xref: /linux/arch/powerpc/kernel/rtas.c (revision eca590f402332ab873d13f2d8d00fa0b91cfff36)
1 /*
2  *
3  * Procedures for interfacing to the RTAS on CHRP machines.
4  *
5  * Peter Bergner, IBM	March 2001.
6  * Copyright (C) 2001 IBM.
7  *
8  *      This program is free software; you can redistribute it and/or
9  *      modify it under the terms of the GNU General Public License
10  *      as published by the Free Software Foundation; either version
11  *      2 of the License, or (at your option) any later version.
12  */
13 
14 #include <stdarg.h>
15 #include <linux/kernel.h>
16 #include <linux/types.h>
17 #include <linux/spinlock.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/capability.h>
21 #include <linux/delay.h>
22 #include <linux/smp.h>
23 #include <linux/completion.h>
24 #include <linux/cpumask.h>
25 #include <linux/memblock.h>
26 #include <linux/slab.h>
27 
28 #include <asm/prom.h>
29 #include <asm/rtas.h>
30 #include <asm/hvcall.h>
31 #include <asm/machdep.h>
32 #include <asm/firmware.h>
33 #include <asm/page.h>
34 #include <asm/param.h>
35 #include <asm/system.h>
36 #include <asm/delay.h>
37 #include <asm/uaccess.h>
38 #include <asm/udbg.h>
39 #include <asm/syscalls.h>
40 #include <asm/smp.h>
41 #include <asm/atomic.h>
42 #include <asm/time.h>
43 #include <asm/mmu.h>
44 #include <asm/topology.h>
45 
46 struct rtas_t rtas = {
47 	.lock = __ARCH_SPIN_LOCK_UNLOCKED
48 };
49 EXPORT_SYMBOL(rtas);
50 
51 DEFINE_SPINLOCK(rtas_data_buf_lock);
52 EXPORT_SYMBOL(rtas_data_buf_lock);
53 
54 char rtas_data_buf[RTAS_DATA_BUF_SIZE] __cacheline_aligned;
55 EXPORT_SYMBOL(rtas_data_buf);
56 
57 unsigned long rtas_rmo_buf;
58 
59 /*
60  * If non-NULL, this gets called when the kernel terminates.
61  * This is done like this so rtas_flash can be a module.
62  */
63 void (*rtas_flash_term_hook)(int);
64 EXPORT_SYMBOL(rtas_flash_term_hook);
65 
66 /* RTAS use home made raw locking instead of spin_lock_irqsave
67  * because those can be called from within really nasty contexts
68  * such as having the timebase stopped which would lockup with
69  * normal locks and spinlock debugging enabled
70  */
71 static unsigned long lock_rtas(void)
72 {
73 	unsigned long flags;
74 
75 	local_irq_save(flags);
76 	preempt_disable();
77 	arch_spin_lock_flags(&rtas.lock, flags);
78 	return flags;
79 }
80 
81 static void unlock_rtas(unsigned long flags)
82 {
83 	arch_spin_unlock(&rtas.lock);
84 	local_irq_restore(flags);
85 	preempt_enable();
86 }
87 
88 /*
89  * call_rtas_display_status and call_rtas_display_status_delay
90  * are designed only for very early low-level debugging, which
91  * is why the token is hard-coded to 10.
92  */
93 static void call_rtas_display_status(char c)
94 {
95 	struct rtas_args *args = &rtas.args;
96 	unsigned long s;
97 
98 	if (!rtas.base)
99 		return;
100 	s = lock_rtas();
101 
102 	args->token = 10;
103 	args->nargs = 1;
104 	args->nret  = 1;
105 	args->rets  = (rtas_arg_t *)&(args->args[1]);
106 	args->args[0] = (unsigned char)c;
107 
108 	enter_rtas(__pa(args));
109 
110 	unlock_rtas(s);
111 }
112 
113 static void call_rtas_display_status_delay(char c)
114 {
115 	static int pending_newline = 0;  /* did last write end with unprinted newline? */
116 	static int width = 16;
117 
118 	if (c == '\n') {
119 		while (width-- > 0)
120 			call_rtas_display_status(' ');
121 		width = 16;
122 		mdelay(500);
123 		pending_newline = 1;
124 	} else {
125 		if (pending_newline) {
126 			call_rtas_display_status('\r');
127 			call_rtas_display_status('\n');
128 		}
129 		pending_newline = 0;
130 		if (width--) {
131 			call_rtas_display_status(c);
132 			udelay(10000);
133 		}
134 	}
135 }
136 
137 void __init udbg_init_rtas_panel(void)
138 {
139 	udbg_putc = call_rtas_display_status_delay;
140 }
141 
142 #ifdef CONFIG_UDBG_RTAS_CONSOLE
143 
144 /* If you think you're dying before early_init_dt_scan_rtas() does its
145  * work, you can hard code the token values for your firmware here and
146  * hardcode rtas.base/entry etc.
