_FNIDX__IBM_CONFIGURE_BRIDGE] = {
173 .name = "ibm,configure-bridge",
174 },
175 [RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR] = {
176 .name = "ibm,configure-connector",
177 .filter = &(const struct rtas_filter) {
178 .buf_idx1 = 0, .size_idx1 = -1,
179 .buf_idx2 = 1, .size_idx2 = -1,
180 .fixed_size = 4096,
181 },
182 },
183 [RTAS_FNIDX__IBM_CONFIGURE_KERNEL_DUMP] = {
184 .name = "ibm,configure-kernel-dump",
185 },
186 [RTAS_FNIDX__IBM_CONFIGURE_PE] = {
187 .name = "ibm,configure-pe",
188 },
189 [RTAS_FNIDX__IBM_CREATE_PE_DMA_WINDOW] = {
190 .name = "ibm,create-pe-dma-window",
191 },
192 [RTAS_FNIDX__IBM_DISPLAY_MESSAGE] = {
193 .name = "ibm,display-message",
194 .filter = &(const struct rtas_filter) {
195 .buf_idx1 = 0, .size_idx1 = -1,
196 .buf_idx2 = -1, .size_idx2 = -1,
197 },
198 },
199 [RTAS_FNIDX__IBM_ERRINJCT] = {
200 .name = "ibm,errinjct",
201 .filter = &(const struct rtas_filter) {
202 .buf_idx1 = 2, .size_idx1 = -1,
203 .buf_idx2 = -1, .size_idx2 = -1,
204 .fixed_size = 1024,
205 },
206 },
207 [RTAS_FNIDX__IBM_EXTI2C] = {
208 .name = "ibm,exti2c",
209 },
210 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO] = {
211 .name = "ibm,get-config-addr-info",
212 },
213 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO2] = {
214 .name = "ibm,get-config-addr-info2",
215 .filter = &(const struct rtas_filter) {
216 .buf_idx1 = -1, .size_idx1 = -1,
217 .buf_idx2 = -1, .size_idx2 = -1,
218 },
219 },
220 [RTAS_FNIDX__IBM_GET_DYNAMIC_SENSOR_STATE] = {
221 .name = "ibm,get-dynamic-sensor-state",
222 .filter = &(const struct rtas_filter) {
223 .buf_idx1 = 1, .size_idx1 = -1,
224 .buf_idx2 = -1, .size_idx2 = -1,
225 },
226 /*
227 * PAPR+ v2.13 R1–7.3.19–3 is explicit that the OS
228 * must not call ibm,get-dynamic-sensor-state with
229 * different inputs until a non-retry status has been
230 * returned.
231 */
232 .lock = &rtas_ibm_get_dynamic_sensor_state_lock,
233 },
234 [RTAS_FNIDX__IBM_GET_INDICES] = {
235 .name = "ibm,get-indices",
236 .filter = &(const struct rtas_filter) {
237 .buf_idx1 = 2, .size_idx1 = 3,
238 .buf_idx2 = -1, .size_idx2 = -1,
239 },
240 /*
241 * PAPR+ v2.13 R1–7.3.17–2 says that the OS must not
242 * interleave ibm,get-indices call sequences with
243 * different inputs.
244 */
245 .lock = &rtas_ibm_get_indices_lock,
246 },
247 [RTAS_FNIDX__IBM_GET_RIO_TOPOLOGY] = {
248 .name = "ibm,get-rio-topology",
249 },
250 [RTAS_FNIDX__IBM_GET_SYSTEM_PARAMETER] = {
251 .name = "ibm,get-system-parameter",
252 .filter = &(const struct rtas_filter) {
253 .buf_idx1 = 1, .size_idx1 = 2,
254 .buf_idx2 = -1, .size_idx2 = -1,
255 },
256 },
257 [RTAS_FNIDX__IBM_GET_VPD] = {
258 .name = "ibm,get-vpd",
259 .filter = &(const struct rtas_filter) {
260 .buf_idx1 = 0, .size_idx1 = -1,
261 .buf_idx2 = 1, .size_idx2 = 2,
262 },
263 /*
264 * PAPR+ v2.13 R1–7.3.20–4 indicates that sequences
265 * should not be allowed to interleave.
266 */
267 .lock = &rtas_ibm_get_vpd_lock,
268 },
269 [RTAS_FNIDX__IBM_GET_XIVE] = {
270 .name = "ibm,get-xive",
271 },
272 [RTAS_FNIDX__IBM_INT_OFF] = {
273 .name = "ibm,int-off",
274 },
275 [RTAS_FNIDX__IBM_INT_ON] = {
276 .name = "ibm,int-on",
277 },
278 [RTAS_FNIDX__IBM_IO_QUIESCE_ACK] = {
279 .name = "ibm,io-quiesce-ack",
280 },
281 [RTAS_FNIDX__IBM_LPAR_PERFTOOLS] = {
282 .name = "ibm,lpar-perftools",
283 .filter = &(const struct rtas_filter) {
284 .buf_idx1 = 2, .size_idx1 = 3,
285 .buf_idx2 = -1, .size_idx2 = -1,
286 },
287 /*
288 * PAPR+ v2.13 R1–7.3.26–6 says the OS should allow
289 * only one call sequence in progress at a time.
290 */
291 .lock = &rtas_ibm_lpar_perftools_lock,
292 },
293 [RTAS_FNIDX__IBM_MANAGE_FLASH_IMAGE] = {
294 .name = "ibm,manage-flash-image",
295 },
296 [RTAS_FNIDX__IBM_MANAGE_STORAGE_PRESERVATION] = {
297 .name = "ibm,manage-storage-preservation",
298 },
299 [RTAS_FNIDX__IBM_NMI_INTERLOCK] = {
300 .name = "ibm,nmi-interlock",
301 },
302 [RTAS_FNIDX__IBM_NMI_REGISTER] = {
303 .name = "ibm,nmi-register",
304 },
305 [RTAS_FNIDX__IBM_OPEN_ERRINJCT] = {
306 .name = "ibm,open-errinjct",
307 .filter = &(const struct rtas_filter) {
308 .buf_idx1 = -1, .size_idx1 = -1,
309 .buf_idx2 = -1, .size_idx2 = -1,
310 },
311 },
312 [RTAS_FNIDX__IBM_OPEN_SRIOV_ALLOW_UNFREEZE] = {
313 .name = "ibm,open-sriov-allow-unfreeze",
314 },
315 [RTAS_FNIDX__IBM_OPEN_SRIOV_MAP_PE_NUMBER] = {
316 .name = "ibm,open-sriov-map-pe-number",
317 },
318 [RTAS_FNIDX__IBM_OS_TERM] = {
319 .name = "ibm,os-term",
320 },
321 [RTAS_FNIDX__IBM_PARTNER_CONTROL] = {
322 .name = "ibm,partner-control",
323 },
324 [RTAS_FNIDX__IBM_PHYSICAL_ATTESTATION] = {
325 .name = "ibm,physical-attestation",
326 .filter = &(const struct rtas_filter) {
327 .buf_idx1 = 0, .size_idx1 = 1,
328 .buf_idx2 = -1, .size_idx2 = -1,
329 },
330 /*
331 * This follows a sequence-based pattern similar to
332 * ibm,get-vpd et al. Since PAPR+ restricts
333 * interleaving call sequences for other functions of
334 * this style, assume the restriction applies here,
335 * even though it's not explicit in the spec.
336 */
337 .lock = &rtas_ibm_physical_attestation_lock,
338 },
339 [RTAS_FNIDX__IBM_PLATFORM_DUMP] = {
340 .name = "ibm,platform-dump",
341 .filter = &(const struct rtas_filter) {
342 .buf_idx1 = 4, .size_idx1 = 5,
343 .buf_idx2 = -1, .size_idx2 = -1,
344 },
345 /*
346 * PAPR+ v2.13 7.3.3.4.1 indicates that concurrent
347 * sequences of ibm,platform-dump are allowed if they
348 * are operating on different dump tags. So leave the
349 * lock pointer unset for now. This may need
350 * reconsideration if kernel-internal users appear.
351 */
352 },
353 [RTAS_FNIDX__IBM_POWER_OFF_UPS] = {
354 .name = "ibm,power-off-ups",
355 },
356 [RTAS_FNIDX__IBM_QUERY_INTERRUPT_SOURCE_NUMBER] = {
357 .name = "ibm,query-interrupt-source-number",
358 },
359 [RTAS_FNIDX__IBM_QUERY_PE_DMA_WINDOW] = {
360 .name = "ibm,query-pe-dma-window",
361 },
362 [RTAS_FNIDX__IBM_READ_PCI_CONFIG] = {
363 .name = "ibm,read-pci-config",
364 },
365 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE] = {
366 .name = "ibm,read-slot-reset-state",
367 .filter = &(const struct rtas_filter) {
368 .buf_idx1 = -1, .size_idx1 = -1,
369 .buf_idx2 = -1, .size_idx2 = -1,
370 },
371 },
372 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE2] = {
373 .name = "ibm,read-slot-reset-state2",
374 },
375 [RTAS_FNIDX__IBM_REMOVE_PE_DMA_WINDOW] = {
376 .name = "ibm,remove-pe-dma-window",
377 },
378 [RTAS_FNIDX__IBM_RESET_PE_DMA_WINDOW] = {
379 /*
380 * Note: PAPR+ v2.13 7.3.31.4.1 spells this as
381 * "ibm,reset-pe-dma-windows" (plural), but RTAS
382 * implementations use the singular form in practice.
