xref: /openbmc/linux/arch/powerpc/kernel/rtas.c (revision c5f41752)
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 #include <linux/reboot.h>
28 
29 #include <asm/prom.h>
30 #include <asm/rtas.h>
31 #include <asm/hvcall.h>
32 #include <asm/machdep.h>
33 #include <asm/firmware.h>
34 #include <asm/page.h>
35 #include <asm/param.h>
36 #include <asm/system.h>
37 #include <asm/delay.h>
38 #include <asm/uaccess.h>
39 #include <asm/udbg.h>
40 #include <asm/syscalls.h>
41 #include <asm/smp.h>
42 #include <asm/atomic.h>
43 #include <asm/time.h>
44 #include <asm/mmu.h>
45 #include <asm/topology.h>
46 #include <asm/pSeries_reconfig.h>
47 
48 struct rtas_t rtas = {
49 	.lock = __ARCH_SPIN_LOCK_UNLOCKED
50 };
51 EXPORT_SYMBOL(rtas);
52 
53 DEFINE_SPINLOCK(rtas_data_buf_lock);
54 EXPORT_SYMBOL(rtas_data_buf_lock);
55 
56 char rtas_data_buf[RTAS_DATA_BUF_SIZE] __cacheline_aligned;
57 EXPORT_SYMBOL(rtas_data_buf);
58 
59 unsigned long rtas_rmo_buf;
60 
61 /*
62  * If non-NULL, this gets called when the kernel terminates.
63  * This is done like this so rtas_flash can be a module.
64  */
65 void (*rtas_flash_term_hook)(int);
66 EXPORT_SYMBOL(rtas_flash_term_hook);
67 
68 /* RTAS use home made raw locking instead of spin_lock_irqsave
69  * because those can be called from within really nasty contexts
70  * such as having the timebase stopped which would lockup with
71  * normal locks and spinlock debugging enabled
72  */
73 static unsigned long lock_rtas(void)
74 {
75 	unsigned long flags;
76 
77 	local_irq_save(flags);
78 	preempt_disable();
79 	arch_spin_lock_flags(&rtas.lock, flags);
80 	return flags;
81 }
82 
83 static void unlock_rtas(unsigned long flags)
84 {
85 	arch_spin_unlock(&rtas.lock);
86 	local_irq_restore(flags);
87 	preempt_enable();
88 }
89 
90 /*
91  * call_rtas_display_status and call_rtas_display_status_delay
92  * are designed only for very early low-level debugging, which
93  * is why the token is hard-coded to 10.
94  */
95 static void call_rtas_display_status(char c)
96 {
97 	struct rtas_args *args = &rtas.args;
98 	unsigned long s;
99 
100 	if (!rtas.base)
101 		return;
102 	s = lock_rtas();
103 
104 	args->token = 10;
105 	args->nargs = 1;
106 	args->nret  = 1;
107 	args->rets  = (rtas_arg_t *)&(args->args[1]);
108 	args->args[0] = (unsigned char)c;
109 
110 	enter_rtas(__pa(args));
111 
112 	unlock_rtas(s);
113 }
114 
115 static void call_rtas_display_status_delay(char c)
116 {
117 	static int pending_newline = 0;  /* did last write end with unprinted newline? */
118 	static int width = 16;
119 
120 	if (c == '\n') {
121 		while (width-- > 0)
122 			call_rtas_display_status(' ');
123 		width = 16;
124 		mdelay(500);
125 		pending_newline = 1;
126 	} else {
127 		if (pending_newline) {
128 			call_rtas_display_status('\r');
129 			call_rtas_display_status('\n');
130 		}
131 		pending_newline = 0;
132 		if (width--) {
133 			call_rtas_display_status(c);
134 			udelay(10000);
135 		}
136 	}
137 }
138 
139 void __init udbg_init_rtas_panel(void)
140 {
141 	udbg_putc = call_rtas_display_status_delay;
142 }
143 
144 #ifdef CONFIG_UDBG_RTAS_CONSOLE
145 
146 /* If you think you're dying before early_init_dt_scan_rtas() does its
147  * work, you can hard code the token values for your firmware here and
148  * hardcode rtas.base/entry etc.