147  */
148 static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
149 static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
150 
151 static void udbg_rtascon_putc(char c)
152 {
153 	int tries;
154 
155 	if (!rtas.base)
156 		return;
157 
158 	/* Add CRs before LFs */
159 	if (c == '\n')
160 		udbg_rtascon_putc('\r');
161 
162 	/* if there is more than one character to be displayed, wait a bit */
163 	for (tries = 0; tries < 16; tries++) {
164 		if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
165 			break;
166 		udelay(1000);
167 	}
168 }
169 
170 static int udbg_rtascon_getc_poll(void)
171 {
172 	int c;
173 
174 	if (!rtas.base)
175 		return -1;
176 
177 	if (rtas_call(rtas_getchar_token, 0, 2, &c))
178 		return -1;
179 
180 	return c;
181 }
182 
183 static int udbg_rtascon_getc(void)
184 {
185 	int c;
186 
187 	while ((c = udbg_rtascon_getc_poll()) == -1)
188 		;
189 
190 	return c;
191 }
192 
193 
194 void __init udbg_init_rtas_console(void)
195 {
196 	udbg_putc = udbg_rtascon_putc;
197 	udbg_getc = udbg_rtascon_getc;
198 	udbg_getc_poll = udbg_rtascon_getc_poll;
199 }
200 #endif /* CONFIG_UDBG_RTAS_CONSOLE */
201 
202 void rtas_progress(char *s, unsigned short hex)
203 {
204 	struct device_node *root;
205 	int width;
206 	const int *p;
207 	char *os;
208 	static int display_character, set_indicator;
209 	static int display_width, display_lines, form_feed;
210 	static const int *row_width;
211 	static DEFINE_SPINLOCK(progress_lock);
212 	static int current_line;
213 	static int pending_newline = 0;  /* did last write end with unprinted newline? */
214 
215 	if (!rtas.base)
216 		return;
217 
218 	if (display_width == 0) {
219 		display_width = 0x10;
220 		if ((root = of_find_node_by_path("/rtas"))) {
221 			if ((p = of_get_property(root,
222 					"ibm,display-line-length", NULL)))
223 				display_width = *p;
224 			if ((p = of_get_property(root,
225 					"ibm,form-feed", NULL)))
226 				form_feed = *p;
227 			if ((p = of_get_property(root,
228 					"ibm,display-number-of-lines", NULL)))
229 				display_lines = *p;
230 			row_width = of_get_property(root,
231 					"ibm,display-truncation-length", NULL);
232 			of_node_put(root);
233 		}
234 		display_character = rtas_token("display-character");
235 		set_indicator = rtas_token("set-indicator");
236 	}
237 
238 	if (display_character == RTAS_UNKNOWN_SERVICE) {
239 		/* use hex display if available */
240 		if (set_indicator != RTAS_UNKNOWN_SERVICE)
241 			rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
242 		return;
243 	}
244 
245 	spin_lock(&progress_lock);
246 
247 	/*
248 	 * Last write ended with newline, but we didn't print it since
249 	 * it would just clear the bottom line of output. Print it now
250 	 * instead.
251 	 *
252 	 * If no newline is pending and form feed is supported, clear the
253 	 * display with a form feed; otherwise, print a CR to start output
254 	 * at the beginning of the line.
255 	 */
256 	if (pending_newline) {
257 		rtas_call(display_character, 1, 1, NULL, '\r');
258 		rtas_call(display_character, 1, 1, NULL, '\n');
259 		pending_newline = 0;
260 	} else {
261 		current_line = 0;
262 		if (form_feed)
263 			rtas_call(display_character, 1, 1, NULL,
264 				  (char)form_feed);
265 		else
266 			rtas_call(display_character, 1, 1, NULL, '\r');
267 	}
268 
269 	if (row_width)
270 		width = row_width[current_line];
271 	else
272 		width = display_width;
273 	os = s;
274 	while (*os) {
275 		if (*os == '\n' || *os == '\r') {
276 			/* If newline is the last character, save it
277 			 * until next call to avoid bumping up the
278 			 * display output.
