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/config.h>
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/smp_lock.h>
19 #include <linux/bootmem.h>
20 #include <linux/acpi.h>
21 #include <linux/timer.h>
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/smp.h>
25 #include <linux/workqueue.h>
28 #include <asm/delay.h>
29 #include <asm/machvec.h>
31 #include <asm/ptrace.h>
32 #include <asm/system.h>
37 #include <asm/hw_irq.h>
41 /* max size of SAL error record (default) */
42 static int sal_rec_max = 10000;
44 /* from mca_drv_asm.S */
45 extern void *mca_handler_bhhook(void);
47 static DEFINE_SPINLOCK(mca_bh_lock);
54 #define MAX_PAGE_ISOLATE 1024
56 static struct page *page_isolate[MAX_PAGE_ISOLATE];
57 static int num_page_isolate = 0;
66 MCA_NOT_RECOVERED = 0,
71 * This pool keeps pointers to the section part of SAL error record
74 slidx_list_t *buffer; /* section pointer list pool */
75 int cur_idx; /* Current index of section pointer list pool */
76 int max_idx; /* Maximum index of section pointer list pool */
80 fatal_mca(const char *fmt, ...)
88 return MCA_NOT_RECOVERED;
92 * mca_page_isolate - isolate a poisoned page in order not to use it later
93 * @paddr: poisoned memory location
96 * one of isolate_status_t, ISOLATE_OK/NG/NONE.
99 static isolate_status_t
100 mca_page_isolate(unsigned long paddr)
105 /* whether physical address is valid or not */
106 if (!ia64_phys_addr_valid(paddr))
109 if (!pfn_valid(paddr >> PAGE_SHIFT))
112 /* convert physical address to physical page number */
113 p = pfn_to_page(paddr>>PAGE_SHIFT);
115 /* check whether a page number have been already registered or not */
116 for (i = 0; i < num_page_isolate; i++)
117 if (page_isolate[i] == p)
118 return ISOLATE_OK; /* already listed */
120 /* limitation check */
121 if (num_page_isolate == MAX_PAGE_ISOLATE)
124 /* kick pages having attribute 'SLAB' or 'Reserved' */
125 if (PageSlab(p) || PageReserved(p))
128 /* add attribute 'Reserved' and register the page */
131 page_isolate[num_page_isolate++] = p;
137 * mca_hanlder_bh - Kill the process which occurred memory read error
138 * @paddr: poisoned address received from MCA Handler
142 mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr)
144 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, "
145 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n",
146 raw_smp_processor_id(), current->pid, current->uid,
147 iip, ipsr, paddr, current->comm);
149 spin_lock(&mca_bh_lock);
150 switch (mca_page_isolate(paddr)) {
152 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr);
155 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr);
160 spin_unlock(&mca_bh_lock);
162 /* This process is about to be killed itself */
167 * mca_make_peidx - Make index of processor error section
168 * @slpi: pointer to record of processor error section
169 * @peidx: pointer to index of processor error section
173 mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx)
176 * calculate the start address of
177 * "struct cpuid_info" and "sal_processor_static_info_t".
179 u64 total_check_num = slpi->valid.num_cache_check
180 + slpi->valid.num_tlb_check
181 + slpi->valid.num_bus_check
182 + slpi->valid.num_reg_file_check
183 + slpi->valid.num_ms_check;
184 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num
185 + sizeof(sal_log_processor_info_t);
186 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info);
188 peidx_head(peidx) = slpi;
189 peidx_mid(peidx) = (struct sal_cpuid_info *)
190 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL);
191 peidx_bottom(peidx) = (sal_processor_static_info_t *)
192 (slpi->valid.psi_static_struct ?