149  */
150 static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
151 static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
152 
153 static void udbg_rtascon_putc(char c)
154 {
155 	int tries;
156 
157 	if (!rtas.base)
158 		return;
159 
160 	/* Add CRs before LFs */
161 	if (c == '\n')
162 		udbg_rtascon_putc('\r');
163 
164 	/* if there is more than one character to be displayed, wait a bit */
165 	for (tries = 0; tries < 16; tries++) {
166 		if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
167 			break;
168 		udelay(1000);
169 	}
170 }
171 
172 static int udbg_rtascon_getc_poll(void)
173 {
174 	int c;
175 
176 	if (!rtas.base)
177 		return -1;
178 
179 	if (rtas_call(rtas_getchar_token, 0, 2, &c))
180 		return -1;
181 
182 	return c;
183 }
184 
185 static int udbg_rtascon_getc(void)
186 {
187 	int c;
188 
189 	while ((c = udbg_rtascon_getc_poll()) == -1)
190 		;
191 
192 	return c;
193 }
194 
195 
196 void __init udbg_init_rtas_console(void)
197 {
198 	udbg_putc = udbg_rtascon_putc;
199 	udbg_getc = udbg_rtascon_getc;
200 	udbg_getc_poll = udbg_rtascon_getc_poll;
201 }
202 #endif /* CONFIG_UDBG_RTAS_CONSOLE */
203 
204 void rtas_progress(char *s, unsigned short hex)
205 {
206 	struct device_node *root;
207 	int width;
208 	const int *p;
209 	char *os;
210 	static int display_character, set_indicator;
211 	static int display_width, display_lines, form_feed;
212 	static const int *row_width;
213 	static DEFINE_SPINLOCK(progress_lock);
214 	static int current_line;
215 	static int pending_newline = 0;  /* did last write end with unprinted newline? */
216 
217 	if (!rtas.base)
218 		return;
219 
220 	if (display_width == 0) {
221 		display_width = 0x10;
222 		if ((root = of_find_node_by_path("/rtas"))) {
223 			if ((p = of_get_property(root,
224 					"ibm,display-line-length", NULL)))
225 				display_width = *p;
226 			if ((p = of_get_property(root,
227 					"ibm,form-feed", NULL)))
228 				form_feed = *p;
229 			if ((p = of_get_property(root,
230 					"ibm,display-number-of-lines", NULL)))
231 				display_lines = *p;
232 			row_width = of_get_property(root,
233 					"ibm,display-truncation-length", NULL);
234 			of_node_put(root);
235 		}
236 		display_character = rtas_token("display-character");
237 		set_indicator = rtas_token("set-indicator");
238 	}
239 
240 	if (display_character == RTAS_UNKNOWN_SERVICE) {
241 		/* use hex display if available */
242 		if (set_indicator != RTAS_UNKNOWN_SERVICE)
243 			rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
244 		return;
245 	}
246 
247 	spin_lock(&progress_lock);
248 
249 	/*
250 	 * Last write ended with newline, but we didn't print it since
251 	 * it would just clear the bottom line of output. Print it now
252 	 * instead.
253 	 *
254 	 * If no newline is pending and form feed is supported, clear the
255 	 * display with a form feed; otherwise, print a CR to start output
256 	 * at the beginning of the line.
257 	 */
258 	if (pending_newline) {
259 		rtas_call(display_character, 1, 1, NULL, '\r');
260 		rtas_call(display_character, 1, 1, NULL, '\n');
261 		pending_newline = 0;
262 	} else {
263 		current_line = 0;
264 		if (form_feed)
265 			rtas_call(display_character, 1, 1, NULL,
266 				  (char)form_feed);
267 		else
268 			rtas_call(display_character, 1, 1, NULL, '\r');
269 	}
270 
271 	if (row_width)
272 		width = row_width[current_line];
273 	else
274 		width = display_width;
275 	os = s;
276 	while (*os) {
277 		if (*os == '\n' || *os == '\r') {
278 			/* If newline is the last character, save it
279 			 * until next call to avoid bumping up the
280 			 * display output.
281 			 */
282 			if (*os == '\n' && !os[1]) {
283 				pending_newline = 1;
284 				current_line++;
285 				if (current_line > display_lines-1)
286 					current_line = display_lines-1;
287 				spin_unlock(&progress_lock);
288 				return;
289 			}
290 
291 			/* RTAS wants CR-LF, not just LF */
292 
293 			if (*os == '\n') {
294 				rtas_call(display_character, 1, 1, NULL, '\r');
295 				rtas_call(display_character, 1, 1, NULL, '\n');
296 			} else {
297 				/* CR might be used to re-draw a line, so we'll
298 				 * leave it alone and not add LF.