279 			 */
280 			if (*os == '\n' && !os[1]) {
281 				pending_newline = 1;
282 				current_line++;
283 				if (current_line > display_lines-1)
284 					current_line = display_lines-1;
285 				spin_unlock(&progress_lock);
286 				return;
287 			}
288 
289 			/* RTAS wants CR-LF, not just LF */
290 
291 			if (*os == '\n') {
292 				rtas_call(display_character, 1, 1, NULL, '\r');
293 				rtas_call(display_character, 1, 1, NULL, '\n');
294 			} else {
295 				/* CR might be used to re-draw a line, so we'll
296 				 * leave it alone and not add LF.
297 				 */
298 				rtas_call(display_character, 1, 1, NULL, *os);
299 			}
300 
301 			if (row_width)
302 				width = row_width[current_line];
303 			else
304 				width = display_width;
305 		} else {
306 			width--;
307 			rtas_call(display_character, 1, 1, NULL, *os);
308 		}
309 
310 		os++;
311 
312 		/* if we overwrite the screen length */
313 		if (width <= 0)
314 			while ((*os != 0) && (*os != '\n') && (*os != '\r'))
315 				os++;
316 	}
317 
318 	spin_unlock(&progress_lock);
319 }
320 EXPORT_SYMBOL(rtas_progress);		/* needed by rtas_flash module */
321 
322 int rtas_token(const char *service)
323 {
324 	const int *tokp;
325 	if (rtas.dev == NULL)
326 		return RTAS_UNKNOWN_SERVICE;
327 	tokp = of_get_property(rtas.dev, service, NULL);
328 	return tokp ? *tokp : RTAS_UNKNOWN_SERVICE;
329 }
330 EXPORT_SYMBOL(rtas_token);
331 
332 int rtas_service_present(const char *service)
333 {
334 	return rtas_token(service) != RTAS_UNKNOWN_SERVICE;
335 }
336 EXPORT_SYMBOL(rtas_service_present);
337 
338 #ifdef CONFIG_RTAS_ERROR_LOGGING
339 /*
340  * Return the firmware-specified size of the error log buffer
341  *  for all rtas calls that require an error buffer argument.
342  *  This includes 'check-exception' and 'rtas-last-error'.
343  */
344 int rtas_get_error_log_max(void)
345 {
346 	static int rtas_error_log_max;
347 	if (rtas_error_log_max)
348 		return rtas_error_log_max;
349 
350 	rtas_error_log_max = rtas_token ("rtas-error-log-max");
351 	if ((rtas_error_log_max == RTAS_UNKNOWN_SERVICE) ||
352 	    (rtas_error_log_max > RTAS_ERROR_LOG_MAX)) {
353 		printk (KERN_WARNING "RTAS: bad log buffer size %d\n",
354 			rtas_error_log_max);
355 		rtas_error_log_max = RTAS_ERROR_LOG_MAX;
356 	}
357 	return rtas_error_log_max;
358 }
359 EXPORT_SYMBOL(rtas_get_error_log_max);
360 
361 
362 static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
363 static int rtas_last_error_token;
364 
365 /** Return a copy of the detailed error text associated with the
366  *  most recent failed call to rtas.  Because the error text
367  *  might go stale if there are any other intervening rtas calls,
368  *  this routine must be called atomically with whatever produced
369  *  the error (i.e. with rtas.lock still held from the previous call).
