1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 *
4 * Procedures for interfacing to the RTAS on CHRP machines.
5 *
6 * Peter Bergner, IBM March 2001.
7 * Copyright (C) 2001 IBM.
8 */
9
10 #define pr_fmt(fmt) "rtas: " fmt
11
12 #include <linux/bsearch.h>
13 #include <linux/capability.h>
14 #include <linux/delay.h>
15 #include <linux/export.h>
16 #include <linux/init.h>
17 #include <linux/kconfig.h>
18 #include <linux/kernel.h>
19 #include <linux/lockdep.h>
20 #include <linux/memblock.h>
21 #include <linux/mutex.h>
22 #include <linux/nospec.h>
23 #include <linux/of.h>
24 #include <linux/of_fdt.h>
25 #include <linux/reboot.h>
26 #include <linux/sched.h>
27 #include <linux/security.h>
28 #include <linux/slab.h>
29 #include <linux/spinlock.h>
30 #include <linux/stdarg.h>
31 #include <linux/syscalls.h>
32 #include <linux/types.h>
33 #include <linux/uaccess.h>
34 #include <linux/xarray.h>
35
36 #include <asm/delay.h>
37 #include <asm/firmware.h>
38 #include <asm/interrupt.h>
39 #include <asm/machdep.h>
40 #include <asm/mmu.h>
41 #include <asm/page.h>
42 #include <asm/rtas-work-area.h>
43 #include <asm/rtas.h>
44 #include <asm/time.h>
45 #include <asm/trace.h>
46 #include <asm/udbg.h>
47
48 struct rtas_filter {
49 /* Indexes into the args buffer, -1 if not used */
50 const int buf_idx1;
51 const int size_idx1;
52 const int buf_idx2;
53 const int size_idx2;
54 /*
55 * Assumed buffer size per the spec if the function does not
56 * have a size parameter, e.g. ibm,errinjct. 0 if unused.
57 */
58 const int fixed_size;
59 };
60
61 /**
62 * struct rtas_function - Descriptor for RTAS functions.
63 *
64 * @token: Value of @name if it exists under the /rtas node.
65 * @name: Function name.
66 * @filter: If non-NULL, invoking this function via the rtas syscall is
67 * generally allowed, and @filter describes constraints on the
68 * arguments. See also @banned_for_syscall_on_le.
69 * @banned_for_syscall_on_le: Set when call via sys_rtas is generally allowed
70 * but specifically restricted on ppc64le. Such
71 * functions are believed to have no users on
72 * ppc64le, and we want to keep it that way. It does
73 * not make sense for this to be set when @filter
74 * is NULL.
75 * @lock: Pointer to an optional dedicated per-function mutex. This
76 * should be set for functions that require multiple calls in
77 * sequence to complete a single operation, and such sequences
78 * will disrupt each other if allowed to interleave. Users of
79 * this function are required to hold the associated lock for
80 * the duration of the call sequence. Add an explanatory
81 * comment to the function table entry if setting this member.
82 */
83 struct rtas_function {
84 s32 token;
85 const bool banned_for_syscall_on_le:1;
86 const char * const name;
87 const struct rtas_filter *filter;
88 struct mutex *lock;
89 };
90
91 /*
92 * Per-function locks for sequence-based RTAS functions.
93 */
94 static DEFINE_MUTEX(rtas_ibm_activate_firmware_lock);
95 static DEFINE_MUTEX(rtas_ibm_get_dynamic_sensor_state_lock);
96 static DEFINE_MUTEX(rtas_ibm_get_indices_lock);
97 static DEFINE_MUTEX(rtas_ibm_lpar_perftools_lock);
98 static DEFINE_MUTEX(rtas_ibm_physical_attestation_lock);
99 static DEFINE_MUTEX(rtas_ibm_set_dynamic_indicator_lock);
100 DEFINE_MUTEX(rtas_ibm_get_vpd_lock);
101
102 static struct rtas_function rtas_function_table[] __ro_after_init = {
103 [RTAS_FNIDX__CHECK_EXCEPTION] = {
104 .name = "check-exception",
105 },
106 [RTAS_FNIDX__DISPLAY_CHARACTER] = {
107 .name = "display-character",
108 .filter = &(const struct rtas_filter) {
109 .buf_idx1 = -1, .size_idx1 = -1,
110 .buf_idx2 = -1, .size_idx2 = -1,
111 },
112 },
113 [RTAS_FNIDX__EVENT_SCAN] = {
114 .name = "event-scan",
115 },
116 [RTAS_FNIDX__FREEZE_TIME_BASE] = {
117 .name = "freeze-time-base",
118 },
119 [RTAS_FNIDX__GET_POWER_LEVEL] = {
120 .name = "get-power-level",
121 .filter = &(const struct rtas_filter) {
122 .buf_idx1 = -1, .size_idx1 = -1,
123 .buf_idx2 = -1, .size_idx2 = -1,
124 },
125 },
126 [RTAS_FNIDX__GET_SENSOR_STATE] = {
127 .name = "get-sensor-state",
128 .filter = &(const struct rtas_filter) {
129 .buf_idx1 = -1, .size_idx1 = -1,
130 .buf_idx2 = -1, .size_idx2 = -1,
131 },
132 },
133 [RTAS_FNIDX__GET_TERM_CHAR] = {
134 .name = "get-term-char",
135 },
136 [RTAS_FNIDX__GET_TIME_OF_DAY] = {
137 .name = "get-time-of-day",
138 .filter = &(const struct rtas_filter) {
139 .buf_idx1 = -1, .size_idx1 = -1,
140 .buf_idx2 = -1, .size_idx2 = -1,
141 },
142 },
143 [RTAS_FNIDX__IBM_ACTIVATE_FIRMWARE] = {
144 .name = "ibm,activate-firmware",
145 .filter = &(const struct rtas_filter) {
146 .buf_idx1 = -1, .size_idx1 = -1,
147 .buf_idx2 = -1, .size_idx2 = -1,
148 },
149 /*
150 * PAPR+ as of v2.13 doesn't explicitly impose any
151 * restriction, but this typically requires multiple
152 * calls before success, and there's no reason to
153 * allow sequences to interleave.
154 */
155 .lock = &rtas_ibm_activate_firmware_lock,
156 },
157 [RTAS_FNIDX__IBM_CBE_START_PTCAL] = {
158 .name = "ibm,cbe-start-ptcal",
159 },
160 [RTAS_FNIDX__IBM_CBE_STOP_PTCAL] = {
161 .name = "ibm,cbe-stop-ptcal",
162 },
163 [RTAS_FNIDX__IBM_CHANGE_MSI] = {
164 .name = "ibm,change-msi",
165 },
166 [RTAS_FNIDX__IBM_CLOSE_ERRINJCT] = {
167 .name = "ibm,close-errinjct",
168 .filter = &(const struct rtas_filter) {
169 .buf_idx1 = -1, .size_idx1 = -1,
170 .buf_idx2 = -1, .size_idx2 = -1,
171 },
172 },
173 [RTAS_FNIDX__IBM_CONFIGURE_BRIDGE] = {
174 .name = "ibm,configure-bridge",
175 },
176 [RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR] = {
177 .name = "ibm,configure-connector",
178 .filter = &(const struct rtas_filter) {
179 .buf_idx1 = 0, .size_idx1 = -1,
180 .buf_idx2 = 1, .size_idx2 = -1,
181 .fixed_size = 4096,
182 },
183 },
184 [RTAS_FNIDX__IBM_CONFIGURE_KERNEL_DUMP] = {
185 .name = "ibm,configure-kernel-dump",
186 },
187 [RTAS_FNIDX__IBM_CONFIGURE_PE] = {
188 .name = "ibm,configure-pe",
189 },
190 [RTAS_FNIDX__IBM_CREATE_PE_DMA_WINDOW] = {
191 .name = "ibm,create-pe-dma-window",
192 },
193 [RTAS_FNIDX__IBM_DISPLAY_MESSAGE] = {
194 .name = "ibm,display-message",
195 .filter = &(const struct rtas_filter) {
196 .buf_idx1 = 0, .size_idx1 = -1,
197 .buf_idx2 = -1, .size_idx2 = -1,
198 },
199 },
200 [RTAS_FNIDX__IBM_ERRINJCT] = {
201 .name = "ibm,errinjct",
202 .filter = &(const struct rtas_filter) {
203 .buf_idx1 = 2, .size_idx1 = -1,
204 .buf_idx2 = -1, .size_idx2 = -1,
205 .fixed_size = 1024,
206 },
207 },
208 [RTAS_FNIDX__IBM_EXTI2C] = {
209 .name = "ibm,exti2c",
210 },
211 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO] = {
212 .name = "ibm,get-config-addr-info",
213 },
214 [RTAS_FNIDX__IBM_GET_CONFIG_ADDR_INFO2] = {
215 .name = "ibm,get-config-addr-info2",
216 .filter = &(const struct rtas_filter) {
217 .buf_idx1 = -1, .size_idx1 = -1,
218 .buf_idx2 = -1, .size_idx2 = -1,
219 },
220 },
221 [RTAS_FNIDX__IBM_GET_DYNAMIC_SENSOR_STATE] = {
222 .name = "ibm,get-dynamic-sensor-state",
223 .filter = &(const struct rtas_filter) {
224 .buf_idx1 = 1, .size_idx1 = -1,
225 .buf_idx2 = -1, .size_idx2 = -1,
226 },
227 /*
228 * PAPR+ v2.13 R1–7.3.19–3 is explicit that the OS
229 * must not call ibm,get-dynamic-sensor-state with
230 * different inputs until a non-retry status has been
231 * returned.
232 */
233 .lock = &rtas_ibm_get_dynamic_sensor_state_lock,
234 },
235 [RTAS_FNIDX__IBM_GET_INDICES] = {
236 .name = "ibm,get-indices",
237 .filter = &(const struct rtas_filter) {
238 .buf_idx1 = 2, .size_idx1 = 3,
239 .buf_idx2 = -1, .size_idx2 = -1,
240 },
241 /*
242 * PAPR+ v2.13 R1–7.3.17–2 says that the OS must not
243 * interleave ibm,get-indices call sequences with
244 * different inputs.