193 ((char*)slpi + head_size + mid_size) : NULL);
197 * mca_make_slidx - Make index of SAL error record
198 * @buffer: pointer to SAL error record
199 * @slidx: pointer to index of SAL error record
202 * 1 if record has platform error / 0 if not
204 #define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \
205 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \
207 list_add(&hl->list, &(sect)); \
208 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; }
211 mca_make_slidx(void *buffer, slidx_table_t *slidx)
213 int platform_err = 0;
214 int record_len = ((sal_log_record_header_t*)buffer)->len;
217 sal_log_section_hdr_t *sp;
220 * Initialize index referring current record
222 INIT_LIST_HEAD(&(slidx->proc_err));
223 INIT_LIST_HEAD(&(slidx->mem_dev_err));
224 INIT_LIST_HEAD(&(slidx->sel_dev_err));
225 INIT_LIST_HEAD(&(slidx->pci_bus_err));
226 INIT_LIST_HEAD(&(slidx->smbios_dev_err));
227 INIT_LIST_HEAD(&(slidx->pci_comp_err));
228 INIT_LIST_HEAD(&(slidx->plat_specific_err));
229 INIT_LIST_HEAD(&(slidx->host_ctlr_err));
230 INIT_LIST_HEAD(&(slidx->plat_bus_err));
231 INIT_LIST_HEAD(&(slidx->unsupported));
234 * Extract a Record Header
236 slidx->header = buffer;
239 * Extract each section records
240 * (arranged from "int ia64_log_platform_info_print()")
242 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0;
243 ercd_pos < record_len; ercd_pos += sp->len, sects++) {
244 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos);
245 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) {
246 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp);
247 } else if (!efi_guidcmp(sp->guid,
248 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) {
250 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp);
251 } else if (!efi_guidcmp(sp->guid,
252 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) {
254 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp);
255 } else if (!efi_guidcmp(sp->guid,
256 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) {
258 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp);
259 } else if (!efi_guidcmp(sp->guid,
260 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) {
262 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp);
263 } else if (!efi_guidcmp(sp->guid,
264 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) {
266 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp);
267 } else if (!efi_guidcmp(sp->guid,
268 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) {
270 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp);
271 } else if (!efi_guidcmp(sp->guid,
272 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) {
274 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp);
275 } else if (!efi_guidcmp(sp->guid,
276 SAL_PLAT_BUS_ERR_SECT_GUID)) {
278 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp);
280 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp);
283 slidx->n_sections = sects;
289 * init_record_index_pools - Initialize pool of lists for SAL record index
292 * 0 on Success / -ENOMEM on Failure
295 init_record_index_pools(void)
298 int rec_max_size; /* Maximum size of SAL error records */
299 int sect_min_size; /* Minimum size of SAL error sections */
300 /* minimum size table of each section */
301 static int sal_log_sect_min_sizes[] = {
302 sizeof(sal_log_processor_info_t)
303 + sizeof(sal_processor_static_info_t),
304 sizeof(sal_log_mem_dev_err_info_t),
305 sizeof(sal_log_sel_dev_err_info_t),
306 sizeof(sal_log_pci_bus_err_info_t),
307 sizeof(sal_log_smbios_dev_err_info_t),
308 sizeof(sal_log_pci_comp_err_info_t),
309 sizeof(sal_log_plat_specific_err_info_t),
310 sizeof(sal_log_host_ctlr_err_info_t),
311 sizeof(sal_log_plat_bus_err_info_t),
315 * MCA handler cannot allocate new memory on flight,
316 * so we preallocate enough memory to handle a SAL record.
318 * Initialize a handling set of slidx_pool:
319 * 1. Pick up the max size of SAL error records
320 * 2. Pick up the min size of SAL error sections
321 * 3. Allocate the pool as enough to 2 SAL records
322 * (now we can estimate the maxinum of section in a record.)
326 rec_max_size = sal_rec_max;
329 sect_min_size = sal_log_sect_min_sizes[0];
330 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++)
331 if (sect_min_size > sal_log_sect_min_sizes[i])
332 sect_min_size = sal_log_sect_min_sizes[i];
335 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1;
336 slidx_pool.buffer = (slidx_list_t *)
337 kmalloc(slidx_pool.max_idx * sizeof(slidx_list_t), GFP_KERNEL);
339 return slidx_pool.buffer ? 0 : -ENOMEM;
343 /*****************************************************************************
344 * Recovery functions *
345 *****************************************************************************/
348 * is_mca_global - Check whether this MCA is global or not
349 * @peidx: pointer of index of processor error section
350 * @pbci: pointer to pal_bus_check_info_t
351 * @sos: pointer to hand off struct between SAL and OS
354 * MCA_IS_LOCAL / MCA_IS_GLOBAL
358 is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
359 struct ia64_sal_os_state *sos)
361 pal_processor_state_info_t *psp =
362 (pal_processor_state_info_t*)peidx_psp(peidx);
365 * PAL can request a rendezvous, if the MCA has a global scope.
366 * If "rz_always" flag is set, SAL requests MCA rendezvous
367 * in spite of global MCA.
368 * Therefore it is local MCA when rendezvous has not been requested.
369 * Failed to rendezvous, the system must be down.
