Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * File: mca_drv.c
4 * Purpose: Generic MCA handling layer
5 *
6 * Copyright (C) 2004 FUJITSU LIMITED
7 * Copyright (C) 2004 Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
8 * Copyright (C) 2005 Silicon Graphics, Inc
9 * Copyright (C) 2005 Keith Owens <kaos@sgi.com>
10 * Copyright (C) 2006 Russ Anderson <rja@sgi.com>
11 */
12#include <linux/types.h>
13#include <linux/init.h>
14#include <linux/sched.h>
15#include <linux/interrupt.h>
16#include <linux/irq.h>
17#include <linux/kallsyms.h>
18#include <linux/memblock.h>
19#include <linux/acpi.h>
20#include <linux/timer.h>
21#include <linux/module.h>
22#include <linux/kernel.h>
23#include <linux/smp.h>
24#include <linux/workqueue.h>
25#include <linux/mm.h>
26#include <linux/slab.h>
27
28#include <asm/delay.h>
29#include <asm/page.h>
30#include <asm/ptrace.h>
31#include <asm/sal.h>
32#include <asm/mca.h>
33
34#include <asm/irq.h>
35#include <asm/hw_irq.h>
36
37#include "mca_drv.h"
38
39/* max size of SAL error record (default) */
40static int sal_rec_max = 10000;
41
42/* from mca_drv_asm.S */
43extern void *mca_handler_bhhook(void);
44
45static DEFINE_SPINLOCK(mca_bh_lock);
46
47typedef enum {
48 MCA_IS_LOCAL = 0,
49 MCA_IS_GLOBAL = 1
50} mca_type_t;
51
52#define MAX_PAGE_ISOLATE 1024
53
54static struct page *page_isolate[MAX_PAGE_ISOLATE];
55static int num_page_isolate = 0;
56
57typedef enum {
58 ISOLATE_NG,
59 ISOLATE_OK,
60 ISOLATE_NONE
61} isolate_status_t;
62
63typedef enum {
64 MCA_NOT_RECOVERED = 0,
65 MCA_RECOVERED = 1
66} recovery_status_t;
67
68/*
69 * This pool keeps pointers to the section part of SAL error record
70 */
71static struct {
72 slidx_list_t *buffer; /* section pointer list pool */
73 int cur_idx; /* Current index of section pointer list pool */
74 int max_idx; /* Maximum index of section pointer list pool */
75} slidx_pool;
76
77static int
78fatal_mca(const char *fmt, ...)
79{
80 va_list args;
81 char buf[256];
82
83 va_start(args, fmt);
84 vsnprintf(buf, sizeof(buf), fmt, args);
85 va_end(args);
86 ia64_mca_printk(KERN_ALERT "MCA: %s\n", buf);
87
88 return MCA_NOT_RECOVERED;
89}
90
91static int
92mca_recovered(const char *fmt, ...)
93{
94 va_list args;
95 char buf[256];
96
97 va_start(args, fmt);
98 vsnprintf(buf, sizeof(buf), fmt, args);
99 va_end(args);
100 ia64_mca_printk(KERN_INFO "MCA: %s\n", buf);
101
102 return MCA_RECOVERED;
103}
104
105/**
106 * mca_page_isolate - isolate a poisoned page in order not to use it later
107 * @paddr: poisoned memory location
108 *
109 * Return value:
110 * one of isolate_status_t, ISOLATE_OK/NG/NONE.
111 */
112
113static isolate_status_t
114mca_page_isolate(unsigned long paddr)
115{
116 int i;
117 struct page *p;
118
119 /* whether physical address is valid or not */
120 if (!ia64_phys_addr_valid(paddr))
121 return ISOLATE_NONE;
122
123 if (!pfn_valid(paddr >> PAGE_SHIFT))
124 return ISOLATE_NONE;
125
126 /* convert physical address to physical page number */
127 p = pfn_to_page(paddr>>PAGE_SHIFT);
128
129 /* check whether a page number have been already registered or not */
130 for (i = 0; i < num_page_isolate; i++)
131 if (page_isolate[i] == p)
132 return ISOLATE_OK; /* already listed */
133
134 /* limitation check */
135 if (num_page_isolate == MAX_PAGE_ISOLATE)
136 return ISOLATE_NG;
137
138 /* kick pages having attribute 'SLAB' or 'Reserved' */
139 if (PageSlab(p) || PageReserved(p))
140 return ISOLATE_NG;
141
142 /* add attribute 'Reserved' and register the page */
143 get_page(p);
144 SetPageReserved(p);
145 page_isolate[num_page_isolate++] = p;
146
147 return ISOLATE_OK;
148}
149
150/**
151 * mca_hanlder_bh - Kill the process which occurred memory read error
152 * @paddr: poisoned address received from MCA Handler
153 */
154
155void
156mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr)
157{
158 ia64_mlogbuf_dump();
159 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, "
160 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n",
161 raw_smp_processor_id(), current->pid,
162 from_kuid(&init_user_ns, current_uid()),
163 iip, ipsr, paddr, current->comm);
164
165 spin_lock(&mca_bh_lock);
166 switch (mca_page_isolate(paddr)) {
167 case ISOLATE_OK:
168 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr);
169 break;
170 case ISOLATE_NG:
171 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr);
172 break;
173 default:
174 break;
175 }
176 spin_unlock(&mca_bh_lock);
177
178 /* This process is about to be killed itself */
179 do_exit(SIGKILL);
180}
181
182/**
183 * mca_make_peidx - Make index of processor error section
184 * @slpi: pointer to record of processor error section
185 * @peidx: pointer to index of processor error section
186 */
187
188static void
189mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx)
190{
191 /*
192 * calculate the start address of
193 * "struct cpuid_info" and "sal_processor_static_info_t".
194 */
195 u64 total_check_num = slpi->valid.num_cache_check
196 + slpi->valid.num_tlb_check
197 + slpi->valid.num_bus_check
198 + slpi->valid.num_reg_file_check
199 + slpi->valid.num_ms_check;
200 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num
201 + sizeof(sal_log_processor_info_t);
202 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info);
203
204 peidx_head(peidx) = slpi;
205 peidx_mid(peidx) = (struct sal_cpuid_info *)
206 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL);
207 peidx_bottom(peidx) = (sal_processor_static_info_t *)
208 (slpi->valid.psi_static_struct ?