245 */
246 .lock = &rtas_ibm_get_indices_lock,
247 },
248 [RTAS_FNIDX__IBM_GET_RIO_TOPOLOGY] = {
249 .name = "ibm,get-rio-topology",
250 },
251 [RTAS_FNIDX__IBM_GET_SYSTEM_PARAMETER] = {
252 .name = "ibm,get-system-parameter",
253 .filter = &(const struct rtas_filter) {
254 .buf_idx1 = 1, .size_idx1 = 2,
255 .buf_idx2 = -1, .size_idx2 = -1,
256 },
257 },
258 [RTAS_FNIDX__IBM_GET_VPD] = {
259 .name = "ibm,get-vpd",
260 .filter = &(const struct rtas_filter) {
261 .buf_idx1 = 0, .size_idx1 = -1,
262 .buf_idx2 = 1, .size_idx2 = 2,
263 },
264 /*
265 * PAPR+ v2.13 R1–7.3.20–4 indicates that sequences
266 * should not be allowed to interleave.
267 */
268 .lock = &rtas_ibm_get_vpd_lock,
269 },
270 [RTAS_FNIDX__IBM_GET_XIVE] = {
271 .name = "ibm,get-xive",
272 },
273 [RTAS_FNIDX__IBM_INT_OFF] = {
274 .name = "ibm,int-off",
275 },
276 [RTAS_FNIDX__IBM_INT_ON] = {
277 .name = "ibm,int-on",
278 },
279 [RTAS_FNIDX__IBM_IO_QUIESCE_ACK] = {
280 .name = "ibm,io-quiesce-ack",
281 },
282 [RTAS_FNIDX__IBM_LPAR_PERFTOOLS] = {
283 .name = "ibm,lpar-perftools",
284 .filter = &(const struct rtas_filter) {
285 .buf_idx1 = 2, .size_idx1 = 3,
286 .buf_idx2 = -1, .size_idx2 = -1,
287 },
288 /*
289 * PAPR+ v2.13 R1–7.3.26–6 says the OS should allow
290 * only one call sequence in progress at a time.
291 */
292 .lock = &rtas_ibm_lpar_perftools_lock,
293 },
294 [RTAS_FNIDX__IBM_MANAGE_FLASH_IMAGE] = {
295 .name = "ibm,manage-flash-image",
296 },
297 [RTAS_FNIDX__IBM_MANAGE_STORAGE_PRESERVATION] = {
298 .name = "ibm,manage-storage-preservation",
299 },
300 [RTAS_FNIDX__IBM_NMI_INTERLOCK] = {
301 .name = "ibm,nmi-interlock",
302 },
303 [RTAS_FNIDX__IBM_NMI_REGISTER] = {
304 .name = "ibm,nmi-register",
305 },
306 [RTAS_FNIDX__IBM_OPEN_ERRINJCT] = {
307 .name = "ibm,open-errinjct",
308 .filter = &(const struct rtas_filter) {
309 .buf_idx1 = -1, .size_idx1 = -1,
310 .buf_idx2 = -1, .size_idx2 = -1,
311 },
312 },
313 [RTAS_FNIDX__IBM_OPEN_SRIOV_ALLOW_UNFREEZE] = {
314 .name = "ibm,open-sriov-allow-unfreeze",
315 },
316 [RTAS_FNIDX__IBM_OPEN_SRIOV_MAP_PE_NUMBER] = {
317 .name = "ibm,open-sriov-map-pe-number",
318 },
319 [RTAS_FNIDX__IBM_OS_TERM] = {
320 .name = "ibm,os-term",
321 },
322 [RTAS_FNIDX__IBM_PARTNER_CONTROL] = {
323 .name = "ibm,partner-control",
324 },
325 [RTAS_FNIDX__IBM_PHYSICAL_ATTESTATION] = {
326 .name = "ibm,physical-attestation",
327 .filter = &(const struct rtas_filter) {
328 .buf_idx1 = 0, .size_idx1 = 1,
329 .buf_idx2 = -1, .size_idx2 = -1,
330 },
331 /*
332 * This follows a sequence-based pattern similar to
333 * ibm,get-vpd et al. Since PAPR+ restricts
334 * interleaving call sequences for other functions of
335 * this style, assume the restriction applies here,
336 * even though it's not explicit in the spec.
337 */
338 .lock = &rtas_ibm_physical_attestation_lock,
339 },
340 [RTAS_FNIDX__IBM_PLATFORM_DUMP] = {
341 .name = "ibm,platform-dump",
342 .filter = &(const struct rtas_filter) {
343 .buf_idx1 = 4, .size_idx1 = 5,
344 .buf_idx2 = -1, .size_idx2 = -1,
345 },
346 /*
347 * PAPR+ v2.13 7.3.3.4.1 indicates that concurrent
348 * sequences of ibm,platform-dump are allowed if they
349 * are operating on different dump tags. So leave the
350 * lock pointer unset for now. This may need
351 * reconsideration if kernel-internal users appear.
352 */
353 },
354 [RTAS_FNIDX__IBM_POWER_OFF_UPS] = {
355 .name = "ibm,power-off-ups",
356 },
357 [RTAS_FNIDX__IBM_QUERY_INTERRUPT_SOURCE_NUMBER] = {
358 .name = "ibm,query-interrupt-source-number",
359 },
360 [RTAS_FNIDX__IBM_QUERY_PE_DMA_WINDOW] = {
361 .name = "ibm,query-pe-dma-window",
362 },
363 [RTAS_FNIDX__IBM_READ_PCI_CONFIG] = {
364 .name = "ibm,read-pci-config",
365 },
366 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE] = {
367 .name = "ibm,read-slot-reset-state",
368 .filter = &(const struct rtas_filter) {
369 .buf_idx1 = -1, .size_idx1 = -1,
370 .buf_idx2 = -1, .size_idx2 = -1,
371 },
372 },
373 [RTAS_FNIDX__IBM_READ_SLOT_RESET_STATE2] = {
374 .name = "ibm,read-slot-reset-state2",
375 },
376 [RTAS_FNIDX__IBM_REMOVE_PE_DMA_WINDOW] = {
377 .name = "ibm,remove-pe-dma-window",
378 },
379 [RTAS_FNIDX__IBM_RESET_PE_DMA_WINDOW] = {
380 /*
381 * Note: PAPR+ v2.13 7.3.31.4.1 spells this as
382 * "ibm,reset-pe-dma-windows" (plural), but RTAS
383 * implementations use the singular form in practice.
384 */
385 .name = "ibm,reset-pe-dma-window",
386 },
387 [RTAS_FNIDX__IBM_SCAN_LOG_DUMP] = {
388 .name = "ibm,scan-log-dump",
389 .filter = &(const struct rtas_filter) {
390 .buf_idx1 = 0, .size_idx1 = 1,
391 .buf_idx2 = -1, .size_idx2 = -1,
392 },
393 },
394 [RTAS_FNIDX__IBM_SET_DYNAMIC_INDICATOR] = {
395 .name = "ibm,set-dynamic-indicator",
396 .filter = &(const struct rtas_filter) {
397 .buf_idx1 = 2, .size_idx1 = -1,
398 .buf_idx2 = -1, .size_idx2 = -1,
399 },
400 /*
401 * PAPR+ v2.13 R1–7.3.18–3 says the OS must not call
402 * this function with different inputs until a
403 * non-retry status has been returned.
404 */
405 .lock = &rtas_ibm_set_dynamic_indicator_lock,
406 },
407 [RTAS_FNIDX__IBM_SET_EEH_OPTION] = {
408 .name = "ibm,set-eeh-option",
409 .filter = &(const struct rtas_filter) {
410 .buf_idx1 = -1, .size_idx1 = -1,
411 .buf_idx2 = -1, .size_idx2 = -1,
412 },
413 },
414 [RTAS_FNIDX__IBM_SET_SLOT_RESET] = {
415 .name = "ibm,set-slot-reset",
416 },
417 [RTAS_FNIDX__IBM_SET_SYSTEM_PARAMETER] = {
418 .name = "ibm,set-system-parameter",
419 .filter = &(const struct rtas_filter) {
420 .buf_idx1 = 1, .size_idx1 = -1,
421 .buf_idx2 = -1, .size_idx2 = -1,
422 },
423 },
424 [RTAS_FNIDX__IBM_SET_XIVE] = {
425 .name = "ibm,set-xive",
426 },
427 [RTAS_FNIDX__IBM_SLOT_ERROR_DETAIL] = {
428 .name = "ibm,slot-error-detail",
429 },
430 [RTAS_FNIDX__IBM_SUSPEND_ME] = {
431 .name = "ibm,suspend-me",
432 .banned_for_syscall_on_le = true,
433 .filter = &(const struct rtas_filter) {
434 .buf_idx1 = -1, .size_idx1 = -1,
435 .buf_idx2 = -1, .size_idx2 = -1,
436 },
437 },
438 [RTAS_FNIDX__IBM_TUNE_DMA_PARMS] = {
439 .name = "ibm,tune-dma-parms",
440 },
441 [RTAS_FNIDX__IBM_UPDATE_FLASH_64_AND_REBOOT] = {
442 .name = "ibm,update-flash-64-and-reboot",
443 },
444 [RTAS_FNIDX__IBM_UPDATE_NODES] = {
445 .name = "ibm,update-nodes",
446 .banned_for_syscall_on_le = true,
447 .filter = &(const struct rtas_filter) {
448 .buf_idx1 = 0, .size_idx1 = -1,
449 .buf_idx2 = -1, .size_idx2 = -1,
450 .fixed_size = 4096,
451 },
452 },
453 [RTAS_FNIDX__IBM_UPDATE_PROPERTIES] = {
454 .name = "ibm,update-properties",
455 .banned_for_syscall_on_le = true,
456 .filter = &(const struct rtas_filter) {
457 .buf_idx1 = 0, .size_idx1 = -1,
458 .buf_idx2 = -1, .size_idx2 = -1,
459 .fixed_size = 4096,
460 },
461 },
462 [RTAS_FNIDX__IBM_VALIDATE_FLASH_IMAGE] = {
463 .name = "ibm,validate-flash-image",
464 },
465 [RTAS_FNIDX__IBM_WRITE_PCI_CONFIG] = {
466 .name = "ibm,write-pci-config",
467 },
468 [RTAS_FNIDX__NVRAM_FETCH] = {
469 .name = "nvram-fetch",
470 },
471 [RTAS_FNIDX__NVRAM_STORE] = {
472 .name = "nvram-store",
473 },
474 [RTAS_FNIDX__POWER_OFF] = {
475 .name = "power-off",
476 },
477 [RTAS_FNIDX__PUT_TERM_CHAR] = {
478 .name = "put-term-char",
479 },
480 [RTAS_FNIDX__QUERY_CPU_STOPPED_STATE] = {
481 .name = "query-cpu-stopped-state",
482 },
483 [RTAS_FNIDX__READ_PCI_CONFIG] = {
484 .name = "read-pci-config",
485 },
486 [RTAS_FNIDX__RTAS_LAST_ERROR] = {
487 .name = "rtas-last-error",
488 },
489 [RTAS_FNIDX__SET_INDICATOR] = {
490 .name = "set-indicator",
491 .filter = &(const struct rtas_filter) {
492 .buf_idx1 = -1, .size_idx1 = -1,
493 .buf_idx2 = -1, .size_idx2 = -1,
494 },
495 },
496 [RTAS_FNIDX__SET_POWER_LEVEL] = {
497 .name = "set-power-level",
498 .filter = &(const struct rtas_filter) {
499 .buf_idx1 = -1, .size_idx1 = -1,
500 .buf_idx2 = -1, .size_idx2 = -1,
501 },
502 },
503 [RTAS_FNIDX__SET_TIME_FOR_POWER_ON] = {
504 .name = "set-time-for-power-on",
505 .filter = &(const struct rtas_filter) {
506 .buf_idx1 = -1, .size_idx1 = -1,
507 .buf_idx2 = -1, .size_idx2 = -1,
508 },
509 },
510 [RTAS_FNIDX__SET_TIME_OF_DAY] = {
511 .name = "set-time-of-day",
512 .filter = &(const struct rtas_filter) {
513 .buf_idx1 = -1, .size_idx1 = -1,
514 .buf_idx2 = -1, .size_idx2 = -1,
515 },
516 },
517 [RTAS_FNIDX__START_CPU] = {
518 .name = "start-cpu",
519 },
520 [RTAS_FNIDX__STOP_SELF] = {
521 .name = "stop-self",
522 },
523 [RTAS_FNIDX__SYSTEM_REBOOT] = {
524 .name = "system-reboot",
525 },
526 [RTAS_FNIDX__THAW_TIME_BASE] = {
527 .name = "thaw-time-base",
528 },
529 [RTAS_FNIDX__WRITE_PCI_CONFIG] = {
530 .name = "write-pci-config",
531 },
532 };
533
534 #define for_each_rtas_function(funcp) \
535 for (funcp = &rtas_function_table[0]; \
536 funcp < &rtas_function_table[ARRAY_SIZE(rtas_function_table)]; \
537 ++funcp)
538
539 /*
540 * Nearly all RTAS calls need to be serialized. All uses of the
541 * default rtas_args block must hold rtas_lock.