299 				 */
300 				rtas_call(display_character, 1, 1, NULL, *os);
301 			}
302 
303 			if (row_width)
304 				width = row_width[current_line];
305 			else
306 				width = display_width;
307 		} else {
308 			width--;
309 			rtas_call(display_character, 1, 1, NULL, *os);
310 		}
311 
312 		os++;
313 
314 		/* if we overwrite the screen length */
315 		if (width <= 0)
316 			while ((*os != 0) && (*os != '\n') && (*os != '\r'))
317 				os++;
318 	}
319 
320 	spin_unlock(&progress_lock);
321 }
322 EXPORT_SYMBOL(rtas_progress);		/* needed by rtas_flash module */
323 
324 int rtas_token(const char *service)
325 {
326 	const int *tokp;
327 	if (rtas.dev == NULL)
328 		return RTAS_UNKNOWN_SERVICE;
329 	tokp = of_get_property(rtas.dev, service, NULL);
330 	return tokp ? *tokp : RTAS_UNKNOWN_SERVICE;
331 }
332 EXPORT_SYMBOL(rtas_token);
333 
334 int rtas_service_present(const char *service)
335 {
336 	return rtas_token(service) != RTAS_UNKNOWN_SERVICE;
337 }
338 EXPORT_SYMBOL(rtas_service_present);
339 
340 #ifdef CONFIG_RTAS_ERROR_LOGGING
341 /*
342  * Return the firmware-specified size of the error log buffer
343  *  for all rtas calls that require an error buffer argument.
344  *  This includes 'check-exception' and 'rtas-last-error'.
345  */
346 int rtas_get_error_log_max(void)
347 {
348 	static int rtas_error_log_max;
349 	if (rtas_error_log_max)
350 		return rtas_error_log_max;
351 
352 	rtas_error_log_max = rtas_token ("rtas-error-log-max");
353 	if ((rtas_error_log_max == RTAS_UNKNOWN_SERVICE) ||
354 	    (rtas_error_log_max > RTAS_ERROR_LOG_MAX)) {
355 		printk (KERN_WARNING "RTAS: bad log buffer size %d\n",
356 			rtas_error_log_max);
357 		rtas_error_log_max = RTAS_ERROR_LOG_MAX;
358 	}
359 	return rtas_error_log_max;
360 }
361 EXPORT_SYMBOL(rtas_get_error_log_max);
362 
363 
364 static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
365 static int rtas_last_error_token;
366 
367 /** Return a copy of the detailed error text associated with the
368  *  most recent failed call to rtas.  Because the error text
369  *  might go stale if there are any other intervening rtas calls,
370  *  this routine must be called atomically with whatever produced
371  *  the error (i.e. with rtas.lock still held from the previous call).
372  */
373 static char *__fetch_rtas_last_error(char *altbuf)
374 {
375 	struct rtas_args err_args, save_args;
376 	u32 bufsz;
377 	char *buf = NULL;
378 
379 	if (rtas_last_error_token == -1)
380 		return NULL;
381 
382 	bufsz = rtas_get_error_log_max();
383 
384 	err_args.token = rtas_last_error_token;
385 	err_args.nargs = 2;
386 	err_args.nret = 1;
387 	err_args.args[0] = (rtas_arg_t)__pa(rtas_err_buf);
388 	err_args.args[1] = bufsz;
389 	err_args.args[2] = 0;
390 
391 	save_args = rtas.args;
392 	rtas.args = err_args;
393 
394 	enter_rtas(__pa(&rtas.args));
395 
396 	err_args = rtas.args;
397 	rtas.args = save_args;
398 
399 	/* Log the error in the unlikely case that there was one. */
400 	if (unlikely(err_args.args[2] == 0)) {
401 		if (altbuf) {
402 			buf = altbuf;
403 		} else {
404 			buf = rtas_err_buf;
405 			if (mem_init_done)
406 				buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
407 		}
408 		if (buf)
409 			memcpy(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
410 	}
411 
412 	return buf;
413 }
414 
415 #define get_errorlog_buffer()	kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
416 
417 #else /* CONFIG_RTAS_ERROR_LOGGING */
418 #define __fetch_rtas_last_error(x)	NULL
419 #define get_errorlog_buffer()		NULL
420 #endif
421 
422 int rtas_call(int token, int nargs, int nret, int *outputs, ...)
