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>
10 #include <linux/config.h>
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 * This pool keeps pointers to the section part of SAL error record
68 slidx_list_t *buffer; /* section pointer list pool */
69 int cur_idx; /* Current index of section pointer list pool */
70 int max_idx; /* Maximum index of section pointer list pool */
74 * mca_page_isolate - isolate a poisoned page in order not to use it later
75 * @paddr: poisoned memory location
78 * one of isolate_status_t, ISOLATE_OK/NG/NONE.
81 static isolate_status_t
82 mca_page_isolate(unsigned long paddr)
87 /* whether physical address is valid or not */
88 if (!ia64_phys_addr_valid(paddr))
91 if (!pfn_valid(paddr))
94 /* convert physical address to physical page number */
95 p = pfn_to_page(paddr>>PAGE_SHIFT);
97 /* check whether a page number have been already registered or not */
98 for (i = 0; i < num_page_isolate; i++)
99 if (page_isolate[i] == p)
100 return ISOLATE_OK; /* already listed */
102 /* limitation check */
103 if (num_page_isolate == MAX_PAGE_ISOLATE)
106 /* kick pages having attribute 'SLAB' or 'Reserved' */
107 if (PageSlab(p) || PageReserved(p))
110 /* add attribute 'Reserved' and register the page */
112 page_isolate[num_page_isolate++] = p;
118 * mca_hanlder_bh - Kill the process which occurred memory read error
119 * @paddr: poisoned address received from MCA Handler
123 mca_handler_bh(unsigned long paddr)
125 printk(KERN_DEBUG "OS_MCA: process [pid: %d](%s) encounters MCA.\n",
126 current->pid, current->comm);
128 spin_lock(&mca_bh_lock);
129 switch (mca_page_isolate(paddr)) {
131 printk(KERN_DEBUG "Page isolation: ( %lx ) success.\n", paddr);
134 printk(KERN_DEBUG "Page isolation: ( %lx ) failure.\n", paddr);
139 spin_unlock(&mca_bh_lock);
141 /* This process is about to be killed itself */
146 * mca_make_peidx - Make index of processor error section
147 * @slpi: pointer to record of processor error section
148 * @peidx: pointer to index of processor error section
152 mca_make_peidx(sal_log_processor_info_t *slpi, peidx_table_t *peidx)
155 * calculate the start address of
156 * "struct cpuid_info" and "sal_processor_static_info_t".
158 u64 total_check_num = slpi->valid.num_cache_check
159 + slpi->valid.num_tlb_check
160 + slpi->valid.num_bus_check
161 + slpi->valid.num_reg_file_check
162 + slpi->valid.num_ms_check;
163 u64 head_size = sizeof(sal_log_mod_error_info_t) * total_check_num
164 + sizeof(sal_log_processor_info_t);
165 u64 mid_size = slpi->valid.cpuid_info * sizeof(struct sal_cpuid_info);
167 peidx_head(peidx) = slpi;
168 peidx_mid(peidx) = (struct sal_cpuid_info *)
169 (slpi->valid.cpuid_info ? ((char*)slpi + head_size) : NULL);
170 peidx_bottom(peidx) = (sal_processor_static_info_t *)
171 (slpi->valid.psi_static_struct ?
