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 * This pool keeps pointers to the section part of SAL error record
69 slidx_list_t *buffer; /* section pointer list pool */
70 int cur_idx; /* Current index of section pointer list pool */
71 int max_idx; /* Maximum index of section pointer list pool */
75 * mca_page_isolate - isolate a poisoned page in order not to use it later
76 * @paddr: poisoned memory location
79 * one of isolate_status_t, ISOLATE_OK/NG/NONE.
82 static isolate_status_t
83 mca_page_isolate(unsigned long paddr)
88 /* whether physical address is valid or not */
89 if (!ia64_phys_addr_valid(paddr))
92 if (!pfn_valid(paddr >> PAGE_SHIFT))
95 /* convert physical address to physical page number */
96 p = pfn_to_page(paddr>>PAGE_SHIFT);
98 /* check whether a page number have been already registered or not */
99 for (i = 0; i < num_page_isolate; i++)
100 if (page_isolate[i] == p)
101 return ISOLATE_OK; /* already listed */
103 /* limitation check */
104 if (num_page_isolate == MAX_PAGE_ISOLATE)
107 /* kick pages having attribute 'SLAB' or 'Reserved' */
108 if (PageSlab(p) || PageReserved(p))
111 /* add attribute 'Reserved' and register the page */
114 page_isolate[num_page_isolate++] = p;
120 * mca_hanlder_bh - Kill the process which occurred memory read error
121 * @paddr: poisoned address received from MCA Handler
125 mca_handler_bh(unsigned long paddr, void *iip, unsigned long ipsr)
127 printk(KERN_ERR "OS_MCA: process [cpu %d, pid: %d, uid: %d, "
128 "iip: %p, psr: 0x%lx,paddr: 0x%lx](%s) encounters MCA.\n",
129 raw_smp_processor_id(), current->pid, current->uid,
130 iip, ipsr, paddr, current->comm);
132 spin_lock(&mca_bh_lock);
133 switch (mca_page_isolate(paddr)) {
135 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr);
138 printk(KERN_CRIT "Page isolation: ( %lx ) failure.\n", paddr);
143 spin_unlock(&mca_bh_lock);
145 /* This process is about to be killed itself */
150 * mca_make_peidx - Make index of processor error section
151 * @slpi: pointer to record of processor error section
152 * @peidx: pointer to index of processor error section
156 mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx)
159 * calculate the start address of
160 * "struct cpuid_info" and "sal_processor_static_info_t".
162 u64 total_check_num = slpi->valid.num_cache_check
163 + slpi->valid.num_tlb_check
164 + slpi->valid.num_bus_check
165 + slpi->valid.num_reg_file_check
166 + slpi->valid.num_ms_check;
167 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num
168 + sizeof(sal_log_processor_info_t);
169 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info);
171 peidx_head(peidx) = slpi;
172 peidx_mid(peidx) = (struct sal_cpuid_info *)
173 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL);
174 peidx_bottom(peidx) = (sal_processor_static_info_t *)
175 (slpi->valid.psi_static_struct ?
