err_titan.c

linux-2.6.15.6

void
titan_machine_check(u64 vector, u64 la_ptr, struct pt_regs *regs)
{
	struct el_common *mchk_header = (struct el_common *)la_ptr;
	struct el_TITAN_sysdata_mcheck *tmchk =
		(struct el_TITAN_sysdata_mcheck *)
		((unsigned long)mchk_header + mchk_header->sys_offset);
	u64 irqmask;

	/*
	 * Mask of Titan interrupt sources which are reported as machine checks
	 *
	 * 63 - CChip Error
	 * 62 - PChip 0 H_Error
	 * 61 - PChip 1 H_Error
	 * 60 - PChip 0 C_Error
	 * 59 - PChip 1 C_Error
	 */
#define TITAN_MCHECK_INTERRUPT_MASK	0xF800000000000000UL

	/*
	 * Sync the processor
	 */
	mb();
	draina();
	
	/*
	 * Only handle system errors here 
	 */
	if ((vector != SCB_Q_SYSMCHK) && (vector != SCB_Q_SYSERR)) {
		ev6_machine_check(vector, la_ptr, regs);
		return;
	}

	/* 
	 * It's a system error, handle it here
	 *
	 * The PALcode has already cleared the error, so just parse it
	 */
	
	/* 
	 * Parse the logout frame without printing first. If the only error(s)
	 * found are classified as "dismissable", then just dismiss them and
	 * don't print any message
	 */
	if (titan_process_logout_frame(mchk_header, 0) != 
	    MCHK_DISPOSITION_DISMISS) {
		char *saved_err_prefix = err_print_prefix;
		err_print_prefix = KERN_CRIT;

		/*
		 * Either a nondismissable error was detected or no
		 * recognized error was detected  in the logout frame 
		 * -- report the error in either case
		 */
		printk("%s"
		       "*System %s Error (Vector 0x%x) reported on CPU %d:\n", 
		       err_print_prefix,
		       (vector == SCB_Q_SYSERR)?"Correctable":"Uncorrectable",
		       (unsigned int)vector, (int)smp_processor_id());
		
#ifdef CONFIG_VERBOSE_MCHECK
		titan_process_logout_frame(mchk_header, alpha_verbose_mcheck);
		if (alpha_verbose_mcheck)
			dik_show_regs(regs, NULL);
#endif /* CONFIG_VERBOSE_MCHECK */

		err_print_prefix = saved_err_prefix;

		/*
		 * Convert any pending interrupts which report as system
		 * machine checks to interrupts
		 */
		irqmask = tmchk->c_dirx & TITAN_MCHECK_INTERRUPT_MASK;
		titan_dispatch_irqs(irqmask, regs);
	}	


	/* 
	 * Release the logout frame 
	 */
	wrmces(0x7);
	mb();
}

/*
 * Subpacket Annotations
 */
static char *el_titan_pchip0_extended_annotation[] = {
	"Subpacket Header", 	"P0_SCTL",	"P0_SERREN",
	"P0_APCTL",		"P0_APERREN",	"P0_AGPERREN",
	"P0_ASPRST",		"P0_AWSBA0",	"P0_AWSBA1",
	"P0_AWSBA2",		"P0_AWSBA3",	"P0_AWSM0",
	"P0_AWSM1",		"P0_AWSM2",	"P0_AWSM3",
	"P0_ATBA0",		"P0_ATBA1",	"P0_ATBA2",
	"P0_ATBA3",		"P0_GPCTL",	"P0_GPERREN",
	"P0_GSPRST",		"P0_GWSBA0",	"P0_GWSBA1",
	"P0_GWSBA2",		"P0_GWSBA3",	"P0_GWSM0",
	"P0_GWSM1",		"P0_GWSM2",	"P0_GWSM3",
	"P0_GTBA0",		"P0_GTBA1",	"P0_GTBA2",
	"P0_GTBA3",		NULL 
};
static char *el_titan_pchip1_extended_annotation[] = {
	"Subpacket Header", 	"P1_SCTL",	"P1_SERREN",
	"P1_APCTL",		"P1_APERREN",	"P1_AGPERREN",
	"P1_ASPRST",		"P1_AWSBA0",	"P1_AWSBA1",
	"P1_AWSBA2",		"P1_AWSBA3",	"P1_AWSM0",
	"P1_AWSM1",		"P1_AWSM2",	"P1_AWSM3",
	"P1_ATBA0",		"P1_ATBA1",	"P1_ATBA2",
	"P1_ATBA3",		"P1_GPCTL",	"P1_GPERREN",
	"P1_GSPRST",		"P1_GWSBA0",	"P1_GWSBA1",
	"P1_GWSBA2",		"P1_GWSBA3",	"P1_GWSM0",
	"P1_GWSM1",		"P1_GWSM2",	"P1_GWSM3",
	"P1_GTBA0",		"P1_GTBA1",	"P1_GTBA2",
	"P1_GTBA3",		NULL 
};
static char *el_titan_memory_extended_annotation[] = {
	"Subpacket Header", 	"AAR0",		"AAR1",
	"AAR2",			"AAR3",		"P0_SCTL",
	"P0_GPCTL",		"P0_APCTL",	"P1_SCTL",
	"P1_GPCTL",		"P1_SCTL",	NULL 
};