370  */
371 static char *__fetch_rtas_last_error(char *altbuf)
372 {
373 	struct rtas_args err_args, save_args;
374 	u32 bufsz;
375 	char *buf = NULL;
376 
377 	if (rtas_last_error_token == -1)
378 		return NULL;
379 
380 	bufsz = rtas_get_error_log_max();
381 
382 	err_args.token = rtas_last_error_token;
383 	err_args.nargs = 2;
384 	err_args.nret = 1;
385 	err_args.args[0] = (rtas_arg_t)__pa(rtas_err_buf);
386 	err_args.args[1] = bufsz;
387 	err_args.args[2] = 0;
388 
389 	save_args = rtas.args;
390 	rtas.args = err_args;
391 
392 	enter_rtas(__pa(&rtas.args));
393 
394 	err_args = rtas.args;
395 	rtas.args = save_args;
396 
397 	/* Log the error in the unlikely case that there was one. */
398 	if (unlikely(err_args.args[2] == 0)) {
399 		if (altbuf) {
400 			buf = altbuf;
401 		} else {
402 			buf = rtas_err_buf;
403 			if (mem_init_done)
404 				buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
405 		}
406 		if (buf)
407 			memcpy(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
408 	}
409 
410 	return buf;
411 }
412 
413 #define get_errorlog_buffer()	kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
414 
415 #else /* CONFIG_RTAS_ERROR_LOGGING */
416 #define __fetch_rtas_last_error(x)	NULL
417 #define get_errorlog_buffer()		NULL
418 #endif
419 
420 int rtas_call(int token, int nargs, int nret, int *outputs, ...)
421 {
422 	va_list list;
423 	int i;
424 	unsigned long s;
425 	struct rtas_args *rtas_args;
426 	char *buff_copy = NULL;
427 	int ret;
428 
429 	if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
430 		return -1;
431 
432 	s = lock_rtas();
433 	rtas_args = &rtas.args;
434 
435 	rtas_args->token = token;
436 	rtas_args->nargs = nargs;
437 	rtas_args->nret  = nret;
438 	rtas_args->rets  = (rtas_arg_t *)&(rtas_args->args[nargs]);
439 	va_start(list, outputs);
440 	for (i = 0; i < nargs; ++i)
441 		rtas_args->args[i] = va_arg(list, rtas_arg_t);
442 	va_end(list);
443 
444 	for (i = 0; i < nret; ++i)
445 		rtas_args->rets[i] = 0;
446 
447 	enter_rtas(__pa(rtas_args));
448 
449 	/* A -1 return code indicates that the last command couldn't
450 	   be completed due to a hardware error. */
451 	if (rtas_args->rets[0] == -1)
452 		buff_copy = __fetch_rtas_last_error(NULL);
453 
454 	if (nret > 1 && outputs != NULL)
455 		for (i = 0; i < nret-1; ++i)
456 			outputs[i] = rtas_args->rets[i+1];
457 	ret = (nret > 0)? rtas_args->rets[0]: 0;
458 
459 	unlock_rtas(s);
460 
461 	if (buff_copy) {
462 		log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
463 		if (mem_init_done)
464 			kfree(buff_copy);
465 	}
466 	return ret;
467 }
468 EXPORT_SYMBOL(rtas_call);
469 
470 /* For RTAS_BUSY (-2), delay for 1 millisecond.  For an extended busy status
471  * code of 990n, perform the hinted delay of 10^n (last digit) milliseconds.
472  */
473 unsigned int rtas_busy_delay_time(int status)
474 {
475 	int order;
476 	unsigned int ms = 0;
477 
478 	if (status == RTAS_BUSY) {
479 		ms = 1;
480 	} else if (status >= 9900 && status <= 9905) {
481 		order = status - 9900;
482 		for (ms = 1; order > 0; order--)
483 			ms *= 10;
484 	}
485 
486 	return ms;
487 }
488 EXPORT_SYMBOL(rtas_busy_delay_time);
489 
490 /* For an RTAS busy status code, perform the hinted delay. */
491 unsigned int rtas_busy_delay(int status)
492 {
493 	unsigned int ms;
494 
495 	might_sleep();
496 	ms = rtas_busy_delay_time(status);
497 	if (ms && need_resched())
498 		msleep(ms);
499 
500 	return ms;
501 }
502 EXPORT_SYMBOL(rtas_busy_delay);
503 
504 static int rtas_error_rc(int rtas_rc)
505 {
506 	int rc;
507 
508 	switch (rtas_rc) {
509 		case -1: 		/* Hardware Error */
510 			rc = -EIO;
511 			break;
512 		case -3:		/* Bad indicator/domain/etc */
513 			rc = -EINVAL;
514 			break;
515 		case -9000:		/* Isolation error */
516 			rc = -EFAULT;
517 			break;
518 		case -9001:		/* Outstanding TCE/PTE */
519 			rc = -EEXIST;
520 			break;
521 		case -9002:		/* No usable slot */
522 			rc = -ENODEV;
523 			break;
524 		default:
525 			printk(KERN_ERR "%s: unexpected RTAS error %d\n",
526 					__func__, rtas_rc);
527 			rc = -ERANGE;
528 			break;