371 switch (sos->rv_rc) {
372 case -1: /* SAL rendezvous unsuccessful */
373 return MCA_IS_GLOBAL;
374 case 0: /* SAL rendezvous not required */
376 case 1: /* SAL rendezvous successful int */
377 case 2: /* SAL rendezvous successful int with init */
383 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
384 * it would be a local MCA. (i.e. processor internal error)
386 if (psp->tc || psp->cc || psp->rc || psp->uc)
390 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
391 * would be a global MCA. (e.g. a system bus address parity error)
393 if (!pbci || pbci->ib)
394 return MCA_IS_GLOBAL;
397 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
398 * could be either a local MCA or a global MCA.
400 * Referring Bus_Check.bsi:
401 * 0: Unknown/unclassified
405 * (FIXME: Are these SGI specific or generic bsi values?)
410 /* e.g. a load from poisoned memory */
415 return MCA_IS_GLOBAL;
418 return MCA_IS_GLOBAL;
422 * recover_from_read_error - Try to recover the errors which type are "read"s.
423 * @slidx: pointer of index of SAL error record
424 * @peidx: pointer of index of processor error section
425 * @pbci: pointer of pal_bus_check_info
426 * @sos: pointer to hand off struct between SAL and OS
429 * 1 on Success / 0 on Failure
433 recover_from_read_error(slidx_table_t *slidx,
434 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
435 struct ia64_sal_os_state *sos)
437 sal_log_mod_error_info_t *smei;
438 pal_min_state_area_t *pmsa;
439 struct ia64_psr *psr1, *psr2;
440 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;
442 /* Is target address valid? */
444 return fatal_mca(KERN_ALERT "MCA: target address not valid\n");
447 * cpu read or memory-mapped io read
449 * offending process affected process OS MCA do
450 * kernel mode kernel mode down system
451 * kernel mode user mode kill the process
452 * user mode kernel mode down system (*)
453 * user mode user mode kill the process
455 * (*) You could terminate offending user-mode process
456 * if (pbci->pv && pbci->pl != 0) *and* if you sure
457 * the process not have any locks of kernel.
460 /* Is minstate valid? */
461 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate))
462 return fatal_mca(KERN_ALERT "MCA: minstate not valid\n");
463 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
464 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr);
467 * Check the privilege level of interrupted context.
468 * If it is user-mode, then terminate affected process.
471 pmsa = sos->pal_min_state;
472 if (psr1->cpl != 0 ||
473 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) {
474 smei = peidx_bus_check(peidx, 0);
475 if (smei->valid.target_identifier) {
477 * setup for resume to bottom half of MCA,
478 * "mca_handler_bhhook"
480 /* pass to bhhook as argument (gr8, ...) */
481 pmsa->pmsa_gr[8-1] = smei->target_identifier;
482 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip;
483 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr;
484 /* set interrupted return address (but no use) */
485 pmsa->pmsa_br0 = pmsa->pmsa_iip;
486 /* change resume address to bottom half */
487 pmsa->pmsa_iip = mca_hdlr_bh->fp;
488 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
489 /* set cpl with kernel mode */
490 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
496 return MCA_RECOVERED;
501 return fatal_mca(KERN_ALERT "MCA: kernel context not recovered,"
502 " iip 0x%lx\n", pmsa->pmsa_iip);
506 * recover_from_platform_error - Recover from platform error.
507 * @slidx: pointer of index of SAL error record
508 * @peidx: pointer of index of processor error section
509 * @pbci: pointer of pal_bus_check_info
510 * @sos: pointer to hand off struct between SAL and OS
513 * 1 on Success / 0 on Failure
517 recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
518 pal_bus_check_info_t *pbci,
519 struct ia64_sal_os_state *sos)
522 pal_processor_state_info_t *psp =
523 (pal_processor_state_info_t*)peidx_psp(peidx);
525 if (psp->bc && pbci->eb && pbci->bsi == 0) {
527 case 1: /* partial read */
528 case 3: /* full line(cpu) read */
529 case 9: /* I/O space read */
530 status = recover_from_read_error(slidx, peidx, pbci,
533 case 0: /* unknown */
534 case 2: /* partial write */
535 case 4: /* full line write */
536 case 5: /* implicit or explicit write-back operation */
537 case 6: /* snoop probe */
538 case 7: /* incoming or outgoing ptc.g */
539 case 8: /* write coalescing transactions */
540 case 10: /* I/O space write */
541 case 11: /* inter-processor interrupt message(IPI) */
542 case 12: /* interrupt acknowledge or
543 external task priority cycle */
553 * recover_from_processor_error
554 * @platform: whether there are some platform error section or not
555 * @slidx: pointer of index of SAL error record
556 * @peidx: pointer of index of processor error section
557 * @pbci: pointer of pal_bus_check_info
558 * @sos: pointer to hand off struct between SAL and OS
561 * 1 on Success / 0 on Failure
564 * Later we try to recover when below all conditions are satisfied.