383 */
384 .name = "ibm,reset-pe-dma-window",
385 },
386 [RTAS_FNIDX__IBM_SCAN_LOG_DUMP] = {
387 .name = "ibm,scan-log-dump",
388 .filter = &(const struct rtas_filter) {
389 .buf_idx1 = 0, .size_idx1 = 1,
390 .buf_idx2 = -1, .size_idx2 = -1,
391 },
392 },
393 [RTAS_FNIDX__IBM_SET_DYNAMIC_INDICATOR] = {
394 .name = "ibm,set-dynamic-indicator",
395 .filter = &(const struct rtas_filter) {
396 .buf_idx1 = 2, .size_idx1 = -1,
397 .buf_idx2 = -1, .size_idx2 = -1,
398 },
399 /*
400 * PAPR+ v2.13 R1–7.3.18–3 says the OS must not call
401 * this function with different inputs until a
402 * non-retry status has been returned.
403 */
404 .lock = &rtas_ibm_set_dynamic_indicator_lock,
405 },
406 [RTAS_FNIDX__IBM_SET_EEH_OPTION] = {
407 .name = "ibm,set-eeh-option",
408 .filter = &(const struct rtas_filter) {
409 .buf_idx1 = -1, .size_idx1 = -1,
410 .buf_idx2 = -1, .size_idx2 = -1,
411 },
412 },
413 [RTAS_FNIDX__IBM_SET_SLOT_RESET] = {
414 .name = "ibm,set-slot-reset",
415 },
416 [RTAS_FNIDX__IBM_SET_SYSTEM_PARAMETER] = {
417 .name = "ibm,set-system-parameter",
418 .filter = &(const struct rtas_filter) {
419 .buf_idx1 = 1, .size_idx1 = -1,
420 .buf_idx2 = -1, .size_idx2 = -1,
421 },
422 },
423 [RTAS_FNIDX__IBM_SET_XIVE] = {
424 .name = "ibm,set-xive",
425 },
426 [RTAS_FNIDX__IBM_SLOT_ERROR_DETAIL] = {
427 .name = "ibm,slot-error-detail",
428 },
429 [RTAS_FNIDX__IBM_SUSPEND_ME] = {
430 .name = "ibm,suspend-me",
431 .banned_for_syscall_on_le = true,
432 .filter = &(const struct rtas_filter) {
433 .buf_idx1 = -1, .size_idx1 = -1,
434 .buf_idx2 = -1, .size_idx2 = -1,
435 },
436 },
437 [RTAS_FNIDX__IBM_TUNE_DMA_PARMS] = {
438 .name = "ibm,tune-dma-parms",
439 },
440 [RTAS_FNIDX__IBM_UPDATE_FLASH_64_AND_REBOOT] = {
441 .name = "ibm,update-flash-64-and-reboot",
442 },
443 [RTAS_FNIDX__IBM_UPDATE_NODES] = {
444 .name = "ibm,update-nodes",
445 .banned_for_syscall_on_le = true,
446 .filter = &(const struct rtas_filter) {
447 .buf_idx1 = 0, .size_idx1 = -1,
448 .buf_idx2 = -1, .size_idx2 = -1,
449 .fixed_size = 4096,
450 },
451 },
452 [RTAS_FNIDX__IBM_UPDATE_PROPERTIES] = {
453 .name = "ibm,update-properties",
454 .banned_for_syscall_on_le = true,
455 .filter = &(const struct rtas_filter) {
456 .buf_idx1 = 0, .size_idx1 = -1,
457 .buf_idx2 = -1, .size_idx2 = -1,
458 .fixed_size = 4096,
459 },
460 },
461 [RTAS_FNIDX__IBM_VALIDATE_FLASH_IMAGE] = {
462 .name = "ibm,validate-flash-image",
463 },
464 [RTAS_FNIDX__IBM_WRITE_PCI_CONFIG] = {
465 .name = "ibm,write-pci-config",
466 },
467 [RTAS_FNIDX__NVRAM_FETCH] = {
468 .name = "nvram-fetch",
469 },
470 [RTAS_FNIDX__NVRAM_STORE] = {
471 .name = "nvram-store",
472 },
473 [RTAS_FNIDX__POWER_OFF] = {
474 .name = "power-off",
475 },
476 [RTAS_FNIDX__PUT_TERM_CHAR] = {
477 .name = "put-term-char",
478 },
479 [RTAS_FNIDX__QUERY_CPU_STOPPED_STATE] = {
480 .name = "query-cpu-stopped-state",
481 },
482 [RTAS_FNIDX__READ_PCI_CONFIG] = {
483 .name = "read-pci-config",
484 },
485 [RTAS_FNIDX__RTAS_LAST_ERROR] = {
486 .name = "rtas-last-error",
487 },
488 [RTAS_FNIDX__SET_INDICATOR] = {
489 .name = "set-indicator",
490 .filter = &(const struct rtas_filter) {
491 .buf_idx1 = -1, .size_idx1 = -1,
492 .buf_idx2 = -1, .size_idx2 = -1,
493 },
494 },
495 [RTAS_FNIDX__SET_POWER_LEVEL] = {
496 .name = "set-power-level",
497 .filter = &(const struct rtas_filter) {
498 .buf_idx1 = -1, .size_idx1 = -1,
499 .buf_idx2 = -1, .size_idx2 = -1,
500 },
501 },
502 [RTAS_FNIDX__SET_TIME_FOR_POWER_ON] = {
503 .name = "set-time-for-power-on",
504 .filter = &(const struct rtas_filter) {
505 .buf_idx1 = -1, .size_idx1 = -1,
506 .buf_idx2 = -1, .size_idx2 = -1,
507 },
508 },
509 [RTAS_FNIDX__SET_TIME_OF_DAY] = {
510 .name = "set-time-of-day",
511 .filter = &(const struct rtas_filter) {
512 .buf_idx1 = -1, .size_idx1 = -1,
513 .buf_idx2 = -1, .size_idx2 = -1,
514 },
515 },
516 [RTAS_FNIDX__START_CPU] = {
517 .name = "start-cpu",
518 },
519 [RTAS_FNIDX__STOP_SELF] = {
520 .name = "stop-self",
521 },
522 [RTAS_FNIDX__SYSTEM_REBOOT] = {
523 .name = "system-reboot",
524 },
525 [RTAS_FNIDX__THAW_TIME_BASE] = {
526 .name = "thaw-time-base",
527 },
528 [RTAS_FNIDX__WRITE_PCI_CONFIG] = {
529 .name = "write-pci-config",
530 },
531};
532
533#define for_each_rtas_function(funcp) \
534 for (funcp = &rtas_function_table[0]; \
535 funcp < &rtas_function_table[ARRAY_SIZE(rtas_function_table)]; \
536 ++funcp)
537
538/*
539 * Nearly all RTAS calls need to be serialized. All uses of the
540 * default rtas_args block must hold rtas_lock.
541 *
542 * Exceptions to the RTAS serialization requirement (e.g. stop-self)
543 * must use a separate rtas_args structure.
544 */
545static DEFINE_RAW_SPINLOCK(rtas_lock);
546static struct rtas_args rtas_args;
547
548/**
549 * rtas_function_token() - RTAS function token lookup.
550 * @handle: Function handle, e.g. RTAS_FN_EVENT_SCAN.
551 *
552 * Context: Any context.
553 * Return: the token value for the function if implemented by this platform,
554 * otherwise RTAS_UNKNOWN_SERVICE.
555 */
556s32 rtas_function_token(const rtas_fn_handle_t handle)
557{
558 const size_t index = handle.index;
559 const bool out_of_bounds = index >= ARRAY_SIZE(rtas_function_table);
560
561 if (WARN_ONCE(out_of_bounds, "invalid function index %zu", index))
562 return RTAS_UNKNOWN_SERVICE;
563 /*
564 * Various drivers attempt token lookups on non-RTAS
565 * platforms.
566 */
567 if (!rtas.dev)
568 return RTAS_UNKNOWN_SERVICE;
569
570 return rtas_function_table[index].token;
571}
572EXPORT_SYMBOL_GPL(rtas_function_token);
573
574static int rtas_function_cmp(const void *a, const void *b)
575{
576 const struct rtas_function *f1 = a;
577 const struct rtas_function *f2 = b;
578
579 return strcmp(f1->name, f2->name);
580}
581
582/*
583 * Boot-time initialization of the function table needs the lookup to
584 * return a non-const-qualified object. Use rtas_name_to_function()
585 * in all other contexts.
