usb-storage filesystem corruption under FC2, kernel ...-435

Fernando Perez Fernando.Perez at colorado.edu
Wed Jun 16 01:35:08 UTC 2004


T. 'Nifty New Hat' Mitchell wrote:
> On Mon, Jun 14, 2004 at 06:28:22PM -0600, Fernando Perez wrote:
> 
>>just reporting that even under today's kernel, the usb-storage problems are 
>>still present (at least under some configurations).  I've been having this 
>>problem ever since installing FC2.
> 
> ....
> 
>>Jun 14 15:44:02 planck kernel: updfstab: Using deprecated /dev/sg mechanism 
>>instead of SG_IO on the actual device
> 
> 
> Two quick questions.
>   * Why are you using a deprecated mechanism?

I didn't configure the kernel to use /dev/sg instead of SG_IO.  I don't even 
know _how_ to make that choice.  I just plug in the usb device, mount it, and 
fire an rsync against it.

 >   * What type of file system is used on the USB device?

ext2

As I said, this process works just fine every day on an identical machine 
running FC1 on the next desk to mine.  In fact, that's how I'm using it now, 
because this little USB disk is my home/office sync-in-a-pocket system.  So 
now instead of plugging it into my own desktop I do it into my officemate's, 
which we have not upgraded to FC2 yet (and wont until it's actually usable).

> With hyperthreading on, you need symmetric multiprocessing support in
> the kernel.  Just guessing, deprecated mechanisms are less likely to be
> multi processor safe than the primary recommended mechanism.

I've reported the bug is present both with both smp and up kernels, so HT is 
not the issue here.

> Regardless there is a bug and it is not likely to be fixed on a
> deprecated functionality.  Pursue it main line if you can.  I suspect,
> like you, that it is related to hyperthreading.  If you cannot suffer
> the errors or time to debug ... turn off hyperthreading(SMP) as the
> most likely problem and when you see reports that it is fixed turn it
> back on and test.

If anyone can suggest what to modify so it doesn't use /dev/sg and use SG_IO 
instead, I'll gladly repeat the test.  It's easy enough to do it, since it 
fails every time.

Thanks,

f.





More information about the fedora-list mailing list