423 {
424 	va_list list;
425 	int i;
426 	unsigned long s;
427 	struct rtas_args *rtas_args;
428 	char *buff_copy = NULL;
429 	int ret;
430 
431 	if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
432 		return -1;
433 
434 	s = lock_rtas();
435 	rtas_args = &rtas.args;
436 
437 	rtas_args->token = token;
438 	rtas_args->nargs = nargs;
439 	rtas_args->nret  = nret;
440 	rtas_args->rets  = (rtas_arg_t *)&(rtas_args->args[nargs]);
441 	va_start(list, outputs);
442 	for (i = 0; i < nargs; ++i)
443 		rtas_args->args[i] = va_arg(list, rtas_arg_t);
444 	va_end(list);
445 
446 	for (i = 0; i < nret; ++i)
447 		rtas_args->rets[i] = 0;
448 
449 	enter_rtas(__pa(rtas_args));
450 
451 	/* A -1 return code indicates that the last command couldn't
452 	   be completed due to a hardware error. */
453 	if (rtas_args->rets[0] == -1)
454 		buff_copy = __fetch_rtas_last_error(NULL);
455 
456 	if (nret > 1 && outputs != NULL)
457 		for (i = 0; i < nret-1; ++i)
458 			outputs[i] = rtas_args->rets[i+1];
459 	ret = (nret > 0)? rtas_args->rets[0]: 0;
460 
461 	unlock_rtas(s);
462 
463 	if (buff_copy) {
464 		log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
465 		if (mem_init_done)
466 			kfree(buff_copy);
467 	}
468 	return ret;
469 }
470 EXPORT_SYMBOL(rtas_call);
471 
472 /* For RTAS_BUSY (-2), delay for 1 millisecond.  For an extended busy status
473  * code of 990n, perform the hinted delay of 10^n (last digit) milliseconds.
474  */
475 unsigned int rtas_busy_delay_time(int status)
476 {
477 	int order;
478 	unsigned int ms = 0;
479 
480 	if (status == RTAS_BUSY) {
481 		ms = 1;
482 	} else if (status >= 9900 && status <= 9905) {
483 		order = status - 9900;
484 		for (ms = 1; order > 0; order--)
485 			ms *= 10;
486 	}
487 
488 	return ms;
489 }
490 EXPORT_SYMBOL(rtas_busy_delay_time);
491 
492 /* For an RTAS busy status code, perform the hinted delay. */
493 unsigned int rtas_busy_delay(int status)
494 {
495 	unsigned int ms;
496 
497 	might_sleep();
498 	ms = rtas_busy_delay_time(status);
499 	if (ms && need_resched())
500 		msleep(ms);
501 
502 	return ms;
503 }
504 EXPORT_SYMBOL(rtas_busy_delay);
505 
506 static int rtas_error_rc(int rtas_rc)
507 {
508 	int rc;
509 
510 	switch (rtas_rc) {
511 		case -1: 		/* Hardware Error */
512 			rc = -EIO;
513 			break;
514 		case -3:		/* Bad indicator/domain/etc */
515 			rc = -EINVAL;
516 			break;
517 		case -9000:		/* Isolation error */
518 			rc = -EFAULT;
519 			break;
520 		case -9001:		/* Outstanding TCE/PTE */
521 			rc = -EEXIST;
522 			break;
523 		case -9002:		/* No usable slot */
524 			rc = -ENODEV;
525 			break;
526 		default:
527 			printk(KERN_ERR "%s: unexpected RTAS error %d\n",
528 					__func__, rtas_rc);
529 			rc = -ERANGE;
530 			break;
531 	}
532 	return rc;
533 }
534 
535 int rtas_get_power_level(int powerdomain, int *level)
536 {
537 	int token = rtas_token("get-power-level");
538 	int rc;
539 
540 	if (token == RTAS_UNKNOWN_SERVICE)
541 		return -ENOENT;
542 
543 	while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
544 		udelay(1);
545 
546 	if (rc < 0)
547 		return rtas_error_rc(rc);
548 	return rc;
549 }
550 EXPORT_SYMBOL(rtas_get_power_level);
551 
552 int rtas_set_power_level(int powerdomain, int level, int *setlevel)
553 {
554 	int token = rtas_token("set-power-level");
555 	int rc;
556 
557 	if (token == RTAS_UNKNOWN_SERVICE)
558 		return -ENOENT;
559 
560 	do {
