Re: req->nr_phys_segments > queue_max_segments (was Re: kernel BUG at drivers/block/virtio_blk.c:172!)

["Thomas D." <whissi@whissi.de>]

Hi,

Jens Axboe wrote:
> Looking at the specific virtio_blk case, it does seem that it is
> checking the segment count before mapping. Does the below fix the
> problem, or does the BUG_ON() still trigger?

Have I understood you correctly, that I should test your patch against
linux-4.1.8?



PS: I upgraded several other systems (they are using dm with LUKS too
but are all ext4, no XFS like this VM) to linux-4.1.9, too and they all
freezed while running (booting was never a problem like with the system
using XFS). Some of them were running for hours until they freezed, some
freezed just seconds after start up finished.

Because I haven't seen the BUG_ON on any of these systems before with
linux-4.1.8 I am not sure if this is the same problem (and the system
with XFS from where I showed you the trace seems to boot fine with
linux-4.1.9 but now freezes, too). Any idea how to debug the freeze?


PS2: On Google I found
http://oss.sgi.com/pipermail/xfs/2014-November/039105.html -- that's why
I am mentioning that I am using XFS (and except to the file system the
VM is identical with the other VMs which didn't show any problems until
linux-4.1.9). I cannot find the patch from
http://oss.sgi.com/pipermail/xfs/2014-November/039149.html in the kernel
mainline. Is it worth to check that direction, too or doesn't the trace
indicate any relationship?


Thanks!


-Thomas