542 *
543 * Exceptions to the RTAS serialization requirement (e.g. stop-self)
544 * must use a separate rtas_args structure.
545 */
546 static DEFINE_RAW_SPINLOCK(rtas_lock);
547 static struct rtas_args rtas_args;
548
549 /**
550 * rtas_function_token() - RTAS function token lookup.
551 * @handle: Function handle, e.g. RTAS_FN_EVENT_SCAN.
552 *
553 * Context: Any context.
554 * Return: the token value for the function if implemented by this platform,
555 * otherwise RTAS_UNKNOWN_SERVICE.
556 */
557 s32 rtas_function_token(const rtas_fn_handle_t handle)
558 {
559 const size_t index = handle.index;
560 const bool out_of_bounds = index >= ARRAY_SIZE(rtas_function_table);
561
562 if (WARN_ONCE(out_of_bounds, "invalid function index %zu", index))
563 return RTAS_UNKNOWN_SERVICE;
564 /*
565 * Various drivers attempt token lookups on non-RTAS
566 * platforms.
567 */
568 if (!rtas.dev)
569 return RTAS_UNKNOWN_SERVICE;
570
571 return rtas_function_table[index].token;
572 }
573 EXPORT_SYMBOL_GPL(rtas_function_token);
574
575 static int rtas_function_cmp(const void *a, const void *b)
576 {
577 const struct rtas_function *f1 = a;
578 const struct rtas_function *f2 = b;
579
580 return strcmp(f1->name, f2->name);
581 }
582
583 /*
584 * Boot-time initialization of the function table needs the lookup to
585 * return a non-const-qualified object. Use rtas_name_to_function()
586 * in all other contexts.
587 */
588 static struct rtas_function *__rtas_name_to_function(const char *name)
589 {
590 const struct rtas_function key = {
591 .name = name,
592 };
593 struct rtas_function *found;
594
595 found = bsearch(&key, rtas_function_table, ARRAY_SIZE(rtas_function_table),
596 sizeof(rtas_function_table[0]), rtas_function_cmp);
597
598 return found;
599 }
600
601 static const struct rtas_function *rtas_name_to_function(const char *name)
602 {
603 return __rtas_name_to_function(name);
604 }
605
606 static DEFINE_XARRAY(rtas_token_to_function_xarray);
607
608 static int __init rtas_token_to_function_xarray_init(void)
609 {
610 const struct rtas_function *func;
611 int err = 0;
612
613 for_each_rtas_function(func) {
614 const s32 token = func->token;
615
616 if (token == RTAS_UNKNOWN_SERVICE)
617 continue;
618
619 err = xa_err(xa_store(&rtas_token_to_function_xarray,
620 token, (void *)func, GFP_KERNEL));
621 if (err)
622 break;
623 }
624
625 return err;
626 }
627 arch_initcall(rtas_token_to_function_xarray_init);
628
629 /*
630 * For use by sys_rtas(), where the token value is provided by user
631 * space and we don't want to warn on failed lookups.
632 */
633 static const struct rtas_function *rtas_token_to_function_untrusted(s32 token)
634 {
635 return xa_load(&rtas_token_to_function_xarray, token);
636 }
637
638 /*
639 * Reverse lookup for deriving the function descriptor from a
640 * known-good token value in contexts where the former is not already
641 * available. @token must be valid, e.g. derived from the result of a
642 * prior lookup against the function table.
643 */
644 static const struct rtas_function *rtas_token_to_function(s32 token)
645 {
646 const struct rtas_function *func;
647
648 if (WARN_ONCE(token < 0, "invalid token %d", token))
649 return NULL;
650
651 func = rtas_token_to_function_untrusted(token);
652 if (func)
653 return func;
654 /*
655 * Fall back to linear scan in case the reverse mapping hasn't
656 * been initialized yet.
657 */
658 if (xa_empty(&rtas_token_to_function_xarray)) {
659 for_each_rtas_function(func) {
660 if (func->token == token)
661 return func;
662 }
663 }
664
665 WARN_ONCE(true, "unexpected failed lookup for token %d", token);
666 return NULL;
667 }
668
669 /* This is here deliberately so it's only used in this file */
670 void enter_rtas(unsigned long);
671
672 static void __do_enter_rtas(struct rtas_args *args)
673 {
674 enter_rtas(__pa(args));
675 srr_regs_clobbered(); /* rtas uses SRRs, invalidate */
676 }
677
678 static void __do_enter_rtas_trace(struct rtas_args *args)
679 {
680 const struct rtas_function *func = rtas_token_to_function(be32_to_cpu(args->token));
681
682 /*
683 * If there is a per-function lock, it must be held by the
684 * caller.
685 */
686 if (func->lock)
687 lockdep_assert_held(func->lock);
688
689 if (args == &rtas_args)
690 lockdep_assert_held(&rtas_lock);
691
692 trace_rtas_input(args, func->name);
693 trace_rtas_ll_entry(args);
694
695 __do_enter_rtas(args);
696
697 trace_rtas_ll_exit(args);
698 trace_rtas_output(args, func->name);
699 }
700
701 static void do_enter_rtas(struct rtas_args *args)
702 {
703 const unsigned long msr = mfmsr();
704 /*
705 * Situations where we want to skip any active tracepoints for
706 * safety reasons:
707 *
708 * 1. The last code executed on an offline CPU as it stops,
709 * i.e. we're about to call stop-self. The tracepoints'
710 * function name lookup uses xarray, which uses RCU, which
711 * isn't valid to call on an offline CPU. Any events
712 * emitted on an offline CPU will be discarded anyway.
713 *
714 * 2. In real mode, as when invoking ibm,nmi-interlock from
715 * the pseries MCE handler. We cannot count on trace
716 * buffers or the entries in rtas_token_to_function_xarray
717 * to be contained in the RMO.
718 */
719 const unsigned long mask = MSR_IR | MSR_DR;
720 const bool can_trace = likely(cpu_online(raw_smp_processor_id()) &&
721 (msr & mask) == mask);
722 /*
723 * Make sure MSR[RI] is currently enabled as it will be forced later
724 * in enter_rtas.
725 */
726 BUG_ON(!(msr & MSR_RI));
727
728 BUG_ON(!irqs_disabled());
729
730 hard_irq_disable(); /* Ensure MSR[EE] is disabled on PPC64 */
731
732 if (can_trace)
733 __do_enter_rtas_trace(args);
734 else
735 __do_enter_rtas(args);
736 }
737
738 struct rtas_t rtas;
739
740 DEFINE_SPINLOCK(rtas_data_buf_lock);
741 EXPORT_SYMBOL_GPL(rtas_data_buf_lock);
742
743 char rtas_data_buf[RTAS_DATA_BUF_SIZE] __aligned(SZ_4K);
744 EXPORT_SYMBOL_GPL(rtas_data_buf);
745
746 unsigned long rtas_rmo_buf;
747
748 /*
749 * If non-NULL, this gets called when the kernel terminates.
750 * This is done like this so rtas_flash can be a module.
751 */
752 void (*rtas_flash_term_hook)(int);
753 EXPORT_SYMBOL_GPL(rtas_flash_term_hook);
754
755 /*
756 * call_rtas_display_status and call_rtas_display_status_delay
757 * are designed only for very early low-level debugging, which
758 * is why the token is hard-coded to 10.