209 ((char*)slpi + head_size + mid_size) : NULL);
210}
211
212/**
213 * mca_make_slidx - Make index of SAL error record
214 * @buffer: pointer to SAL error record
215 * @slidx: pointer to index of SAL error record
216 *
217 * Return value:
218 * 1 if record has platform error / 0 if not
219 */
220#define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \
221 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \
222 hl->hdr = ptr; \
223 list_add(&hl->list, &(sect)); \
224 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; }
225
226static int
227mca_make_slidx(void *buffer, slidx_table_t *slidx)
228{
229 int platform_err = 0;
230 int record_len = ((sal_log_record_header_t*)buffer)->len;
231 u32 ercd_pos;
232 int sects;
233 sal_log_section_hdr_t *sp;
234
235 /*
236 * Initialize index referring current record
237 */
238 INIT_LIST_HEAD(&(slidx->proc_err));
239 INIT_LIST_HEAD(&(slidx->mem_dev_err));
240 INIT_LIST_HEAD(&(slidx->sel_dev_err));
241 INIT_LIST_HEAD(&(slidx->pci_bus_err));
242 INIT_LIST_HEAD(&(slidx->smbios_dev_err));
243 INIT_LIST_HEAD(&(slidx->pci_comp_err));
244 INIT_LIST_HEAD(&(slidx->plat_specific_err));
245 INIT_LIST_HEAD(&(slidx->host_ctlr_err));
246 INIT_LIST_HEAD(&(slidx->plat_bus_err));
247 INIT_LIST_HEAD(&(slidx->unsupported));
248
249 /*
250 * Extract a Record Header
251 */
252 slidx->header = buffer;
253
254 /*
255 * Extract each section records
256 * (arranged from "int ia64_log_platform_info_print()")
257 */
258 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0;
259 ercd_pos < record_len; ercd_pos += sp->len, sects++) {
260 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos);
261 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) {
262 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp);
263 } else if (!efi_guidcmp(sp->guid,
264 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) {
265 platform_err = 1;
266 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp);
267 } else if (!efi_guidcmp(sp->guid,
268 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) {
269 platform_err = 1;
270 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp);
271 } else if (!efi_guidcmp(sp->guid,
272 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) {
273 platform_err = 1;
274 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp);
275 } else if (!efi_guidcmp(sp->guid,
276 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) {
277 platform_err = 1;
278 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp);
279 } else if (!efi_guidcmp(sp->guid,
280 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) {
281 platform_err = 1;
282 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp);
283 } else if (!efi_guidcmp(sp->guid,
284 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) {
285 platform_err = 1;
286 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp);
287 } else if (!efi_guidcmp(sp->guid,
288 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) {
289 platform_err = 1;
290 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp);
291 } else if (!efi_guidcmp(sp->guid,
292 SAL_PLAT_BUS_ERR_SECT_GUID)) {
293 platform_err = 1;
294 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp);
295 } else {
296 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp);
297 }
298 }
299 slidx->n_sections = sects;
300
301 return platform_err;
302}
303
304/**
305 * init_record_index_pools - Initialize pool of lists for SAL record index
306 *
307 * Return value:
308 * 0 on Success / -ENOMEM on Failure
309 */
310static int
311init_record_index_pools(void)
312{
313 int i;
314 int rec_max_size; /* Maximum size of SAL error records */
315 int sect_min_size; /* Minimum size of SAL error sections */
316 /* minimum size table of each section */
317 static int sal_log_sect_min_sizes[] = {
318 sizeof(sal_log_processor_info_t)
319 + sizeof(sal_processor_static_info_t),
320 sizeof(sal_log_mem_dev_err_info_t),
321 sizeof(sal_log_sel_dev_err_info_t),
322 sizeof(sal_log_pci_bus_err_info_t),
323 sizeof(sal_log_smbios_dev_err_info_t),
324 sizeof(sal_log_pci_comp_err_info_t),
325 sizeof(sal_log_plat_specific_err_info_t),
326 sizeof(sal_log_host_ctlr_err_info_t),
327 sizeof(sal_log_plat_bus_err_info_t),
328 };
329
330 /*
331 * MCA handler cannot allocate new memory on flight,
332 * so we preallocate enough memory to handle a SAL record.
333 *
334 * Initialize a handling set of slidx_pool:
335 * 1. Pick up the max size of SAL error records
336 * 2. Pick up the min size of SAL error sections
337 * 3. Allocate the pool as enough to 2 SAL records
338 * (now we can estimate the maxinum of section in a record.)
339 */
340
341 /* - 1 - */
342 rec_max_size = sal_rec_max;
343
344 /* - 2 - */
345 sect_min_size = sal_log_sect_min_sizes[0];
346 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++)
347 if (sect_min_size > sal_log_sect_min_sizes[i])
348 sect_min_size = sal_log_sect_min_sizes[i];
349
350 /* - 3 - */
351 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1;
352 slidx_pool.buffer =
353 kmalloc_array(slidx_pool.max_idx, sizeof(slidx_list_t),
354 GFP_KERNEL);
355
356 return slidx_pool.buffer ? 0 : -ENOMEM;
357}
358
359
360/*****************************************************************************
361 * Recovery functions *
362 *****************************************************************************/
363
364/**
365 * is_mca_global - Check whether this MCA is global or not
366 * @peidx: pointer of index of processor error section
367 * @pbci: pointer to pal_bus_check_info_t
368 * @sos: pointer to hand off struct between SAL and OS
369 *
370 * Return value:
371 * MCA_IS_LOCAL / MCA_IS_GLOBAL
372 */
373
374static mca_type_t
375is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
376 struct ia64_sal_os_state *sos)
377{
378 pal_processor_state_info_t *psp =
379 (pal_processor_state_info_t*)peidx_psp(peidx);
380
381 /*
382 * PAL can request a rendezvous, if the MCA has a global scope.
383 * If "rz_always" flag is set, SAL requests MCA rendezvous
384 * in spite of global MCA.
385 * Therefore it is local MCA when rendezvous has not been requested.
386 * Failed to rendezvous, the system must be down.