529 	}
530 	return rc;
531 }
532 
533 int rtas_get_power_level(int powerdomain, int *level)
534 {
535 	int token = rtas_token("get-power-level");
536 	int rc;
537 
538 	if (token == RTAS_UNKNOWN_SERVICE)
539 		return -ENOENT;
540 
541 	while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
542 		udelay(1);
543 
544 	if (rc < 0)
545 		return rtas_error_rc(rc);
546 	return rc;
547 }
548 EXPORT_SYMBOL(rtas_get_power_level);
549 
550 int rtas_set_power_level(int powerdomain, int level, int *setlevel)
551 {
552 	int token = rtas_token("set-power-level");
553 	int rc;
554 
555 	if (token == RTAS_UNKNOWN_SERVICE)
556 		return -ENOENT;
557 
558 	do {
559 		rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
560 	} while (rtas_busy_delay(rc));
561 
562 	if (rc < 0)
563 		return rtas_error_rc(rc);
564 	return rc;
565 }
566 EXPORT_SYMBOL(rtas_set_power_level);
567 
568 int rtas_get_sensor(int sensor, int index, int *state)
569 {
570 	int token = rtas_token("get-sensor-state");
571 	int rc;
572 
573 	if (token == RTAS_UNKNOWN_SERVICE)
574 		return -ENOENT;
575 
576 	do {
577 		rc = rtas_call(token, 2, 2, state, sensor, index);
578 	} while (rtas_busy_delay(rc));
579 
580 	if (rc < 0)
581 		return rtas_error_rc(rc);
582 	return rc;
583 }
584 EXPORT_SYMBOL(rtas_get_sensor);
585 
586 bool rtas_indicator_present(int token, int *maxindex)
587 {
588 	int proplen, count, i;
589 	const struct indicator_elem {
590 		u32 token;
591 		u32 maxindex;
592 	} *indicators;
593 
594 	indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
595 	if (!indicators)
596 		return false;
597 
598 	count = proplen / sizeof(struct indicator_elem);
599 
600 	for (i = 0; i < count; i++) {
601 		if (indicators[i].token != token)
602 			continue;
603 		if (maxindex)
604 			*maxindex = indicators[i].maxindex;
605 		return true;
606 	}
607 
608 	return false;
609 }
610 EXPORT_SYMBOL(rtas_indicator_present);
611 
612 int rtas_set_indicator(int indicator, int index, int new_value)
613 {
614 	int token = rtas_token("set-indicator");
615 	int rc;
616 
617 	if (token == RTAS_UNKNOWN_SERVICE)
618 		return -ENOENT;
619 
620 	do {
621 		rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
622 	} while (rtas_busy_delay(rc));
623 
624 	if (rc < 0)
625 		return rtas_error_rc(rc);
626 	return rc;
627 }
628 EXPORT_SYMBOL(rtas_set_indicator);
629 
630 /*
631  * Ignoring RTAS extended delay
632  */
633 int rtas_set_indicator_fast(int indicator, int index, int new_value)
634 {
635 	int rc;
636 	int token = rtas_token("set-indicator");
637 
638 	if (token == RTAS_UNKNOWN_SERVICE)
639 		return -ENOENT;
640 
641 	rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
642 
643 	WARN_ON(rc == -2 || (rc >= 9900 && rc <= 9905));
644 
645 	if (rc < 0)
646 		return rtas_error_rc(rc);
647 
648 	return rc;
649 }
650 
651 void rtas_restart(char *cmd)
652 {
653 	if (rtas_flash_term_hook)
654 		rtas_flash_term_hook(SYS_RESTART);
655 	printk("RTAS system-reboot returned %d\n",
656 	       rtas_call(rtas_token("system-reboot"), 0, 1, NULL));
657 	for (;;);
658 }
659 
660 void rtas_power_off(void)
661 {
662 	if (rtas_flash_term_hook)
663 		rtas_flash_term_hook(SYS_POWER_OFF);
664 	/* allow power on only with power button press */
665 	printk("RTAS power-off returned %d\n",
666 	       rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
667 	for (;;);
668 }
669 
670 void rtas_halt(void)
671 {
672 	if (rtas_flash_term_hook)
673 		rtas_flash_term_hook(SYS_HALT);
674 	/* allow power on only with power button press */
675 	printk("RTAS power-off returned %d\n",
676 	       rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
677 	for (;;);
678 }
679 
680 /* Must be in the RMO region, so we place it here */
681 static char rtas_os_term_buf[2048];
682 
683 void rtas_os_term(char *str)
684 {
685 	int status;
686 
687 	/*
688 	 * Firmware with the ibm,extended-os-term property is guaranteed
689 	 * to always return from an ibm,os-term call. Earlier versions without
690 	 * this property may terminate the partition which we want to avoid
691 	 * since it interferes with panic_timeout.