759 */
760 static void call_rtas_display_status(unsigned char c)
761 {
762 unsigned long flags;
763
764 if (!rtas.base)
765 return;
766
767 raw_spin_lock_irqsave(&rtas_lock, flags);
768 rtas_call_unlocked(&rtas_args, 10, 1, 1, NULL, c);
769 raw_spin_unlock_irqrestore(&rtas_lock, flags);
770 }
771
772 static void call_rtas_display_status_delay(char c)
773 {
774 static int pending_newline = 0; /* did last write end with unprinted newline? */
775 static int width = 16;
776
777 if (c == '\n') {
778 while (width-- > 0)
779 call_rtas_display_status(' ');
780 width = 16;
781 mdelay(500);
782 pending_newline = 1;
783 } else {
784 if (pending_newline) {
785 call_rtas_display_status('\r');
786 call_rtas_display_status('\n');
787 }
788 pending_newline = 0;
789 if (width--) {
790 call_rtas_display_status(c);
791 udelay(10000);
792 }
793 }
794 }
795
796 void __init udbg_init_rtas_panel(void)
797 {
798 udbg_putc = call_rtas_display_status_delay;
799 }
800
801 #ifdef CONFIG_UDBG_RTAS_CONSOLE
802
803 /* If you think you're dying before early_init_dt_scan_rtas() does its
804 * work, you can hard code the token values for your firmware here and
805 * hardcode rtas.base/entry etc.
806 */
807 static unsigned int rtas_putchar_token = RTAS_UNKNOWN_SERVICE;
808 static unsigned int rtas_getchar_token = RTAS_UNKNOWN_SERVICE;
809
810 static void udbg_rtascon_putc(char c)
811 {
812 int tries;
813
814 if (!rtas.base)
815 return;
816
817 /* Add CRs before LFs */
818 if (c == '\n')
819 udbg_rtascon_putc('\r');
820
821 /* if there is more than one character to be displayed, wait a bit */
822 for (tries = 0; tries < 16; tries++) {
823 if (rtas_call(rtas_putchar_token, 1, 1, NULL, c) == 0)
824 break;
825 udelay(1000);
826 }
827 }
828
829 static int udbg_rtascon_getc_poll(void)
830 {
831 int c;
832
833 if (!rtas.base)
834 return -1;
835
836 if (rtas_call(rtas_getchar_token, 0, 2, &c))
837 return -1;
838
839 return c;
840 }
841
842 static int udbg_rtascon_getc(void)
843 {
844 int c;
845
846 while ((c = udbg_rtascon_getc_poll()) == -1)
847 ;
848
849 return c;
850 }
851
852
853 void __init udbg_init_rtas_console(void)
854 {
855 udbg_putc = udbg_rtascon_putc;
856 udbg_getc = udbg_rtascon_getc;
857 udbg_getc_poll = udbg_rtascon_getc_poll;
858 }
859 #endif /* CONFIG_UDBG_RTAS_CONSOLE */
860
861 void rtas_progress(char *s, unsigned short hex)
862 {
863 struct device_node *root;
864 int width;
865 const __be32 *p;
866 char *os;
867 static int display_character, set_indicator;
868 static int display_width, display_lines, form_feed;
869 static const int *row_width;
870 static DEFINE_SPINLOCK(progress_lock);
871 static int current_line;
872 static int pending_newline = 0; /* did last write end with unprinted newline? */
873
874 if (!rtas.base)
875 return;
876
877 if (display_width == 0) {
878 display_width = 0x10;
879 if ((root = of_find_node_by_path("/rtas"))) {
880 if ((p = of_get_property(root,
881 "ibm,display-line-length", NULL)))
882 display_width = be32_to_cpu(*p);
883 if ((p = of_get_property(root,
884 "ibm,form-feed", NULL)))
885 form_feed = be32_to_cpu(*p);
886 if ((p = of_get_property(root,
887 "ibm,display-number-of-lines", NULL)))
888 display_lines = be32_to_cpu(*p);
889 row_width = of_get_property(root,
890 "ibm,display-truncation-length", NULL);
891 of_node_put(root);
892 }
893 display_character = rtas_function_token(RTAS_FN_DISPLAY_CHARACTER);
894 set_indicator = rtas_function_token(RTAS_FN_SET_INDICATOR);
895 }
896
897 if (display_character == RTAS_UNKNOWN_SERVICE) {
898 /* use hex display if available */
899 if (set_indicator != RTAS_UNKNOWN_SERVICE)
900 rtas_call(set_indicator, 3, 1, NULL, 6, 0, hex);
901 return;
902 }
903
904 spin_lock(&progress_lock);
905
906 /*
907 * Last write ended with newline, but we didn't print it since
908 * it would just clear the bottom line of output. Print it now
909 * instead.
910 *
911 * If no newline is pending and form feed is supported, clear the
912 * display with a form feed; otherwise, print a CR to start output
913 * at the beginning of the line.
914 */
915 if (pending_newline) {
916 rtas_call(display_character, 1, 1, NULL, '\r');
917 rtas_call(display_character, 1, 1, NULL, '\n');
918 pending_newline = 0;
919 } else {
920 current_line = 0;
921 if (form_feed)
922 rtas_call(display_character, 1, 1, NULL,
923 (char)form_feed);
924 else
925 rtas_call(display_character, 1, 1, NULL, '\r');
926 }
927
928 if (row_width)
929 width = row_width[current_line];
930 else
931 width = display_width;
932 os = s;
933 while (*os) {
934 if (*os == '\n' || *os == '\r') {
935 /* If newline is the last character, save it
936 * until next call to avoid bumping up the
937 * display output.
938 */
939 if (*os == '\n' && !os[1]) {
940 pending_newline = 1;
941 current_line++;
942 if (current_line > display_lines-1)
943 current_line = display_lines-1;
944 spin_unlock(&progress_lock);
945 return;
946 }
947
948 /* RTAS wants CR-LF, not just LF */
949
950 if (*os == '\n') {
951 rtas_call(display_character, 1, 1, NULL, '\r');
952 rtas_call(display_character, 1, 1, NULL, '\n');
953 } else {
954 /* CR might be used to re-draw a line, so we'll
955 * leave it alone and not add LF.
956 */
957 rtas_call(display_character, 1, 1, NULL, *os);
958 }
959
960 if (row_width)
961 width = row_width[current_line];
962 else
963 width = display_width;
964 } else {
965 width--;
966 rtas_call(display_character, 1, 1, NULL, *os);
967 }
968
969 os++;
970
971 /* if we overwrite the screen length */
972 if (width <= 0)
973 while ((*os != 0) && (*os != '\n') && (*os != '\r'))
974 os++;
975 }
976
977 spin_unlock(&progress_lock);
978 }
979 EXPORT_SYMBOL_GPL(rtas_progress); /* needed by rtas_flash module */
980
981 int rtas_token(const char *service)
982 {
983 const struct rtas_function *func;
984 const __be32 *tokp;
985
986 if (rtas.dev == NULL)
987 return RTAS_UNKNOWN_SERVICE;
988
989 func = rtas_name_to_function(service);
990 if (func)
991 return func->token;
992 /*
993 * The caller is looking up a name that is not known to be an
994 * RTAS function. Either it's a function that needs to be
995 * added to the table, or they're misusing rtas_token() to
996 * access non-function properties of the /rtas node. Warn and
997 * fall back to the legacy behavior.
998 */
999 WARN_ONCE(1, "unknown function `%s`, should it be added to rtas_function_table?\n",
1000 service);
1001
1002 tokp = of_get_property(rtas.dev, service, NULL);
1003 return tokp ? be32_to_cpu(*tokp) : RTAS_UNKNOWN_SERVICE;
1004 }
1005 EXPORT_SYMBOL_GPL(rtas_token);
1006
1007 #ifdef CONFIG_RTAS_ERROR_LOGGING
1008
1009 static u32 rtas_error_log_max __ro_after_init = RTAS_ERROR_LOG_MAX;
1010
1011 /*
1012 * Return the firmware-specified size of the error log buffer
1013 * for all rtas calls that require an error buffer argument.
1014 * This includes 'check-exception' and 'rtas-last-error'.
1015 */
1016 int rtas_get_error_log_max(void)
1017 {
1018 return rtas_error_log_max;
1019 }
1020
1021 static void __init init_error_log_max(void)
1022 {
1023 static const char propname[] __initconst = "rtas-error-log-max";
1024 u32 max;
1025
1026 if (of_property_read_u32(rtas.dev, propname, &max)) {
1027 pr_warn("%s not found, using default of %u\n",
1028 propname, RTAS_ERROR_LOG_MAX);
1029 max = RTAS_ERROR_LOG_MAX;
1030 }
1031
1032 if (max > RTAS_ERROR_LOG_MAX) {
1033 pr_warn("%s = %u, clamping max error log size to %u\n",
1034 propname, max, RTAS_ERROR_LOG_MAX);
1035 max = RTAS_ERROR_LOG_MAX;
1036 }
1037
1038 rtas_error_log_max = max;
1039 }
1040
1041
1042 static char rtas_err_buf[RTAS_ERROR_LOG_MAX];
1043
1044 /** Return a copy of the detailed error text associated with the
1045 * most recent failed call to rtas. Because the error text
1046 * might go stale if there are any other intervening rtas calls,
1047 * this routine must be called atomically with whatever produced
1048 * the error (i.e. with rtas_lock still held from the previous call).
