3 * Purpose: Generic MCA handling layer
5 * Copyright (C) 2004 FUJITSU LIMITED
6 * Copyright (C) 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>
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/smp_lock.h>
18 #include <linux/bootmem.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>
27 #include <asm/delay.h>
28 #include <asm/machvec.h>
30 #include <asm/ptrace.h>
31 #include <asm/system.h>
36 #include <asm/hw_irq.h>
40 /* max size of SAL error record (default) */
41 static int sal_rec_max = 10000;
43 /* from mca_drv_asm.S */
44 extern void *mca_handler_bhhook(void);
46 static DEFINE_SPINLOCK(mca_bh_lock);
53 #define MAX_PAGE_ISOLATE 1024
55 static struct page *page_isolate[MAX_PAGE_ISOLATE];
56 static int num_page_isolate = 0;
65 MCA_NOT_RECOVERED = 0,
70 * This pool keeps pointers to the section part of SAL error record
73 slidx_list_t *buffer; /* section pointer list pool */
74 int cur_idx; /* Current index of section pointer list pool */
75 int max_idx; /* Maximum index of section pointer list pool */
79 fatal_mca(const char *fmt, ...)
85 vsnprintf(buf, sizeof(buf), fmt, args);
87 ia64_mca_printk(KERN_ALERT "MCA: %s\n", buf);
89 return MCA_NOT_RECOVERED;
93 mca_recovered(const char *fmt, ...)
99 vsnprintf(buf, sizeof(buf), fmt, args);
101 ia64_mca_printk(KERN_INFO "MCA: %s\n", buf);
103 return MCA_RECOVERED;
107 * mca_page_isolate - isolate a poisoned page in order not to use it later
108 * @paddr: poisoned memory location
111 * one of isolate_status_t, ISOLATE_OK/NG/NONE.
114 static isolate_status_t
115 mca_page_isolate(unsigned long paddr)
120 /* whether physical address is valid or not */
121 if (!ia64_phys_addr_valid(paddr))
124 if (!pfn_valid(paddr >> PAGE_SHIFT))
127 /* convert physical address to physical page number */
128 p = pfn_to_page(paddr>>PAGE_SHIFT);
130 /* check whether a page number have been already registered or not */
131 for (i = 0; i < num_page_isolate; i++)
132 if (page_isolate[i] == p)
133 return ISOLATE_OK; /* already listed */
135 /* limitation check */
136 if (num_page_isolate == MAX_PAGE_ISOLATE)
139 /* kick pages having attribute 'SLAB' or 'Reserved' */
140 if (PageSlab(p) || PageReserved(p))
143 /* add attribute 'Reserved' and register the page */
146 page_isolate[num_page_isolate++] = p;
152 * mca_hanlder_bh - Kill the process which occurred memory read error
153 * @paddr: poisoned address received from MCA Handler
157 mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr)
160 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, "
161 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n",
162 raw_smp_processor_id(), current->pid, current->uid,
163 iip, ipsr, paddr, current->comm);
165 spin_lock(&mca_bh_lock);
166 switch (mca_page_isolate(paddr)) {
168 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr);
171 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr);
176 spin_unlock(&mca_bh_lock);
178 /* This process is about to be killed itself */
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
189 mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx)
192 * calculate the start address of
193 * "struct cpuid_info" and "sal_processor_static_info_t".
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);
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);
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
218 * 1 if record has platform error / 0 if not
220 #define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \
221 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \
223 list_add(&hl->list, &(sect)); \
224 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; }
227 mca_make_slidx(void *buffer, slidx_table_t *slidx)
229 int platform_err = 0;
230 int record_len = ((sal_log_record_header_t*)buffer)->len;
233 sal_log_section_hdr_t *sp;
236 * Initialize index referring current record
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));
250 * Extract a Record Header
252 slidx->header = buffer;
255 * Extract each section records
256 * (arranged from "int ia64_log_platform_info_print()")
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)) {
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)) {
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)) {
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)) {
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)) {
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)) {
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)) {
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)) {
294 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp);
296 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp);
299 slidx->n_sections = sects;
305 * init_record_index_pools - Initialize pool of lists for SAL record index
308 * 0 on Success / -ENOMEM on Failure
311 init_record_index_pools(void)
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),
331 * MCA handler cannot allocate new memory on flight,
332 * so we preallocate enough memory to handle a SAL record.
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.)