387 */
388 switch (sos->rv_rc) {
389 case -1: /* SAL rendezvous unsuccessful */
390 return MCA_IS_GLOBAL;
391 case 0: /* SAL rendezvous not required */
392 return MCA_IS_LOCAL;
393 case 1: /* SAL rendezvous successful int */
394 case 2: /* SAL rendezvous successful int with init */
395 default:
396 break;
397 }
398
399 /*
400 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
401 * it would be a local MCA. (i.e. processor internal error)
402 */
403 if (psp->tc || psp->cc || psp->rc || psp->uc)
404 return MCA_IS_LOCAL;
405
406 /*
407 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
408 * would be a global MCA. (e.g. a system bus address parity error)
409 */
410 if (!pbci || pbci->ib)
411 return MCA_IS_GLOBAL;
412
413 /*
414 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
415 * could be either a local MCA or a global MCA.
416 *
417 * Referring Bus_Check.bsi:
418 * 0: Unknown/unclassified
419 * 1: BERR#
420 * 2: BINIT#
421 * 3: Hard Fail
422 * (FIXME: Are these SGI specific or generic bsi values?)
423 */
424 if (pbci->eb)
425 switch (pbci->bsi) {
426 case 0:
427 /* e.g. a load from poisoned memory */
428 return MCA_IS_LOCAL;
429 case 1:
430 case 2:
431 case 3:
432 return MCA_IS_GLOBAL;
433 }
434
435 return MCA_IS_GLOBAL;
436}
437
438/**
439 * get_target_identifier - Get the valid Cache or Bus check target identifier.
440 * @peidx: pointer of index of processor error section
441 *
442 * Return value:
443 * target address on Success / 0 on Failure
444 */
445static u64
446get_target_identifier(peidx_table_t *peidx)
447{
448 u64 target_address = 0;
449 sal_log_mod_error_info_t *smei;
450 pal_cache_check_info_t *pcci;
451 int i, level = 9;
452
453 /*
454 * Look through the cache checks for a valid target identifier
455 * If more than one valid target identifier, return the one
456 * with the lowest cache level.
457 */
458 for (i = 0; i < peidx_cache_check_num(peidx); i++) {
459 smei = (sal_log_mod_error_info_t *)peidx_cache_check(peidx, i);
460 if (smei->valid.target_identifier && smei->target_identifier) {
461 pcci = (pal_cache_check_info_t *)&(smei->check_info);
462 if (!target_address || (pcci->level < level)) {
463 target_address = smei->target_identifier;
464 level = pcci->level;
465 continue;
466 }
467 }
468 }
469 if (target_address)
470 return target_address;
471
472 /*
473 * Look at the bus check for a valid target identifier
474 */
475 smei = peidx_bus_check(peidx, 0);
476 if (smei && smei->valid.target_identifier)
477 return smei->target_identifier;
478
479 return 0;
480}
481
482/**
483 * recover_from_read_error - Try to recover the errors which type are "read"s.
484 * @slidx: pointer of index of SAL error record
485 * @peidx: pointer of index of processor error section
486 * @pbci: pointer of pal_bus_check_info
487 * @sos: pointer to hand off struct between SAL and OS
488 *
489 * Return value:
490 * 1 on Success / 0 on Failure
491 */
492
493static int
494recover_from_read_error(slidx_table_t *slidx,
495 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
496 struct ia64_sal_os_state *sos)
497{
498 u64 target_identifier;
499 pal_min_state_area_t *pmsa;
500 struct ia64_psr *psr1, *psr2;
501 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;
502
503 /* Is target address valid? */
504 target_identifier = get_target_identifier(peidx);
505 if (!target_identifier)
506 return fatal_mca("target address not valid");
507
508 /*
509 * cpu read or memory-mapped io read
510 *
511 * offending process affected process OS MCA do
512 * kernel mode kernel mode down system
513 * kernel mode user mode kill the process
514 * user mode kernel mode down system (*)
515 * user mode user mode kill the process
516 *
517 * (*) You could terminate offending user-mode process
518 * if (pbci->pv && pbci->pl != 0) *and* if you sure
519 * the process not have any locks of kernel.
520 */
521
522 /* Is minstate valid? */
523 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate))
524 return fatal_mca("minstate not valid");
525 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
526 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr);
527
528 /*
529 * Check the privilege level of interrupted context.
530 * If it is user-mode, then terminate affected process.
531 */
532
533 pmsa = sos->pal_min_state;
534 if (psr1->cpl != 0 ||
535 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) {
536 /*
537 * setup for resume to bottom half of MCA,
538 * "mca_handler_bhhook"
539 */
540 /* pass to bhhook as argument (gr8, ...) */
541 pmsa->pmsa_gr[8-1] = target_identifier;
542 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip;
543 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr;
544 /* set interrupted return address (but no use) */
545 pmsa->pmsa_br0 = pmsa->pmsa_iip;
546 /* change resume address to bottom half */
547 pmsa->pmsa_iip = mca_hdlr_bh->fp;
548 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
549 /* set cpl with kernel mode */
550 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
551 psr2->cpl = 0;
552 psr2->ri = 0;
553 psr2->bn = 1;
554 psr2->i = 0;
555
556 return mca_recovered("user memory corruption. "
557 "kill affected process - recovered.");
558 }
559
560 return fatal_mca("kernel context not recovered, iip 0x%lx\n",
561 pmsa->pmsa_iip);
562}
563
564/**
565 * recover_from_platform_error - Recover from platform error.