586 */
587static struct rtas_function *__rtas_name_to_function(const char *name)
588{
589 const struct rtas_function key = {
590 .name = name,
591 };
592 struct rtas_function *found;
593
594 found = bsearch(&key, rtas_function_table, ARRAY_SIZE(rtas_function_table),
595 sizeof(rtas_function_table[0]), rtas_function_cmp);
596
597 return found;
598}
599
600static const struct rtas_function *rtas_name_to_function(const char *name)
601{
602 return __rtas_name_to_function(name);
603}
604
605static DEFINE_XARRAY(rtas_token_to_function_xarray);
606
607static int __init rtas_token_to_function_xarray_init(void)
608{
609 const struct rtas_function *func;
610 int err = 0;
611
612 for_each_rtas_function(func) {
613 const s32 token = func->token;
614
615 if (token == RTAS_UNKNOWN_SERVICE)
616 continue;
617
618 err = xa_err(xa_store(&rtas_token_to_function_xarray,
619 token, (void *)func, GFP_KERNEL));
620 if (err)
621 break;
622 }
623
624 return err;
625}
626arch_initcall(rtas_token_to_function_xarray_init);
627
628/*
629 * For use by sys_rtas(), where the token value is provided by user
630 * space and we don't want to warn on failed lookups.
631 */
632static const struct rtas_function *rtas_token_to_function_untrusted(s32 token)
633{
634 return xa_load(&rtas_token_to_function_xarray, token);
635}
636
637/*
638 * Reverse lookup for deriving the function descriptor from a
639 * known-good token value in contexts where the former is not already
640 * available. @token must be valid, e.g. derived from the result of a
641 * prior lookup against the function table.
642 */
643static const struct rtas_function *rtas_token_to_function(s32 token)
644{
645 const struct rtas_function *func;
646
647 if (WARN_ONCE(token < 0, "invalid token %d", token))
648 return NULL;
649
650 func = rtas_token_to_function_untrusted(token);
651 if (func)
652 return func;
653 /*
654 * Fall back to linear scan in case the reverse mapping hasn't
655 * been initialized yet.
656 */
657 if (xa_empty(&rtas_token_to_function_xarray)) {
658 for_each_rtas_function(func) {
659 if (func->token == token)
660 return func;
661 }
662 }
663
664 WARN_ONCE(true, "unexpected failed lookup for token %d", token);
665 return NULL;
666}
667
668/* This is here deliberately so it's only used in this file */
669void enter_rtas(unsigned long);
670
671static void __do_enter_rtas(struct rtas_args *args)
672{
673 enter_rtas(__pa(args));
674 srr_regs_clobbered(); /* rtas uses SRRs, invalidate */
675}
676
677static void __do_enter_rtas_trace(struct rtas_args *args)
678{
679 const struct rtas_function *func = rtas_token_to_function(be32_to_cpu(args->token));
680
681 /*
682 * If there is a per-function lock, it must be held by the
683 * caller.
684 */
685 if (func->lock)
686 lockdep_assert_held(func->lock);
687
688 if (args == &rtas_args)
689 lockdep_assert_held(&rtas_lock);
690
691 trace_rtas_input(args, func->name);
692 trace_rtas_ll_entry(args);
693
694 __do_enter_rtas(args);
695
696 trace_rtas_ll_exit(args);
697 trace_rtas_output(args, func->name);
698}
699
700static void do_enter_rtas(struct rtas_args *args)
701{
702 const unsigned long msr = mfmsr();
703 /*
704 * Situations where we want to skip any active tracepoints for
705 * safety reasons:
706 *
707 * 1. The last code executed on an offline CPU as it stops,
708 * i.e. we're about to call stop-self. The tracepoints'
709 * function name lookup uses xarray, which uses RCU, which
710 * isn't valid to call on an offline CPU. Any events
711 * emitted on an offline CPU will be discarded anyway.
712 *
713 * 2. In real mode, as when invoking ibm,nmi-interlock from
714 * the pseries MCE handler. We cannot count on trace
715 * buffers or the entries in rtas_token_to_function_xarray
716 * to be contained in the RMO.
717 */
718 const unsigned long mask = MSR_IR | MSR_DR;
719 const bool can_trace = likely(cpu_online(raw_smp_processor_id()) &&
720 (msr & mask) == mask);
721 /*
722 * Make sure MSR[RI] is currently enabled as it will be forced later
723 * in enter_rtas.
724 */
725 BUG_ON(!(msr & MSR_RI));
726
727 BUG_ON(!irqs_disabled());
728
729 hard_irq_disable(); /* Ensure MSR[EE] is disabled on PPC64 */
730
731 if (can_trace)
732 __do_enter_rtas_trace(args);
733 else
734 __do_enter_rtas(args);
735}
736
737struct rtas_t rtas;
738
739DEFINE_SPINLOCK(rtas_data_buf_lock);
740EXPORT_SYMBOL_GPL(rtas_data_buf_lock);
741
742char rtas_data_buf[RTAS_DATA_BUF_SIZE] __aligned(SZ_4K);
743EXPORT_SYMBOL_GPL(rtas_data_buf);
744
745unsigned long rtas_rmo_buf;
746
747/*
748 * If non-NULL, this gets called when the kernel terminates.
749 * This is done like this so rtas_flash can be a module.
750 */
751void (*rtas_flash_term_hook)(int);
752EXPORT_SYMBOL_GPL(rtas_flash_term_hook);
753
754/*
755 * call_rtas_display_status and call_rtas_display_status_delay
756 * are designed only for very early low-level debugging, which
757 * is why the token is hard-coded to 10.
758 */
759static void call_rtas_display_status(unsigned char c)
760{
761 unsigned long flags;
762
763 if (!rtas.base)
764 return;
765
766 raw_spin_lock_irqsave(&rtas_lock, flags);
767 rtas_call_unlocked(&rtas_args, 10, 1, 1, NULL, c);
768 raw_spin_unlock_irqrestore(&rtas_lock, flags);
769}
770
771static void call_rtas_display_status_delay(char c)
772{
773 static int pending_newline = 0; /* did last write end with unprinted newline? */
774 static int width = 16;
775
776 if (c == '\n') {
777 while (width-- > 0)
778 call_rtas_display_status(' ');
779 width = 16;
780 mdelay(500);
781 pending_newline = 1;
782 } else {
783 if (pending_newline) {
784 call_rtas_display_status('\r');
785 call_rtas_display_status('\n');
786 }
787 pending_newline = 0;
788 if (width--) {
789 call_rtas_display_status(c);
790 udelay(10000);
791 }
792 }
793}
794
795void __init udbg_init_rtas_panel(void)
796{
797 udbg_putc = call_rtas_display_status_delay;
798}
799
800#ifdef CONFIG_UDBG_RTAS_CONSOLE
801
802/* If you think you're dying before early_init_dt_scan_rtas() does its
803 * work, you can hard code the token values for your firmware here and
804 * hardcode rtas.base/entry etc.
805 */
806static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
807static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
808
809static void udbg_rtascon_putc(char c)
810{
811 int tries;
812
813 if (!rtas.base)
814 return;
815
816 /* Add CRs before LFs */
817 if (c == '\n')
818 udbg_rtascon_putc('\r');
819
820 /* if there is more than one character to be displayed, wait a bit */
821 for (tries = 0; tries < 16; tries++) {
822 if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
823 break;
824 udelay(1000);
825 }
826}
827
828static int udbg_rtascon_getc_poll(void)
829{
830 int c;
831
832 if (!rtas.base)
833 return -1;
834
835 if (rtas_call(rtas_getchar_token, 0, 2, &c))
836 return -1;
837
838 return c;
839}
840
841static int udbg_rtascon_getc(void)
842{
843 int c;
844
845 while ((c = udbg_rtascon_getc_poll()) == -1)
846 ;
847
848 return c;
849}
850
851
852void __init udbg_init_rtas_console(void)
853{
854 udbg_putc = udbg_rtascon_putc;
855 udbg_getc = udbg_rtascon_getc;
856 udbg_getc_poll = udbg_rtascon_getc_poll;
857}
858#endif /* CONFIG_UDBG_RTAS_CONSOLE */
859
860void rtas_progress(char *s, unsigned short hex)
861{
862 struct device_node *root;
863 int width;
864 const __be32 *p;
865 char *os;
866 static int display_character, set_indicator;
867 static int display_width, display_lines, form_feed;
868 static const int *row_width;
869 static DEFINE_SPINLOCK(progress_lock);
870 static int current_line;
871 static int pending_newline = 0; /* did last write end with unprinted newline? */
872
873 if (!rtas.base)
874 return;
875
876 if (display_width == 0) {
877 display_width = 0x10;
878 if ((root = of_find_node_by_path("/rtas"))) {
879 if ((p = of_get_property(root,
880 "ibm,display-line-length", NULL)))
881 display_width = be32_to_cpu(*p);
882 if ((p = of_get_property(root,
883 "ibm,form-feed", NULL)))
884 form_feed = be32_to_cpu(*p);
885 if ((p = of_get_property(root,
886 "ibm,display-number-of-lines", NULL)))
887 display_lines = be32_to_cpu(*p);
888 row_width = of_get_property(root,
889 "ibm,display-truncation-length", NULL);
890 of_node_put(root);
891 }
892 display_character = rtas_function_token(RTAS_FN_DISPLAY_CHARACTER);
893 set_indicator = rtas_function_token(RTAS_FN_SET_INDICATOR);
894 }
895
896 if (display_character == RTAS_UNKNOWN_SERVICE) {
897 /* use hex display if available */
898 if (set_indicator != RTAS_UNKNOWN_SERVICE)
899 rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
900 return;
901 }
902
903 spin_lock(&progress_lock);
904
905 /*
906 * Last write ended with newline, but we didn't print it since
907 * it would just clear the bottom line of output. Print it now
908 * instead.
