Re: [PATCH 1/1] tpm: fix cause of SMP stack traces

[Kylene Hall <kjhall@us.ibm.com>]

On Tue, 2005-01-18 at 16:37, Chris Wright wrote:
> * Kylene Hall (kjhall@us.ibm.com) wrote:
> > There were misplaced spinlock acquires and releases in the probe, open, 
> > close and release paths which were causing might_sleep and schedule while 
> > atomic error messages accompanied by stack traces when the kernel was 
> > compiled with SMP support. Bug reported by Reben Jenster 
> > <ruben@hotheads.de>
> > 
> > Signed-off-by: Kylene Hall <kjhall@us.ibm.com>
> > ---
> > diff -uprN linux-2.6.10/drivers/char/tpm/tpm.c linux-2.6.10-tpm/drivers/char/tpm/tpm.c
> > --- linux-2.6.10/drivers/char/tpm/tpm.c	2005-01-18 16:42:17.000000000 -0600
> > +++ linux-2.6.10-tpm/drivers/char/tpm/tpm.c	2005-01-18 12:52:53.000000000 -0600
> > @@ -373,8 +372,9 @@ int tpm_open(struct inode *inode, struct
> >  {
> >  	int rc = 0, minor = iminor(inode);
> >  	struct tpm_chip *chip = NULL, *pos;
> > +	unsigned long flags;
> >  
> > -	spin_lock(&driver_lock);
> > +	spin_lock_irqsave(&driver_lock, flags);
> 
> Hmm, unless I'm missing something, this is only worse (for might sleep
> warnings).  Now you've disabled irq's too.

I actually had to move the location of some of the locks to remove the
might sleep warnings.  Since I didn't know much about the might sleep
warnings before, my first course of action was to try using the disable
irq mechanism and I went ahead and just left them in once it was working
with the new lock placements.  I assume you believe they shouldn't be
necessary at all?

Thanks,
Kylie   

> 
> thanks,
> -chris