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, ...)
87 return MCA_NOT_RECOVERED;
91 * mca_page_isolate - isolate a poisoned page in order not to use it later
92 * @paddr: poisoned memory location
95 * one of isolate_status_t, ISOLATE_OK/NG/NONE.
98 static isolate_status_t
99 mca_page_isolate(unsigned long paddr)
104 /* whether physical address is valid or not */
105 if (!ia64_phys_addr_valid(paddr))
108 if (!pfn_valid(paddr >> PAGE_SHIFT))
111 /* convert physical address to physical page number */
112 p = pfn_to_page(paddr>>PAGE_SHIFT);
114 /* check whether a page number have been already registered or not */
115 for (i = 0; i < num_page_isolate; i++)
116 if (page_isolate[i] == p)
117 return ISOLATE_OK; /* already listed */
119 /* limitation check */
120 if (num_page_isolate == MAX_PAGE_ISOLATE)
123 /* kick pages having attribute 'SLAB' or 'Reserved' */
124 if (PageSlab(p) || PageReserved(p))
127 /* add attribute 'Reserved' and register the page */
130 page_isolate[num_page_isolate++] = p;
136 * mca_hanlder_bh - Kill the process which occurred memory read error
137 * @paddr: poisoned address received from MCA Handler
141 mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr)
143 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, "
144 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n",
145 raw_smp_processor_id(), current->pid, current->uid,
146 iip, ipsr, paddr, current->comm);
148 spin_lock(&mca_bh_lock);
149 switch (mca_page_isolate(paddr)) {
151 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr);
154 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr);
159 spin_unlock(&mca_bh_lock);
161 /* This process is about to be killed itself */
166 * mca_make_peidx - Make index of processor error section
167 * @slpi: pointer to record of processor error section
168 * @peidx: pointer to index of processor error section
172 mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx)
175 * calculate the start address of
176 * "struct cpuid_info" and "sal_processor_static_info_t".
178 u64 total_check_num = slpi->valid.num_cache_check
179 + slpi->valid.num_tlb_check
180 + slpi->valid.num_bus_check
181 + slpi->valid.num_reg_file_check
182 + slpi->valid.num_ms_check;
183 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num
184 + sizeof(sal_log_processor_info_t);
185 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info);
187 peidx_head(peidx) = slpi;
188 peidx_mid(peidx) = (struct sal_cpuid_info *)
189 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL);
190 peidx_bottom(peidx) = (sal_processor_static_info_t *)
191 (slpi->valid.psi_static_struct ?
192 ((char*)slpi + head_size + mid_size) : NULL);
196 * mca_make_slidx - Make index of SAL error record
197 * @buffer: pointer to SAL error record
198 * @slidx: pointer to index of SAL error record
201 * 1 if record has platform error / 0 if not
203 #define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \
204 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \
206 list_add(&hl->list, &(sect)); \
207 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; }
210 mca_make_slidx(void *buffer, slidx_table_t *slidx)
212 int platform_err = 0;
213 int record_len = ((sal_log_record_header_t*)buffer)->len;
216 sal_log_section_hdr_t *sp;
219 * Initialize index referring current record
221 INIT_LIST_HEAD(&(slidx->proc_err));
222 INIT_LIST_HEAD(&(slidx->mem_dev_err));
223 INIT_LIST_HEAD(&(slidx->sel_dev_err));
224 INIT_LIST_HEAD(&(slidx->pci_bus_err));
225 INIT_LIST_HEAD(&(slidx->smbios_dev_err));
226 INIT_LIST_HEAD(&(slidx->pci_comp_err));
227 INIT_LIST_HEAD(&(slidx->plat_specific_err));
228 INIT_LIST_HEAD(&(slidx->host_ctlr_err));
229 INIT_LIST_HEAD(&(slidx->plat_bus_err));
230 INIT_LIST_HEAD(&(slidx->unsupported));
233 * Extract a Record Header
235 slidx->header = buffer;
238 * Extract each section records
239 * (arranged from "int ia64_log_platform_info_print()")
241 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0;
242 ercd_pos < record_len; ercd_pos += sp->len, sects++) {
243 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos);
244 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) {
245 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp);
246 } else if (!efi_guidcmp(sp->guid,
247 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) {
249 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp);
250 } else if (!efi_guidcmp(sp->guid,