909 *
910 * If no newline is pending and form feed is supported, clear the
911 * display with a form feed; otherwise, print a CR to start output
912 * at the beginning of the line.
913 */
914 if (pending_newline) {
915 rtas_call(display_character, 1, 1, NULL, '\r');
916 rtas_call(display_character, 1, 1, NULL, '\n');
917 pending_newline = 0;
918 } else {
919 current_line = 0;
920 if (form_feed)
921 rtas_call(display_character, 1, 1, NULL,
922 (char)form_feed);
923 else
924 rtas_call(display_character, 1, 1, NULL, '\r');
925 }
926
927 if (row_width)
928 width = row_width[current_line];
929 else
930 width = display_width;
931 os = s;
932 while (*os) {
933 if (*os == '\n' || *os == '\r') {
934 /* If newline is the last character, save it
935 * until next call to avoid bumping up the
936 * display output.
937 */
938 if (*os == '\n' && !os[1]) {
939 pending_newline = 1;
940 current_line++;
941 if (current_line > display_lines-1)
942 current_line = display_lines-1;
943 spin_unlock(&progress_lock);
944 return;
945 }
946
947 /* RTAS wants CR-LF, not just LF */
948
949 if (*os == '\n') {
950 rtas_call(display_character, 1, 1, NULL, '\r');
951 rtas_call(display_character, 1, 1, NULL, '\n');
952 } else {
953 /* CR might be used to re-draw a line, so we'll
954 * leave it alone and not add LF.
955 */
956 rtas_call(display_character, 1, 1, NULL, *os);
957 }
958
959 if (row_width)
960 width = row_width[current_line];
961 else
962 width = display_width;
963 } else {
964 width--;
965 rtas_call(display_character, 1, 1, NULL, *os);
966 }
967
968 os++;
969
970 /* if we overwrite the screen length */
971 if (width <= 0)
972 while ((*os != 0) && (*os != '\n') && (*os != '\r'))
973 os++;
974 }
975
976 spin_unlock(&progress_lock);
977}
978EXPORT_SYMBOL_GPL(rtas_progress); /* needed by rtas_flash module */
979
980int rtas_token(const char *service)
981{
982 const struct rtas_function *func;
983 const __be32 *tokp;
984
985 if (rtas.dev == NULL)
986 return RTAS_UNKNOWN_SERVICE;
987
988 func = rtas_name_to_function(service);
989 if (func)
990 return func->token;
991 /*
992 * The caller is looking up a name that is not known to be an
993 * RTAS function. Either it's a function that needs to be
994 * added to the table, or they're misusing rtas_token() to
995 * access non-function properties of the /rtas node. Warn and
996 * fall back to the legacy behavior.
997 */
998 WARN_ONCE(1, "unknown function `%s`, should it be added to rtas_function_table?\n",
999 service);
1000
1001 tokp = of_get_property(rtas.dev, service, NULL);
1002 return tokp ? be32_to_cpu(*tokp) : RTAS_UNKNOWN_SERVICE;
1003}
1004EXPORT_SYMBOL_GPL(rtas_token);
1005
1006#ifdef CONFIG_RTAS_ERROR_LOGGING
1007
1008static u32 rtas_error_log_max __ro_after_init = RTAS_ERROR_LOG_MAX;
1009
1010/*
1011 * Return the firmware-specified size of the error log buffer
1012 * for all rtas calls that require an error buffer argument.
1013 * This includes 'check-exception' and 'rtas-last-error'.
1014 */
1015int rtas_get_error_log_max(void)
1016{
1017 return rtas_error_log_max;
1018}
1019
1020static void __init init_error_log_max(void)
1021{
1022 static const char propname[] __initconst = "rtas-error-log-max";
1023 u32 max;
1024
1025 if (of_property_read_u32(rtas.dev, propname, &max)) {
1026 pr_warn("%s not found, using default of %u\n",
1027 propname, RTAS_ERROR_LOG_MAX);
1028 max = RTAS_ERROR_LOG_MAX;
1029 }
1030
1031 if (max > RTAS_ERROR_LOG_MAX) {
1032 pr_warn("%s = %u, clamping max error log size to %u\n",
1033 propname, max, RTAS_ERROR_LOG_MAX);
1034 max = RTAS_ERROR_LOG_MAX;
1035 }
1036
1037 rtas_error_log_max = max;
1038}
1039
1040
1041static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
1042
1043/** Return a copy of the detailed error text associated with the
1044 * most recent failed call to rtas. Because the error text
1045 * might go stale if there are any other intervening rtas calls,
1046 * this routine must be called atomically with whatever produced
1047 * the error (i.e. with rtas_lock still held from the previous call).
1048 */
1049static char *__fetch_rtas_last_error(char *altbuf)
1050{
1051 const s32 token = rtas_function_token(RTAS_FN_RTAS_LAST_ERROR);
1052 struct rtas_args err_args, save_args;
1053 u32 bufsz;
1054 char *buf = NULL;
1055
1056 lockdep_assert_held(&rtas_lock);
1057
1058 if (token == -1)
1059 return NULL;
1060
1061 bufsz = rtas_get_error_log_max();
1062
1063 err_args.token = cpu_to_be32(token);
1064 err_args.nargs = cpu_to_be32(2);
1065 err_args.nret = cpu_to_be32(1);
1066 err_args.args[0] = cpu_to_be32(__pa(rtas_err_buf));
1067 err_args.args[1] = cpu_to_be32(bufsz);
1068 err_args.args[2] = 0;
1069
1070 save_args = rtas_args;
1071 rtas_args = err_args;
1072
1073 do_enter_rtas(&rtas_args);
1074
1075 err_args = rtas_args;
1076 rtas_args = save_args;
1077
1078 /* Log the error in the unlikely case that there was one. */
1079 if (unlikely(err_args.args[2] == 0)) {
1080 if (altbuf) {
1081 buf = altbuf;
1082 } else {
1083 buf = rtas_err_buf;
1084 if (slab_is_available())
1085 buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
1086 }
1087 if (buf)
1088 memmove(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
1089 }
1090
1091 return buf;
1092}
1093
1094#define get_errorlog_buffer() kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
1095
1096#else /* CONFIG_RTAS_ERROR_LOGGING */
1097#define __fetch_rtas_last_error(x) NULL
1098#define get_errorlog_buffer() NULL
1099static void __init init_error_log_max(void) {}
1100#endif
1101
1102
1103static void
1104va_rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret,
1105 va_list list)
1106{
1107 int i;
1108
1109 args->token = cpu_to_be32(token);
1110 args->nargs = cpu_to_be32(nargs);
1111 args->nret = cpu_to_be32(nret);
1112 args->rets = &(args->args[nargs]);
1113
1114 for (i = 0; i < nargs; ++i)
1115 args->args[i] = cpu_to_be32(va_arg(list, __u32));
1116
1117 for (i = 0; i < nret; ++i)
1118 args->rets[i] = 0;
1119
1120 do_enter_rtas(args);
1121}
1122
1123/**
1124 * rtas_call_unlocked() - Invoke an RTAS firmware function without synchronization.
1125 * @args: RTAS parameter block to be used for the call, must obey RTAS addressing
1126 * constraints.
1127 * @token: Identifies the function being invoked.
1128 * @nargs: Number of input parameters. Does not include token.
1129 * @nret: Number of output parameters, including the call status.
1130 * @....: List of @nargs input parameters.
1131 *
1132 * Invokes the RTAS function indicated by @token, which the caller
1133 * should obtain via rtas_function_token().
1134 *
1135 * This function is similar to rtas_call(), but must be used with a
1136 * limited set of RTAS calls specifically exempted from the general
1137 * requirement that only one RTAS call may be in progress at any
1138 * time. Examples include stop-self and ibm,nmi-interlock.
1139 */
1140void rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret, ...)
1141{
1142 va_list list;
1143
1144 va_start(list, nret);
1145 va_rtas_call_unlocked(args, token, nargs, nret, list);
1146 va_end(list);
1147}
1148
1149static bool token_is_restricted_errinjct(s32 token)
1150{
1151 return token == rtas_function_token(RTAS_FN_IBM_OPEN_ERRINJCT) ||
1152 token == rtas_function_token(RTAS_FN_IBM_ERRINJCT);
1153}
1154
1155/**
1156 * rtas_call() - Invoke an RTAS firmware function.