176 ((char*)slpi + head_size + mid_size) : NULL);
180 * mca_make_slidx - Make index of SAL error record
181 * @buffer: pointer to SAL error record
182 * @slidx: pointer to index of SAL error record
185 * 1 if record has platform error / 0 if not
187 #define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \
188 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \
190 list_add(&hl->list, &(sect)); \
191 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; }
194 mca_make_slidx(void *buffer, slidx_table_t *slidx)
196 int platform_err = 0;
197 int record_len = ((sal_log_record_header_t*)buffer)->len;
200 sal_log_section_hdr_t *sp;
203 * Initialize index referring current record
205 INIT_LIST_HEAD(&(slidx->proc_err));
206 INIT_LIST_HEAD(&(slidx->mem_dev_err));
207 INIT_LIST_HEAD(&(slidx->sel_dev_err));
208 INIT_LIST_HEAD(&(slidx->pci_bus_err));
209 INIT_LIST_HEAD(&(slidx->smbios_dev_err));
210 INIT_LIST_HEAD(&(slidx->pci_comp_err));
211 INIT_LIST_HEAD(&(slidx->plat_specific_err));
212 INIT_LIST_HEAD(&(slidx->host_ctlr_err));
213 INIT_LIST_HEAD(&(slidx->plat_bus_err));
214 INIT_LIST_HEAD(&(slidx->unsupported));
217 * Extract a Record Header
219 slidx->header = buffer;
222 * Extract each section records
223 * (arranged from "int ia64_log_platform_info_print()")
225 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0;
226 ercd_pos < record_len; ercd_pos += sp->len, sects++) {
227 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos);
228 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) {
229 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp);
230 } else if (!efi_guidcmp(sp->guid,
231 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) {
233 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp);
234 } else if (!efi_guidcmp(sp->guid,
235 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) {
237 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp);
238 } else if (!efi_guidcmp(sp->guid,
239 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) {
241 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp);
242 } else if (!efi_guidcmp(sp->guid,
243 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) {
245 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp);
246 } else if (!efi_guidcmp(sp->guid,
247 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) {
249 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp);
250 } else if (!efi_guidcmp(sp->guid,
251 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) {
253 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp);
254 } else if (!efi_guidcmp(sp->guid,
255 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) {
257 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp);
258 } else if (!efi_guidcmp(sp->guid,
259 SAL_PLAT_BUS_ERR_SECT_GUID)) {
261 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp);
263 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp);
266 slidx->n_sections = sects;
272 * init_record_index_pools - Initialize pool of lists for SAL record index
275 * 0 on Success / -ENOMEM on Failure
278 init_record_index_pools(void)
281 int rec_max_size; /* Maximum size of SAL error records */
282 int sect_min_size; /* Minimum size of SAL error sections */
283 /* minimum size table of each section */
284 static int sal_log_sect_min_sizes[] = {
285 sizeof(sal_log_processor_info_t)
286 + sizeof(sal_processor_static_info_t),
287 sizeof(sal_log_mem_dev_err_info_t),
288 sizeof(sal_log_sel_dev_err_info_t),
289 sizeof(sal_log_pci_bus_err_info_t),
290 sizeof(sal_log_smbios_dev_err_info_t),
291 sizeof(sal_log_pci_comp_err_info_t),
292 sizeof(sal_log_plat_specific_err_info_t),
293 sizeof(sal_log_host_ctlr_err_info_t),
294 sizeof(sal_log_plat_bus_err_info_t),
298 * MCA handler cannot allocate new memory on flight,
299 * so we preallocate enough memory to handle a SAL record.
301 * Initialize a handling set of slidx_pool:
302 * 1. Pick up the max size of SAL error records
303 * 2. Pick up the min size of SAL error sections
304 * 3. Allocate the pool as enough to 2 SAL records
305 * (now we can estimate the maxinum of section in a record.)
309 rec_max_size = sal_rec_max;
312 sect_min_size = sal_log_sect_min_sizes[0];
313 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++)
314 if (sect_min_size > sal_log_sect_min_sizes[i])
315 sect_min_size = sal_log_sect_min_sizes[i];
318 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1;
319 slidx_pool.buffer = (slidx_list_t *)
320 kmalloc(slidx_pool.max_idx * sizeof(slidx_list_t), GFP_KERNEL);
322 return slidx_pool.buffer ? 0 : -ENOMEM;
326 /*****************************************************************************
327 * Recovery functions *
328 *****************************************************************************/
331 * is_mca_global - Check whether this MCA is global or not
332 * @peidx: pointer of index of processor error section
333 * @pbci: pointer to pal_bus_check_info_t
334 * @sos: pointer to hand off struct between SAL and OS
337 * MCA_IS_LOCAL / MCA_IS_GLOBAL
341 is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
342 struct ia64_sal_os_state *sos)
344 pal_processor_state_info_t *psp =
345 (pal_processor_state_info_t*)peidx_psp(peidx);
348 * PAL can request a rendezvous, if the MCA has a global scope.
349 * If "rz_always" flag is set, SAL requests MCA rendezvous
350 * in spite of global MCA.