692 	 */
693 	if (RTAS_UNKNOWN_SERVICE == rtas_token("ibm,os-term") ||
694 	    RTAS_UNKNOWN_SERVICE == rtas_token("ibm,extended-os-term"))
695 		return;
696 
697 	snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
698 
699 	do {
700 		status = rtas_call(rtas_token("ibm,os-term"), 1, 1, NULL,
701 				   __pa(rtas_os_term_buf));
702 	} while (rtas_busy_delay(status));
703 
704 	if (status != 0)
705 		printk(KERN_EMERG "ibm,os-term call failed %d\n", status);
706 }
707 
708 static int ibm_suspend_me_token = RTAS_UNKNOWN_SERVICE;
709 #ifdef CONFIG_PPC_PSERIES
710 static int __rtas_suspend_last_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
711 {
712 	u16 slb_size = mmu_slb_size;
713 	int rc = H_MULTI_THREADS_ACTIVE;
714 	int cpu;
715 
716 	slb_set_size(SLB_MIN_SIZE);
717 	stop_topology_update();
718 	printk(KERN_DEBUG "calling ibm,suspend-me on cpu %i\n", smp_processor_id());
719 
720 	while (rc == H_MULTI_THREADS_ACTIVE && !atomic_read(&data->done) &&
721 	       !atomic_read(&data->error))
722 		rc = rtas_call(data->token, 0, 1, NULL);
723 
724 	if (rc || atomic_read(&data->error)) {
725 		printk(KERN_DEBUG "ibm,suspend-me returned %d\n", rc);
726 		slb_set_size(slb_size);
727 	}
728 
729 	if (atomic_read(&data->error))
730 		rc = atomic_read(&data->error);
731 
732 	atomic_set(&data->error, rc);
733 	start_topology_update();
734 
735 	if (wake_when_done) {
736 		atomic_set(&data->done, 1);
737 
738 		for_each_online_cpu(cpu)
739 			plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
740 	}
741 
742 	if (atomic_dec_return(&data->working) == 0)
743 		complete(data->complete);
744 
745 	return rc;
746 }
747 
748 int rtas_suspend_last_cpu(struct rtas_suspend_me_data *data)
749 {
750 	atomic_inc(&data->working);
751 	return __rtas_suspend_last_cpu(data, 0);
752 }
753 
754 static int __rtas_suspend_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
755 {
756 	long rc = H_SUCCESS;
757 	unsigned long msr_save;
758 	int cpu;
759 
760 	atomic_inc(&data->working);
761 
762 	/* really need to ensure MSR.EE is off for H_JOIN */
763 	msr_save = mfmsr();
764 	mtmsr(msr_save & ~(MSR_EE));
765 
766 	while (rc == H_SUCCESS && !atomic_read(&data->done) && !atomic_read(&data->error))
767 		rc = plpar_hcall_norets(H_JOIN);
768 
769 	mtmsr(msr_save);
770 
771 	if (rc == H_SUCCESS) {
772 		/* This cpu was prodded and the suspend is complete. */
773 		goto out;
774 	} else if (rc == H_CONTINUE) {
775 		/* All other cpus are in H_JOIN, this cpu does
776 		 * the suspend.
777 		 */
778 		return __rtas_suspend_last_cpu(data, wake_when_done);
779 	} else {
780 		printk(KERN_ERR "H_JOIN on cpu %i failed with rc = %ld\n",
781 		       smp_processor_id(), rc);
782 		atomic_set(&data->error, rc);
783 	}
784 
785 	if (wake_when_done) {
786 		atomic_set(&data->done, 1);
787 
788 		/* This cpu did the suspend or got an error; in either case,
789 		 * we need to prod all other other cpus out of join state.
790 		 * Extra prods are harmless.