static struct el_subpacket_annotation el_titan_annotations[] = {
	SUBPACKET_ANNOTATION(EL_CLASS__REGATTA_FAMILY,
			     EL_TYPE__REGATTA__TITAN_PCHIP0_EXTENDED,
			     1,
			     "Titan PChip 0 Extended Frame",
			     el_titan_pchip0_extended_annotation),
	SUBPACKET_ANNOTATION(EL_CLASS__REGATTA_FAMILY,
			     EL_TYPE__REGATTA__TITAN_PCHIP1_EXTENDED,
			     1,
			     "Titan PChip 1 Extended Frame",
			     el_titan_pchip1_extended_annotation),
	SUBPACKET_ANNOTATION(EL_CLASS__REGATTA_FAMILY,
			     EL_TYPE__REGATTA__TITAN_MEMORY_EXTENDED,
			     1,
			     "Titan Memory Extended Frame",
			     el_titan_memory_extended_annotation),
	SUBPACKET_ANNOTATION(EL_CLASS__REGATTA_FAMILY,
			     EL_TYPE__TERMINATION__TERMINATION,
			     1,
			     "Termination Subpacket",
			     NULL)
};

static struct el_subpacket *
el_process_regatta_subpacket(struct el_subpacket *header)
{
	int status;

	if (header->class != EL_CLASS__REGATTA_FAMILY) {
		printk("%s  ** Unexpected header CLASS %d TYPE %d, aborting\n",
		       err_print_prefix,
		       header->class, header->type);
		return NULL;
	}

	switch(header->type) {
	case EL_TYPE__REGATTA__PROCESSOR_ERROR_FRAME:
	case EL_TYPE__REGATTA__SYSTEM_ERROR_FRAME:
	case EL_TYPE__REGATTA__ENVIRONMENTAL_FRAME:
	case EL_TYPE__REGATTA__PROCESSOR_DBL_ERROR_HALT:
	case EL_TYPE__REGATTA__SYSTEM_DBL_ERROR_HALT:
		printk("%s  ** Occurred on CPU %d:\n", 
		       err_print_prefix,
		       (int)header->by_type.regatta_frame.cpuid);
		status = privateer_process_logout_frame((struct el_common *)
			header->by_type.regatta_frame.data_start, 1);
		break;
	default:
		printk("%s  ** REGATTA TYPE %d SUBPACKET\n", 
		       err_print_prefix, header->type);
		el_annotate_subpacket(header);
		break;
	}


	return (struct el_subpacket *)((unsigned long)header + header->length);
} 

static struct el_subpacket_handler titan_subpacket_handler = 
	SUBPACKET_HANDLER_INIT(EL_CLASS__REGATTA_FAMILY, 
			       el_process_regatta_subpacket);

void
titan_register_error_handlers(void)
{
	size_t i;

	for (i = 0; i < ARRAY_SIZE (el_titan_annotations); i++)
		cdl_register_subpacket_annotation(&el_titan_annotations[i]);

	cdl_register_subpacket_handler(&titan_subpacket_handler);

	ev6_register_error_handlers();
}


/*
 * Privateer
 */

static int
privateer_process_680_frame(struct el_common *mchk_header, int print)
{
	int status = MCHK_DISPOSITION_UNKNOWN_ERROR;
#ifdef CONFIG_VERBOSE_MCHECK
	struct el_PRIVATEER_envdata_mcheck *emchk =
		(struct el_PRIVATEER_envdata_mcheck *)
		((unsigned long)mchk_header + mchk_header->sys_offset);

	/* TODO - catagorize errors, for now, no error */

	if (!print)
		return status;

	/* TODO - decode instead of just dumping... */
	printk("%s  Summary Flags:         %016lx\n"
 	         "  CChip DIRx:            %016lx\n"
		 "  System Management IR:  %016lx\n"
		 "  CPU IR:                %016lx\n"
		 "  Power Supply IR:       %016lx\n"
		 "  LM78 Fault Status:     %016lx\n"
		 "  System Doors:          %016lx\n"
		 "  Temperature Warning:   %016lx\n"
		 "  Fan Control:           %016lx\n"
		 "  Fatal Power Down Code: %016lx\n",
	       err_print_prefix,
	       emchk->summary,
	       emchk->c_dirx,
	       emchk->smir,
	       emchk->cpuir,
	       emchk->psir,
	       emchk->fault,
	       emchk->sys_doors,
	       emchk->temp_warn,
	       emchk->fan_ctrl,
	       emchk->code);
#endif /* CONFIG_VERBOSE_MCHECK */

	return status;
}

int
privateer_process_logout_frame(struct el_common *mchk_header, int print)
{
	struct el_common_EV6_mcheck *ev6mchk = 
		(struct el_common_EV6_mcheck *)mchk_header;
	int status = MCHK_DISPOSITION_UNKNOWN_ERROR;