351 * Therefore it is local MCA when rendezvous has not been requested.
352 * Failed to rendezvous, the system must be down.
354 switch (sos->rv_rc) {
355 case -1: /* SAL rendezvous unsuccessful */
356 return MCA_IS_GLOBAL;
357 case 0: /* SAL rendezvous not required */
359 case 1: /* SAL rendezvous successful int */
360 case 2: /* SAL rendezvous successful int with init */
366 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
367 * it would be a local MCA. (i.e. processor internal error)
369 if (psp->tc || psp->cc || psp->rc || psp->uc)
373 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
374 * would be a global MCA. (e.g. a system bus address parity error)
376 if (!pbci || pbci->ib)
377 return MCA_IS_GLOBAL;
380 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
381 * could be either a local MCA or a global MCA.
383 * Referring Bus_Check.bsi:
384 * 0: Unknown/unclassified
388 * (FIXME: Are these SGI specific or generic bsi values?)
393 /* e.g. a load from poisoned memory */
398 return MCA_IS_GLOBAL;
401 return MCA_IS_GLOBAL;
405 * recover_from_read_error - Try to recover the errors which type are "read"s.
406 * @slidx: pointer of index of SAL error record
407 * @peidx: pointer of index of processor error section
408 * @pbci: pointer of pal_bus_check_info
409 * @sos: pointer to hand off struct between SAL and OS
412 * 1 on Success / 0 on Failure
416 recover_from_read_error(slidx_table_t *slidx,
417 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
418 struct ia64_sal_os_state *sos)
420 sal_log_mod_error_info_t *smei;
421 pal_min_state_area_t *pmsa;
422 struct ia64_psr *psr1, *psr2;
423 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;
425 /* Is target address valid? */
430 * cpu read or memory-mapped io read
432 * offending process affected process OS MCA do
433 * kernel mode kernel mode down system
434 * kernel mode user mode kill the process
435 * user mode kernel mode down system (*)
436 * user mode user mode kill the process
438 * (*) You could terminate offending user-mode process
439 * if (pbci->pv && pbci->pl != 0) *and* if you sure
440 * the process not have any locks of kernel.
443 /* Is minstate valid? */
444 if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate))
446 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
447 psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr);
450 * Check the privilege level of interrupted context.
451 * If it is user-mode, then terminate affected process.
454 pmsa = sos->pal_min_state;
455 if (psr1->cpl != 0 ||
456 ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) {
457 smei = peidx_bus_check(peidx, 0);
458 if (smei->valid.target_identifier) {
460 * setup for resume to bottom half of MCA,
461 * "mca_handler_bhhook"
463 /* pass to bhhook as argument (gr8, ...) */
464 pmsa->pmsa_gr[8-1] = smei->target_identifier;
465 pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip;
466 pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr;
467 /* set interrupted return address (but no use) */
468 pmsa->pmsa_br0 = pmsa->pmsa_iip;
469 /* change resume address to bottom half */
470 pmsa->pmsa_iip = mca_hdlr_bh->fp;
471 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
472 /* set cpl with kernel mode */
473 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
488 * recover_from_platform_error - Recover from platform error.
489 * @slidx: pointer of index of SAL error record
490 * @peidx: pointer of index of processor error section
491 * @pbci: pointer of pal_bus_check_info
492 * @sos: pointer to hand off struct between SAL and OS
495 * 1 on Success / 0 on Failure
499 recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
500 pal_bus_check_info_t *pbci,
501 struct ia64_sal_os_state *sos)
504 pal_processor_state_info_t *psp =
505 (pal_processor_state_info_t*)peidx_psp(peidx);
507 if (psp->bc && pbci->eb && pbci->bsi == 0) {
509 case 1: /* partial read */
510 case 3: /* full line(cpu) read */
511 case 9: /* I/O space read */
512 status = recover_from_read_error(slidx, peidx, pbci,
515 case 0: /* unknown */
516 case 2: /* partial write */
517 case 4: /* full line write */
518 case 5: /* implicit or explicit write-back operation */
519 case 6: /* snoop probe */
520 case 7: /* incoming or outgoing ptc.g */
521 case 8: /* write coalescing transactions */
522 case 10: /* I/O space write */
523 case 11: /* inter-processor interrupt message(IPI) */
524 case 12: /* interrupt acknowledge or
525 external task priority cycle */
535 * recover_from_processor_error
536 * @platform: whether there are some platform error section or not
537 * @slidx: pointer of index of SAL error record
538 * @peidx: pointer of index of processor error section
539 * @pbci: pointer of pal_bus_check_info
540 * @sos: pointer to hand off struct between SAL and OS
543 * 1 on Success / 0 on Failure
546 * Later we try to recover when below all conditions are satisfied.