1157 * @token: Identifies the function being invoked.
1158 * @nargs: Number of input parameters. Does not include token.
1159 * @nret: Number of output parameters, including the call status.
1160 * @outputs: Array of @nret output words.
1161 * @....: List of @nargs input parameters.
1162 *
1163 * Invokes the RTAS function indicated by @token, which the caller
1164 * should obtain via rtas_function_token().
1165 *
1166 * The @nargs and @nret arguments must match the number of input and
1167 * output parameters specified for the RTAS function.
1168 *
1169 * rtas_call() returns RTAS status codes, not conventional Linux errno
1170 * values. Callers must translate any failure to an appropriate errno
1171 * in syscall context. Most callers of RTAS functions that can return
1172 * -2 or 990x should use rtas_busy_delay() to correctly handle those
1173 * statuses before calling again.
1174 *
1175 * The return value descriptions are adapted from 7.2.8 [RTAS] Return
1176 * Codes of the PAPR and CHRP specifications.
1177 *
1178 * Context: Process context preferably, interrupt context if
1179 * necessary. Acquires an internal spinlock and may perform
1180 * GFP_ATOMIC slab allocation in error path. Unsafe for NMI
1181 * context.
1182 * Return:
1183 * * 0 - RTAS function call succeeded.
1184 * * -1 - RTAS function encountered a hardware or
1185 * platform error, or the token is invalid,
1186 * or the function is restricted by kernel policy.
1187 * * -2 - Specs say "A necessary hardware device was busy,
1188 * and the requested function could not be
1189 * performed. The operation should be retried at
1190 * a later time." This is misleading, at least with
1191 * respect to current RTAS implementations. What it
1192 * usually means in practice is that the function
1193 * could not be completed while meeting RTAS's
1194 * deadline for returning control to the OS (250us
1195 * for PAPR/PowerVM, typically), but the call may be
1196 * immediately reattempted to resume work on it.
1197 * * -3 - Parameter error.
1198 * * -7 - Unexpected state change.
1199 * * 9000...9899 - Vendor-specific success codes.
1200 * * 9900...9905 - Advisory extended delay. Caller should try
1201 * again after ~10^x ms has elapsed, where x is
1202 * the last digit of the status [0-5]. Again going
1203 * beyond the PAPR text, 990x on PowerVM indicates
1204 * contention for RTAS-internal resources. Other
1205 * RTAS call sequences in progress should be
1206 * allowed to complete before reattempting the
1207 * call.
1208 * * -9000 - Multi-level isolation error.
1209 * * -9999...-9004 - Vendor-specific error codes.
1210 * * Additional negative values - Function-specific error.
1211 * * Additional positive values - Function-specific success.
1212 */
1213int rtas_call(int token, int nargs, int nret, int *outputs, ...)
1214{
1215 struct pin_cookie cookie;
1216 va_list list;
1217 int i;
1218 unsigned long flags;
1219 struct rtas_args *args;
1220 char *buff_copy = NULL;
1221 int ret;
1222
1223 if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
1224 return -1;
1225
1226 if (token_is_restricted_errinjct(token)) {
1227 /*
1228 * It would be nicer to not discard the error value
1229 * from security_locked_down(), but callers expect an
1230 * RTAS status, not an errno.
1231 */
1232 if (security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION))
1233 return -1;
1234 }
1235
1236 if ((mfmsr() & (MSR_IR|MSR_DR)) != (MSR_IR|MSR_DR)) {
1237 WARN_ON_ONCE(1);
1238 return -1;
1239 }
1240
1241 raw_spin_lock_irqsave(&rtas_lock, flags);
1242 cookie = lockdep_pin_lock(&rtas_lock);
1243
1244 /* We use the global rtas args buffer */
1245 args = &rtas_args;
1246
1247 va_start(list, outputs);
1248 va_rtas_call_unlocked(args, token, nargs, nret, list);
1249 va_end(list);
1250
1251 /* A -1 return code indicates that the last command couldn't
1252 be completed due to a hardware error. */
1253 if (be32_to_cpu(args->rets[0]) == -1)
1254 buff_copy = __fetch_rtas_last_error(NULL);
1255
1256 if (nret > 1 && outputs != NULL)
1257 for (i = 0; i < nret-1; ++i)
1258 outputs[i] = be32_to_cpu(args->rets[i + 1]);
1259 ret = (nret > 0) ? be32_to_cpu(args->rets[0]) : 0;
1260
1261 lockdep_unpin_lock(&rtas_lock, cookie);
1262 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1263
1264 if (buff_copy) {
1265 log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
1266 if (slab_is_available())
1267 kfree(buff_copy);
1268 }
1269 return ret;
1270}
1271EXPORT_SYMBOL_GPL(rtas_call);
1272
1273/**
1274 * rtas_busy_delay_time() - From an RTAS status value, calculate the
1275 * suggested delay time in milliseconds.
1276 *
1277 * @status: a value returned from rtas_call() or similar APIs which return
1278 * the status of a RTAS function call.
1279 *
1280 * Context: Any context.
1281 *
1282 * Return:
1283 * * 100000 - If @status is 9905.
1284 * * 10000 - If @status is 9904.
1285 * * 1000 - If @status is 9903.
1286 * * 100 - If @status is 9902.
1287 * * 10 - If @status is 9901.
1288 * * 1 - If @status is either 9900 or -2. This is "wrong" for -2, but
1289 * some callers depend on this behavior, and the worst outcome
1290 * is that they will delay for longer than necessary.
1291 * * 0 - If @status is not a busy or extended delay value.
1292 */
1293unsigned int rtas_busy_delay_time(int status)
1294{
1295 int order;
1296 unsigned int ms = 0;
1297
1298 if (status == RTAS_BUSY) {
1299 ms = 1;
1300 } else if (status >= RTAS_EXTENDED_DELAY_MIN &&
1301 status <= RTAS_EXTENDED_DELAY_MAX) {
1302 order = status - RTAS_EXTENDED_DELAY_MIN;
1303 for (ms = 1; order > 0; order--)
1304 ms *= 10;
1305 }
1306
1307 return ms;
1308}
1309
1310/*
1311 * Early boot fallback for rtas_busy_delay().
1312 */
1313static bool __init rtas_busy_delay_early(int status)
1314{
1315 static size_t successive_ext_delays __initdata;
1316 bool retry;
1317
1318 switch (status) {
1319 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1320 /*
1321 * In the unlikely case that we receive an extended
1322 * delay status in early boot, the OS is probably not
1323 * the cause, and there's nothing we can do to clear
1324 * the condition. Best we can do is delay for a bit
1325 * and hope it's transient. Lie to the caller if it
1326 * seems like we're stuck in a retry loop.
1327 */
1328 mdelay(1);
1329 retry = true;
1330 successive_ext_delays += 1;
1331 if (successive_ext_delays > 1000) {
1332 pr_err("too many extended delays, giving up\n");
1333 dump_stack();
1334 retry = false;
1335 successive_ext_delays = 0;
1336 }
1337 break;
1338 case RTAS_BUSY:
1339 retry = true;
1340 successive_ext_delays = 0;
1341 break;
1342 default:
1343 retry = false;
1344 successive_ext_delays = 0;
1345 break;
1346 }
1347
1348 return retry;
1349}
1350
1351/**
1352 * rtas_busy_delay() - helper for RTAS busy and extended delay statuses
1353 *
1354 * @status: a value returned from rtas_call() or similar APIs which return
1355 * the status of a RTAS function call.
1356 *
1357 * Context: Process context. May sleep or schedule.
1358 *
1359 * Return:
1360 * * true - @status is RTAS_BUSY or an extended delay hint. The
1361 * caller may assume that the CPU has been yielded if necessary,
1362 * and that an appropriate delay for @status has elapsed.
1363 * Generally the caller should reattempt the RTAS call which
1364 * yielded @status.
1365 *
1366 * * false - @status is not @RTAS_BUSY nor an extended delay hint. The
1367 * caller is responsible for handling @status.
1368 */
1369bool __ref rtas_busy_delay(int status)
1370{
1371 unsigned int ms;
1372 bool ret;
1373
1374 /*
1375 * Can't do timed sleeps before timekeeping is up.
1376 */
1377 if (system_state < SYSTEM_SCHEDULING)
1378 return rtas_busy_delay_early(status);
1379
1380 switch (status) {
1381 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1382 ret = true;
1383 ms = rtas_busy_delay_time(status);
1384 /*
1385 * The extended delay hint can be as high as 100 seconds.
1386 * Surely any function returning such a status is either
1387 * buggy or isn't going to be significantly slowed by us
1388 * polling at 1HZ. Clamp the sleep time to one second.