791 		 */
792 		for_each_online_cpu(cpu)
793 			plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
794 	}
795 out:
796 	if (atomic_dec_return(&data->working) == 0)
797 		complete(data->complete);
798 	return rc;
799 }
800 
801 int rtas_suspend_cpu(struct rtas_suspend_me_data *data)
802 {
803 	return __rtas_suspend_cpu(data, 0);
804 }
805 
806 static void rtas_percpu_suspend_me(void *info)
807 {
808 	__rtas_suspend_cpu((struct rtas_suspend_me_data *)info, 1);
809 }
810 
811 int rtas_ibm_suspend_me(struct rtas_args *args)
812 {
813 	long state;
814 	long rc;
815 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
816 	struct rtas_suspend_me_data data;
817 	DECLARE_COMPLETION_ONSTACK(done);
818 
819 	if (!rtas_service_present("ibm,suspend-me"))
820 		return -ENOSYS;
821 
822 	/* Make sure the state is valid */
823 	rc = plpar_hcall(H_VASI_STATE, retbuf,
824 			 ((u64)args->args[0] << 32) | args->args[1]);
825 
826 	state = retbuf[0];
827 
828 	if (rc) {
829 		printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned %ld\n",rc);
830 		return rc;
831 	} else if (state == H_VASI_ENABLED) {
832 		args->args[args->nargs] = RTAS_NOT_SUSPENDABLE;
833 		return 0;
834 	} else if (state != H_VASI_SUSPENDING) {
835 		printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned state %ld\n",
836 		       state);
837 		args->args[args->nargs] = -1;
838 		return 0;
839 	}
840 
841 	atomic_set(&data.working, 0);
842 	atomic_set(&data.done, 0);
843 	atomic_set(&data.error, 0);
844 	data.token = rtas_token("ibm,suspend-me");
845 	data.complete = &done;
846 
847 	/* Call function on all CPUs.  One of us will make the
848 	 * rtas call
849 	 */
850 	if (on_each_cpu(rtas_percpu_suspend_me, &data, 0))
851 		atomic_set(&data.error, -EINVAL);
852 
853 	wait_for_completion(&done);
854 
855 	if (atomic_read(&data.error) != 0)
856 		printk(KERN_ERR "Error doing global join\n");
857 
858 	return atomic_read(&data.error);
859 }
860 #else /* CONFIG_PPC_PSERIES */
861 int rtas_ibm_suspend_me(struct rtas_args *args)
862 {
863 	return -ENOSYS;
864 }
865 #endif
866 
867 asmlinkage int ppc_rtas(struct rtas_args __user *uargs)
868 {
869 	struct rtas_args args;
870 	unsigned long flags;
871 	char *buff_copy, *errbuf = NULL;
872 	int nargs;
873 	int rc;
874 
875 	if (!capable(CAP_SYS_ADMIN))
876 		return -EPERM;
877 
878 	if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
879 		return -EFAULT;
880 
881 	nargs = args.nargs;
882 	if (nargs > ARRAY_SIZE(args.args)
883 	    || args.nret > ARRAY_SIZE(args.args)
884 	    || nargs + args.nret > ARRAY_SIZE(args.args))
885 		return -EINVAL;
886 
887 	/* Copy in args. */
888 	if (copy_from_user(args.args, uargs->args,
889 			   nargs * sizeof(rtas_arg_t)) != 0)
890 		return -EFAULT;
891 
892 	if (args.token == RTAS_UNKNOWN_SERVICE)
893 		return -EINVAL;
894 
895 	args.rets = &args.args[nargs];
896 	memset(args.rets, 0, args.nret * sizeof(rtas_arg_t));
897 
898 	/* Need to handle ibm,suspend_me call specially */
899 	if (args.token == ibm_suspend_me_token) {
900 		rc = rtas_ibm_suspend_me(&args);
901 		if (rc)
902 			return rc;
903 		goto copy_return;
904 	}
905 
906 	buff_copy = get_errorlog_buffer();
907 
908 	flags = lock_rtas();
909 
910 	rtas.args = args;
911 	enter_rtas(__pa(&rtas.args));
912 	args = rtas.args;
913 
914 	/* A -1 return code indicates that the last command couldn't
915 	   be completed due to a hardware error. */
916 	if (args.rets[0] == -1)
917 		errbuf = __fetch_rtas_last_error(buff_copy);
918 
919 	unlock_rtas(flags);
920 
921 	if (buff_copy) {
922 		if (errbuf)
923 			log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
924 		kfree(buff_copy);
925 	}
926 
927  copy_return:
928 	/* Copy out args. */
929 	if (copy_to_user(uargs->args + nargs,
930 			 args.args + nargs,
931 			 args.nret * sizeof(rtas_arg_t)) != 0)
932 		return -EFAULT;
933 
934 	return 0;
935 }
936 
937 /*
938  * Call early during boot, before mem init or bootmem, to retrieve the RTAS
939  * informations from the device-tree and allocate the RMO buffer for userland
940  * accesses.