342 rec_max_size = sal_rec_max;
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];
351 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1;
352 slidx_pool.buffer = (slidx_list_t *)
353 kmalloc(slidx_pool.max_idx * sizeof(slidx_list_t), GFP_KERNEL);
355 return slidx_pool.buffer ? 0 : -ENOMEM;
359 /*****************************************************************************
360 * Recovery functions *
361 *****************************************************************************/
364 * is_mca_global - Check whether this MCA is global or not
365 * @peidx: pointer of index of processor error section
366 * @pbci: pointer to pal_bus_check_info_t
367 * @sos: pointer to hand off struct between SAL and OS
370 * MCA_IS_LOCAL / MCA_IS_GLOBAL
374 is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
375 struct ia64_sal_os_state *sos)
377 pal_processor_state_info_t *psp =
378 (pal_processor_state_info_t*)peidx_psp(peidx);
381 * PAL can request a rendezvous, if the MCA has a global scope.
382 * If "rz_always" flag is set, SAL requests MCA rendezvous
383 * in spite of global MCA.
384 * Therefore it is local MCA when rendezvous has not been requested.
385 * Failed to rendezvous, the system must be down.
387 switch (sos->rv_rc) {
388 case -1: /* SAL rendezvous unsuccessful */
389 return MCA_IS_GLOBAL;
390 case 0: /* SAL rendezvous not required */
392 case 1: /* SAL rendezvous successful int */
393 case 2: /* SAL rendezvous successful int with init */
399 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
400 * it would be a local MCA. (i.e. processor internal error)
402 if (psp->tc || psp->cc || psp->rc || psp->uc)
406 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
407 * would be a global MCA. (e.g. a system bus address parity error)
409 if (!pbci || pbci->ib)
410 return MCA_IS_GLOBAL;
413 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
414 * could be either a local MCA or a global MCA.
416 * Referring Bus_Check.bsi:
417 * 0: Unknown/unclassified
421 * (FIXME: Are these SGI specific or generic bsi values?)
426 /* e.g. a load from poisoned memory */
431 return MCA_IS_GLOBAL;
434 return MCA_IS_GLOBAL;
438 * recover_from_read_error - Try to recover the errors which type are "read"s.
439 * @slidx: pointer of index of SAL error record
440 * @peidx: pointer of index of processor error section
441 * @pbci: pointer of pal_bus_check_info
442 * @sos: pointer to hand off struct between SAL and OS
445 * 1 on Success / 0 on Failure
449 recover_from_read_error(slidx_table_t *slidx,
450 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
451 struct ia64_sal_os_state *sos)
453 sal_log_mod_error_info_t *smei;
454 pal_min_state_area_t *pmsa;
455 struct ia64_psr *psr1, *psr2;
456 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;
458 /* Is target address valid? */
460 return fatal_mca("target address not valid");
463 * cpu read or memory-mapped io read
465 * offending process affected process OS MCA do
466 * kernel mode kernel mode down system
467 * kernel mode user mode kill the process
468 * user mode kernel mode down system (*)
469 * user mode user mode kill the process
471 * (*) You could terminate offending user-mode process
472 * if (pbci->pv && pbci->pl != 0) *and* if you sure
473 * the process not have any locks of kernel.
476 /* Is minstate valid? */
477 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate))
478 return fatal_mca("minstate not valid");
479 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
480 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr);
483 * Check the privilege level of interrupted context.
484 * If it is user-mode, then terminate affected process.
487 pmsa = sos->pal_min_state;
488 if (psr1->cpl != 0 ||
489 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) {
490 smei = peidx_bus_check(peidx, 0);
491 if (smei->valid.target_identifier) {
493 * setup for resume to bottom half of MCA,
494 * "mca_handler_bhhook"
496 /* pass to bhhook as argument (gr8, ...) */
497 pmsa->pmsa_gr[8-1] = smei->target_identifier;
498 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip;
499 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr;
500 /* set interrupted return address (but no use) */
501 pmsa->pmsa_br0 = pmsa->pmsa_iip;
502 /* change resume address to bottom half */
503 pmsa->pmsa_iip = mca_hdlr_bh->fp;
504 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
505 /* set cpl with kernel mode */
506 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
512 return mca_recovered("user memory corruption. "
513 "kill affected process - recovered.");
518 return fatal_mca("kernel context not recovered, iip 0x%lx\n",
523 * recover_from_platform_error - Recover from platform error.