172 ((char*)slpi + head_size + mid_size) : NULL);
176 * mca_make_slidx - Make index of SAL error record
177 * @buffer: pointer to SAL error record
178 * @slidx: pointer to index of SAL error record
181 * 1 if record has platform error / 0 if not
183 #define LOG_INDEX_ADD_SECT_PTR(sect, ptr) \
184 {slidx_list_t *hl = &slidx_pool.buffer[slidx_pool.cur_idx]; \
186 list_add(&hl->list, &(sect)); \
187 slidx_pool.cur_idx = (slidx_pool.cur_idx + 1)%slidx_pool.max_idx; }
190 mca_make_slidx(void *buffer, slidx_table_t *slidx)
192 int platform_err = 0;
193 int record_len = ((sal_log_record_header_t*)buffer)->len;
196 sal_log_section_hdr_t *sp;
199 * Initialize index referring current record
201 INIT_LIST_HEAD(&(slidx->proc_err));
202 INIT_LIST_HEAD(&(slidx->mem_dev_err));
203 INIT_LIST_HEAD(&(slidx->sel_dev_err));
204 INIT_LIST_HEAD(&(slidx->pci_bus_err));
205 INIT_LIST_HEAD(&(slidx->smbios_dev_err));
206 INIT_LIST_HEAD(&(slidx->pci_comp_err));
207 INIT_LIST_HEAD(&(slidx->plat_specific_err));
208 INIT_LIST_HEAD(&(slidx->host_ctlr_err));
209 INIT_LIST_HEAD(&(slidx->plat_bus_err));
210 INIT_LIST_HEAD(&(slidx->unsupported));
213 * Extract a Record Header
215 slidx->header = buffer;
218 * Extract each section records
219 * (arranged from "int ia64_log_platform_info_print()")
221 for (ercd_pos = sizeof(sal_log_record_header_t), sects = 0;
222 ercd_pos < record_len; ercd_pos += sp->len, sects++) {
223 sp = (sal_log_section_hdr_t *)((char*)buffer + ercd_pos);
224 if (!efi_guidcmp(sp->guid, SAL_PROC_DEV_ERR_SECT_GUID)) {
225 LOG_INDEX_ADD_SECT_PTR(slidx->proc_err, sp);
226 } else if (!efi_guidcmp(sp->guid,
227 SAL_PLAT_MEM_DEV_ERR_SECT_GUID)) {
229 LOG_INDEX_ADD_SECT_PTR(slidx->mem_dev_err, sp);
230 } else if (!efi_guidcmp(sp->guid,
231 SAL_PLAT_SEL_DEV_ERR_SECT_GUID)) {
233 LOG_INDEX_ADD_SECT_PTR(slidx->sel_dev_err, sp);
234 } else if (!efi_guidcmp(sp->guid,
235 SAL_PLAT_PCI_BUS_ERR_SECT_GUID)) {
237 LOG_INDEX_ADD_SECT_PTR(slidx->pci_bus_err, sp);
238 } else if (!efi_guidcmp(sp->guid,
239 SAL_PLAT_SMBIOS_DEV_ERR_SECT_GUID)) {
241 LOG_INDEX_ADD_SECT_PTR(slidx->smbios_dev_err, sp);
242 } else if (!efi_guidcmp(sp->guid,
243 SAL_PLAT_PCI_COMP_ERR_SECT_GUID)) {
245 LOG_INDEX_ADD_SECT_PTR(slidx->pci_comp_err, sp);
246 } else if (!efi_guidcmp(sp->guid,
247 SAL_PLAT_SPECIFIC_ERR_SECT_GUID)) {
249 LOG_INDEX_ADD_SECT_PTR(slidx->plat_specific_err, sp);
250 } else if (!efi_guidcmp(sp->guid,
251 SAL_PLAT_HOST_CTLR_ERR_SECT_GUID)) {
253 LOG_INDEX_ADD_SECT_PTR(slidx->host_ctlr_err, sp);
254 } else if (!efi_guidcmp(sp->guid,
255 SAL_PLAT_BUS_ERR_SECT_GUID)) {
257 LOG_INDEX_ADD_SECT_PTR(slidx->plat_bus_err, sp);
259 LOG_INDEX_ADD_SECT_PTR(slidx->unsupported, sp);
262 slidx->n_sections = sects;
268 * init_record_index_pools - Initialize pool of lists for SAL record index
271 * 0 on Success / -ENOMEM on Failure
274 init_record_index_pools(void)
277 int rec_max_size; /* Maximum size of SAL error records */
278 int sect_min_size; /* Minimum size of SAL error sections */
279 /* minimum size table of each section */
280 static int sal_log_sect_min_sizes[] = {
281 sizeof(sal_log_processor_info_t)
282 + sizeof(sal_processor_static_info_t),
283 sizeof(sal_log_mem_dev_err_info_t),
284 sizeof(sal_log_sel_dev_err_info_t),
285 sizeof(sal_log_pci_bus_err_info_t),
286 sizeof(sal_log_smbios_dev_err_info_t),
287 sizeof(sal_log_pci_comp_err_info_t),
288 sizeof(sal_log_plat_specific_err_info_t),
289 sizeof(sal_log_host_ctlr_err_info_t),
290 sizeof(sal_log_plat_bus_err_info_t),
294 * MCA handler cannot allocate new memory on flight,
295 * so we preallocate enough memory to handle a SAL record.
297 * Initialize a handling set of slidx_pool:
298 * 1. Pick up the max size of SAL error records
299 * 2. Pick up the min size of SAL error sections
300 * 3. Allocate the pool as enough to 2 SAL records
301 * (now we can estimate the maxinum of section in a record.)
