mca_drv.c

LINUX 2.6.17.4的源码

	      struct ia64_sal_os_state *sos)
{
	pal_processor_state_info_t *psp =
		(pal_processor_state_info_t*)peidx_psp(peidx);

	/*
	 * PAL can request a rendezvous, if the MCA has a global scope.
	 * If "rz_always" flag is set, SAL requests MCA rendezvous
	 * in spite of global MCA.
	 * Therefore it is local MCA when rendezvous has not been requested.
	 * Failed to rendezvous, the system must be down.
	 */
	switch (sos->rv_rc) {
		case -1: /* SAL rendezvous unsuccessful */
			return MCA_IS_GLOBAL;
		case  0: /* SAL rendezvous not required */
			return MCA_IS_LOCAL;
		case  1: /* SAL rendezvous successful int */
		case  2: /* SAL rendezvous successful int with init */
		default:
			break;
	}

	/*
	 * If One or more Cache/TLB/Reg_File/Uarch_Check is here,
	 * it would be a local MCA. (i.e. processor internal error)
	 */
	if (psp->tc || psp->cc || psp->rc || psp->uc)
		return MCA_IS_LOCAL;
	
	/*
	 * Bus_Check structure with Bus_Check.ib (internal bus error) flag set
	 * would be a global MCA. (e.g. a system bus address parity error)
	 */
	if (!pbci || pbci->ib)
		return MCA_IS_GLOBAL;

	/*
	 * Bus_Check structure with Bus_Check.eb (external bus error) flag set
	 * could be either a local MCA or a global MCA.
	 *
	 * Referring Bus_Check.bsi:
	 *   0: Unknown/unclassified
	 *   1: BERR#
	 *   2: BINIT#
	 *   3: Hard Fail
	 * (FIXME: Are these SGI specific or generic bsi values?)
	 */
	if (pbci->eb)
		switch (pbci->bsi) {
			case 0:
				/* e.g. a load from poisoned memory */
				return MCA_IS_LOCAL;
			case 1:
			case 2:
			case 3:
				return MCA_IS_GLOBAL;
		}

	return MCA_IS_GLOBAL;
}

/**
 * recover_from_read_error - Try to recover the errors which type are "read"s.
 * @slidx:	pointer of index of SAL error record
 * @peidx:	pointer of index of processor error section
 * @pbci:	pointer of pal_bus_check_info
 * @sos:	pointer to hand off struct between SAL and OS
 *
 * Return value:
 *	1 on Success / 0 on Failure
 */

static int
recover_from_read_error(slidx_table_t *slidx,
			peidx_table_t *peidx, pal_bus_check_info_t *pbci,
			struct ia64_sal_os_state *sos)
{
	sal_log_mod_error_info_t *smei;
	pal_min_state_area_t *pmsa;
	struct ia64_psr *psr1, *psr2;
	ia64_fptr_t *mca_hdlr_bh = (ia64_fptr_t*)mca_handler_bhhook;

	/* Is target address valid? */
	if (!pbci->tv)
		return fatal_mca(KERN_ALERT "MCA: target address not valid\n");

	/*
	 * cpu read or memory-mapped io read
	 *
	 *    offending process  affected process  OS MCA do
	 *     kernel mode        kernel mode       down system
	 *     kernel mode        user   mode       kill the process
	 *     user   mode        kernel mode       down system (*)
	 *     user   mode        user   mode       kill the process
	 *
	 * (*) You could terminate offending user-mode process
	 *    if (pbci->pv && pbci->pl != 0) *and* if you sure
	 *    the process not have any locks of kernel.
	 */

	/* Is minstate valid? */
	if (!peidx_bottom(peidx) || !(peidx_bottom(peidx)->valid.minstate))
		return fatal_mca(KERN_ALERT "MCA: minstate not valid\n");
	psr1 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_ipsr);
	psr2 =(struct ia64_psr *)&(peidx_minstate_area(peidx)->pmsa_xpsr);

