Re: NWFS 2.2.15 Oops Fixed with LINUX_BUFFER_CACHE

From: Jeff V. Merkey (jmerkey@timpanogas.com)
Date: Thu Mar 30 2000 - 01:47:27 EST


Stephen,

Did a lot of stress testing this evening on the changes we discussed
today. I have discovered that unless lock_kernel() is also held over
calls to wait_on_buffer() after calling ll_rw_block(), I start getting
DMA timeouts on the IDE controller (indicating it's getting memory
that's not 512-byte aligned). If lock_kernel() is held over this call,
no problem.

A few more hours of heavy stress testing to go .....

Jeff

"Jeff V. Merkey" wrote:
>
> Thanks. THis is what I needed to know. More changes and I will get it
> up. I will post nwfs0329.tar.gz this evening -- a little more code
> cleanup for 2.4 to do.
>
> Jeff
>
> "Stephen C. Tweedie" wrote:
> >
> > Hi,
> >
> > On Wed, Mar 29, 2000 at 04:51:39PM -0700, Jeff V. Merkey wrote:
> > >
> > > Do I need to also call lock_kernel() around calls to ll_rw_block() as
> > > well (they seem to be SMP safe but thought I would ask)?
> >
> > Yes. The ll_rw_block stuff is SMP safe only with respect to completion
> > of IOs. In other words, you can take device driver interrupts without
> > the big lock, and ll_rw_block is properly threaded with respect to the
> > interrupts.
> >
> > However, the submission of requests in the first place is currently
> > always made with the big lock. ll_rw_block itself probably won't
> > complain if you don't have it, but the device drivers assume that you
> > already hold it so you don't want to risk changing that.
> >
> > --Stephen

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Fri Mar 31 2000 - 21:00:26 EST