561 		rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
562 	} while (rtas_busy_delay(rc));
563 
564 	if (rc < 0)
565 		return rtas_error_rc(rc);
566 	return rc;
567 }
568 EXPORT_SYMBOL(rtas_set_power_level);
569 
570 int rtas_get_sensor(int sensor, int index, int *state)
571 {
572 	int token = rtas_token("get-sensor-state");
573 	int rc;
574 
575 	if (token == RTAS_UNKNOWN_SERVICE)
576 		return -ENOENT;
577 
578 	do {
579 		rc = rtas_call(token, 2, 2, state, sensor, index);
580 	} while (rtas_busy_delay(rc));
581 
582 	if (rc < 0)
583 		return rtas_error_rc(rc);
584 	return rc;
585 }
586 EXPORT_SYMBOL(rtas_get_sensor);
587 
588 bool rtas_indicator_present(int token, int *maxindex)
589 {
590 	int proplen, count, i;
591 	const struct indicator_elem {
592 		u32 token;
593 		u32 maxindex;
594 	} *indicators;
595 
596 	indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
597 	if (!indicators)
598 		return false;
599 
600 	count = proplen / sizeof(struct indicator_elem);
601 
602 	for (i = 0; i < count; i++) {
603 		if (indicators[i].token != token)
604 			continue;
605 		if (maxindex)
606 			*maxindex = indicators[i].maxindex;
607 		return true;
608 	}
609 
610 	return false;
611 }
612 EXPORT_SYMBOL(rtas_indicator_present);
613 
614 int rtas_set_indicator(int indicator, int index, int new_value)
615 {
616 	int token = rtas_token("set-indicator");
617 	int rc;
618 
619 	if (token == RTAS_UNKNOWN_SERVICE)
620 		return -ENOENT;
621 
622 	do {
623 		rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
624 	} while (rtas_busy_delay(rc));
625 
626 	if (rc < 0)
627 		return rtas_error_rc(rc);
628 	return rc;
629 }
630 EXPORT_SYMBOL(rtas_set_indicator);
631 
632 /*
633  * Ignoring RTAS extended delay
634  */
635 int rtas_set_indicator_fast(int indicator, int index, int new_value)
636 {
637 	int rc;
638 	int token = rtas_token("set-indicator");
639 
640 	if (token == RTAS_UNKNOWN_SERVICE)
641 		return -ENOENT;
642 
643 	rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
644 
645 	WARN_ON(rc == -2 || (rc >= 9900 && rc <= 9905));
646 
647 	if (rc < 0)
648 		return rtas_error_rc(rc);
649 
650 	return rc;
651 }
652 
653 void rtas_restart(char *cmd)
654 {
655 	if (rtas_flash_term_hook)
656 		rtas_flash_term_hook(SYS_RESTART);
657 	printk("RTAS system-reboot returned %d\n",
658 	       rtas_call(rtas_token("system-reboot"), 0, 1, NULL));
659 	for (;;);
660 }
661 
662 void rtas_power_off(void)
663 {
664 	if (rtas_flash_term_hook)
665 		rtas_flash_term_hook(SYS_POWER_OFF);
666 	/* allow power on only with power button press */
667 	printk("RTAS power-off returned %d\n",
668 	       rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
669 	for (;;);
670 }
671 
672 void rtas_halt(void)
673 {
674 	if (rtas_flash_term_hook)
675 		rtas_flash_term_hook(SYS_HALT);
676 	/* allow power on only with power button press */
677 	printk("RTAS power-off returned %d\n",
678 	       rtas_call(rtas_token("power-off"), 2, 1, NULL, -1, -1));
679 	for (;;);
680 }
681 
682 /* Must be in the RMO region, so we place it here */
683 static char rtas_os_term_buf[2048];
684 
685 void rtas_os_term(char *str)
686 {
687 	int status;
688 
689 	/*
690 	 * Firmware with the ibm,extended-os-term property is guaranteed
691 	 * to always return from an ibm,os-term call. Earlier versions without
692 	 * this property may terminate the partition which we want to avoid
693 	 * since it interferes with panic_timeout.