	/*
	 *  Check the privilege level of interrupted context.
	 *   If it is user-mode, then terminate affected process.
	 */

	pmsa = sos->pal_min_state;
	if (psr1->cpl != 0 ||
	   ((psr2->cpl != 0) && mca_recover_range(pmsa->pmsa_iip))) {
		smei = peidx_bus_check(peidx, 0);
		if (smei->valid.target_identifier) {
			/*
			 *  setup for resume to bottom half of MCA,
			 * "mca_handler_bhhook"
			 */
			/* pass to bhhook as argument (gr8, ...) */
			pmsa->pmsa_gr[8-1] = smei->target_identifier;
			pmsa->pmsa_gr[9-1] = pmsa->pmsa_iip;
			pmsa->pmsa_gr[10-1] = pmsa->pmsa_ipsr;
			/* set interrupted return address (but no use) */
			pmsa->pmsa_br0 = pmsa->pmsa_iip;
			/* change resume address to bottom half */
			pmsa->pmsa_iip = mca_hdlr_bh->fp;
			pmsa->pmsa_gr[1-1] = mca_hdlr_bh->gp;
			/* set cpl with kernel mode */
			psr2 = (struct ia64_psr *)&pmsa->pmsa_ipsr;
			psr2->cpl = 0;
			psr2->ri  = 0;
			psr2->bn  = 1;
			psr2->i  = 0;

			return MCA_RECOVERED;
		}

	}

	return fatal_mca(KERN_ALERT "MCA: kernel context not recovered,"
			  " iip 0x%lx\n", pmsa->pmsa_iip);
}

/**
 * recover_from_platform_error - Recover from platform error.
 * @slidx:	pointer of index of SAL error record
 * @peidx:	pointer of index of processor error section
 * @pbci:	pointer of pal_bus_check_info
 * @sos:	pointer to hand off struct between SAL and OS
 *
 * Return value:
 *	1 on Success / 0 on Failure
 */

static int
recover_from_platform_error(slidx_table_t *slidx, peidx_table_t *peidx,
			    pal_bus_check_info_t *pbci,
			    struct ia64_sal_os_state *sos)
{
	int status = 0;
	pal_processor_state_info_t *psp =
		(pal_processor_state_info_t*)peidx_psp(peidx);

	if (psp->bc && pbci->eb && pbci->bsi == 0) {
		switch(pbci->type) {
		case 1: /* partial read */
		case 3: /* full line(cpu) read */
		case 9: /* I/O space read */
			status = recover_from_read_error(slidx, peidx, pbci,
							 sos);
			break;
		case 0: /* unknown */
		case 2: /* partial write */
		case 4: /* full line write */
		case 5: /* implicit or explicit write-back operation */
		case 6: /* snoop probe */
		case 7: /* incoming or outgoing ptc.g */
		case 8: /* write coalescing transactions */
		case 10: /* I/O space write */
		case 11: /* inter-processor interrupt message(IPI) */
		case 12: /* interrupt acknowledge or
				external task priority cycle */
		default:
			break;
		}
	}

	return status;
}

/**
 * recover_from_processor_error
 * @platform:	whether there are some platform error section or not
 * @slidx:	pointer of index of SAL error record
 * @peidx:	pointer of index of processor error section
 * @pbci:	pointer of pal_bus_check_info
 * @sos:	pointer to hand off struct between SAL and OS
 *
 * Return value:
 *	1 on Success / 0 on Failure
 */
/*
 *  Later we try to recover when below all conditions are satisfied.
 *   1. Only one processor error section is exist.
 *   2. BUS_CHECK is exist and the others are not exist.(Except TLB_CHECK)
 *   3. The entry of BUS_CHECK_INFO is 1.
 *   4. "External bus error" flag is set and the others are not set.
 */

static int
recover_from_processor_error(int platform, slidx_table_t *slidx,
			     peidx_table_t *peidx, pal_bus_check_info_t *pbci,
			     struct ia64_sal_os_state *sos)
{
	pal_processor_state_info_t *psp =
		(pal_processor_state_info_t*)peidx_psp(peidx);

	/*
	 * Processor recovery status must key off of the PAL recovery
	 * status in the Processor State Parameter.
	 */

	/*
	 * The machine check is corrected.
	 */
	if (psp->cm == 1)
		return MCA_RECOVERED;

