Re: [Qemu-devel] [PATCH RFT 0/3] iscsi: fix NULL dereferences / races between task completion and abort

[Paolo Bonzini]

> > When performing the UNMAPS, maybe setting timeout to something
> > really big, and at the same time also dropping queue-depth to something
> > really small so that the guest kernel will not be able to send so
> > many UNMAPs concurrently.
> 
> How is this relevant to the fact, that i can't even work with SSH any
> longer with paolo's patch while UNMAPs are running? With my patchset
> you can still work on the machine.

It is relevant because if you tweak the timeouts you will not hit the
cancel at all.

I would like to get a trace of the commands that are sent, so that we
can attack the problem in the guest kernel.  For example, if it's sending
UNMAPs that span 100GB or so, we could enforce a limit in the guest kernel
on the maximum number of discarded blocks per UNMAP.  But if the problem
is the _number_ of UNMAPs rather than the size, it would need a different
heuristic.

Paolo