694 	 */
695 	if (RTAS_UNKNOWN_SERVICE == rtas_token("ibm,os-term") ||
696 	    RTAS_UNKNOWN_SERVICE == rtas_token("ibm,extended-os-term"))
697 		return;
698 
699 	snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
700 
701 	do {
702 		status = rtas_call(rtas_token("ibm,os-term"), 1, 1, NULL,
703 				   __pa(rtas_os_term_buf));
704 	} while (rtas_busy_delay(status));
705 
706 	if (status != 0)
707 		printk(KERN_EMERG "ibm,os-term call failed %d\n", status);
708 }
709 
710 static int ibm_suspend_me_token = RTAS_UNKNOWN_SERVICE;
711 #ifdef CONFIG_PPC_PSERIES
712 static int __rtas_suspend_last_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
713 {
714 	u16 slb_size = mmu_slb_size;
715 	int rc = H_MULTI_THREADS_ACTIVE;
716 	int cpu;
717 
718 	slb_set_size(SLB_MIN_SIZE);
719 	stop_topology_update();
720 	printk(KERN_DEBUG "calling ibm,suspend-me on cpu %i\n", smp_processor_id());
721 
722 	while (rc == H_MULTI_THREADS_ACTIVE && !atomic_read(&data->done) &&
723 	       !atomic_read(&data->error))
724 		rc = rtas_call(data->token, 0, 1, NULL);
725 
726 	if (rc || atomic_read(&data->error)) {
727 		printk(KERN_DEBUG "ibm,suspend-me returned %d\n", rc);
728 		slb_set_size(slb_size);
729 	}
730 
731 	if (atomic_read(&data->error))
732 		rc = atomic_read(&data->error);
733 
734 	atomic_set(&data->error, rc);
735 	start_topology_update();
736 	pSeries_coalesce_init();
737 
738 	if (wake_when_done) {
739 		atomic_set(&data->done, 1);
740 
741 		for_each_online_cpu(cpu)
742 			plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
743 	}
744 
745 	if (atomic_dec_return(&data->working) == 0)
746 		complete(data->complete);
747 
748 	return rc;
749 }
750 
751 int rtas_suspend_last_cpu(struct rtas_suspend_me_data *data)
752 {
753 	atomic_inc(&data->working);
754 	return __rtas_suspend_last_cpu(data, 0);
755 }
756 
757 static int __rtas_suspend_cpu(struct rtas_suspend_me_data *data, int wake_when_done)
758 {
759 	long rc = H_SUCCESS;
760 	unsigned long msr_save;
761 	int cpu;
762 
763 	atomic_inc(&data->working);
764 
765 	/* really need to ensure MSR.EE is off for H_JOIN */
766 	msr_save = mfmsr();
767 	mtmsr(msr_save & ~(MSR_EE));
768 
769 	while (rc == H_SUCCESS && !atomic_read(&data->done) && !atomic_read(&data->error))
770 		rc = plpar_hcall_norets(H_JOIN);
771 
772 	mtmsr(msr_save);
773 
774 	if (rc == H_SUCCESS) {
775 		/* This cpu was prodded and the suspend is complete. */
776 		goto out;
777 	} else if (rc == H_CONTINUE) {
778 		/* All other cpus are in H_JOIN, this cpu does
779 		 * the suspend.
780 		 */
781 		return __rtas_suspend_last_cpu(data, wake_when_done);
782 	} else {
783 		printk(KERN_ERR "H_JOIN on cpu %i failed with rc = %ld\n",
784 		       smp_processor_id(), rc);
785 		atomic_set(&data->error, rc);
786 	}
787 
788 	if (wake_when_done) {
789 		atomic_set(&data->done, 1);
790 
791 		/* This cpu did the suspend or got an error; in either case,
792 		 * we need to prod all other other cpus out of join state.
793 		 * Extra prods are harmless.