1049 */
1050 static char *__fetch_rtas_last_error(char *altbuf)
1051 {
1052 const s32 token = rtas_function_token(RTAS_FN_RTAS_LAST_ERROR);
1053 struct rtas_args err_args, save_args;
1054 u32 bufsz;
1055 char *buf = NULL;
1056
1057 lockdep_assert_held(&rtas_lock);
1058
1059 if (token == -1)
1060 return NULL;
1061
1062 bufsz = rtas_get_error_log_max();
1063
1064 err_args.token = cpu_to_be32(token);
1065 err_args.nargs = cpu_to_be32(2);
1066 err_args.nret = cpu_to_be32(1);
1067 err_args.args[0] = cpu_to_be32(__pa(rtas_err_buf));
1068 err_args.args[1] = cpu_to_be32(bufsz);
1069 err_args.args[2] = 0;
1070
1071 save_args = rtas_args;
1072 rtas_args = err_args;
1073
1074 do_enter_rtas(&rtas_args);
1075
1076 err_args = rtas_args;
1077 rtas_args = save_args;
1078
1079 /* Log the error in the unlikely case that there was one. */
1080 if (unlikely(err_args.args[2] == 0)) {
1081 if (altbuf) {
1082 buf = altbuf;
1083 } else {
1084 buf = rtas_err_buf;
1085 if (slab_is_available())
1086 buf = kmalloc(RTAS_ERROR_LOG_MAX, GFP_ATOMIC);
1087 }
1088 if (buf)
1089 memmove(buf, rtas_err_buf, RTAS_ERROR_LOG_MAX);
1090 }
1091
1092 return buf;
1093 }
1094
1095 #define get_errorlog_buffer() kmalloc(RTAS_ERROR_LOG_MAX, GFP_KERNEL)
1096
1097 #else /* CONFIG_RTAS_ERROR_LOGGING */
1098 #define __fetch_rtas_last_error(x) NULL
1099 #define get_errorlog_buffer() NULL
1100 static void __init init_error_log_max(void) {}
1101 #endif
1102
1103
1104 static void
1105 va_rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret,
1106 va_list list)
1107 {
1108 int i;
1109
1110 args->token = cpu_to_be32(token);
1111 args->nargs = cpu_to_be32(nargs);
1112 args->nret = cpu_to_be32(nret);
1113 args->rets = &(args->args[nargs]);
1114
1115 for (i = 0; i < nargs; ++i)
1116 args->args[i] = cpu_to_be32(va_arg(list, __u32));
1117
1118 for (i = 0; i < nret; ++i)
1119 args->rets[i] = 0;
1120
1121 do_enter_rtas(args);
1122 }
1123
1124 /**
1125 * rtas_call_unlocked() - Invoke an RTAS firmware function without synchronization.
1126 * @args: RTAS parameter block to be used for the call, must obey RTAS addressing
1127 * constraints.
1128 * @token: Identifies the function being invoked.
1129 * @nargs: Number of input parameters. Does not include token.
1130 * @nret: Number of output parameters, including the call status.
1131 * @....: List of @nargs input parameters.
1132 *
1133 * Invokes the RTAS function indicated by @token, which the caller
1134 * should obtain via rtas_function_token().
1135 *
1136 * This function is similar to rtas_call(), but must be used with a
1137 * limited set of RTAS calls specifically exempted from the general
1138 * requirement that only one RTAS call may be in progress at any
1139 * time. Examples include stop-self and ibm,nmi-interlock.
1140 */
1141 void rtas_call_unlocked(struct rtas_args *args, int token, int nargs, int nret, ...)
1142 {
1143 va_list list;
1144
1145 va_start(list, nret);
1146 va_rtas_call_unlocked(args, token, nargs, nret, list);
1147 va_end(list);
1148 }
1149
1150 static bool token_is_restricted_errinjct(s32 token)
1151 {
1152 return token == rtas_function_token(RTAS_FN_IBM_OPEN_ERRINJCT) ||
1153 token == rtas_function_token(RTAS_FN_IBM_ERRINJCT);
1154 }
1155
1156 /**
1157 * rtas_call() - Invoke an RTAS firmware function.
1158 * @token: Identifies the function being invoked.
1159 * @nargs: Number of input parameters. Does not include token.
1160 * @nret: Number of output parameters, including the call status.
1161 * @outputs: Array of @nret output words.
1162 * @....: List of @nargs input parameters.
1163 *
1164 * Invokes the RTAS function indicated by @token, which the caller
1165 * should obtain via rtas_function_token().
1166 *
1167 * The @nargs and @nret arguments must match the number of input and
1168 * output parameters specified for the RTAS function.
1169 *
1170 * rtas_call() returns RTAS status codes, not conventional Linux errno
1171 * values. Callers must translate any failure to an appropriate errno
1172 * in syscall context. Most callers of RTAS functions that can return
1173 * -2 or 990x should use rtas_busy_delay() to correctly handle those
1174 * statuses before calling again.
1175 *
1176 * The return value descriptions are adapted from 7.2.8 [RTAS] Return
1177 * Codes of the PAPR and CHRP specifications.
1178 *
1179 * Context: Process context preferably, interrupt context if
1180 * necessary. Acquires an internal spinlock and may perform
1181 * GFP_ATOMIC slab allocation in error path. Unsafe for NMI
1182 * context.
1183 * Return:
1184 * * 0 - RTAS function call succeeded.
1185 * * -1 - RTAS function encountered a hardware or
1186 * platform error, or the token is invalid,
1187 * or the function is restricted by kernel policy.
1188 * * -2 - Specs say "A necessary hardware device was busy,
1189 * and the requested function could not be
1190 * performed. The operation should be retried at
1191 * a later time." This is misleading, at least with
1192 * respect to current RTAS implementations. What it
1193 * usually means in practice is that the function
1194 * could not be completed while meeting RTAS's
1195 * deadline for returning control to the OS (250us
1196 * for PAPR/PowerVM, typically), but the call may be
1197 * immediately reattempted to resume work on it.
1198 * * -3 - Parameter error.
1199 * * -7 - Unexpected state change.
1200 * * 9000...9899 - Vendor-specific success codes.
1201 * * 9900...9905 - Advisory extended delay. Caller should try
1202 * again after ~10^x ms has elapsed, where x is
1203 * the last digit of the status [0-5]. Again going
1204 * beyond the PAPR text, 990x on PowerVM indicates
1205 * contention for RTAS-internal resources. Other
1206 * RTAS call sequences in progress should be
1207 * allowed to complete before reattempting the
1208 * call.
1209 * * -9000 - Multi-level isolation error.
1210 * * -9999...-9004 - Vendor-specific error codes.
1211 * * Additional negative values - Function-specific error.
1212 * * Additional positive values - Function-specific success.
1213 */
1214 int rtas_call(int token, int nargs, int nret, int *outputs, ...)
1215 {
1216 struct pin_cookie cookie;
1217 va_list list;
1218 int i;
1219 unsigned long flags;
1220 struct rtas_args *args;
1221 char *buff_copy = NULL;
1222 int ret;
1223
1224 if (!rtas.entry || token == RTAS_UNKNOWN_SERVICE)
1225 return -1;
1226
1227 if (token_is_restricted_errinjct(token)) {
1228 /*
1229 * It would be nicer to not discard the error value
1230 * from security_locked_down(), but callers expect an
1231 * RTAS status, not an errno.
1232 */
1233 if (security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION))
1234 return -1;
1235 }
1236
1237 if ((mfmsr() & (MSR_IR|MSR_DR)) != (MSR_IR|MSR_DR)) {
1238 WARN_ON_ONCE(1);
1239 return -1;
1240 }
1241
1242 raw_spin_lock_irqsave(&rtas_lock, flags);
1243 cookie = lockdep_pin_lock(&rtas_lock);
1244
1245 /* We use the global rtas args buffer */
1246 args = &rtas_args;
1247
1248 va_start(list, outputs);
1249 va_rtas_call_unlocked(args, token, nargs, nret, list);
1250 va_end(list);
1251
1252 /* A -1 return code indicates that the last command couldn't
1253 be completed due to a hardware error. */
1254 if (be32_to_cpu(args->rets[0]) == -1)
1255 buff_copy = __fetch_rtas_last_error(NULL);
1256
1257 if (nret > 1 && outputs != NULL)
1258 for (i = 0; i < nret-1; ++i)
1259 outputs[i] = be32_to_cpu(args->rets[i + 1]);
1260 ret = (nret > 0) ? be32_to_cpu(args->rets[0]) : 0;
1261
1262 lockdep_unpin_lock(&rtas_lock, cookie);
1263 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1264
1265 if (buff_copy) {
1266 log_error(buff_copy, ERR_TYPE_RTAS_LOG, 0);
1267 if (slab_is_available())
1268 kfree(buff_copy);
1269 }
1270 return ret;
1271 }
1272 EXPORT_SYMBOL_GPL(rtas_call);
1273
1274 /**
1275 * rtas_busy_delay_time() - From an RTAS status value, calculate the
1276 * suggested delay time in milliseconds.
1277 *
1278 * @status: a value returned from rtas_call() or similar APIs which return
1279 * the status of a RTAS function call.
1280 *
1281 * Context: Any context.
1282 *
1283 * Return:
1284 * * 100000 - If @status is 9905.
1285 * * 10000 - If @status is 9904.
1286 * * 1000 - If @status is 9903.
1287 * * 100 - If @status is 9902.
1288 * * 10 - If @status is 9901.
1289 * * 1 - If @status is either 9900 or -2. This is "wrong" for -2, but
1290 * some callers depend on this behavior, and the worst outcome
1291 * is that they will delay for longer than necessary.
1292 * * 0 - If @status is not a busy or extended delay value.
1293 */
1294 unsigned int rtas_busy_delay_time(int status)
1295 {
1296 int order;
1297 unsigned int ms = 0;
1298
1299 if (status == RTAS_BUSY) {
1300 ms = 1;
1301 } else if (status >= RTAS_EXTENDED_DELAY_MIN &&
1302 status <= RTAS_EXTENDED_DELAY_MAX) {
1303 order = status - RTAS_EXTENDED_DELAY_MIN;
1304 for (ms = 1; order > 0; order--)
1305 ms *= 10;
1306 }
1307
1308 return ms;
1309 }
1310
1311 /*
1312 * Early boot fallback for rtas_busy_delay().
1313 */
1314 static bool __init rtas_busy_delay_early(int status)
1315 {
1316 static size_t successive_ext_delays __initdata;
1317 bool retry;
1318
1319 switch (status) {
1320 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1321 /*
1322 * In the unlikely case that we receive an extended
1323 * delay status in early boot, the OS is probably not
1324 * the cause, and there's nothing we can do to clear
1325 * the condition. Best we can do is delay for a bit
1326 * and hope it's transient. Lie to the caller if it
1327 * seems like we're stuck in a retry loop.
1328 */
1329 mdelay(1);
1330 retry = true;
1331 successive_ext_delays += 1;
1332 if (successive_ext_delays > 1000) {
1333 pr_err("too many extended delays, giving up\n");
1334 dump_stack();
1335 retry = false;
1336 successive_ext_delays = 0;
1337 }
1338 break;
1339 case RTAS_BUSY:
1340 retry = true;
1341 successive_ext_delays = 0;
1342 break;
1343 default:
1344 retry = false;
1345 successive_ext_delays = 0;
1346 break;
1347 }
1348
1349 return retry;
1350 }
1351
1352 /**
1353 * rtas_busy_delay() - helper for RTAS busy and extended delay statuses
1354 *
1355 * @status: a value returned from rtas_call() or similar APIs which return
1356 * the status of a RTAS function call.