547 * 1. Only one processor error section is exist.
548 * 2. BUS_CHECK is exist and the others are not exist.(Except TLB_CHECK)
549 * 3. The entry of BUS_CHECK_INFO is 1.
550 * 4. "External bus error" flag is set and the others are not set.
554 recover_from_processor_error(int platform, slidx_table_t *slidx,
555 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
556 struct ia64_sal_os_state *sos)
558 pal_processor_state_info_t *psp =
559 (pal_processor_state_info_t*)peidx_psp(peidx);
562 * Processor recovery status must key off of the PAL recovery
563 * status in the Processor State Parameter.
567 * The machine check is corrected.
573 * The error was not contained. Software must be reset.
575 if (psp->us || psp->ci == 0)
579 * The cache check and bus check bits have four possible states
581 * 0 0 Weird record, not recovered
582 * 1 0 Cache error, not recovered
583 * 0 1 I/O error, attempt recovery
584 * 1 1 Memory error, attempt recovery
586 if (psp->bc == 0 || pbci == NULL)
590 * Sorry, we cannot handle so many.
592 if (peidx_bus_check_num(peidx) > 1)
595 * Well, here is only one bus error.
597 if (pbci->ib || pbci->cc)
599 if (pbci->eb && pbci->bsi > 0)
603 * This is a local MCA and estimated as recoverble external bus error.
604 * (e.g. a load from poisoned memory)
605 * This means "there are some platform errors".
608 return recover_from_platform_error(slidx, peidx, pbci, sos);
610 * On account of strange SAL error record, we cannot recover.
616 * mca_try_to_recover - Try to recover from MCA
617 * @rec: pointer to a SAL error record
618 * @sos: pointer to hand off struct between SAL and OS
621 * 1 on Success / 0 on Failure
625 mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
631 pal_bus_check_info_t pbci;
633 /* Make index of SAL error record */
634 platform_err = mca_make_slidx(rec, &slidx);
636 /* Count processor error sections */
637 n_proc_err = slidx_count(&slidx, proc_err);
639 /* Now, OS can recover when there is one processor error section */
642 else if (n_proc_err == 0) {
643 /* Weird SAL record ... We need not to recover */
648 /* Make index of processor error section */
649 mca_make_peidx((sal_log_processor_info_t*)
650 slidx_first_entry(&slidx.proc_err)->hdr, &peidx);
652 /* Extract Processor BUS_CHECK[0] */
653 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
655 /* Check whether MCA is global or not */
656 if (is_mca_global(&peidx, &pbci, sos))
659 /* Try to recover a processor error */
660 return recover_from_processor_error(platform_err, &slidx, &peidx,
665 * =============================================================================
668 int __init mca_external_handler_init(void)
670 if (init_record_index_pools())
673 /* register external mca handlers */
674 if (ia64_reg_MCA_extension(mca_try_to_recover)) {
675 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
676 kfree(slidx_pool.buffer);
682 void __exit mca_external_handler_exit(void)
684 /* unregister external mca handlers */
685 ia64_unreg_MCA_extension();
686 kfree(slidx_pool.buffer);
689 module_init(mca_external_handler_init);
690 module_exit(mca_external_handler_exit);
692 module_param(sal_rec_max, int, 0644);
693 MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");
695 MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
696 MODULE_LICENSE("GPL");