794 		 */
795 		for_each_online_cpu(cpu)
796 			plpar_hcall_norets(H_PROD, get_hard_smp_processor_id(cpu));
797 	}
798 out:
799 	if (atomic_dec_return(&data->working) == 0)
800 		complete(data->complete);
801 	return rc;
802 }
803 
804 int rtas_suspend_cpu(struct rtas_suspend_me_data *data)
805 {
806 	return __rtas_suspend_cpu(data, 0);
807 }
808 
809 static void rtas_percpu_suspend_me(void *info)
810 {
811 	__rtas_suspend_cpu((struct rtas_suspend_me_data *)info, 1);
812 }
813 
814 int rtas_ibm_suspend_me(struct rtas_args *args)
815 {
816 	long state;
817 	long rc;
818 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
819 	struct rtas_suspend_me_data data;
820 	DECLARE_COMPLETION_ONSTACK(done);
821 
822 	if (!rtas_service_present("ibm,suspend-me"))
823 		return -ENOSYS;
824 
825 	/* Make sure the state is valid */
826 	rc = plpar_hcall(H_VASI_STATE, retbuf,
827 			 ((u64)args->args[0] << 32) | args->args[1]);
828 
829 	state = retbuf[0];
830 
831 	if (rc) {
832 		printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned %ld\n",rc);
833 		return rc;
834 	} else if (state == H_VASI_ENABLED) {
835 		args->args[args->nargs] = RTAS_NOT_SUSPENDABLE;
836 		return 0;
837 	} else if (state != H_VASI_SUSPENDING) {
838 		printk(KERN_ERR "rtas_ibm_suspend_me: vasi_state returned state %ld\n",
839 		       state);
840 		args->args[args->nargs] = -1;
841 		return 0;
842 	}
843 
844 	atomic_set(&data.working, 0);
845 	atomic_set(&data.done, 0);
846 	atomic_set(&data.error, 0);
847 	data.token = rtas_token("ibm,suspend-me");
848 	data.complete = &done;
849 
850 	/* Call function on all CPUs.  One of us will make the
851 	 * rtas call
852 	 */
853 	if (on_each_cpu(rtas_percpu_suspend_me, &data, 0))
854 		atomic_set(&data.error, -EINVAL);
855 
856 	wait_for_completion(&done);
857 
858 	if (atomic_read(&data.error) != 0)
859 		printk(KERN_ERR "Error doing global join\n");
860 
861 	return atomic_read(&data.error);
862 }
863 #else /* CONFIG_PPC_PSERIES */
864 int rtas_ibm_suspend_me(struct rtas_args *args)
865 {
866 	return -ENOSYS;
867 }
868 #endif
869 
870 asmlinkage int ppc_rtas(struct rtas_args __user *uargs)
871 {
872 	struct rtas_args args;
873 	unsigned long flags;
874 	char *buff_copy, *errbuf = NULL;
875 	int nargs;
876 	int rc;
877 
878 	if (!capable(CAP_SYS_ADMIN))
879 		return -EPERM;
880 
881 	if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
882 		return -EFAULT;
883 
884 	nargs = args.nargs;
885 	if (nargs > ARRAY_SIZE(args.args)
886 	    || args.nret > ARRAY_SIZE(args.args)
887 	    || nargs + args.nret > ARRAY_SIZE(args.args))
888 		return -EINVAL;
889 
890 	/* Copy in args. */
891 	if (copy_from_user(args.args, uargs->args,
892 			   nargs * sizeof(rtas_arg_t)) != 0)
893 		return -EFAULT;
894 
895 	if (args.token == RTAS_UNKNOWN_SERVICE)
896 		return -EINVAL;
897 
898 	args.rets = &args.args[nargs];
899 	memset(args.rets, 0, args.nret * sizeof(rtas_arg_t));
900 
901 	/* Need to handle ibm,suspend_me call specially */
902 	if (args.token == ibm_suspend_me_token) {
903 		rc = rtas_ibm_suspend_me(&args);
904 		if (rc)
905 			return rc;
906 		goto copy_return;
907 	}
908 
909 	buff_copy = get_errorlog_buffer();
910 
911 	flags = lock_rtas();
912 
913 	rtas.args = args;
914 	enter_rtas(__pa(&rtas.args));
915 	args = rtas.args;
916 
917 	/* A -1 return code indicates that the last command couldn't
918 	   be completed due to a hardware error. */
919 	if (args.rets[0] == -1)
920 		errbuf = __fetch_rtas_last_error(buff_copy);
921 
922 	unlock_rtas(flags);
923 
924 	if (buff_copy) {
925 		if (errbuf)
926 			log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
927 		kfree(buff_copy);
928 	}
929 
930  copy_return:
931 	/* Copy out args. */
932 	if (copy_to_user(uargs->args + nargs,
933 			 args.args + nargs,
934 			 args.nret * sizeof(rtas_arg_t)) != 0)
935 		return -EFAULT;
936 
937 	return 0;
938 }
939 
940 /*
941  * Call early during boot, before mem init or bootmem, to retrieve the RTAS
942  * informations from the device-tree and allocate the RMO buffer for userland
943  * accesses.