	/*
	 * The error was not contained.  Software must be reset.
	 */
	if (psp->us || psp->ci == 0)
		return fatal_mca(KERN_ALERT "MCA: error not contained\n");

	/*
	 * The cache check and bus check bits have four possible states
	 *   cc bc
	 *    0  0	Weird record, not recovered
	 *    1  0	Cache error, not recovered
	 *    0  1	I/O error, attempt recovery
	 *    1  1	Memory error, attempt recovery
	 */
	if (psp->bc == 0 || pbci == NULL)
		return fatal_mca(KERN_ALERT "MCA: No bus check\n");

	/*
	 * Sorry, we cannot handle so many.
	 */
	if (peidx_bus_check_num(peidx) > 1)
		return fatal_mca(KERN_ALERT "MCA: Too many bus checks\n");
	/*
	 * Well, here is only one bus error.
	 */
	if (pbci->ib)
		return fatal_mca(KERN_ALERT "MCA: Internal Bus error\n");
	if (pbci->cc)
		return fatal_mca(KERN_ALERT "MCA: Cache-cache error\n");
	if (pbci->eb && pbci->bsi > 0)
		return fatal_mca(KERN_ALERT "MCA: External bus check fatal status\n");

	/*
	 * This is a local MCA and estimated as recoverble external bus error.
	 * (e.g. a load from poisoned memory)
	 * This means "there are some platform errors".
	 */
	if (platform)
		return recover_from_platform_error(slidx, peidx, pbci, sos);
	/*
	 * On account of strange SAL error record, we cannot recover.
	 */
	return fatal_mca(KERN_ALERT "MCA: Strange SAL record\n");
}

/**
 * mca_try_to_recover - Try to recover from MCA
 * @rec:	pointer to a SAL error record
 * @sos:	pointer to hand off struct between SAL and OS
 *
 * Return value:
 *	1 on Success / 0 on Failure
 */

static int
mca_try_to_recover(void *rec, struct ia64_sal_os_state *sos)
{
	int platform_err;
	int n_proc_err;
	slidx_table_t slidx;
	peidx_table_t peidx;
	pal_bus_check_info_t pbci;

	/* Make index of SAL error record */
	platform_err = mca_make_slidx(rec, &slidx);

	/* Count processor error sections */
	n_proc_err = slidx_count(&slidx, proc_err);

	 /* Now, OS can recover when there is one processor error section */
	if (n_proc_err > 1)
		return fatal_mca(KERN_ALERT "MCA: Too Many Errors\n");
	else if (n_proc_err == 0)
		/* Weird SAL record ... We need not to recover */
		return fatal_mca(KERN_ALERT "MCA: Weird SAL record\n");

	/* Make index of processor error section */
	mca_make_peidx((sal_log_processor_info_t*)
		slidx_first_entry(&slidx.proc_err)->hdr, &peidx);

	/* Extract Processor BUS_CHECK[0] */
	*((u64*)&pbci) = peidx_check_info(&peidx, bus_check, 0);

	/* Check whether MCA is global or not */
	if (is_mca_global(&peidx, &pbci, sos))
		return fatal_mca(KERN_ALERT "MCA: global MCA\n");
	
	/* Try to recover a processor error */
	return recover_from_processor_error(platform_err, &slidx, &peidx,
					    &pbci, sos);
}

/*
 * =============================================================================
 */

int __init mca_external_handler_init(void)
{
	if (init_record_index_pools())
		return -ENOMEM;

	/* register external mca handlers */
	if (ia64_reg_MCA_extension(mca_try_to_recover)) {	
		printk(KERN_ERR "ia64_reg_MCA_extension failed.\n");
		kfree(slidx_pool.buffer);
		return -EFAULT;
	}
	return 0;
}

void __exit mca_external_handler_exit(void)
{
	/* unregister external mca handlers */
	ia64_unreg_MCA_extension();
	kfree(slidx_pool.buffer);
}

module_init(mca_external_handler_init);
module_exit(mca_external_handler_exit);

module_param(sal_rec_max, int, 0644);
MODULE_PARM_DESC(sal_rec_max, "Max size of SAL error record");

MODULE_DESCRIPTION("ia64 platform dependent mca handler driver");
MODULE_LICENSE("GPL");