	/*
	 * Machine check codes
	 */
#define PRIVATEER_MCHK__CORR_ECC		0x86	/* 630 */
#define PRIVATEER_MCHK__DC_TAG_PERR		0x9E	/* 630 */
#define PRIVATEER_MCHK__PAL_BUGCHECK		0x8E	/* 670 */
#define PRIVATEER_MCHK__OS_BUGCHECK		0x90	/* 670 */
#define PRIVATEER_MCHK__PROC_HRD_ERR		0x98	/* 670 */
#define PRIVATEER_MCHK__ISTREAM_CMOV_PRX	0xA0	/* 670 */
#define PRIVATEER_MCHK__ISTREAM_CMOV_FLT	0xA2	/* 670 */
#define PRIVATEER_MCHK__SYS_HRD_ERR		0x202	/* 660 */
#define PRIVATEER_MCHK__SYS_CORR_ERR		0x204	/* 620 */
#define PRIVATEER_MCHK__SYS_ENVIRON		0x206	/* 680 */

	switch(ev6mchk->MCHK_Code) {
	/*
	 * Vector 630 - Processor, Correctable
	 */
	case PRIVATEER_MCHK__CORR_ECC:
	case PRIVATEER_MCHK__DC_TAG_PERR:
		/*
		 * Fall through to vector 670 for processing...
		 */
	/*
	 * Vector 670 - Processor, Uncorrectable
	 */
	case PRIVATEER_MCHK__PAL_BUGCHECK:
	case PRIVATEER_MCHK__OS_BUGCHECK:
	case PRIVATEER_MCHK__PROC_HRD_ERR:
	case PRIVATEER_MCHK__ISTREAM_CMOV_PRX:
	case PRIVATEER_MCHK__ISTREAM_CMOV_FLT:
		status |= ev6_process_logout_frame(mchk_header, print);
		break;

	/*
	 * Vector 620 - System, Correctable
	 */
	case PRIVATEER_MCHK__SYS_CORR_ERR:
		/*
		 * Fall through to vector 660 for processing...
		 */
	/*
	 * Vector 660 - System, Uncorrectable
	 */
	case PRIVATEER_MCHK__SYS_HRD_ERR:
		status |= titan_process_logout_frame(mchk_header, print);
		break;

	/* 
	 * Vector 680 - System, Environmental
	 */
	case PRIVATEER_MCHK__SYS_ENVIRON:	/* System, Environmental */
		status |= privateer_process_680_frame(mchk_header, print);
		break;

	/* 
	 * Unknown
	 */
	default:
		status |= MCHK_DISPOSITION_REPORT;
		if (print) {
			printk("%s** Unknown Error, frame follows\n", 
			       err_print_prefix);
			mchk_dump_logout_frame(mchk_header);
		}

	}

	return status;
}

void
privateer_machine_check(u64 vector, u64 la_ptr, struct pt_regs *regs)
{
	struct el_common *mchk_header = (struct el_common *)la_ptr;
	struct el_TITAN_sysdata_mcheck *tmchk =
		(struct el_TITAN_sysdata_mcheck *)
		(la_ptr + mchk_header->sys_offset);
	u64 irqmask;
	char *saved_err_prefix = err_print_prefix;

#define PRIVATEER_680_INTERRUPT_MASK		(0xE00UL)
#define PRIVATEER_HOTPLUG_INTERRUPT_MASK	(0xE00UL)

	/*
	 * Sync the processor.
	 */
	mb();
	draina();

	/* 
	 * Only handle system events here.
	 */
	if (vector != SCB_Q_SYSEVENT) 
		return titan_machine_check(vector, la_ptr, regs);

	/*
	 * Report the event - System Events should be reported even if no
	 * error is indicated since the event could indicate the return
	 * to normal status.
	 */
	err_print_prefix = KERN_CRIT;
	printk("%s*System Event (Vector 0x%x) reported on CPU %d:\n", 
	       err_print_prefix,
	       (unsigned int)vector, (int)smp_processor_id());
	privateer_process_680_frame(mchk_header, 1);
	err_print_prefix = saved_err_prefix;
	
	/* 
	 * Convert any pending interrupts which report as 680 machine
	 * checks to interrupts.
	 */
	irqmask = tmchk->c_dirx & PRIVATEER_680_INTERRUPT_MASK;

	/*
	 * Dispatch the interrupt(s).
	 */
	titan_dispatch_irqs(irqmask, regs);

	/* 
	 * Release the logout frame.
	 */
	wrmces(0x7);
	mb();
}