944  */
945 void __init rtas_initialize(void)
946 {
947 	unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
948 
949 	/* Get RTAS dev node and fill up our "rtas" structure with infos
950 	 * about it.
951 	 */
952 	rtas.dev = of_find_node_by_name(NULL, "rtas");
953 	if (rtas.dev) {
954 		const u32 *basep, *entryp, *sizep;
955 
956 		basep = of_get_property(rtas.dev, "linux,rtas-base", NULL);
957 		sizep = of_get_property(rtas.dev, "rtas-size", NULL);
958 		if (basep != NULL && sizep != NULL) {
959 			rtas.base = *basep;
960 			rtas.size = *sizep;
961 			entryp = of_get_property(rtas.dev,
962 					"linux,rtas-entry", NULL);
963 			if (entryp == NULL) /* Ugh */
964 				rtas.entry = rtas.base;
965 			else
966 				rtas.entry = *entryp;
967 		} else
968 			rtas.dev = NULL;
969 	}
970 	if (!rtas.dev)
971 		return;
972 
973 	/* If RTAS was found, allocate the RMO buffer for it and look for
974 	 * the stop-self token if any
975 	 */
976 #ifdef CONFIG_PPC64
977 	if (machine_is(pseries) && firmware_has_feature(FW_FEATURE_LPAR)) {
978 		rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
979 		ibm_suspend_me_token = rtas_token("ibm,suspend-me");
980 	}
981 #endif
982 	rtas_rmo_buf = memblock_alloc_base(RTAS_RMOBUF_MAX, PAGE_SIZE, rtas_region);
983 
984 #ifdef CONFIG_RTAS_ERROR_LOGGING
985 	rtas_last_error_token = rtas_token("rtas-last-error");
986 #endif
987 }
988 
989 int __init early_init_dt_scan_rtas(unsigned long node,
990 		const char *uname, int depth, void *data)
991 {
992 	u32 *basep, *entryp, *sizep;
993 
994 	if (depth != 1 || strcmp(uname, "rtas") != 0)
995 		return 0;
996 
997 	basep  = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
998 	entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
999 	sizep  = of_get_flat_dt_prop(node, "rtas-size", NULL);
1000 
1001 	if (basep && entryp && sizep) {
1002 		rtas.base = *basep;
1003 		rtas.entry = *entryp;
1004 		rtas.size = *sizep;
1005 	}
1006 
1007 #ifdef CONFIG_UDBG_RTAS_CONSOLE
1008 	basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
1009 	if (basep)
1010 		rtas_putchar_token = *basep;
1011 
1012 	basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
1013 	if (basep)
1014 		rtas_getchar_token = *basep;
1015 
1016 	if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
1017 	    rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
1018 		udbg_init_rtas_console();
1019 
1020 #endif
1021 
1022 	/* break now */
1023 	return 1;
1024 }
1025 
1026 static arch_spinlock_t timebase_lock;
1027 static u64 timebase = 0;
1028 
1029 void __cpuinit rtas_give_timebase(void)
1030 {
1031 	unsigned long flags;
1032 
1033 	local_irq_save(flags);
1034 	hard_irq_disable();
1035 	arch_spin_lock(&timebase_lock);
1036 	rtas_call(rtas_token("freeze-time-base"), 0, 1, NULL);
1037 	timebase = get_tb();
1038 	arch_spin_unlock(&timebase_lock);
1039 
1040 	while (timebase)
1041 		barrier();
1042 	rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
1043 	local_irq_restore(flags);
1044 }
1045 
1046 void __cpuinit rtas_take_timebase(void)
1047 {
1048 	while (!timebase)
1049 		barrier();
1050 	arch_spin_lock(&timebase_lock);
1051 	set_tb(timebase >> 32, timebase & 0xffffffff);
1052 	timebase = 0;
1053 	arch_spin_unlock(&timebase_lock);
1054 }
1055