566 * @slidx: pointer of index of SAL error record
567 * @peidx: pointer of index of processor error section
568 * @pbci: pointer of pal_bus_check_info
569 * @sos: pointer to hand off struct between SAL and OS
570 *
571 * Return value:
572 * 1 on Success / 0 on Failure
573 */
574
575static int
576recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
577 pal_bus_check_info_t *pbci,
578 struct ia64_sal_os_state *sos)
579{
580 int status = 0;
581 pal_processor_state_info_t *psp =
582 (pal_processor_state_info_t*)peidx_psp(peidx);
583
584 if (psp->bc && pbci->eb && pbci->bsi == 0) {
585 switch(pbci->type) {
586 case 1: /* partial read */
587 case 3: /* full line(cpu) read */
588 case 9: /* I/O space read */
589 status = recover_from_read_error(slidx, peidx, pbci,
590 sos);
591 break;
592 case 0: /* unknown */
593 case 2: /* partial write */
594 case 4: /* full line write */
595 case 5: /* implicit or explicit write-back operation */
596 case 6: /* snoop probe */
597 case 7: /* incoming or outgoing ptc.g */
598 case 8: /* write coalescing transactions */
599 case 10: /* I/O space write */
600 case 11: /* inter-processor interrupt message(IPI) */
601 case 12: /* interrupt acknowledge or
602 external task priority cycle */
603 default:
604 break;
605 }
606 } else if (psp->cc && !psp->bc) { /* Cache error */
607 status = recover_from_read_error(slidx, peidx, pbci, sos);
608 }
609
610 return status;
611}
612
613/*
614 * recover_from_tlb_check
615 * @peidx: pointer of index of processor error section
616 *
617 * Return value:
618 * 1 on Success / 0 on Failure
619 */
620static int
621recover_from_tlb_check(peidx_table_t *peidx)
622{
623 sal_log_mod_error_info_t *smei;
624 pal_tlb_check_info_t *ptci;
625
626 smei = (sal_log_mod_error_info_t *)peidx_tlb_check(peidx, 0);
627 ptci = (pal_tlb_check_info_t *)&(smei->check_info);
628
629 /*
630 * Look for signature of a duplicate TLB DTC entry, which is
631 * a SW bug and always fatal.
632 */
633 if (ptci->op == PAL_TLB_CHECK_OP_PURGE
634 && !(ptci->itr || ptci->dtc || ptci->itc))
635 return fatal_mca("Duplicate TLB entry");
636
637 return mca_recovered("TLB check recovered");
638}
639
640/**
641 * recover_from_processor_error
642 * @platform: whether there are some platform error section or not
643 * @slidx: pointer of index of SAL error record
644 * @peidx: pointer of index of processor error section
645 * @pbci: pointer of pal_bus_check_info
646 * @sos: pointer to hand off struct between SAL and OS
647 *
648 * Return value:
649 * 1 on Success / 0 on Failure
650 */
651
652static int
653recover_from_processor_error(int platform, slidx_table_t *slidx,
654 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
655 struct ia64_sal_os_state *sos)
656{
657 pal_processor_state_info_t *psp =
658 (pal_processor_state_info_t*)peidx_psp(peidx);
659
660 /*
661 * Processor recovery status must key off of the PAL recovery
662 * status in the Processor State Parameter.
663 */
664
665 /*
666 * The machine check is corrected.
667 */
668 if (psp->cm == 1)
669 return mca_recovered("machine check is already corrected.");
670
671 /*
672 * The error was not contained. Software must be reset.
673 */
674 if (psp->us || psp->ci == 0)
675 return fatal_mca("error not contained");
676
677 /*
678 * Look for recoverable TLB check
679 */
680 if (psp->tc && !(psp->cc || psp->bc || psp->rc || psp->uc))
681 return recover_from_tlb_check(peidx);
682
683 /*
684 * The cache check and bus check bits have four possible states
685 * cc bc
686 * 1 1 Memory error, attempt recovery
687 * 1 0 Cache error, attempt recovery
688 * 0 1 I/O error, attempt recovery
689 * 0 0 Other error type, not recovered
690 */
691 if (psp->cc == 0 && (psp->bc == 0 || pbci == NULL))
692 return fatal_mca("No cache or bus check");
693
694 /*
695 * Cannot handle more than one bus check.
696 */
697 if (peidx_bus_check_num(peidx) > 1)
698 return fatal_mca("Too many bus checks");
699
700 if (pbci->ib)
701 return fatal_mca("Internal Bus error");
702 if (pbci->eb && pbci->bsi > 0)
703 return fatal_mca("External bus check fatal status");
704
705 /*
706 * This is a local MCA and estimated as a recoverable error.
707 */
708 if (platform)
709 return recover_from_platform_error(slidx, peidx, pbci, sos);
710
711 /*
712 * On account of strange SAL error record, we cannot recover.
713 */
714 return fatal_mca("Strange SAL record");
715}
716
717/**
718 * mca_try_to_recover - Try to recover from MCA
719 * @rec: pointer to a SAL error record
720 * @sos: pointer to hand off struct between SAL and OS
721 *
722 * Return value:
723 * 1 on Success / 0 on Failure
724 */
725
726static int
727mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
728{
729 int platform_err;
730 int n_proc_err;
731 slidx_table_t slidx;
732 peidx_table_t peidx;
733 pal_bus_check_info_t pbci;
734
735 /* Make index of SAL error record */
736 platform_err = mca_make_slidx(rec, &slidx);
737
738 /* Count processor error sections */
739 n_proc_err = slidx_count(&slidx, proc_err);
740
741 /* Now, OS can recover when there is one processor error section */
742 if (n_proc_err > 1)
743 return fatal_mca("Too Many Errors");
744 else if (n_proc_err == 0)
745 /* Weird SAL record ... We can't do anything */
746 return fatal_mca("Weird SAL record");
747
748 /* Make index of processor error section */
749 mca_make_peidx((sal_log_processor_info_t*)
750 slidx_first_entry(&slidx.proc_err)->hdr, &peidx);
751
752 /* Extract Processor BUS_CHECK[0] */
753 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
754
755 /* Check whether MCA is global or not */
756 if (is_mca_global(&peidx, &pbci, sos))
757 return fatal_mca("global MCA");
758
759 /* Try to recover a processor error */
760 return recover_from_processor_error(platform_err, &slidx, &peidx,
761 &pbci, sos);
762}
763
764/*
765 * =============================================================================
766 */
767
768int __init mca_external_handler_init(void)
769{
770 if (init_record_index_pools())
771 return -ENOMEM;
772
773 /* register external mca handlers */