1357 *
1358 * Context: Process context. May sleep or schedule.
1359 *
1360 * Return:
1361 * * true - @status is RTAS_BUSY or an extended delay hint. The
1362 * caller may assume that the CPU has been yielded if necessary,
1363 * and that an appropriate delay for @status has elapsed.
1364 * Generally the caller should reattempt the RTAS call which
1365 * yielded @status.
1366 *
1367 * * false - @status is not @RTAS_BUSY nor an extended delay hint. The
1368 * caller is responsible for handling @status.
1369 */
1370 bool __ref rtas_busy_delay(int status)
1371 {
1372 unsigned int ms;
1373 bool ret;
1374
1375 /*
1376 * Can't do timed sleeps before timekeeping is up.
1377 */
1378 if (system_state < SYSTEM_SCHEDULING)
1379 return rtas_busy_delay_early(status);
1380
1381 switch (status) {
1382 case RTAS_EXTENDED_DELAY_MIN...RTAS_EXTENDED_DELAY_MAX:
1383 ret = true;
1384 ms = rtas_busy_delay_time(status);
1385 /*
1386 * The extended delay hint can be as high as 100 seconds.
1387 * Surely any function returning such a status is either
1388 * buggy or isn't going to be significantly slowed by us
1389 * polling at 1HZ. Clamp the sleep time to one second.
1390 */
1391 ms = clamp(ms, 1U, 1000U);
1392 /*
1393 * The delay hint is an order-of-magnitude suggestion, not
1394 * a minimum. It is fine, possibly even advantageous, for
1395 * us to pause for less time than hinted. For small values,
1396 * use usleep_range() to ensure we don't sleep much longer
1397 * than actually needed.
1398 *
1399 * See Documentation/timers/timers-howto.rst for
1400 * explanation of the threshold used here. In effect we use
1401 * usleep_range() for 9900 and 9901, msleep() for
1402 * 9902-9905.
1403 */
1404 if (ms <= 20)
1405 usleep_range(ms * 100, ms * 1000);
1406 else
1407 msleep(ms);
1408 break;
1409 case RTAS_BUSY:
1410 ret = true;
1411 /*
1412 * We should call again immediately if there's no other
1413 * work to do.
1414 */
1415 cond_resched();
1416 break;
1417 default:
1418 ret = false;
1419 /*
1420 * Not a busy or extended delay status; the caller should
1421 * handle @status itself. Ensure we warn on misuses in
1422 * atomic context regardless.
1423 */
1424 might_sleep();
1425 break;
1426 }
1427
1428 return ret;
1429 }
1430 EXPORT_SYMBOL_GPL(rtas_busy_delay);
1431
1432 int rtas_error_rc(int rtas_rc)
1433 {
1434 int rc;
1435
1436 switch (rtas_rc) {
1437 case RTAS_HARDWARE_ERROR: /* Hardware Error */
1438 rc = -EIO;
1439 break;
1440 case RTAS_INVALID_PARAMETER: /* Bad indicator/domain/etc */
1441 rc = -EINVAL;
1442 break;
1443 case -9000: /* Isolation error */
1444 rc = -EFAULT;
1445 break;
1446 case -9001: /* Outstanding TCE/PTE */
1447 rc = -EEXIST;
1448 break;
1449 case -9002: /* No usable slot */
1450 rc = -ENODEV;
1451 break;
1452 default:
1453 pr_err("%s: unexpected error %d\n", __func__, rtas_rc);
1454 rc = -ERANGE;
1455 break;
1456 }
1457 return rc;
1458 }
1459 EXPORT_SYMBOL_GPL(rtas_error_rc);
1460
1461 int rtas_get_power_level(int powerdomain, int *level)
1462 {
1463 int token = rtas_function_token(RTAS_FN_GET_POWER_LEVEL);
1464 int rc;
1465
1466 if (token == RTAS_UNKNOWN_SERVICE)
1467 return -ENOENT;
1468
1469 while ((rc = rtas_call(token, 1, 2, level, powerdomain)) == RTAS_BUSY)
1470 udelay(1);
1471
1472 if (rc < 0)
1473 return rtas_error_rc(rc);
1474 return rc;
1475 }
1476 EXPORT_SYMBOL_GPL(rtas_get_power_level);
1477
1478 int rtas_set_power_level(int powerdomain, int level, int *setlevel)
1479 {
1480 int token = rtas_function_token(RTAS_FN_SET_POWER_LEVEL);
1481 int rc;
1482
1483 if (token == RTAS_UNKNOWN_SERVICE)
1484 return -ENOENT;
1485
1486 do {
1487 rc = rtas_call(token, 2, 2, setlevel, powerdomain, level);
1488 } while (rtas_busy_delay(rc));
1489
1490 if (rc < 0)
1491 return rtas_error_rc(rc);
1492 return rc;
1493 }
1494 EXPORT_SYMBOL_GPL(rtas_set_power_level);
1495
1496 int rtas_get_sensor(int sensor, int index, int *state)
1497 {
1498 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1499 int rc;
1500
1501 if (token == RTAS_UNKNOWN_SERVICE)
1502 return -ENOENT;
1503
1504 do {
1505 rc = rtas_call(token, 2, 2, state, sensor, index);
1506 } while (rtas_busy_delay(rc));
1507
1508 if (rc < 0)
1509 return rtas_error_rc(rc);
1510 return rc;
1511 }
1512 EXPORT_SYMBOL_GPL(rtas_get_sensor);
1513
1514 int rtas_get_sensor_fast(int sensor, int index, int *state)
1515 {
1516 int token = rtas_function_token(RTAS_FN_GET_SENSOR_STATE);
1517 int rc;
1518
1519 if (token == RTAS_UNKNOWN_SERVICE)
1520 return -ENOENT;
1521
1522 rc = rtas_call(token, 2, 2, state, sensor, index);
1523 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1524 rc <= RTAS_EXTENDED_DELAY_MAX));
1525
1526 if (rc < 0)
1527 return rtas_error_rc(rc);
1528 return rc;
1529 }
1530
1531 bool rtas_indicator_present(int token, int *maxindex)
1532 {
1533 int proplen, count, i;
1534 const struct indicator_elem {
1535 __be32 token;
1536 __be32 maxindex;
1537 } *indicators;
1538
1539 indicators = of_get_property(rtas.dev, "rtas-indicators", &proplen);
1540 if (!indicators)
1541 return false;
1542
1543 count = proplen / sizeof(struct indicator_elem);
1544
1545 for (i = 0; i < count; i++) {
1546 if (__be32_to_cpu(indicators[i].token) != token)
1547 continue;
1548 if (maxindex)
1549 *maxindex = __be32_to_cpu(indicators[i].maxindex);
1550 return true;
1551 }
1552
1553 return false;
1554 }
1555
1556 int rtas_set_indicator(int indicator, int index, int new_value)
1557 {
1558 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1559 int rc;
1560
1561 if (token == RTAS_UNKNOWN_SERVICE)
1562 return -ENOENT;
1563
1564 do {
1565 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1566 } while (rtas_busy_delay(rc));
1567
1568 if (rc < 0)
1569 return rtas_error_rc(rc);
1570 return rc;
1571 }
1572 EXPORT_SYMBOL_GPL(rtas_set_indicator);
1573
1574 /*
1575 * Ignoring RTAS extended delay
1576 */
1577 int rtas_set_indicator_fast(int indicator, int index, int new_value)
1578 {
1579 int token = rtas_function_token(RTAS_FN_SET_INDICATOR);
1580 int rc;
1581
1582 if (token == RTAS_UNKNOWN_SERVICE)
1583 return -ENOENT;
1584
1585 rc = rtas_call(token, 3, 1, NULL, indicator, index, new_value);
1586
1587 WARN_ON(rc == RTAS_BUSY || (rc >= RTAS_EXTENDED_DELAY_MIN &&
1588 rc <= RTAS_EXTENDED_DELAY_MAX));
1589
1590 if (rc < 0)
1591 return rtas_error_rc(rc);
1592
1593 return rc;
1594 }
1595
1596 /**
1597 * rtas_ibm_suspend_me() - Call ibm,suspend-me to suspend the LPAR.
1598 *
1599 * @fw_status: RTAS call status will be placed here if not NULL.
1600 *
1601 * rtas_ibm_suspend_me() should be called only on a CPU which has
1602 * received H_CONTINUE from the H_JOIN hcall. All other active CPUs
1603 * should be waiting to return from H_JOIN.
1604 *
1605 * rtas_ibm_suspend_me() may suspend execution of the OS
1606 * indefinitely. Callers should take appropriate measures upon return, such as
1607 * resetting watchdog facilities.
1608 *
1609 * Callers may choose to retry this call if @fw_status is
1610 * %RTAS_THREADS_ACTIVE.
1611 *
1612 * Return:
1613 * 0 - The partition has resumed from suspend, possibly after
1614 * migration to a different host.
1615 * -ECANCELED - The operation was aborted.
1616 * -EAGAIN - There were other CPUs not in H_JOIN at the time of the call.
1617 * -EBUSY - Some other condition prevented the suspend from succeeding.
1618 * -EIO - Hardware/platform error.