941  */
942 void __init rtas_initialize(void)
943 {
944 	unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
945 
946 	/* Get RTAS dev node and fill up our "rtas" structure with infos
947 	 * about it.
948 	 */
949 	rtas.dev = of_find_node_by_name(NULL, "rtas");
950 	if (rtas.dev) {
951 		const u32 *basep, *entryp, *sizep;
952 
953 		basep = of_get_property(rtas.dev, "linux,rtas-base", NULL);
954 		sizep = of_get_property(rtas.dev, "rtas-size", NULL);
955 		if (basep != NULL && sizep != NULL) {
956 			rtas.base = *basep;
957 			rtas.size = *sizep;
958 			entryp = of_get_property(rtas.dev,
959 					"linux,rtas-entry", NULL);
960 			if (entryp == NULL) /* Ugh */
961 				rtas.entry = rtas.base;
962 			else
963 				rtas.entry = *entryp;
964 		} else
965 			rtas.dev = NULL;
966 	}
967 	if (!rtas.dev)
968 		return;
969 
970 	/* If RTAS was found, allocate the RMO buffer for it and look for
971 	 * the stop-self token if any
972 	 */
973 #ifdef CONFIG_PPC64
974 	if (machine_is(pseries) && firmware_has_feature(FW_FEATURE_LPAR)) {
975 		rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
976 		ibm_suspend_me_token = rtas_token("ibm,suspend-me");
977 	}
978 #endif
979 	rtas_rmo_buf = memblock_alloc_base(RTAS_RMOBUF_MAX, PAGE_SIZE, rtas_region);
980 
981 #ifdef CONFIG_RTAS_ERROR_LOGGING
982 	rtas_last_error_token = rtas_token("rtas-last-error");
983 #endif
984 }
985 
986 int __init early_init_dt_scan_rtas(unsigned long node,
987 		const char *uname, int depth, void *data)
988 {
989 	u32 *basep, *entryp, *sizep;
990 
991 	if (depth != 1 || strcmp(uname, "rtas") != 0)
992 		return 0;
993 
994 	basep  = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
995 	entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
996 	sizep  = of_get_flat_dt_prop(node, "rtas-size", NULL);
997 
998 	if (basep && entryp && sizep) {
999 		rtas.base = *basep;
1000 		rtas.entry = *entryp;
1001 		rtas.size = *sizep;
1002 	}
1003 
1004 #ifdef CONFIG_UDBG_RTAS_CONSOLE
1005 	basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
1006 	if (basep)
1007 		rtas_putchar_token = *basep;
1008 
1009 	basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
1010 	if (basep)
1011 		rtas_getchar_token = *basep;
1012 
1013 	if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
1014 	    rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
1015 		udbg_init_rtas_console();
1016 
1017 #endif
1018 
1019 	/* break now */
1020 	return 1;
1021 }
1022 
1023 static arch_spinlock_t timebase_lock;
1024 static u64 timebase = 0;
1025 
1026 void __cpuinit rtas_give_timebase(void)
1027 {
1028 	unsigned long flags;
1029 
1030 	local_irq_save(flags);
1031 	hard_irq_disable();
1032 	arch_spin_lock(&timebase_lock);
1033 	rtas_call(rtas_token("freeze-time-base"), 0, 1, NULL);
1034 	timebase = get_tb();
1035 	arch_spin_unlock(&timebase_lock);
1036 
1037 	while (timebase)
1038 		barrier();
1039 	rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
1040 	local_irq_restore(flags);
1041 }
1042 
1043 void __cpuinit rtas_take_timebase(void)
1044 {
1045 	while (!timebase)
1046 		barrier();
1047 	arch_spin_lock(&timebase_lock);
1048 	set_tb(timebase >> 32, timebase & 0xffffffff);
1049 	timebase = 0;
1050 	arch_spin_unlock(&timebase_lock);
1051 }
1052