774 if (ia64_reg_MCA_extension(mca_try_to_recover)) {
775 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
776 kfree(slidx_pool.buffer);
777 return -EFAULT;
778 }
779 return 0;
780}
781
782void __exit mca_external_handler_exit(void)
783{
784 /* unregister external mca handlers */
785 ia64_unreg_MCA_extension();
786 kfree(slidx_pool.buffer);
787}
788
789module_init(mca_external_handler_init);
790module_exit(mca_external_handler_exit);
791
792module_param(sal_rec_max, int, 0644);
793MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");
794
795MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
796MODULE_LICENSE("GPL");
1/*
2 * File: mca_drv.c
3 * Purpose: Generic MCA handling layer
4 *
5 * Copyright (C) 2004 FUJITSU LIMITED
6 * Copyright (C) 2004 Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
7 * Copyright (C) 2005 Silicon Graphics, Inc
8 * Copyright (C) 2005 Keith Owens <kaos@sgi.com>
9 * Copyright (C) 2006 Russ Anderson <rja@sgi.com>
10 */
11#include <linux/types.h>
12#include <linux/init.h>
13#include <linux/sched.h>
14#include <linux/interrupt.h>
15#include <linux/irq.h>
16#include <linux/kallsyms.h>
17#include <linux/bootmem.h>
18#include <linux/acpi.h>
19#include <linux/timer.h>
20#include <linux/module.h>
21#include <linux/kernel.h>
22#include <linux/smp.h>
23#include <linux/workqueue.h>
24#include <linux/mm.h>
25#include <linux/slab.h>
26
27#include <asm/delay.h>
28#include <asm/machvec.h>
29#include <asm/page.h>
30#include <asm/ptrace.h>
31#include <asm/sal.h>
32#include <asm/mca.h>
33
34#include <asm/irq.h>
35#include <asm/hw_irq.h>
36
37#include "mca_drv.h"
38
39/* max size of SAL error record (default) */
40static int sal_rec_max = 10000;
41
42/* from mca_drv_asm.S */
43extern void *mca_handler_bhhook(void);
44
45static DEFINE_SPINLOCK(mca_bh_lock);
46
47typedef enum {
48 MCA_IS_LOCAL = 0,
49 MCA_IS_GLOBAL = 1
50} mca_type_t;
51
52#define MAX_PAGE_ISOLATE 1024
53
54static struct page *page_isolate[MAX_PAGE_ISOLATE];
55static int num_page_isolate = 0;
56
57typedef enum {
58 ISOLATE_NG,
59 ISOLATE_OK,
60 ISOLATE_NONE
61} isolate_status_t;
62
63typedef enum {
64 MCA_NOT_RECOVERED = 0,
65 MCA_RECOVERED = 1
66} recovery_status_t;
67
68/*
69 * This pool keeps pointers to the section part of SAL error record
70 */
71static struct {
72 slidx_list_t *buffer; /* section pointer list pool */
73 int cur_idx; /* Current index of section pointer list pool */
74 int max_idx; /* Maximum index of section pointer list pool */
75} slidx_pool;
76
77static int
78fatal_mca(const char *fmt, ...)
79{
80 va_list args;
81 char buf[256];
82
83 va_start(args, fmt);
84 vsnprintf(buf, sizeof(buf), fmt, args);
85 va_end(args);
86 ia64_mca_printk(KERN_ALERT "MCA: %s\n", buf);
87
88 return MCA_NOT_RECOVERED;
89}
90
91static int
92mca_recovered(const char *fmt, ...)
93{
94 va_list args;
95 char buf[256];
96
97 va_start(args, fmt);
98 vsnprintf(buf, sizeof(buf), fmt, args);
99 va_end(args);
100 ia64_mca_printk(KERN_INFO "MCA: %s\n", buf);
101
102 return MCA_RECOVERED;
103}
104
105/**
106 * mca_page_isolate - isolate a poisoned page in order not to use it later
107 * @paddr: poisoned memory location
108 *
109 * Return value:
110 * one of isolate_status_t, ISOLATE_OK/NG/NONE.
111 */
112
113static isolate_status_t
114mca_page_isolate(unsigned long paddr)
115{
116 int i;
117 struct page *p;
118
119 /* whether physical address is valid or not */
120 if (!ia64_phys_addr_valid(paddr))
121 return ISOLATE_NONE;
122
123 if (!pfn_valid(paddr >> PAGE_SHIFT))
124 return ISOLATE_NONE;
125
126 /* convert physical address to physical page number */
127 p = pfn_to_page(paddr>>PAGE_SHIFT);
128
129 /* check whether a page number have been already registered or not */
130 for (i = 0; i < num_page_isolate; i++)
131 if (page_isolate[i] == p)
132 return ISOLATE_OK; /* already listed */
133
134 /* limitation check */
135 if (num_page_isolate == MAX_PAGE_ISOLATE)
136 return ISOLATE_NG;
137
138 /* kick pages having attribute 'SLAB' or 'Reserved' */
139 if (PageSlab(p) || PageReserved(p))
140 return ISOLATE_NG;
141
142 /* add attribute 'Reserved' and register the page */
143 get_page(p);
144 SetPageReserved(p);
145 page_isolate[num_page_isolate++] = p;
146
147 return ISOLATE_OK;
148}
149
150/**
151 * mca_hanlder_bh - Kill the process which occurred memory read error
152 * @paddr: poisoned address received from MCA Handler
153 */
154
155void
156mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr)
157{
158 ia64_mlogbuf_dump();
159 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, "
160 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n",
161 raw_smp_processor_id(), current->pid, current_uid(),
162 iip, ipsr, paddr, current->comm);
163
164 spin_lock(&mca_bh_lock);
165 switch (mca_page_isolate(paddr)) {
166 case ISOLATE_OK:
167 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr);
168 break;
169 case ISOLATE_NG:
170 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr);
171 break;
172 default:
173 break;
174 }
175 spin_unlock(&mca_bh_lock);
176
177 /* This process is about to be killed itself */
178 do_exit(SIGKILL);
179}
180
181/**
182 * mca_make_peidx - Make index of processor error section
183 * @slpi: pointer to record of processor error section
184 * @peidx: pointer to index of processor error section
185 */
186
187static void
188mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx)
189{
190 /*
191 * calculate the start address of
192 * "struct cpuid_info" and "sal_processor_static_info_t".
193 */
194 u64 total_check_num = slpi->valid.num_cache_check
195 + slpi->valid.num_tlb_check
196 + slpi->valid.num_bus_check
197 + slpi->valid.num_reg_file_check
198 + slpi->valid.num_ms_check;
199 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num
200 + sizeof(sal_log_processor_info_t);
201 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info);
202
203 peidx_head(peidx) = slpi;
204 peidx_mid(peidx) = (struct sal_cpuid_info *)
205 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL);
206 peidx_bottom(peidx) = (sal_processor_static_info_t *)
207 (slpi->valid.psi_static_struct ?