565 * 1. Only one processor error section is exist.
566 * 2. BUS_CHECK is exist and the others are not exist.(Except TLB_CHECK)
567 * 3. The entry of BUS_CHECK_INFO is 1.
568 * 4. "External bus error" flag is set and the others are not set.
572 recover_from_processor_error(int platform, slidx_table_t *slidx,
573 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
574 struct ia64_sal_os_state *sos)
576 pal_processor_state_info_t *psp =
577 (pal_processor_state_info_t*)peidx_psp(peidx);
580 * Processor recovery status must key off of the PAL recovery
581 * status in the Processor State Parameter.
585 * The machine check is corrected.
588 return MCA_RECOVERED;
591 * The error was not contained. Software must be reset.
593 if (psp->us || psp->ci == 0)
594 return fatal_mca(KERN_ALERT "MCA: error not contained\n");
597 * The cache check and bus check bits have four possible states
599 * 0 0 Weird record, not recovered
600 * 1 0 Cache error, not recovered
601 * 0 1 I/O error, attempt recovery
602 * 1 1 Memory error, attempt recovery
604 if (psp->bc == 0 || pbci == NULL)
605 return fatal_mca(KERN_ALERT "MCA: No bus check\n");
608 * Sorry, we cannot handle so many.
610 if (peidx_bus_check_num(peidx) > 1)
611 return fatal_mca(KERN_ALERT "MCA: Too many bus checks\n");
613 * Well, here is only one bus error.
616 return fatal_mca(KERN_ALERT "MCA: Internal Bus error\n");
618 return fatal_mca(KERN_ALERT "MCA: Cache-cache error\n");
619 if (pbci->eb && pbci->bsi > 0)
620 return fatal_mca(KERN_ALERT "MCA: External bus check fatal status\n");
623 * This is a local MCA and estimated as recoverble external bus error.
624 * (e.g. a load from poisoned memory)
625 * This means "there are some platform errors".
628 return recover_from_platform_error(slidx, peidx, pbci, sos);
630 * On account of strange SAL error record, we cannot recover.
632 return fatal_mca(KERN_ALERT "MCA: Strange SAL record\n");
636 * mca_try_to_recover - Try to recover from MCA
637 * @rec: pointer to a SAL error record
638 * @sos: pointer to hand off struct between SAL and OS
641 * 1 on Success / 0 on Failure
645 mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
651 pal_bus_check_info_t pbci;
653 /* Make index of SAL error record */
654 platform_err = mca_make_slidx(rec, &slidx);
656 /* Count processor error sections */
657 n_proc_err = slidx_count(&slidx, proc_err);
659 /* Now, OS can recover when there is one processor error section */
661 return fatal_mca(KERN_ALERT "MCA: Too Many Errors\n");
662 else if (n_proc_err == 0)
663 /* Weird SAL record ... We need not to recover */
664 return fatal_mca(KERN_ALERT "MCA: Weird SAL record\n");
666 /* Make index of processor error section */
667 mca_make_peidx((sal_log_processor_info_t*)
668 slidx_first_entry(&slidx.proc_err)->hdr, &peidx);
670 /* Extract Processor BUS_CHECK[0] */
671 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
673 /* Check whether MCA is global or not */
674 if (is_mca_global(&peidx, &pbci, sos))
675 return fatal_mca(KERN_ALERT "MCA: global MCA\n");
677 /* Try to recover a processor error */
678 return recover_from_processor_error(platform_err, &slidx, &peidx,
683 * =============================================================================
686 int __init mca_external_handler_init(void)
688 if (init_record_index_pools())
691 /* register external mca handlers */
692 if (ia64_reg_MCA_extension(mca_try_to_recover)) {
693 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
694 kfree(slidx_pool.buffer);
700 void __exit mca_external_handler_exit(void)
702 /* unregister external mca handlers */
703 ia64_unreg_MCA_extension();
704 kfree(slidx_pool.buffer);
707 module_init(mca_external_handler_init);
708 module_exit(mca_external_handler_exit);
710 module_param(sal_rec_max, int, 0644);
711 MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");
713 MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
714 MODULE_LICENSE("GPL");