251 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) {
253 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp);
254 } else if (!efi_guidcmp(sp->guid,
255 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) {
257 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp);
258 } else if (!efi_guidcmp(sp->guid,
259 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) {
261 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp);
262 } else if (!efi_guidcmp(sp->guid,
263 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) {
265 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp);
266 } else if (!efi_guidcmp(sp->guid,
267 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) {
269 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp);
270 } else if (!efi_guidcmp(sp->guid,
271 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) {
273 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp);
274 } else if (!efi_guidcmp(sp->guid,
275 SAL_PLAT_BUS_ERR_SECT_GUID)) {
277 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp);
279 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp);
282 slidx->n_sections = sects;
288 * init_record_index_pools - Initialize pool of lists for SAL record index
291 * 0 on Success / -ENOMEM on Failure
294 init_record_index_pools(void)
297 int rec_max_size; /* Maximum size of SAL error records */
298 int sect_min_size; /* Minimum size of SAL error sections */
299 /* minimum size table of each section */
300 static int sal_log_sect_min_sizes[] = {
301 sizeof(sal_log_processor_info_t)
302 + sizeof(sal_processor_static_info_t),
303 sizeof(sal_log_mem_dev_err_info_t),
304 sizeof(sal_log_sel_dev_err_info_t),
305 sizeof(sal_log_pci_bus_err_info_t),
306 sizeof(sal_log_smbios_dev_err_info_t),
307 sizeof(sal_log_pci_comp_err_info_t),
308 sizeof(sal_log_plat_specific_err_info_t),
309 sizeof(sal_log_host_ctlr_err_info_t),
310 sizeof(sal_log_plat_bus_err_info_t),
314 * MCA handler cannot allocate new memory on flight,
315 * so we preallocate enough memory to handle a SAL record.
317 * Initialize a handling set of slidx_pool:
318 * 1. Pick up the max size of SAL error records
319 * 2. Pick up the min size of SAL error sections
320 * 3. Allocate the pool as enough to 2 SAL records
321 * (now we can estimate the maxinum of section in a record.)
325 rec_max_size = sal_rec_max;
328 sect_min_size = sal_log_sect_min_sizes[0];
329 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++)
330 if (sect_min_size > sal_log_sect_min_sizes[i])
331 sect_min_size = sal_log_sect_min_sizes[i];
334 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1;
335 slidx_pool.buffer = (slidx_list_t *)
336 kmalloc(slidx_pool.max_idx * sizeof(slidx_list_t), GFP_KERNEL);
338 return slidx_pool.buffer ? 0 : -ENOMEM;
342 /*****************************************************************************
343 * Recovery functions *
344 *****************************************************************************/
347 * is_mca_global - Check whether this MCA is global or not
348 * @peidx: pointer of index of processor error section
349 * @pbci: pointer to pal_bus_check_info_t
350 * @sos: pointer to hand off struct between SAL and OS
353 * MCA_IS_LOCAL / MCA_IS_GLOBAL
357 is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
358 struct ia64_sal_os_state *sos)
360 pal_processor_state_info_t *psp =
361 (pal_processor_state_info_t*)peidx_psp(peidx);
364 * PAL can request a rendezvous, if the MCA has a global scope.
365 * If "rz_always" flag is set, SAL requests MCA rendezvous
366 * in spite of global MCA.
367 * Therefore it is local MCA when rendezvous has not been requested.
368 * Failed to rendezvous, the system must be down.
370 switch (sos->rv_rc) {
371 case -1: /* SAL rendezvous unsuccessful */
372 return MCA_IS_GLOBAL;
373 case 0: /* SAL rendezvous not required */
375 case 1: /* SAL rendezvous successful int */
376 case 2: /* SAL rendezvous successful int with init */
382 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
383 * it would be a local MCA. (i.e. processor internal error)
385 if (psp->tc || psp->cc || psp->rc || psp->uc)
389 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
390 * would be a global MCA. (e.g. a system bus address parity error)
392 if (!pbci || pbci->ib)
393 return MCA_IS_GLOBAL;
396 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
397 * could be either a local MCA or a global MCA.