524 * @slidx: pointer of index of SAL error record
525 * @peidx: pointer of index of processor error section
526 * @pbci: pointer of pal_bus_check_info
527 * @sos: pointer to hand off struct between SAL and OS
530 * 1 on Success / 0 on Failure
534 recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
535 pal_bus_check_info_t *pbci,
536 struct ia64_sal_os_state *sos)
539 pal_processor_state_info_t *psp =
540 (pal_processor_state_info_t*)peidx_psp(peidx);
542 if (psp->bc && pbci->eb && pbci->bsi == 0) {
544 case 1: /* partial read */
545 case 3: /* full line(cpu) read */
546 case 9: /* I/O space read */
547 status = recover_from_read_error(slidx, peidx, pbci,
550 case 0: /* unknown */
551 case 2: /* partial write */
552 case 4: /* full line write */
553 case 5: /* implicit or explicit write-back operation */
554 case 6: /* snoop probe */
555 case 7: /* incoming or outgoing ptc.g */
556 case 8: /* write coalescing transactions */
557 case 10: /* I/O space write */
558 case 11: /* inter-processor interrupt message(IPI) */
559 case 12: /* interrupt acknowledge or
560 external task priority cycle */
570 * recover_from_processor_error
571 * @platform: whether there are some platform error section or not
572 * @slidx: pointer of index of SAL error record
573 * @peidx: pointer of index of processor error section
574 * @pbci: pointer of pal_bus_check_info
575 * @sos: pointer to hand off struct between SAL and OS
578 * 1 on Success / 0 on Failure
581 * Later we try to recover when below all conditions are satisfied.
582 * 1. Only one processor error section is exist.
583 * 2. BUS_CHECK is exist and the others are not exist.(Except TLB_CHECK)
584 * 3. The entry of BUS_CHECK_INFO is 1.
585 * 4. "External bus error" flag is set and the others are not set.
589 recover_from_processor_error(int platform, slidx_table_t *slidx,
590 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
591 struct ia64_sal_os_state *sos)
593 pal_processor_state_info_t *psp =
594 (pal_processor_state_info_t*)peidx_psp(peidx);
597 * Processor recovery status must key off of the PAL recovery
598 * status in the Processor State Parameter.
602 * The machine check is corrected.
605 return mca_recovered("machine check is already corrected.");
608 * The error was not contained. Software must be reset.
610 if (psp->us || psp->ci == 0)
611 return fatal_mca("error not contained");
614 * The cache check and bus check bits have four possible states
616 * 0 0 Weird record, not recovered
617 * 1 0 Cache error, not recovered
618 * 0 1 I/O error, attempt recovery
619 * 1 1 Memory error, attempt recovery
621 if (psp->bc == 0 || pbci == NULL)
622 return fatal_mca("No bus check");
625 * Sorry, we cannot handle so many.
627 if (peidx_bus_check_num(peidx) > 1)
628 return fatal_mca("Too many bus checks");
630 * Well, here is only one bus error.
633 return fatal_mca("Internal Bus error");
635 return fatal_mca("Cache-cache error");
636 if (pbci->eb && pbci->bsi > 0)
637 return fatal_mca("External bus check fatal status");
640 * This is a local MCA and estimated as recoverble external bus error.
641 * (e.g. a load from poisoned memory)
642 * This means "there are some platform errors".
645 return recover_from_platform_error(slidx, peidx, pbci, sos);
647 * On account of strange SAL error record, we cannot recover.
649 return fatal_mca("Strange SAL record");
653 * mca_try_to_recover - Try to recover from MCA
654 * @rec: pointer to a SAL error record
655 * @sos: pointer to hand off struct between SAL and OS
658 * 1 on Success / 0 on Failure
662 mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
668 pal_bus_check_info_t pbci;
670 /* Make index of SAL error record */
671 platform_err = mca_make_slidx(rec, &slidx);
673 /* Count processor error sections */
674 n_proc_err = slidx_count(&slidx, proc_err);
676 /* Now, OS can recover when there is one processor error section */
678 return fatal_mca("Too Many Errors");
679 else if (n_proc_err == 0)
680 /* Weird SAL record ... We can't do anything */
681 return fatal_mca("Weird SAL record");
683 /* Make index of processor error section */
684 mca_make_peidx((sal_log_processor_info_t*)
685 slidx_first_entry(&slidx.proc_err)->hdr, &peidx);
687 /* Extract Processor BUS_CHECK[0] */
688 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
690 /* Check whether MCA is global or not */
691 if (is_mca_global(&peidx, &pbci, sos))
692 return fatal_mca("global MCA");
694 /* Try to recover a processor error */
695 return recover_from_processor_error(platform_err, &slidx, &peidx,
700 * =============================================================================
703 int __init mca_external_handler_init(void)
705 if (init_record_index_pools())
708 /* register external mca handlers */
709 if (ia64_reg_MCA_extension(mca_try_to_recover)) {
710 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
711 kfree(slidx_pool.buffer);
717 void __exit mca_external_handler_exit(void)
719 /* unregister external mca handlers */
720 ia64_unreg_MCA_extension();
721 kfree(slidx_pool.buffer);
724 module_init(mca_external_handler_init);
725 module_exit(mca_external_handler_exit);
727 module_param(sal_rec_max, int, 0644);
728 MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");
730 MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
731 MODULE_LICENSE("GPL");