Nishanth Aravamudan wrote:
> Hi,
> 
> At the recommendation of Pat Mansfield, I'm posting this issue to
> linux-scsi. I have run into a big problem with the scsi_debug driver
> which is causing my machine to hang.
> 
> I was finally able to get a large number (10000) of scsi_debug disks via
> modprobe (no partitioning, no mounting, no fs), so I decided to go ahead
> and try to continue the experiment. I found that I was not able to ls in
> a mounted directory from one of the scsi_debug disks. In the chance that
> it was somehow due to the high number of disks present, I tried just
> 
>   modprobe scsi_debug
> 
> so that I would only get one disk. I then ran these commands [0]. I did
> not set the SCSI logging on until the last moment, after I synced the
> existent SCSI disks in the machine (there are two actual SCSI disks)
> [1]. I found that the ls command never fails to hang (although sometimes
> it will not hang immediately, I have to
> 
>   cd lost+found
> 
> first. Interestingly, if the machine does not have highmem (i.e. less
> than 896 MB of RAM) or the kernel is booted with mem=800 (or some other
> number less than 896), the problem dissappears. I had to put the check
> in my patch in for the NULL dereference, as it would cause a
> segmentation fault otherwise. I'm guessing that is where the problem may
> be, as the NULL-check was not there before (an assumption that it
> (scatg2virt) would always work?).

Nishanth,
So this problem seems related to highmem.

The attachment is against the current scsi_debug driver
(at least in lk 2.6.9-rc4). It uses dma_map_sg() and friends
together with phys_to_virt() which comes highly recommended:
"in almost all conceivable cases a device driver should not be
using this function" :-) Could you try the patch.

Perhaps others my be able to answer this: is setting
scsi_host_template::dma_boundary (to some figure other
than 0xffffffff) going to help in this case?

Doug Gilbert