Content-Type: multipart/mixed;
boundary="------------4B08707380476D0EA895C09B"
This is a multi-part message in MIME format.
--------------4B08707380476D0EA895C09B
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
John Cavan wrote:
> Tim Waugh wrote:
> >
> > On Thu, Nov 16, 2000 at 09:50:40PM -0500, John Cavan wrote:
> >
> > > [...] This patch unlocks, allows the lowlevel driver to do it's
> > > probes, and then relocks. It could probably be more granular in the
> > > parport_pc code, but my own home tests show it to be working fine.
> >
> > Is that safe?
Safer than an oops/lockup :-)
> I'm not sure. I know why it causes the NMI lockup, but I'm not enough of
> an expert to sort it out. I've got a pretty good feel for the Zip
> driver, but not the parport or scsi code yet, so I don't know how safe
> it is. The new scsi error stuff does mention that drivers must
> spinunlock/spinlock if it enables interrupts.
>
> > Also, what bit of the parport code is tripping over the lock?
> > Request_module or something?
>
> During the init phase of the parport_pc module it probes and enables the
> IRQ(s) of the parallel port, but the scsi layer has them locked.
John and Tim,
At least using imm on my SMP machine (BP6 dual celery)
I found that I had to go a bit further than John's patch.
Basically I have unlocked the whole of imm_detect().
It was necessary to unblock parport_enumerate()
but not sufficient.
Please see attachment. I don't have a ppa device to test.
Eric Y. makes some comments just before the (*detect())
call in scsi.c relating to low level driver detect routines.
Doug Gilbert
--------------4B08707380476D0EA895C09B
Content-Type: text/plain; charset=us-ascii;
name="imm.cx2_diff"
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
filename="imm.cx2_diff"
--- linux/drivers/scsi/imm.c Mon Nov 20 16:36:19 2000
+++ linux/drivers/scsi/imm.cx2 Mon Nov 20 22:46:10 2000
@@ -122,14 +122,17 @@
struct Scsi_Host *hreg;
int ports;
int i, nhosts, try_again;
- struct parport *pb = parport_enumerate();
+ struct parport *pb;
printk("imm: Version %s\n", IMM_VERSION);
+ spin_unlock_irq(&io_request_lock);
+ pb = parport_enumerate();
nhosts = 0;
try_again = 0;
if (!pb) {
printk("imm: parport reports no devices.\n");
+ spin_lock_irq(&io_request_lock);
return 0;
}
retry_entry:
@@ -154,6 +157,7 @@
printk(KERN_ERR "imm%d: failed to claim parport because a "
"pardevice is owning the port for too longtime!\n",
i);
+ spin_lock_irq(&io_request_lock);
return 0;
}
}
@@ -208,12 +212,16 @@
nhosts++;
}
if (nhosts == 0) {
- if (try_again == 1)
+ if (try_again == 1) {
+ spin_lock_irq(&io_request_lock);
return 0;
+ }
try_again = 1;
goto retry_entry;
- } else
+ } else {
+ spin_lock_irq(&io_request_lock);
return 1; /* return number of hosts detected */
+ }
}
/* This is to give the imm driver a way to modify the timings (and other
--------------4B08707380476D0EA895C09B--
Is the io_request_lock-dropping patch for ppa/imm needed for 2.2.x as
well, or is it just a 2.4.x thing?
Tim.
*/