1389 */
1390 ms = clamp(ms, 1U, 1000U);
1391 /*
1392 * The delay hint is an order-of-magnitude suggestion, not
1393 * a minimum. It is fine, possibly even advantageous, for
1394 * us to pause for less time than hinted. For small values,
1395 * use usleep_range() to ensure we don't sleep much longer
1396 * than actually needed.
1397 *
1398 * See Documentation/timers/timers-howto.rst for
1399 * explanation of the threshold used here. In effect we use
1400 * usleep_range() for 9900 and 9901, msleep() for
1401 * 9902-9905.
1402 */
1403 if (ms <= 20)
1404 usleep_range(ms * 100, ms * 1000);
1405 else
1406 msleep(ms);
1407 break;
1408 case RTAS_BUSY:
1409 ret = true;
1410 /*
1411 * We should call again immediately if there's no other
1412 * work to do.
1413 */
1414 cond_resched();
1415 break;
1416 default:
1417 ret = false;
1418 /*
1419 * Not a busy or extended delay status; the caller should
1420 * handle @status itself. Ensure we warn on misuses in
1421 * atomic context regardless.
1422 */
1423 might_sleep();
1424 break;
1425 }
1426
1427 return ret;
1428}
1429EXPORT_SYMBOL_GPL(rtas_busy_delay);
1430
1431int rtas_error_rc(int rtas_rc)
1432{
1433 int rc;
1434
1435 switch (rtas_rc) {
1436 case RTAS_HARDWARE_ERROR: /* Hardware Error */
1437 rc = -EIO;
1438 break;
1439 case RTAS_INVALID_PARAMETER: /* Bad indicator/domain/etc */
1440 rc = -EINVAL;
1441 break;
1442 case -9000: /* Isolation error */
1443 rc = -EFAULT;
1444 break;
1445 case -9001: /* Outstanding TCE/PTE */
1446 rc = -EEXIST;
1447 break;
1448 case -9002: /* No usable slot */
1449 rc = -ENODEV;
1450 break;
1451 default:
1452 pr_err("%s: unexpected error %d\n", __func__, rtas_rc);
1453 rc = -ERANGE;
1454 break;
1455 }
1456 return rc;
1457}
1458EXPORT_SYMBOL_GPL(rtas_error_rc);
1459
1460int rtas_get_power_level(int powerdomain, int *level)
1461{
1462 int token = rtas_function_token(RTAS_FN_GET_POWER_LEVEL);
1463 int rc;
1464
1465 if (token == RTAS_UNKNOWN_SERVICE)
1466 return -ENOENT;
1467
1468 while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
1469 udelay(1);
1470
1471 if (rc < 0)
1472 return rtas_error_rc(rc);
1473 return rc;
1474}
1475EXPORT_SYMBOL_GPL(rtas_get_power_level);
1476
1477int rtas_set_power_level(int powerdomain, int level, int *setlevel)
1478{
1479 int token = rtas_function_token(RTAS_FN_SET_POWER_LEVEL);
1480 int rc;
1481
1482 if (token == RTAS_UNKNOWN_SERVICE)
1483 return -ENOENT;
1484
1485 do {
1486 rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
1487 } while (rtas_busy_delay(rc));
1488
1489 if (rc < 0)
1490 return rtas_error_rc(rc);
1491 return rc;
1492}
1493EXPORT_SYMBOL_GPL(rtas_set_power_level);
1494
1495int rtas_get_sensor(int sensor, int index, int *state)
1496{
1497 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1498 int rc;
1499
1500 if (token == RTAS_UNKNOWN_SERVICE)
1501 return -ENOENT;
1502
1503 do {
1504 rc = rtas_call(token, 2, 2, state, sensor, index);
1505 } while (rtas_busy_delay(rc));
1506
1507 if (rc < 0)
1508 return rtas_error_rc(rc);
1509 return rc;
1510}
1511EXPORT_SYMBOL_GPL(rtas_get_sensor);
1512
1513int rtas_get_sensor_fast(int sensor, int index, int *state)
1514{
1515 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1516 int rc;
1517
1518 if (token == RTAS_UNKNOWN_SERVICE)
1519 return -ENOENT;
1520
1521 rc = rtas_call(token, 2, 2, state, sensor, index);
1522 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1523 rc <= RTAS_EXTENDED_DELAY_MAX));
1524
1525 if (rc < 0)
1526 return rtas_error_rc(rc);
1527 return rc;
1528}
1529
1530bool rtas_indicator_present(int token, int *maxindex)
1531{
1532 int proplen, count, i;
1533 const struct indicator_elem {
1534 __be32 token;
1535 __be32 maxindex;
1536 } *indicators;
1537
1538 indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
1539 if (!indicators)
1540 return false;
1541
1542 count = proplen / sizeof(struct indicator_elem);
1543
1544 for (i = 0; i < count; i++) {
1545 if (__be32_to_cpu(indicators[i].token) != token)
1546 continue;
1547 if (maxindex)
1548 *maxindex = __be32_to_cpu(indicators[i].maxindex);
1549 return true;
1550 }
1551
1552 return false;
1553}
1554
1555int rtas_set_indicator(int indicator, int index, int new_value)
1556{
1557 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1558 int rc;
1559
1560 if (token == RTAS_UNKNOWN_SERVICE)
1561 return -ENOENT;
1562
1563 do {
1564 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1565 } while (rtas_busy_delay(rc));
1566
1567 if (rc < 0)
1568 return rtas_error_rc(rc);
1569 return rc;
1570}
1571EXPORT_SYMBOL_GPL(rtas_set_indicator);
1572
1573/*
1574 * Ignoring RTAS extended delay
1575 */
1576int rtas_set_indicator_fast(int indicator, int index, int new_value)
1577{
1578 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1579 int rc;
1580
1581 if (token == RTAS_UNKNOWN_SERVICE)
1582 return -ENOENT;
1583
1584 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1585
1586 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1587 rc <= RTAS_EXTENDED_DELAY_MAX));
1588
1589 if (rc < 0)
1590 return rtas_error_rc(rc);
1591
1592 return rc;
1593}
1594
1595/**
1596 * rtas_ibm_suspend_me() - Call ibm,suspend-me to suspend the LPAR.
1597 *
1598 * @fw_status: RTAS call status will be placed here if not NULL.
1599 *
1600 * rtas_ibm_suspend_me() should be called only on a CPU which has
1601 * received H_CONTINUE from the H_JOIN hcall. All other active CPUs
1602 * should be waiting to return from H_JOIN.
1603 *
1604 * rtas_ibm_suspend_me() may suspend execution of the OS
1605 * indefinitely. Callers should take appropriate measures upon return, such as
1606 * resetting watchdog facilities.
1607 *
1608 * Callers may choose to retry this call if @fw_status is
1609 * %RTAS_THREADS_ACTIVE.
1610 *
1611 * Return:
1612 * 0 - The partition has resumed from suspend, possibly after
1613 * migration to a different host.
1614 * -ECANCELED - The operation was aborted.
1615 * -EAGAIN - There were other CPUs not in H_JOIN at the time of the call.
1616 * -EBUSY - Some other condition prevented the suspend from succeeding.
1617 * -EIO - Hardware/platform error.
1618 */
1619int rtas_ibm_suspend_me(int *fw_status)
1620{
1621 int token = rtas_function_token(RTAS_FN_IBM_SUSPEND_ME);
1622 int fwrc;
1623 int ret;
1624
1625 fwrc = rtas_call(token, 0, 1, NULL);
1626
1627 switch (fwrc) {
1628 case 0:
1629 ret = 0;
1630 break;
1631 case RTAS_SUSPEND_ABORTED:
1632 ret = -ECANCELED;
1633 break;
1634 case RTAS_THREADS_ACTIVE:
1635 ret = -EAGAIN;
1636 break;
1637 case RTAS_NOT_SUSPENDABLE:
1638 case RTAS_OUTSTANDING_COPROC:
1639 ret = -EBUSY;
1640 break;
1641 case -1:
1642 default:
1643 ret = -EIO;
1644 break;
1645 }
1646
1647 if (fw_status)
1648 *fw_status = fwrc;
1649
1650 return ret;
1651}
1652
1653void __noreturn rtas_restart(char *cmd)
1654{
1655 if (rtas_flash_term_hook)
1656 rtas_flash_term_hook(SYS_RESTART);
1657 pr_emerg("system-reboot returned %d\n",
1658 rtas_call(rtas_function_token(RTAS_FN_SYSTEM_REBOOT), 0, 1, NULL));
1659 for (;;);
1660}
1661
1662void rtas_power_off(void)
1663{
1664 if (rtas_flash_term_hook)
1665 rtas_flash_term_hook(SYS_POWER_OFF);
1666 /* allow power on only with power button press */
1667 pr_emerg("power-off returned %d\n",
1668 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1669 for (;;);
1670}
1671
1672void __noreturn rtas_halt(void)
1673{
1674 if (rtas_flash_term_hook)
1675 rtas_flash_term_hook(SYS_HALT);
1676 /* allow power on only with power button press */
1677 pr_emerg("power-off returned %d\n",
1678 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1679 for (;;);
1680}
1681
1682/* Must be in the RMO region, so we place it here */
1683static char rtas_os_term_buf[2048];
1684static bool ibm_extended_os_term;
1685
1686void rtas_os_term(char *str)
1687{
1688 s32 token = rtas_function_token(RTAS_FN_IBM_OS_TERM);
1689 static struct rtas_args args;
1690 int status;
1691
1692 /*
1693 * Firmware with the ibm,extended-os-term property is guaranteed
1694 * to always return from an ibm,os-term call. Earlier versions without
1695 * this property may terminate the partition which we want to avoid
1696 * since it interferes with panic_timeout.
