Re: [Linux-usb-users] Possible bug in usb storage (2.6.11 kernel)

[Matthew Dharm]

On Thu, Sep 08, 2005 at 02:28:09PM -0600, Jim Ramsay wrote:
> On 9/8/05, Jim Ramsay <jim.ramsay@xxxxxxxxx> wrote:
> > On 9/8/05, Matthew Dharm <mdharm-kernel@xxxxxxxxxxxxxxxxxx> wrote:
> > > On Thu, Sep 08, 2005 at 11:14:36AM -0600, Jim Ramsay wrote:
> > > > I think I have found a possible bug:
> > > > [...]
> > > > I suppose the scsi code could be changed to guarantee that
> > > > srb->request_buffer is page-aligned or cache-aligned, but that seems
> > > > like the wrong solution for this bug.
> > >
> > > Fixing the SCSI layer is -exactly- the correct solution.  The SCSI layer is
> > > supposed to guarantee us that those buffers are suitable for DMA'ing, and
> > > apparently it's violating that promise.
> > 
> > Thanks, I'll check on what buffer I'm actually getting, where it's
> > allocated, and post back what I find, or how I fixed it.
> 
> More information:
> 
> The error only occurrs during device sensing when the
> srb->request_buffer is assigned as follows, by usb/storage/transport.c
> in the routine usb_stor_invoke_transport:
> 
> old_request_buffer = srb->request_buffer;
> srb->request_buffer = srb->sense_buffer;
> 
> Now, this is a problem because srb->sense_buffer is defined as follows
> in the struct scsi_cmnd:
> 
> #define SCSI_SENSE_BUFFERSIZE   96
>         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
> 
> Since it is not allocated at runtime there is NO WAY the SCSI layer
> can possibly guarantee it is page- or cache-aligned and ready for DMA.
> 
> Any suggestions on best fix for this?  Is it still a SCSI-layer issue?
>  Or should USB step up in this case and ensure this buffer is dma-safe
> itself?

Ah, that buffer doesn't come from SCSI (tho I've long thought they should
provide us with a sense data buffer).  So this is a real usb-storage bug.

Matt

-- 
Matthew Dharm                              Home: mdharm-usb@xxxxxxxxxxxxxxxxxx 
Maintainer, Linux USB Mass Storage Driver

What, are you one of those Microsoft-bashing Linux freaks?
					-- Customer to Greg
User Friendly, 2/10/1999

Attachment: pgp00000.pgp
Description: PGP signature