1619 */
1620 int rtas_ibm_suspend_me(int *fw_status)
1621 {
1622 int token = rtas_function_token(RTAS_FN_IBM_SUSPEND_ME);
1623 int fwrc;
1624 int ret;
1625
1626 fwrc = rtas_call(token, 0, 1, NULL);
1627
1628 switch (fwrc) {
1629 case 0:
1630 ret = 0;
1631 break;
1632 case RTAS_SUSPEND_ABORTED:
1633 ret = -ECANCELED;
1634 break;
1635 case RTAS_THREADS_ACTIVE:
1636 ret = -EAGAIN;
1637 break;
1638 case RTAS_NOT_SUSPENDABLE:
1639 case RTAS_OUTSTANDING_COPROC:
1640 ret = -EBUSY;
1641 break;
1642 case -1:
1643 default:
1644 ret = -EIO;
1645 break;
1646 }
1647
1648 if (fw_status)
1649 *fw_status = fwrc;
1650
1651 return ret;
1652 }
1653
1654 void __noreturn rtas_restart(char *cmd)
1655 {
1656 if (rtas_flash_term_hook)
1657 rtas_flash_term_hook(SYS_RESTART);
1658 pr_emerg("system-reboot returned %d\n",
1659 rtas_call(rtas_function_token(RTAS_FN_SYSTEM_REBOOT), 0, 1, NULL));
1660 for (;;);
1661 }
1662
1663 void rtas_power_off(void)
1664 {
1665 if (rtas_flash_term_hook)
1666 rtas_flash_term_hook(SYS_POWER_OFF);
1667 /* allow power on only with power button press */
1668 pr_emerg("power-off returned %d\n",
1669 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1670 for (;;);
1671 }
1672
1673 void __noreturn rtas_halt(void)
1674 {
1675 if (rtas_flash_term_hook)
1676 rtas_flash_term_hook(SYS_HALT);
1677 /* allow power on only with power button press */
1678 pr_emerg("power-off returned %d\n",
1679 rtas_call(rtas_function_token(RTAS_FN_POWER_OFF), 2, 1, NULL, -1, -1));
1680 for (;;);
1681 }
1682
1683 /* Must be in the RMO region, so we place it here */
1684 static char rtas_os_term_buf[2048];
1685 static bool ibm_extended_os_term;
1686
1687 void rtas_os_term(char *str)
1688 {
1689 s32 token = rtas_function_token(RTAS_FN_IBM_OS_TERM);
1690 static struct rtas_args args;
1691 int status;
1692
1693 /*
1694 * Firmware with the ibm,extended-os-term property is guaranteed
1695 * to always return from an ibm,os-term call. Earlier versions without
1696 * this property may terminate the partition which we want to avoid
1697 * since it interferes with panic_timeout.
1698 */
1699
1700 if (token == RTAS_UNKNOWN_SERVICE || !ibm_extended_os_term)
1701 return;
1702
1703 snprintf(rtas_os_term_buf, 2048, "OS panic: %s", str);
1704
1705 /*
1706 * Keep calling as long as RTAS returns a "try again" status,
1707 * but don't use rtas_busy_delay(), which potentially
1708 * schedules.
1709 */
1710 do {
1711 rtas_call_unlocked(&args, token, 1, 1, NULL, __pa(rtas_os_term_buf));
1712 status = be32_to_cpu(args.rets[0]);
1713 } while (rtas_busy_delay_time(status));
1714
1715 if (status != 0)
1716 pr_emerg("ibm,os-term call failed %d\n", status);
1717 }
1718
1719 /**
1720 * rtas_activate_firmware() - Activate a new version of firmware.
1721 *
1722 * Context: This function may sleep.
1723 *
1724 * Activate a new version of partition firmware. The OS must call this
1725 * after resuming from a partition hibernation or migration in order
1726 * to maintain the ability to perform live firmware updates. It's not
1727 * catastrophic for this method to be absent or to fail; just log the
1728 * condition in that case.
1729 */
1730 void rtas_activate_firmware(void)
1731 {
1732 int token = rtas_function_token(RTAS_FN_IBM_ACTIVATE_FIRMWARE);
1733 int fwrc;
1734
1735 if (token == RTAS_UNKNOWN_SERVICE) {
1736 pr_notice("ibm,activate-firmware method unavailable\n");
1737 return;
1738 }
1739
1740 mutex_lock(&rtas_ibm_activate_firmware_lock);
1741
1742 do {
1743 fwrc = rtas_call(token, 0, 1, NULL);
1744 } while (rtas_busy_delay(fwrc));
1745
1746 mutex_unlock(&rtas_ibm_activate_firmware_lock);
1747
1748 if (fwrc)
1749 pr_err("ibm,activate-firmware failed (%i)\n", fwrc);
1750 }
1751
1752 /**
1753 * get_pseries_errorlog() - Find a specific pseries error log in an RTAS
1754 * extended event log.
1755 * @log: RTAS error/event log
1756 * @section_id: two character section identifier
1757 *
1758 * Return: A pointer to the specified errorlog or NULL if not found.
1759 */
1760 noinstr struct pseries_errorlog *get_pseries_errorlog(struct rtas_error_log *log,
1761 uint16_t section_id)
1762 {
1763 struct rtas_ext_event_log_v6 *ext_log =
1764 (struct rtas_ext_event_log_v6 *)log->buffer;
1765 struct pseries_errorlog *sect;
1766 unsigned char *p, *log_end;
1767 uint32_t ext_log_length = rtas_error_extended_log_length(log);
1768 uint8_t log_format = rtas_ext_event_log_format(ext_log);
1769 uint32_t company_id = rtas_ext_event_company_id(ext_log);
1770
1771 /* Check that we understand the format */
1772 if (ext_log_length < sizeof(struct rtas_ext_event_log_v6) ||
1773 log_format != RTAS_V6EXT_LOG_FORMAT_EVENT_LOG ||
1774 company_id != RTAS_V6EXT_COMPANY_ID_IBM)
1775 return NULL;
1776
1777 log_end = log->buffer + ext_log_length;
1778 p = ext_log->vendor_log;
1779
1780 while (p < log_end) {
1781 sect = (struct pseries_errorlog *)p;
1782 if (pseries_errorlog_id(sect) == section_id)
1783 return sect;
1784 p += pseries_errorlog_length(sect);
1785 }
1786
1787 return NULL;
1788 }
1789
1790 /*
1791 * The sys_rtas syscall, as originally designed, allows root to pass
1792 * arbitrary physical addresses to RTAS calls. A number of RTAS calls
1793 * can be abused to write to arbitrary memory and do other things that
1794 * are potentially harmful to system integrity, and thus should only
1795 * be used inside the kernel and not exposed to userspace.
1796 *
1797 * All known legitimate users of the sys_rtas syscall will only ever
1798 * pass addresses that fall within the RMO buffer, and use a known
1799 * subset of RTAS calls.
1800 *
1801 * Accordingly, we filter RTAS requests to check that the call is
1802 * permitted, and that provided pointers fall within the RMO buffer.
1803 * If a function is allowed to be invoked via the syscall, then its
1804 * entry in the rtas_functions table points to a rtas_filter that
1805 * describes its constraints, with the indexes of the parameters which
1806 * are expected to contain addresses and sizes of buffers allocated
1807 * inside the RMO buffer.
1808 */
1809
1810 static bool in_rmo_buf(u32 base, u32 end)
1811 {
1812 return base >= rtas_rmo_buf &&
1813 base < (rtas_rmo_buf + RTAS_USER_REGION_SIZE) &&
1814 base <= end &&
1815 end >= rtas_rmo_buf &&
1816 end < (rtas_rmo_buf + RTAS_USER_REGION_SIZE);
1817 }
1818
1819 static bool block_rtas_call(const struct rtas_function *func, int nargs,
1820 struct rtas_args *args)
1821 {
1822 const struct rtas_filter *f;
1823 const bool is_platform_dump =
1824 func == &rtas_function_table[RTAS_FNIDX__IBM_PLATFORM_DUMP];
1825 const bool is_config_conn =
1826 func == &rtas_function_table[RTAS_FNIDX__IBM_CONFIGURE_CONNECTOR];
1827 u32 base, size, end;
1828
1829 /*
1830 * Only functions with filters attached are allowed.
1831 */
1832 f = func->filter;
1833 if (!f)
1834 goto err;
1835 /*
1836 * And some functions aren't allowed on LE.
1837 */
1838 if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) && func->banned_for_syscall_on_le)
1839 goto err;
1840
1841 if (f->buf_idx1 != -1) {
1842 base = be32_to_cpu(args->args[f->buf_idx1]);
1843 if (f->size_idx1 != -1)
1844 size = be32_to_cpu(args->args[f->size_idx1]);
1845 else if (f->fixed_size)
1846 size = f->fixed_size;
1847 else
1848 size = 1;
1849
1850 end = base + size - 1;
1851
1852 /*
1853 * Special case for ibm,platform-dump - NULL buffer
1854 * address is used to indicate end of dump processing
1855 */
1856 if (is_platform_dump && base == 0)
1857 return false;
1858
1859 if (!in_rmo_buf(base, end))
1860 goto err;
1861 }
1862
1863 if (f->buf_idx2 != -1) {
1864 base = be32_to_cpu(args->args[f->buf_idx2]);
1865 if (f->size_idx2 != -1)
1866 size = be32_to_cpu(args->args[f->size_idx2]);
1867 else if (f->fixed_size)
1868 size = f->fixed_size;
1869 else
1870 size = 1;
1871 end = base + size - 1;
1872
1873 /*
1874 * Special case for ibm,configure-connector where the
1875 * address can be 0
1876 */
1877 if (is_config_conn && base == 0)
1878 return false;
1879
1880 if (!in_rmo_buf(base, end))
1881 goto err;
1882 }
1883
1884 return false;
1885 err:
1886 pr_err_ratelimited("sys_rtas: RTAS call blocked - exploit attempt?\n");
1887 pr_err_ratelimited("sys_rtas: %s nargs=%d (called by %s)\n",
1888 func->name, nargs, current->comm);
1889 return true;
1890 }
1891
1892 /* We assume to be passed big endian arguments */
1893 SYSCALL_DEFINE1(rtas, struct rtas_args __user *, uargs)
1894 {
1895 const struct rtas_function *func;
1896 struct pin_cookie cookie;
1897 struct rtas_args args;
1898 unsigned long flags;
1899 char *buff_copy, *errbuf = NULL;
1900 int nargs, nret, token;
1901
1902 if (!capable(CAP_SYS_ADMIN))
1903 return -EPERM;
1904
1905 if (!rtas.entry)
1906 return -EINVAL;
1907
1908 if (copy_from_user(&args, uargs, 3 * sizeof(u32)) != 0)
1909 return -EFAULT;
1910
1911 nargs = be32_to_cpu(args.nargs);
1912 nret = be32_to_cpu(args.nret);
1913 token = be32_to_cpu(args.token);
1914
1915 if (nargs >= ARRAY_SIZE(args.args)
1916 || nret > ARRAY_SIZE(args.args)
1917 || nargs + nret > ARRAY_SIZE(args.args))
1918 return -EINVAL;
1919
1920 nargs = array_index_nospec(nargs, ARRAY_SIZE(args.args));
1921 nret = array_index_nospec(nret, ARRAY_SIZE(args.args) - nargs);
1922
1923 /* Copy in args. */
1924 if (copy_from_user(args.args, uargs->args,
1925 nargs * sizeof(rtas_arg_t)) != 0)
1926 return -EFAULT;
1927
1928 /*
1929 * If this token doesn't correspond to a function the kernel
1930 * understands, you're not allowed to call it.