305 rec_max_size = sal_rec_max;
308 sect_min_size = sal_log_sect_min_sizes[0];
309 for (i = 1; i < sizeof sal_log_sect_min_sizes/sizeof(size_t); i++)
310 if (sect_min_size > sal_log_sect_min_sizes[i])
311 sect_min_size = sal_log_sect_min_sizes[i];
314 slidx_pool.max_idx = (rec_max_size/sect_min_size) * 2 + 1;
315 slidx_pool.buffer = (slidx_list_t *)
316 kmalloc(slidx_pool.max_idx * sizeof(slidx_list_t), GFP_KERNEL);
318 return slidx_pool.buffer ? 0 : -ENOMEM;
322 /*****************************************************************************
323 * Recovery functions *
324 *****************************************************************************/
327 * is_mca_global - Check whether this MCA is global or not
328 * @peidx: pointer of index of processor error section
329 * @pbci: pointer to pal_bus_check_info_t
330 * @sos: pointer to hand off struct between SAL and OS
333 * MCA_IS_LOCAL / MCA_IS_GLOBAL
337 is_mca_global(peidx_table_t *peidx, pal_bus_check_info_t *pbci,
338 struct ia64_sal_os_state *sos)
340 pal_processor_state_info_t *psp =
341 (pal_processor_state_info_t*)peidx_psp(peidx);
344 * PAL can request a rendezvous, if the MCA has a global scope.
345 * If "rz_always" flag is set, SAL requests MCA rendezvous
346 * in spite of global MCA.
347 * Therefore it is local MCA when rendezvous has not been requested.
348 * Failed to rendezvous, the system must be down.
350 switch (sos->rv_rc) {
351 case -1: /* SAL rendezvous unsuccessful */
352 return MCA_IS_GLOBAL;
353 case 0: /* SAL rendezvous not required */
355 case 1: /* SAL rendezvous successful int */
356 case 2: /* SAL rendezvous successful int with init */
362 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
363 * it would be a local MCA. (i.e. processor internal error)
365 if (psp->tc || psp->cc || psp->rc || psp->uc)
369 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
370 * would be a global MCA. (e.g. a system bus address parity error)
372 if (!pbci || pbci->ib)
373 return MCA_IS_GLOBAL;
376 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
377 * could be either a local MCA or a global MCA.
379 * Referring Bus_Check.bsi:
380 * 0: Unknown/unclassified
384 * (FIXME: Are these SGI specific or generic bsi values?)
389 /* e.g. a load from poisoned memory */
394 return MCA_IS_GLOBAL;
397 return MCA_IS_GLOBAL;
401 * recover_from_read_error - Try to recover the errors which type are "read"s.
402 * @slidx: pointer of index of SAL error record
403 * @peidx: pointer of index of processor error section
404 * @pbci: pointer of pal_bus_check_info
405 * @sos: pointer to hand off struct between SAL and OS
408 * 1 on Success / 0 on Failure
412 recover_from_read_error(slidx_table_t *slidx,
413 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
414 struct ia64_sal_os_state *sos)
416 sal_log_mod_error_info_t *smei;
417 pal_min_state_area_t *pmsa;
418 struct ia64_psr *psr1, *psr2;
419 ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;
421 /* Is target address valid? */
426 * cpu read or memory-mapped io read
428 * offending process affected process OS MCA do
429 * kernel mode kernel mode down system
430 * kernel mode user mode kill the process
431 * user mode kernel mode down system (*)
432 * user mode user mode kill the process
434 * (*) You could terminate offending user-mode process
435 * if (pbci->pv && pbci->pl != 0) *and* if you sure
436 * the process not have any locks of kernel.
439 psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
442 * Check the privilege level of interrupted context.
443 * If it is user-mode, then terminate affected process.
445 if (psr1->cpl != 0) {
446 smei = peidx_bus_check(peidx, 0);
447 if (smei->valid.target_identifier) {
449 * setup for resume to bottom half of MCA,
450 * "mca_handler_bhhook"
452 pmsa = sos->pal_min_state;
453 /* pass to bhhook as 1st argument (gr8) */
454 pmsa->pmsa_gr[8-1] = smei->target_identifier;
455 /* set interrupted return address (but no use) */
456 pmsa->pmsa_br0 = pmsa->pmsa_iip;
457 /* change resume address to bottom half */
458 pmsa->pmsa_iip = mca_hdlr_bh->fp;
459 pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
460 /* set cpl with kernel mode */
461 psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
475 * recover_from_platform_error - Recover from platform error.