208 ((char*)slpi + head_size + mid_size) : NULL);
209}
210
211/**
212 * mca_make_slidx - Make index of SAL error record
213 * @buffer: pointer to SAL error record
214 * @slidx: pointer to index of SAL error record
215 *
216 * Return value:
217 * 1 if record has platform error / 0 if not
218 */
219#define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \
220 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \
221 hl->hdr = ptr; \
222 list_add(&hl->list, &(sect)); \
223 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; }
224
225static int
226mca_make_slidx(void *buffer, slidx_table_t *slidx)
227{
228 int platform_err = 0;
229 int record_len = ((sal_log_record_header_t*)buffer)->len;
230 u32 ercd_pos;
231 int sects;
232 sal_log_section_hdr_t *sp;
233
234 /*
235 * Initialize index referring current record
236 */
237 INIT_LIST_HEAD(&(slidx->proc_err));
238 INIT_LIST_HEAD(&(slidx->mem_dev_err));
239 INIT_LIST_HEAD(&(slidx->sel_dev_err));
240 INIT_LIST_HEAD(&(slidx->pci_bus_err));
241 INIT_LIST_HEAD(&(slidx->smbios_dev_err));
242 INIT_LIST_HEAD(&(slidx->pci_comp_err));
243 INIT_LIST_HEAD(&(slidx->plat_specific_err));
244 INIT_LIST_HEAD(&(slidx->host_ctlr_err));
245 INIT_LIST_HEAD(&(slidx->plat_bus_err));
246 INIT_LIST_HEAD(&(slidx->unsupported));
247
248 /*
249 * Extract a Record Header
250 */
251 slidx->header = buffer;
252
253 /*
254 * Extract each section records
255 * (arranged from "int ia64_log_platform_info_print()")
256 */
257 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0;
258 ercd_pos < record_len; ercd_pos += sp->len, sects++) {
259 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos);
260 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) {
261 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp);
262 } else if (!efi_guidcmp(sp->guid,
263 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) {
264 platform_err = 1;
265 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp);
266 } else if (!efi_guidcmp(sp->guid,
267 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) {
268 platform_err = 1;
269 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp);
270 } else if (!efi_guidcmp(sp->guid,
271 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) {
272 platform_err = 1;
273 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp);
274 } else if (!efi_guidcmp(sp->guid,
275 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) {
276 platform_err = 1;
277 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp);
278 } else if (!efi_guidcmp(sp->guid,
279 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) {
280 platform_err = 1;
281 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp);
282 } else if (!efi_guidcmp(sp->guid,
283 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) {
284 platform_err = 1;
285 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp);
286 } else if (!efi_guidcmp(sp->guid,
287 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) {
288 platform_err = 1;
289 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp);
290 } else if (!efi_guidcmp(sp->guid,
291 SAL_PLAT_BUS_ERR_SECT_GUID)) {
292 platform_err = 1;
293 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp);
294 } else {
295 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp);
296 }
297 }
298 slidx->n_sections = sects;
299
300 return platform_err;
301}
302
303/**
304 * init_record_index_pools - Initialize pool of lists for SAL record index
305 *
306 * Return value:
307 * 0 on Success / -ENOMEM on Failure
308 */
309static int
310init_record_index_pools(void)
311{
312 int i;
313 int rec_max_size; /* Maximum size of SAL error records */
314 int sect_min_size; /* Minimum size of SAL error sections */
315 /* minimum size table of each section */
316 static int sal_log_sect_min_sizes[] = {
317 sizeof(sal_log_processor_info_t)
318 + sizeof(sal_processor_static_info_t),
319 sizeof(sal_log_mem_dev_err_info_t),
320 sizeof(sal_log_sel_dev_err_info_t),
321 sizeof(sal_log_pci_bus_err_info_t),
322 sizeof(sal_log_smbios_dev_err_info_t),
323 sizeof(sal_log_pci_comp_err_info_t),
324 sizeof(sal_log_plat_specific_err_info_t),
325 sizeof(sal_log_host_ctlr_err_info_t),
326 sizeof(sal_log_plat_bus_err_info_t),
327 };
328
329 /*
330 * MCA handler cannot allocate new memory on flight,
331 * so we preallocate enough memory to handle a SAL record.
332 *
333 * Initialize a handling set of slidx_pool:
334 * 1. Pick up the max size of SAL error records
335 * 2. Pick up the min size of SAL error sections
336 * 3. Allocate the pool as enough to 2 SAL records
337 * (now we can estimate the maxinum of section in a record.)
338 */
339
340 /* - 1 - */
341 rec_max_size = sal_rec_max;
342
343 /* - 2 - */
344 sect_min_size = sal_log_sect_min_sizes[0];
345 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++)
346 if (sect_min_size > sal_log_sect_min_sizes[i])
347 sect_min_size = sal_log_sect_min_sizes[i];
348
349 /* - 3 - */
350 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1;
351 slidx_pool.buffer = (slidx_list_t *)
352 kmalloc(slidx_pool.max_idx * sizeof(slidx_list_t), GFP_KERNEL);
353
354 return slidx_pool.buffer ? 0 : -ENOMEM;
355}
356
357
358/*****************************************************************************
359 * Recovery functions *
360 *****************************************************************************/
361
362/**
363 * is_mca_global - Check whether this MCA is global or not
364 * @peidx: pointer of index of processor error section
365 * @pbci: pointer to pal_bus_check_info_t
366 * @sos: pointer to hand off struct between SAL and OS
367 *
368 * Return value:
369 * MCA_IS_LOCAL / MCA_IS_GLOBAL
370 */
371
372static mca_type_t
373is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
374 struct ia64_sal_os_state *sos)
375{
376 pal_processor_state_info_t *psp =
377 (pal_processor_state_info_t*)peidx_psp(peidx);
378
379 /*
380 * PAL can request a rendezvous, if the MCA has a global scope.
381 * If "rz_always" flag is set, SAL requests MCA rendezvous
382 * in spite of global MCA.
383 * Therefore it is local MCA when rendezvous has not been requested.
384 * Failed to rendezvous, the system must be down.