1931 */
1932 func = rtas_token_to_function_untrusted(token);
1933 if (!func)
1934 return -EINVAL;
1935
1936 args.rets = &args.args[nargs];
1937 memset(args.rets, 0, nret * sizeof(rtas_arg_t));
1938
1939 if (block_rtas_call(func, nargs, &args))
1940 return -EINVAL;
1941
1942 if (token_is_restricted_errinjct(token)) {
1943 int err;
1944
1945 err = security_locked_down(LOCKDOWN_RTAS_ERROR_INJECTION);
1946 if (err)
1947 return err;
1948 }
1949
1950 /* Need to handle ibm,suspend_me call specially */
1951 if (token == rtas_function_token(RTAS_FN_IBM_SUSPEND_ME)) {
1952
1953 /*
1954 * rtas_ibm_suspend_me assumes the streamid handle is in cpu
1955 * endian, or at least the hcall within it requires it.
1956 */
1957 int rc = 0;
1958 u64 handle = ((u64)be32_to_cpu(args.args[0]) << 32)
1959 | be32_to_cpu(args.args[1]);
1960 rc = rtas_syscall_dispatch_ibm_suspend_me(handle);
1961 if (rc == -EAGAIN)
1962 args.rets[0] = cpu_to_be32(RTAS_NOT_SUSPENDABLE);
1963 else if (rc == -EIO)
1964 args.rets[0] = cpu_to_be32(-1);
1965 else if (rc)
1966 return rc;
1967 goto copy_return;
1968 }
1969
1970 buff_copy = get_errorlog_buffer();
1971
1972 /*
1973 * If this function has a mutex assigned to it, we must
1974 * acquire it to avoid interleaving with any kernel-based uses
1975 * of the same function. Kernel-based sequences acquire the
1976 * appropriate mutex explicitly.
1977 */
1978 if (func->lock)
1979 mutex_lock(func->lock);
1980
1981 raw_spin_lock_irqsave(&rtas_lock, flags);
1982 cookie = lockdep_pin_lock(&rtas_lock);
1983
1984 rtas_args = args;
1985 do_enter_rtas(&rtas_args);
1986 args = rtas_args;
1987
1988 /* A -1 return code indicates that the last command couldn't
1989 be completed due to a hardware error. */
1990 if (be32_to_cpu(args.rets[0]) == -1)
1991 errbuf = __fetch_rtas_last_error(buff_copy);
1992
1993 lockdep_unpin_lock(&rtas_lock, cookie);
1994 raw_spin_unlock_irqrestore(&rtas_lock, flags);
1995
1996 if (func->lock)
1997 mutex_unlock(func->lock);
1998
1999 if (buff_copy) {
2000 if (errbuf)
2001 log_error(errbuf, ERR_TYPE_RTAS_LOG, 0);
2002 kfree(buff_copy);
2003 }
2004
2005 copy_return:
2006 /* Copy out args. */
2007 if (copy_to_user(uargs->args + nargs,
2008 args.args + nargs,
2009 nret * sizeof(rtas_arg_t)) != 0)
2010 return -EFAULT;
2011
2012 return 0;
2013 }
2014
2015 static void __init rtas_function_table_init(void)
2016 {
2017 struct property *prop;
2018
2019 for (size_t i = 0; i < ARRAY_SIZE(rtas_function_table); ++i) {
2020 struct rtas_function *curr = &rtas_function_table[i];
2021 struct rtas_function *prior;
2022 int cmp;
2023
2024 curr->token = RTAS_UNKNOWN_SERVICE;
2025
2026 if (i == 0)
2027 continue;
2028 /*
2029 * Ensure table is sorted correctly for binary search
2030 * on function names.
2031 */
2032 prior = &rtas_function_table[i - 1];
2033
2034 cmp = strcmp(prior->name, curr->name);
2035 if (cmp < 0)
2036 continue;
2037
2038 if (cmp == 0) {
2039 pr_err("'%s' has duplicate function table entries\n",
2040 curr->name);
2041 } else {
2042 pr_err("function table unsorted: '%s' wrongly precedes '%s'\n",
2043 prior->name, curr->name);
2044 }
2045 }
2046
2047 for_each_property_of_node(rtas.dev, prop) {
2048 struct rtas_function *func;
2049
2050 if (prop->length != sizeof(u32))
2051 continue;
2052
2053 func = __rtas_name_to_function(prop->name);
2054 if (!func)
2055 continue;
2056
2057 func->token = be32_to_cpup((__be32 *)prop->value);
2058
2059 pr_debug("function %s has token %u\n", func->name, func->token);
2060 }
2061 }
2062
2063 /*
2064 * Call early during boot, before mem init, to retrieve the RTAS
2065 * information from the device-tree and allocate the RMO buffer for userland
2066 * accesses.
2067 */
2068 void __init rtas_initialize(void)
2069 {
2070 unsigned long rtas_region = RTAS_INSTANTIATE_MAX;
2071 u32 base, size, entry;
2072 int no_base, no_size, no_entry;
2073
2074 /* Get RTAS dev node and fill up our "rtas" structure with infos
2075 * about it.
2076 */
2077 rtas.dev = of_find_node_by_name(NULL, "rtas");
2078 if (!rtas.dev)
2079 return;
2080
2081 no_base = of_property_read_u32(rtas.dev, "linux,rtas-base", &base);
2082 no_size = of_property_read_u32(rtas.dev, "rtas-size", &size);
2083 if (no_base || no_size) {
2084 of_node_put(rtas.dev);
2085 rtas.dev = NULL;
2086 return;
2087 }
2088
2089 rtas.base = base;
2090 rtas.size = size;
2091 no_entry = of_property_read_u32(rtas.dev, "linux,rtas-entry", &entry);
2092 rtas.entry = no_entry ? rtas.base : entry;
2093
2094 init_error_log_max();
2095
2096 /* Must be called before any function token lookups */
2097 rtas_function_table_init();
2098
2099 /*
2100 * Discover this now to avoid a device tree lookup in the
2101 * panic path.
2102 */
2103 ibm_extended_os_term = of_property_read_bool(rtas.dev, "ibm,extended-os-term");
2104
2105 /* If RTAS was found, allocate the RMO buffer for it and look for
2106 * the stop-self token if any
2107 */
2108 #ifdef CONFIG_PPC64
2109 if (firmware_has_feature(FW_FEATURE_LPAR))
2110 rtas_region = min(ppc64_rma_size, RTAS_INSTANTIATE_MAX);
2111 #endif
2112 rtas_rmo_buf = memblock_phys_alloc_range(RTAS_USER_REGION_SIZE, PAGE_SIZE,
2113 0, rtas_region);
2114 if (!rtas_rmo_buf)
2115 panic("ERROR: RTAS: Failed to allocate %lx bytes below %pa\n",
2116 PAGE_SIZE, &rtas_region);
2117
2118 rtas_work_area_reserve_arena(rtas_region);
2119 }
2120
2121 int __init early_init_dt_scan_rtas(unsigned long node,
2122 const char *uname, int depth, void *data)
2123 {
2124 const u32 *basep, *entryp, *sizep;
2125
2126 if (depth != 1 || strcmp(uname, "rtas") != 0)
2127 return 0;
2128
2129 basep = of_get_flat_dt_prop(node, "linux,rtas-base", NULL);
2130 entryp = of_get_flat_dt_prop(node, "linux,rtas-entry", NULL);
2131 sizep = of_get_flat_dt_prop(node, "rtas-size", NULL);
2132
2133 #ifdef CONFIG_PPC64
2134 /* need this feature to decide the crashkernel offset */
2135 if (of_get_flat_dt_prop(node, "ibm,hypertas-functions", NULL))
2136 powerpc_firmware_features |= FW_FEATURE_LPAR;
2137 #endif
2138
2139 if (basep && entryp && sizep) {
2140 rtas.base = *basep;
2141 rtas.entry = *entryp;
2142 rtas.size = *sizep;
2143 }
2144
2145 #ifdef CONFIG_UDBG_RTAS_CONSOLE
2146 basep = of_get_flat_dt_prop(node, "put-term-char", NULL);
2147 if (basep)
2148 rtas_putchar_token = *basep;
2149
2150 basep = of_get_flat_dt_prop(node, "get-term-char", NULL);
2151 if (basep)
2152 rtas_getchar_token = *basep;
2153
2154 if (rtas_putchar_token != RTAS_UNKNOWN_SERVICE &&
2155 rtas_getchar_token != RTAS_UNKNOWN_SERVICE)
2156 udbg_init_rtas_console();
2157
2158 #endif
2159
2160 /* break now */
2161 return 1;
2162 }
2163
2164 static DEFINE_RAW_SPINLOCK(timebase_lock);
2165 static u64 timebase = 0;
2166
2167 void rtas_give_timebase(void)
2168 {
2169 unsigned long flags;
2170
2171 raw_spin_lock_irqsave(&timebase_lock, flags);
2172 hard_irq_disable();
2173 rtas_call(rtas_function_token(RTAS_FN_FREEZE_TIME_BASE), 0, 1, NULL);
2174 timebase = get_tb();
2175 raw_spin_unlock(&timebase_lock);
2176
2177 while (timebase)
2178 barrier();
2179 rtas_call(rtas_function_token(RTAS_FN_THAW_TIME_BASE), 0, 1, NULL);
2180 local_irq_restore(flags);
2181 }
2182
2183 void rtas_take_timebase(void)
2184 {
2185 while (!timebase)
2186 barrier();
2187 raw_spin_lock(&timebase_lock);
2188 set_tb(timebase >> 32, timebase & 0xffffffff);
2189 timebase = 0;
2190 raw_spin_unlock(&timebase_lock);
2191 }
2192
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.