476 * @slidx: pointer of index of SAL error record
477 * @peidx: pointer of index of processor error section
478 * @pbci: pointer of pal_bus_check_info
479 * @sos: pointer to hand off struct between SAL and OS
482 * 1 on Success / 0 on Failure
486 recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
487 pal_bus_check_info_t *pbci,
488 struct ia64_sal_os_state *sos)
491 pal_processor_state_info_t *psp =
492 (pal_processor_state_info_t*)peidx_psp(peidx);
494 if (psp->bc && pbci->eb && pbci->bsi == 0) {
496 case 1: /* partial read */
497 case 3: /* full line(cpu) read */
498 case 9: /* I/O space read */
499 status = recover_from_read_error(slidx, peidx, pbci,
502 case 0: /* unknown */
503 case 2: /* partial write */
504 case 4: /* full line write */
505 case 5: /* implicit or explicit write-back operation */
506 case 6: /* snoop probe */
507 case 7: /* incoming or outgoing ptc.g */
508 case 8: /* write coalescing transactions */
509 case 10: /* I/O space write */
510 case 11: /* inter-processor interrupt message(IPI) */
511 case 12: /* interrupt acknowledge or
512 external task priority cycle */
522 * recover_from_processor_error
523 * @platform: whether there are some platform error section or not
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
533 * Later we try to recover when below all conditions are satisfied.
534 * 1. Only one processor error section is exist.
535 * 2. BUS_CHECK is exist and the others are not exist.(Except TLB_CHECK)
536 * 3. The entry of BUS_CHECK_INFO is 1.
537 * 4. "External bus error" flag is set and the others are not set.
541 recover_from_processor_error(int platform, slidx_table_t *slidx,
542 peidx_table_t *peidx, pal_bus_check_info_t *pbci,
543 struct ia64_sal_os_state *sos)
545 pal_processor_state_info_t *psp =
546 (pal_processor_state_info_t*)peidx_psp(peidx);
549 * We cannot recover errors with other than bus_check.
551 if (psp->cc || psp->rc || psp->uc)
555 * If there is no bus error, record is weird but we need not to recover.
557 if (psp->bc == 0 || pbci == NULL)
561 * Sorry, we cannot handle so many.
563 if (peidx_bus_check_num(peidx) > 1)
566 * Well, here is only one bus error.
568 if (pbci->ib || pbci->cc)
570 if (pbci->eb && pbci->bsi > 0)
576 * This is a local MCA and estimated as recoverble external bus error.
577 * (e.g. a load from poisoned memory)
578 * This means "there are some platform errors".
581 return recover_from_platform_error(slidx, peidx, pbci, sos);
583 * On account of strange SAL error record, we cannot recover.
589 * mca_try_to_recover - Try to recover from MCA
590 * @rec: pointer to a SAL error record
591 * @sos: pointer to hand off struct between SAL and OS
594 * 1 on Success / 0 on Failure
598 mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
604 pal_bus_check_info_t pbci;
606 /* Make index of SAL error record */
607 platform_err = mca_make_slidx(rec, &slidx);
609 /* Count processor error sections */
610 n_proc_err = slidx_count(&slidx, proc_err);
612 /* Now, OS can recover when there is one processor error section */
615 else if (n_proc_err == 0) {
616 /* Weird SAL record ... We need not to recover */
621 /* Make index of processor error section */
622 mca_make_peidx((sal_log_processor_info_t*)
623 slidx_first_entry(&slidx.proc_err)->hdr, &peidx);
625 /* Extract Processor BUS_CHECK[0] */
626 *((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);
628 /* Check whether MCA is global or not */
629 if (is_mca_global(&peidx, &pbci, sos))
632 /* Try to recover a processor error */
633 return recover_from_processor_error(platform_err, &slidx, &peidx,
638 * =============================================================================
641 int __init mca_external_handler_init(void)
643 if (init_record_index_pools())
646 /* register external mca handlers */
647 if (ia64_reg_MCA_extension(mca_try_to_recover)) {
648 printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
649 kfree(slidx_pool.buffer);
655 void __exit mca_external_handler_exit(void)
657 /* unregister external mca handlers */
658 ia64_unreg_MCA_extension();
659 kfree(slidx_pool.buffer);
662 module_init(mca_external_handler_init);
663 module_exit(mca_external_handler_exit);
665 module_param(sal_rec_max, int, 0644);
666 MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");
668 MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
669 MODULE_LICENSE("GPL");