1697 */
1698
1699 if (token == RTAS_UNKNOWN_SERVICE || !ibm_extended_os_term)
1700 return;
1701
1702 snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
1703
1704 /*
1705 * Keep calling as long as RTAS returns a "try again" status,
1706 * but don't use rtas_busy_delay(), which potentially
1707 * schedules.
1708 */
1709 do {
1710 rtas_call_unlocked(&args, token, 1, 1, NULL, __pa(rtas_os_term_buf));
1711 status = be32_to_cpu(args.rets[0]);
1712 } while (rtas_busy_delay_time(status));
1713
1714 if (status != 0)
1715 pr_emerg("ibm,os-term call failed %d\n", status);
1716}
1717
1718/**
1719 * rtas_activate_firmware() - Activate a new version of firmware.
1720 *
1721 * Context: This function may sleep.
1722 *
1723 * Activate a new version of partition firmware. The OS must call this
1724 * after resuming from a partition hibernation or migration in order
1725 * to maintain the ability to perform live firmware updates. It's not
1726 * catastrophic for this method to be absent or to fail; just log the
1727 * condition in that case.
1728 */
1729void rtas_activate_firmware(void)
1730{
1731 int token = rtas_function_token(RTAS_FN_IBM_ACTIVATE_FIRMWARE);
1732 int fwrc;
1733
1734 if (token == RTAS_UNKNOWN_SERVICE) {
1735 pr_notice("ibm,activate-firmware method unavailable\n");
1736 return;
1737 }
1738
1739 mutex_lock(&rtas_ibm_activate_firmware_lock);
1740
1741 do {
1742 fwrc = rtas_call(token, 0, 1, NULL);
1743 } while (rtas_busy_delay(fwrc));
1744
1745 mutex_unlock(&rtas_ibm_activate_firmware_lock);
1746
1747 if (fwrc)
1748 pr_err("ibm,activate-firmware failed (%i)\n", fwrc);
1749}
1750
1751/**
1752 * get_pseries_errorlog() - Find a specific pseries error log in an RTAS
1753 * extended event log.
1754 * @log: RTAS error/event log
1755 * @section_id: two character section identifier
1756 *
1757 * Return: A pointer to the specified errorlog or NULL if not found.
1758 */
1759noinstr struct pseries_errorlog *get_pseries_errorlog(struct rtas_error_log *log,
1760 uint16_t section_id)
1761{
1762 struct rtas_ext_event_log_v6 *ext_log =
1763 (struct rtas_ext_event_log_v6 *)log->buffer;
1764 struct pseries_errorlog *sect;
1765 unsigned char *p, *log_end;
1766 uint32_t ext_log_length = rtas_error_extended_log_length(log);
1767 uint8_t log_format = rtas_ext_event_log_format(ext_log);
1768 uint32_t company_id = rtas_ext_event_company_id(ext_log);
1769
1770 /* Check that we understand the format */
1771 if (ext_log_length < sizeof(struct rtas_ext_event_log_v6) ||
1772 log_format != RTAS_V6EXT_LOG_FORMAT_EVENT_LOG ||
1773 company_id != RTAS_V6EXT_COMPANY_ID_IBM)
1774 return NULL;
1775
1776 log_end = log->buffer + ext_log_length;
1777 p = ext_log->vendor_log;
1778
1779 while (p < log_end) {
1780 sect = (struct pseries_errorlog *)p;
1781 if (pseries_errorlog_id(sect) == section_id)
1782 return sect;
1783 p += pseries_errorlog_length(sect);
1784 }
1785
1786 return NULL;
1787}
1788
1789/*
1790 * The sys_rtas syscall, as originally designed, allows root to pass
1791 * arbitrary physical addresses to RTAS calls. A number of RTAS calls
1792 * can be abused to write to arbitrary memory and do other things that
1793 * are potentially harmful to system integrity, and thus should only
1794 * be used inside the kernel and not exposed to userspace.
1795 *
1796 * All known legitimate users of the sys_rtas syscall will only ever
1797 * pass addresses that fall within the RMO buffer, and use a known
1798 * subset of RTAS calls.
1799 *
1800 * Accordingly, we filter RTAS requests to check that the call is
1801 * permitted, and that provided pointers fall within the RMO buffer.
1802 * If a function is allowed to be invoked via the syscall, then its
1803 * entry in the rtas_functions table points to a rtas_filter that
1804 * describes its constraints, with the indexes of the parameters which
1805 * are expected to contain addresses and sizes of buffers allocated
1806 * inside the RMO buffer.
1807 */
1808
1809static bool in_rmo_buf(u32 base, u32 end)
1810{
1811 return base >= rtas_rmo_buf &&
1812 base < (rtas_rmo_buf + RTAS_USER_REGION_SIZE) &&
1813 base <= end &&
1814 end >= rtas_rmo_buf &&
1815 end < (rtas_rmo_buf + RTAS_USER_REGION_SIZE);
1816}
1817
1818static bool block_rtas_call(const struct rtas_function *func, int nargs,
1819 struct rtas_args *args)
1820{
1821 const struct rtas_filter *f;
1822 const bool is_platform_dump =
1823 func == &rtas_function_table[RTAS_FNIDX__IBM_PLATFORM_DUMP];
1824 const bool is_config_conn =
1825 func == &rtas_function_table[RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR];
1826 u32 base, size, end;
1827
1828 /*
1829 * Only functions with filters attached are allowed.
1830 */
1831 f = func->filter;
1832 if (!f)
1833 goto err;
1834 /*
1835 * And some functions aren't allowed on LE.
1836 */
1837 if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) && func->banned_for_syscall_on_le)
1838 goto err;
1839
1840 if (f->buf_idx1 != -1) {
1841 base = be32_to_cpu(args->args[f->buf_idx1]);
1842 if (f->size_idx1 != -1)
1843 size = be32_to_cpu(args->args[f->size_idx1]);
1844 else if (f->fixed_size)
1845 size = f->fixed_size;
1846 else
1847 size = 1;
1848
1849 end = base + size - 1;
1850
1851 /*
1852 * Special case for ibm,platform-dump - NULL buffer
1853 * address is used to indicate end of dump processing
1854 */
1855 if (is_platform_dump && base == 0)
1856 return false;
1857
1858 if (!in_rmo_buf(base, end))
1859 goto err;
1860 }
1861
1862 if (f->buf_idx2 != -1) {
1863 base = be32_to_cpu(args->args[f->buf_idx2]);
1864 if (f->size_idx2 != -1)
1865 size = be32_to_cpu(args->args[f->size_idx2]);
1866 else if (f->fixed_size)
1867 size = f->fixed_size;
1868 else
1869 size = 1;
1870 end = base + size - 1;
1871
1872 /*
1873 * Special case for ibm,configure-connector where the
1874 * address can be 0
1875 */
1876 if (is_config_conn && base == 0)
1877 return false;
1878
1879 if (!in_rmo_buf(base, end))
1880 goto err;
1881 }
1882
1883 return false;
1884err:
1885 pr_err_ratelimited("sys_rtas: RTAS call blocked - exploit attempt?\n");
1886 pr_err_ratelimited("sys_rtas: %s nargs=%d (called by %s)\n",
1887 func->name, nargs, current->comm);
1888 return true;
1889}
1890
1891/* We assume to be passed big endian arguments */
1892SYSCALL_DEFINE1(rtas, struct rtas_args __user *, uargs)
1893{
1894 const struct rtas_function *func;
1895 struct pin_cookie cookie;
1896 struct rtas_args args;
1897 unsigned long flags;
1898 char *buff_copy, *errbuf = NULL;
1899 int nargs, nret, token;
1900
1901 if (!capable(CAP_SYS_ADMIN))
1902 return -EPERM;
1903
1904 if (!rtas.entry)
1905 return -EINVAL;
1906
1907 if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
1908 return -EFAULT;
1909
1910 nargs = be32_to_cpu(args.nargs);
1911 nret = be32_to_cpu(args.nret);
1912 token = be32_to_cpu(args.token);
1913
1914 if (nargs >= ARRAY_SIZE(args.args)
1915 || nret > ARRAY_SIZE(args.args)
1916 || nargs + nret > ARRAY_SIZE(args.args))
1917 return -EINVAL;
1918
1919 /* Copy in args. */
1920 if (copy_from_user(args.args, uargs->args,
1921 nargs * sizeof(rtas_arg_t)) != 0)
1922 return -EFAULT;
1923
1924 /*
1925 * If this token doesn't correspond to a function the kernel
1926 * understands, you're not allowed to call it.