385 */
386 switch (sos->rv_rc) {
387 case -1: /* SAL rendezvous unsuccessful */
388 return MCA_IS_GLOBAL;
389 case 0: /* SAL rendezvous not required */
390 return MCA_IS_LOCAL;
391 case 1: /* SAL rendezvous successful int */
392 case 2: /* SAL rendezvous successful int with init */
393 default:
394 break;
395 }
396
397 /*
398 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
399 * it would be a local MCA. (i.e. processor internal error)
400 */
401 if (psp->tc || psp->cc || psp->rc || psp->uc)
402 return MCA_IS_LOCAL;
403
404 /*
405 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
406 * would be a global MCA. (e.g. a system bus address parity error)
407 */
408 if (!pbci || pbci->ib)
409 return MCA_IS_GLOBAL;
410
411 /*
412 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
413 * could be either a local MCA or a global MCA.
414 *
415 * Referring Bus_Check.bsi:
416 * 0: Unknown/unclassified
417 * 1: BERR#
418 * 2: BINIT#
419 * 3: Hard Fail
420 * (FIXME: Are these SGI specific or generic bsi values?)
421 */
422 if (pbci->eb)
423 switch (pbci->bsi) {
424 case 0:
425 /* e.g. a load from poisoned memory */
426 return MCA_IS_LOCAL;
427 case 1:
428 case 2:
429 case 3:
430 return MCA_IS_GLOBAL;
431 }
432
433 return MCA_IS_GLOBAL;
434}
435
436/**
437 * get_target_identifier - Get the valid Cache or Bus check target identifier.
438 * @peidx: pointer of index of processor error section
439 *
440 * Return value:
441 * target address on Success / 0 on Failure
442 */
443static u64
444get_target_identifier(peidx_table_t *peidx)
445{
446 u64 target_address = 0;
447 sal_log_mod_error_info_t *smei;
448 pal_cache_check_info_t *pcci;
449 int i, level = 9;
450
451 /*
452 * Look through the cache checks for a valid target identifier
453 * If more than one valid target identifier, return the one
454 * with the lowest cache level.
455 */
456 for (i = 0; i < peidx_cache_check_num(peidx); i++) {
457 smei = (sal_log_mod_error_info_t *)peidx_cache_check(peidx, i);
458 if (smei->valid.target_identifier && smei->target_identifier) {
459 pcci = (pal_cache_check_info_t *)&(smei->check_info);
460 if (!target_address || (pcci->level < level)) {
461 target_address = smei->target_identifier;
462 level = pcci->level;
463 continue;
464 }
465 }
466 }
467 if (target_address)
468 return target_address;
469
470 /*
471 * Look at the bus check for a valid target identifier
472 */
473 smei = peidx_bus_check(peidx, 0);
474 if (smei && smei->valid.target_identifier)
475 return smei->target_identifier;
476
477 return 0;
478}
479
480/**
481 * recover_from_read_error - Try to recover the errors which type are "read"s.
482 * @slidx: pointer of index of SAL error record
483 * @peidx: pointer of index of processor error section
484 * @pbci: pointer of pal_bus_check_info
485 * @sos: pointer to hand off struct between SAL and OS
486 *
487 * Return value:
488 * 1 on Success / 0 on Failure
489 */
490
491static int
492recover_from_read_error(slidx_table_t *slidx,
493 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
494 struct ia64_sal_os_state *sos)
495{
496 u64 target_identifier;
497 pal_min_state_area_t *pmsa;
498 struct ia64_psr *psr1, *psr2;
499 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;
500
501 /* Is target address valid? */
502 target_identifier = get_target_identifier(peidx);
503 if (!target_identifier)
504 return fatal_mca("target address not valid");
505
506 /*
507 * cpu read or memory-mapped io read
508 *
509 * offending process affected process OS MCA do
510 * kernel mode kernel mode down system
511 * kernel mode user mode kill the process
512 * user mode kernel mode down system (*)
513 * user mode user mode kill the process
514 *
515 * (*) You could terminate offending user-mode process
516 * if (pbci->pv && pbci->pl != 0) *and* if you sure
517 * the process not have any locks of kernel.
518 */
519
520 /* Is minstate valid? */
521 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate))
522 return fatal_mca("minstate not valid");
523 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
524 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr);
525
526 /*
527 * Check the privilege level of interrupted context.
528 * If it is user-mode, then terminate affected process.
529 */
530
531 pmsa = sos->pal_min_state;
532 if (psr1->cpl != 0 ||
533 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) {
534 /*
535 * setup for resume to bottom half of MCA,
536 * "mca_handler_bhhook"
537 */
538 /* pass to bhhook as argument (gr8, ...) */
539 pmsa->pmsa_gr[8-1] = target_identifier;
540 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip;
541 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr;
542 /* set interrupted return address (but no use) */
543 pmsa->pmsa_br0 = pmsa->pmsa_iip;
544 /* change resume address to bottom half */
545 pmsa->pmsa_iip = mca_hdlr_bh->fp;
546 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
547 /* set cpl with kernel mode */
548 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
549 psr2->cpl = 0;
550 psr2->ri = 0;
551 psr2->bn = 1;
552 psr2->i = 0;
553
554 return mca_recovered("user memory corruption. "
555 "kill affected process - recovered.");
556 }
557
558 return fatal_mca("kernel context not recovered, iip 0x%lx\n",
559 pmsa->pmsa_iip);
560}
561
562/**
563 * recover_from_platform_error - Recover from platform error.