399 * Referring Bus_Check.bsi:
400 * 0: Unknown/unclassified
404 * (FIXME: Are these SGI specific or generic bsi values?)
409 /* e.g. a load from poisoned memory */
414 return MCA_IS_GLOBAL;
417 return MCA_IS_GLOBAL;
421 * recover_from_read_error - Try to recover the errors which type are "read"s.
422 * @slidx: pointer of index of SAL error record
423 * @peidx: pointer of index of processor error section
424 * @pbci: pointer of pal_bus_check_info
425 * @sos: pointer to hand off struct between SAL and OS
428 * 1 on Success / 0 on Failure
432 recover_from_read_error(slidx_table_t *slidx,
433 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
434 struct ia64_sal_os_state *sos)
436 sal_log_mod_error_info_t *smei;
437 pal_min_state_area_t *pmsa;
438 struct ia64_psr *psr1, *psr2;
439 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;
441 /* Is target address valid? */
443 return fatal_mca(KERN_ALERT "MCA: target address not valid\n");
446 * cpu read or memory-mapped io read
448 * offending process affected process OS MCA do
449 * kernel mode kernel mode down system
450 * kernel mode user mode kill the process
451 * user mode kernel mode down system (*)
452 * user mode user mode kill the process
454 * (*) You could terminate offending user-mode process
455 * if (pbci->pv && pbci->pl != 0) *and* if you sure
456 * the process not have any locks of kernel.
459 /* Is minstate valid? */
460 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate))
461 return fatal_mca(KERN_ALERT "MCA: minstate not valid\n");
462 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
463 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr);
466 * Check the privilege level of interrupted context.
467 * If it is user-mode, then terminate affected process.
470 pmsa = sos->pal_min_state;
471 if (psr1->cpl != 0 ||
472 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) {
473 smei = peidx_bus_check(peidx, 0);
474 if (smei->valid.target_identifier) {
476 * setup for resume to bottom half of MCA,
477 * "mca_handler_bhhook"
479 /* pass to bhhook as argument (gr8, ...) */
480 pmsa->pmsa_gr[8-1] = smei->target_identifier;
481 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip;
482 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr;
483 /* set interrupted return address (but no use) */
484 pmsa->pmsa_br0 = pmsa->pmsa_iip;
485 /* change resume address to bottom half */
486 pmsa->pmsa_iip = mca_hdlr_bh->fp;
487 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
488 /* set cpl with kernel mode */
489 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
495 return MCA_RECOVERED;
500 return fatal_mca(KERN_ALERT "MCA: kernel context not recovered,"
501 " iip 0x%lx\n", pmsa->pmsa_iip);
505 * recover_from_platform_error - Recover from platform error.
506 * @slidx: pointer of index of SAL error record
507 * @peidx: pointer of index of processor error section
508 * @pbci: pointer of pal_bus_check_info
509 * @sos: pointer to hand off struct between SAL and OS
512 * 1 on Success / 0 on Failure
516 recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
517 pal_bus_check_info_t *pbci,
518 struct ia64_sal_os_state *sos)
521 pal_processor_state_info_t *psp =
522 (pal_processor_state_info_t*)peidx_psp(peidx);
524 if (psp->bc && pbci->eb && pbci->bsi == 0) {
526 case 1: /* partial read */
527 case 3: /* full line(cpu) read */
528 case 9: /* I/O space read */
529 status = recover_from_read_error(slidx, peidx, pbci,
532 case 0: /* unknown */
533 case 2: /* partial write */
534 case 4: /* full line write */
535 case 5: /* implicit or explicit write-back operation */
536 case 6: /* snoop probe */
537 case 7: /* incoming or outgoing ptc.g */
538 case 8: /* write coalescing transactions */
539 case 10: /* I/O space write */
540 case 11: /* inter-processor interrupt message(IPI) */
541 case 12: /* interrupt acknowledge or
542 external task priority cycle */
552 * recover_from_processor_error
553 * @platform: whether there are some platform error section or not
554 * @slidx: pointer of index of SAL error record
555 * @peidx: pointer of index of processor error section
556 * @pbci: pointer of pal_bus_check_info
557 * @sos: pointer to hand off struct between SAL and OS
560 * 1 on Success / 0 on Failure
563 * Later we try to recover when below all conditions are satisfied.
