On Fri, Oct 20, 2006 at 01:05:10PM -0500, Linas Vepstas wrote:
> Index: linux-2.6.19-rc1-git11/drivers/scsi/sym53c8xx_2/sym_glue.c
> ===================================================================
> --- linux-2.6.19-rc1-git11.orig/drivers/scsi/sym53c8xx_2/sym_glue.c 2006-10-20 12:25:11.000000000 -0500
> +++ linux-2.6.19-rc1-git11/drivers/scsi/sym53c8xx_2/sym_glue.c 2006-10-20 12:41:15.000000000 -0500
> @@ -659,6 +659,11 @@ static irqreturn_t sym53c8xx_intr(int ir
>
> if (DEBUG_FLAGS & DEBUG_TINY) printf_debug ("[");
>
> + /* Avoid spinloop trying to handle interrupts on frozen device */
> + if ((np->s.device->error_state != pci_channel_io_normal) &&
> + (np->s.device->error_state != 0))
> + return IRQ_HANDLED;
> +
This needs to be before the printf_debug call.
> @@ -726,6 +731,19 @@ static int sym_eh_handler(int op, char *
>
> dev_warn(&cmd->device->sdev_gendev, "%s operation started.\n", opname);
>
> + /* We may be in an error condition because the PCI bus
> + * went down. In this case, we need to wait until the
> + * PCI bus is reset, the card is reset, and only then
> + * proceed with the scsi error recovery. There's no
> + * point in hurrying; take a leisurely wait.
> + */
> +#define WAIT_FOR_PCI_RECOVERY 35
> + if ((np->s.device->error_state != pci_channel_io_normal) &&
> + (np->s.device->error_state != 0) &&
> + (wait_for_completion_timeout(&np->s.io_reset_wait,
> + WAIT_FOR_PCI_RECOVERY*HZ) == 0))
> + return SCSI_FAILED;
> +
Is it safe / reasonable / a good idea to sleep for 35 seconds in the EH
handler? I'm not that familiar with how the EH code works. It has its
own thread, so I suppose that's OK.
Are the driver's data structures still intact after a reset?
I generally prefer not to be so perlish in conditionals, ie:
if ((np->s.device->error_state != pci_channel_io_normal) &&
(np->s.device->error_state != 0) {
int timed_out = wait_for_completion_timeout(
&np->s.io_reset_wait, WAIT_FOR_PCI_RECOVERY*HZ);
if (!timed_out)
return SCSI_FAILED;
}
Why is the condition so complicated though? What does 0 mean if it's
not io_normal? At least let's hide that behind a convenience macro:
if (abnormal_error_state(np->s.device->error_state)) {
...
}
> Index: linux-2.6.19-rc1-git11/drivers/scsi/sym53c8xx_2/sym_hipd.c
> ===================================================================
> --- linux-2.6.19-rc1-git11.orig/drivers/scsi/sym53c8xx_2/sym_hipd.c 2006-10-20 12:25:11.000000000 -0500
> +++ linux-2.6.19-rc1-git11/drivers/scsi/sym53c8xx_2/sym_hipd.c 2006-10-20 12:41:16.000000000 -0500
> @@ -2761,6 +2761,7 @@ void sym_interrupt (struct sym_hcb *np)
> u_char istat, istatc;
> u_char dstat;
> u_short sist;
> + u_int icnt;
The cryptic names in this routine are actually register names. Calling
a counter 'icnt' is unhelpful (rather than fitting in with the style).
Just 'i' will do.
> /*
> * interrupt on the fly ?
> @@ -2802,6 +2803,7 @@ void sym_interrupt (struct sym_hcb *np)
> sist = 0;
> dstat = 0;
> istatc = istat;
> + icnt = 0;
> do {
> if (istatc & SIP)
> sist |= INW(np, nc_sist);
> @@ -2809,6 +2811,14 @@ void sym_interrupt (struct sym_hcb *np)
> dstat |= INB(np, nc_dstat);
> istatc = INB(np, nc_istat);
> istat |= istatc;
> +
> + /* Prevent deadlock waiting on a condition that may never clear. */
> + icnt ++;
> + if (icnt > 100) {
> + if ((np->s.device->error_state != pci_channel_io_normal)
> + && (np->s.device->error_state != 0))
> + return;
> + }
> } while (istatc & (SIP|DIP));
Though, since INB and INW will return 0xff and 0xffff, why not use that
as our test rather than using a counter?
if (sist == 0xffff && dstat == 0xff) {
if (abnormal_error_state(np->s.device->error_state)
return;
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