564 * @slidx: pointer of index of SAL error record
565 * @peidx: pointer of index of processor error section
566 * @pbci: pointer of pal_bus_check_info
567 * @sos: pointer to hand off struct between SAL and OS
568 *
569 * Return value:
570 * 1 on Success / 0 on Failure
571 */
572
573static int
574recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
575 pal_bus_check_info_t *pbci,
576 struct ia64_sal_os_state *sos)
577{
578 int status = 0;
579 pal_processor_state_info_t *psp =
580 (pal_processor_state_info_t*)peidx_psp(peidx);
581
582 if (psp->bc && pbci->eb && pbci->bsi == 0) {
583 switch(pbci->type) {
584 case 1: /* partial read */
585 case 3: /* full line(cpu) read */
586 case 9: /* I/O space read */
587 status = recover_from_read_error(slidx, peidx, pbci,
588 sos);
589 break;
590 case 0: /* unknown */
591 case 2: /* partial write */
592 case 4: /* full line write */
593 case 5: /* implicit or explicit write-back operation */
594 case 6: /* snoop probe */
595 case 7: /* incoming or outgoing ptc.g */
596 case 8: /* write coalescing transactions */
597 case 10: /* I/O space write */
598 case 11: /* inter-processor interrupt message(IPI) */
599 case 12: /* interrupt acknowledge or
600 external task priority cycle */
601 default:
602 break;
603 }
604 } else if (psp->cc && !psp->bc) { /* Cache error */
605 status = recover_from_read_error(slidx, peidx, pbci, sos);
606 }
607
608 return status;
609}
610
611/*
612 * recover_from_tlb_check
613 * @peidx: pointer of index of processor error section
614 *
615 * Return value:
616 * 1 on Success / 0 on Failure
617 */
618static int
619recover_from_tlb_check(peidx_table_t *peidx)
620{
621 sal_log_mod_error_info_t *smei;
622 pal_tlb_check_info_t *ptci;
623
624 smei = (sal_log_mod_error_info_t *)peidx_tlb_check(peidx, 0);
625 ptci = (pal_tlb_check_info_t *)&(smei->check_info);
626
627 /*
628 * Look for signature of a duplicate TLB DTC entry, which is
629 * a SW bug and always fatal.
630 */
631 if (ptci->op == PAL_TLB_CHECK_OP_PURGE
632 && !(ptci->itr || ptci->dtc || ptci->itc))
633 return fatal_mca("Duplicate TLB entry");
634
635 return mca_recovered("TLB check recovered");
636}
637
638/**
639 * recover_from_processor_error
640 * @platform: whether there are some platform error section or not
641 * @slidx: pointer of index of SAL error record
642 * @peidx: pointer of index of processor error section
643 * @pbci: pointer of pal_bus_check_info
644 * @sos: pointer to hand off struct between SAL and OS
645 *
646 * Return value:
647 * 1 on Success / 0 on Failure
648 */
649
650static int
651recover_from_processor_error(int platform, slidx_table_t *slidx,
652 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
653 struct ia64_sal_os_state *sos)
654{
655 pal_processor_state_info_t *psp =
656 (pal_processor_state_info_t*)peidx_psp(peidx);
657
658 /*
659 * Processor recovery status must key off of the PAL recovery
660 * status in the Processor State Parameter.
661 */
662
663 /*
664 * The machine check is corrected.
665 */
666 if (psp->cm == 1)
667 return mca_recovered("machine check is already corrected.");
668
669 /*
670 * The error was not contained. Software must be reset.
671 */
672 if (psp->us || psp->ci == 0)
673 return fatal_mca("error not contained");
674
675 /*
676 * Look for recoverable TLB check
677 */
678 if (psp->tc && !(psp->cc || psp->bc || psp->rc || psp->uc))
679 return recover_from_tlb_check(peidx);
680
681 /*
682 * The cache check and bus check bits have four possible states
683 * cc bc
684 * 1 1 Memory error, attempt recovery
685 * 1 0 Cache error, attempt recovery
686 * 0 1 I/O error, attempt recovery
687 * 0 0 Other error type, not recovered
688 */
689 if (psp->cc == 0 && (psp->bc == 0 || pbci == NULL))
690 return fatal_mca("No cache or bus check");
691
692 /*
693 * Cannot handle more than one bus check.
694 */
695 if (peidx_bus_check_num(peidx) > 1)
696 return fatal_mca("Too many bus checks");
697
698 if (pbci->ib)
699 return fatal_mca("Internal Bus error");
700 if (pbci->eb && pbci->bsi > 0)
701 return fatal_mca("External bus check fatal status");
702
703 /*
704 * This is a local MCA and estimated as a recoverable error.
705 */
706 if (platform)
707 return recover_from_platform_error(slidx, peidx, pbci, sos);
708
709 /*
710 * On account of strange SAL error record, we cannot recover.
711 */
712 return fatal_mca("Strange SAL record");
713}
714
715/**
716 * mca_try_to_recover - Try to recover from MCA
717 * @rec: pointer to a SAL error record
718 * @sos: pointer to hand off struct between SAL and OS
719 *
720 * Return value:
721 * 1 on Success / 0 on Failure
722 */
723
724static int
725mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
726{
727 int platform_err;
728 int n_proc_err;
729 slidx_table_t slidx;
730 peidx_table_t peidx;
731 pal_bus_check_info_t pbci;
732
733 /* Make index of SAL error record */
734 platform_err = mca_make_slidx(rec, &slidx);
735
736 /* Count processor error sections */
737 n_proc_err = slidx_count(&slidx, proc_err);
738
739 /* Now, OS can recover when there is one processor error section */
740 if (n_proc_err > 1)
741 return fatal_mca("Too Many Errors");
742 else if (n_proc_err == 0)
743 /* Weird SAL record ... We can't do anything */
744 return fatal_mca("Weird SAL record");
745
746 /* Make index of processor error section */
747 mca_make_peidx((sal_log_processor_info_t*)
748 slidx_first_entry(&slidx.proc_err)->hdr, &peidx);
749
750 /* Extract Processor BUS_CHECK[0] */
751 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
752
753 /* Check whether MCA is global or not */
754 if (is_mca_global(&peidx, &pbci, sos))
755 return fatal_mca("global MCA");
756
757 /* Try to recover a processor error */
758 return recover_from_processor_error(platform_err, &slidx, &peidx,
759 &pbci, sos);
760}
761
762/*
763 * =============================================================================
764 */
765
766int __init mca_external_handler_init(void)
767{
768 if (init_record_index_pools())
769 return -ENOMEM;
770
771 /* register external mca handlers */
772 if (ia64_reg_MCA_extension(mca_try_to_recover)) {
773 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
774 kfree(slidx_pool.buffer);
775 return -EFAULT;
776 }
777 return 0;
778}
779
780void __exit mca_external_handler_exit(void)
781{
782 /* unregister external mca handlers */
783 ia64_unreg_MCA_extension();
784 kfree(slidx_pool.buffer);
785}
786
787module_init(mca_external_handler_init);
788module_exit(mca_external_handler_exit);
789
790module_param(sal_rec_max, int, 0644);
791MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");
792
793MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
794MODULE_LICENSE("GPL");