06.06.2023 11:45, Hanna Czenczek wrote:
On 06.06.23 10:00, Michael Tokarev wrote:
..
This seems to be over-complicated, both of them, no?
I would have preferred to have this discussion while the patch was still on-list for review (this specific version was for two months, counting from
the first version was three). Do you think it is so complicated and thus bug-prone that we must revert this series now and try the other route?
Well. I come across this change only now when reviewing patches applied to
qemu/master.
This one fixes a real bug which people were hitting, which is also quite
difficult to
diagnose and which has a possibility for data corruption and other
"interesting" effects,
so it is quite a natural thing to at least think about back-porting this change
to
previous -stable qemu release. Bugs like this should be fixed in -stable IMHO.
Sadly I haven't noticed this change before, sure I'd have exactly the same
thoughts
by then, and would be glad to help analyzing other parts of the code with
potential
of having issues with IOV_MAX-exceeding vectors.
I can agree that perhaps the other route could have been simpler, but now we already have patches that are reviewed and in master, which solve the
problem. So I won’t spend more time on tackling this issue from another angle. If you are happy to do so, patches are always welcome.
That's a good point too.
Thanks,
/mjt