564 * 1. Only one processor error section is exist.
565 * 2. BUS_CHECK is exist and the others are not exist.(Except TLB_CHECK)
566 * 3. The entry of BUS_CHECK_INFO is 1.
567 * 4. "External bus error" flag is set and the others are not set.
571 recover_from_processor_error(int platform, slidx_table_t *slidx,
572 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
573 struct ia64_sal_os_state *sos)
575 pal_processor_state_info_t *psp =
576 (pal_processor_state_info_t*)peidx_psp(peidx);
579 * Processor recovery status must key off of the PAL recovery
580 * status in the Processor State Parameter.
584 * The machine check is corrected.
587 return MCA_RECOVERED;
590 * The error was not contained. Software must be reset.
592 if (psp->us || psp->ci == 0)
593 return fatal_mca(KERN_ALERT "MCA: error not contained\n");
596 * The cache check and bus check bits have four possible states
598 * 0 0 Weird record, not recovered
599 * 1 0 Cache error, not recovered
600 * 0 1 I/O error, attempt recovery
601 * 1 1 Memory error, attempt recovery
603 if (psp->bc == 0 || pbci == NULL)
604 return fatal_mca(KERN_ALERT "MCA: No bus check\n");
607 * Sorry, we cannot handle so many.
609 if (peidx_bus_check_num(peidx) > 1)
610 return fatal_mca(KERN_ALERT "MCA: Too many bus checks\n");
612 * Well, here is only one bus error.
615 return fatal_mca(KERN_ALERT "MCA: Internal Bus error\n");
617 return fatal_mca(KERN_ALERT "MCA: Cache-cache error\n");
618 if (pbci->eb && pbci->bsi > 0)
619 return fatal_mca(KERN_ALERT "MCA: External bus check fatal status\n");
622 * This is a local MCA and estimated as recoverble external bus error.
623 * (e.g. a load from poisoned memory)
624 * This means "there are some platform errors".
627 return recover_from_platform_error(slidx, peidx, pbci, sos);
629 * On account of strange SAL error record, we cannot recover.
631 return fatal_mca(KERN_ALERT "MCA: Strange SAL record\n");
635 * mca_try_to_recover - Try to recover from MCA
636 * @rec: pointer to a SAL error record
637 * @sos: pointer to hand off struct between SAL and OS
640 * 1 on Success / 0 on Failure
644 mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
650 pal_bus_check_info_t pbci;
652 /* Make index of SAL error record */
653 platform_err = mca_make_slidx(rec, &slidx);
655 /* Count processor error sections */
656 n_proc_err = slidx_count(&slidx, proc_err);
658 /* Now, OS can recover when there is one processor error section */
660 return fatal_mca(KERN_ALERT "MCA: Too Many Errors\n");
661 else if (n_proc_err == 0)
662 /* Weird SAL record ... We need not to recover */
663 return fatal_mca(KERN_ALERT "MCA: Weird SAL record\n");
665 /* Make index of processor error section */
666 mca_make_peidx((sal_log_processor_info_t*)
667 slidx_first_entry(&slidx.proc_err)->hdr, &peidx);
669 /* Extract Processor BUS_CHECK[0] */
670 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
672 /* Check whether MCA is global or not */
673 if (is_mca_global(&peidx, &pbci, sos))
674 return fatal_mca(KERN_ALERT "MCA: global MCA\n");
676 /* Try to recover a processor error */
677 return recover_from_processor_error(platform_err, &slidx, &peidx,
682 * =============================================================================
685 int __init mca_external_handler_init(void)
687 if (init_record_index_pools())
690 /* register external mca handlers */
691 if (ia64_reg_MCA_extension(mca_try_to_recover)) {
692 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
693 kfree(slidx_pool.buffer);
699 void __exit mca_external_handler_exit(void)
701 /* unregister external mca handlers */
702 ia64_unreg_MCA_extension();
703 kfree(slidx_pool.buffer);
706 module_init(mca_external_handler_init);
707 module_exit(mca_external_handler_exit);
709 module_param(sal_rec_max, int, 0644);
710 MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");
712 MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
713 MODULE_LICENSE("GPL");