1927 */
1928 func = rtas_token_to_function_untrusted(token);
1929 if (!func)
1930 return -EINVAL;
1931
1932 args.rets = &args.args[nargs];
1933 memset(args.rets, 0, nret * sizeof(rtas_arg_t));
1934
1935 if (block_rtas_call(func, nargs, &args))
1936 return -EINVAL;
1937
1938 if (token_is_restricted_errinjct(token)) {
1939 int err;
1940
1941 err = security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION);
1942 if (err)
1943 return err;
1944 }
1945
1946 /* Need to handle ibm,suspend_me call specially */
1947 if (token == rtas_function_token(RTAS_FN_IBM_SUSPEND_ME)) {
1948
1949 /*
1950 * rtas_ibm_suspend_me assumes the streamid handle is in cpu
1951 * endian, or at least the hcall within it requires it.
1952 */
1953 int rc = 0;
1954 u64 handle = ((u64)be32_to_cpu(args.args[0]) << 32)
1955 | be32_to_cpu(args.args[1]);
1956 rc = rtas_syscall_dispatch_ibm_suspend_me(handle);
1957 if (rc == -EAGAIN)
1958 args.rets[0] = cpu_to_be32(RTAS_NOT_SUSPENDABLE);
1959 else if (rc == -EIO)
1960 args.rets[0] = cpu_to_be32(-1);
1961 else if (rc)
1962 return rc;
1963 goto copy_return;
1964 }
1965
1966 buff_copy = get_errorlog_buffer();
1967
1968 /*
1969 * If this function has a mutex assigned to it, we must
1970 * acquire it to avoid interleaving with any kernel-based uses
1971 * of the same function. Kernel-based sequences acquire the
1972 * appropriate mutex explicitly.
1973 */
1974 if (func->lock)
1975 mutex_lock(func->lock);
1976
1977 raw_spin_lock_irqsave(&rtas_lock, flags);
1978 cookie = lockdep_pin_lock(&rtas_lock);
1979
1980 rtas_args = args;
1981 do_enter_rtas(&rtas_args);
1982 args = rtas_args;
1983
1984 /* A -1 return code indicates that the last command couldn't
1985 be completed due to a hardware error. */
1986 if (be32_to_cpu(args.rets[0]) == -1)
1987 errbuf = __fetch_rtas_last_error(buff_copy);
1988
1989 lockdep_unpin_lock(&rtas_lock, cookie);
1990 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1991
1992 if (func->lock)
1993 mutex_unlock(func->lock);
1994
1995 if (buff_copy) {
1996 if (errbuf)
1997 log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
1998 kfree(buff_copy);
1999 }
2000
2001 copy_return:
2002 /* Copy out args. */
2003 if (copy_to_user(uargs->args + nargs,
2004 args.args + nargs,
2005 nret * sizeof(rtas_arg_t)) != 0)
2006 return -EFAULT;
2007
2008 return 0;
2009}
2010
2011static void __init rtas_function_table_init(void)
2012{
2013 struct property *prop;
2014
2015 for (size_t i = 0; i < ARRAY_SIZE(rtas_function_table); ++i) {
2016 struct rtas_function *curr = &rtas_function_table[i];
2017 struct rtas_function *prior;
2018 int cmp;
2019
2020 curr->token = RTAS_UNKNOWN_SERVICE;
2021
2022 if (i == 0)
2023 continue;
2024 /*
2025 * Ensure table is sorted correctly for binary search
2026 * on function names.
2027 */
2028 prior = &rtas_function_table[i - 1];
2029
2030 cmp = strcmp(prior->name, curr->name);
2031 if (cmp < 0)
2032 continue;
2033
2034 if (cmp == 0) {
2035 pr_err("'%s' has duplicate function table entries\n",
2036 curr->name);
2037 } else {
2038 pr_err("function table unsorted: '%s' wrongly precedes '%s'\n",
2039 prior->name, curr->name);
2040 }
2041 }
2042
2043 for_each_property_of_node(rtas.dev, prop) {
2044 struct rtas_function *func;
2045
2046 if (prop->length != sizeof(u32))
2047 continue;
2048
2049 func = __rtas_name_to_function(prop->name);
2050 if (!func)
2051 continue;
2052
2053 func->token = be32_to_cpup((__be32 *)prop->value);
2054
2055 pr_debug("function %s has token %u\n", func->name, func->token);
2056 }
2057}
2058
2059/*
2060 * Call early during boot, before mem init, to retrieve the RTAS
2061 * information from the device-tree and allocate the RMO buffer for userland
2062 * accesses.
2063 */
2064void __init rtas_initialize(void)
2065{
2066 unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
2067 u32 base, size, entry;
2068 int no_base, no_size, no_entry;
2069
2070 /* Get RTAS dev node and fill up our "rtas" structure with infos
2071 * about it.
2072 */
2073 rtas.dev = of_find_node_by_name(NULL, "rtas");
2074 if (!rtas.dev)
2075 return;
2076
2077 no_base = of_property_read_u32(rtas.dev, "linux,rtas-base", &base);
2078 no_size = of_property_read_u32(rtas.dev, "rtas-size", &size);
2079 if (no_base || no_size) {
2080 of_node_put(rtas.dev);
2081 rtas.dev = NULL;
2082 return;
2083 }
2084
2085 rtas.base = base;
2086 rtas.size = size;
2087 no_entry = of_property_read_u32(rtas.dev, "linux,rtas-entry", &entry);
2088 rtas.entry = no_entry ? rtas.base : entry;
2089
2090 init_error_log_max();
2091
2092 /* Must be called before any function token lookups */
2093 rtas_function_table_init();
2094
2095 /*
2096 * Discover this now to avoid a device tree lookup in the
2097 * panic path.
2098 */
2099 ibm_extended_os_term = of_property_read_bool(rtas.dev, "ibm,extended-os-term");
2100
2101 /* If RTAS was found, allocate the RMO buffer for it and look for
2102 * the stop-self token if any
2103 */
2104#ifdef CONFIG_PPC64
2105 if (firmware_has_feature(FW_FEATURE_LPAR))
2106 rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
2107#endif
2108 rtas_rmo_buf = memblock_phys_alloc_range(RTAS_USER_REGION_SIZE, PAGE_SIZE,
2109 0, rtas_region);
2110 if (!rtas_rmo_buf)
2111 panic("ERROR: RTAS: Failed to allocate %lx bytes below %pa\n",
2112 PAGE_SIZE, &rtas_region);
2113
2114 rtas_work_area_reserve_arena(rtas_region);
2115}
2116
2117int __init early_init_dt_scan_rtas(unsigned long node,
2118 const char *uname, int depth, void *data)
2119{
2120 const u32 *basep, *entryp, *sizep;
2121
2122 if (depth != 1 || strcmp(uname, "rtas") != 0)
2123 return 0;
2124
2125 basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
2126 entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
2127 sizep = of_get_flat_dt_prop(node, "rtas-size", NULL);
2128
2129#ifdef CONFIG_PPC64
2130 /* need this feature to decide the crashkernel offset */
2131 if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL))
2132 powerpc_firmware_features |= FW_FEATURE_LPAR;
2133#endif
2134
2135 if (basep && entryp && sizep) {
2136 rtas.base = *basep;
2137 rtas.entry = *entryp;
2138 rtas.size = *sizep;
2139 }
2140
2141#ifdef CONFIG_UDBG_RTAS_CONSOLE
2142 basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
2143 if (basep)
2144 rtas_putchar_token = *basep;
2145
2146 basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
2147 if (basep)
2148 rtas_getchar_token = *basep;
2149
2150 if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
2151 rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
2152 udbg_init_rtas_console();
2153
2154#endif
2155
2156 /* break now */
2157 return 1;
2158}
2159
2160static DEFINE_RAW_SPINLOCK(timebase_lock);
2161static u64 timebase = 0;
2162
2163void rtas_give_timebase(void)
2164{
2165 unsigned long flags;
2166
2167 raw_spin_lock_irqsave(&timebase_lock, flags);
2168 hard_irq_disable();
2169 rtas_call(rtas_function_token(RTAS_FN_FREEZE_TIME_BASE), 0, 1, NULL);
2170 timebase = get_tb();
2171 raw_spin_unlock(&timebase_lock);
2172
2173 while (timebase)
2174 barrier();
2175 rtas_call(rtas_function_token(RTAS_FN_THAW_TIME_BASE), 0, 1, NULL);
2176 local_irq_restore(flags);
2177}
2178
2179void rtas_take_timebase(void)
2180{
2181 while (!timebase)
2182 barrier();
2183 raw_spin_lock(&timebase_lock);
2184 set_tb(timebase >> 32, timebase & 0xffffffff);
2185 timebase = 0;
2186 raw_spin_unlock(&timebase_lock);
2187}