Re: Plugin Memory Callback Debugging

["Alex Bennée" <alex.bennee@linaro.org>]

Aaron Lindsay <aaron@os.amperecomputing.com> writes:

> Emilio,
>
> On Dec 18 00:24, Emilio Cota wrote:
>> On Tue, Nov 29, 2022 at 15:37:51 -0500, Aaron Lindsay wrote:
>> (snip)
>> > > Does this hint that there are cases where reset cpu->plugin_mem_cbs to NULL is
>> > > getting optimized away, but not the code to set it in the first place?
>> > 
>> > Is there anyone who could help take a look at this from the code gen
>> > perspective?
>> 
>> Thanks for the report. Just adding assertions was enough to uncover
>> several bugs. I did not reproduce the use-after-free, but by calling
>> reset from a callback it's easy to see how it can occur.
>> 
>> I have fixes in https://github.com/cota/qemu/tree/plugins
>> 
>> Can you please give those a try?
>> 
>> BTW I created an issue on gitlab to track this
>>   https://gitlab.com/qemu-project/qemu/-/issues/1381
>
> Thanks so much for digging into this!
>
> I rebased your plugins branch on top of v7.2.0 and tested with several
> scenarios which reliably triggered the bug for me. None of them
> reproduced the original problem (or hit other bugs!) with your fixes.

Emilio,

Are you going to be able to post the patches soon? I'd like to get the
fixes in as early in the cycle as possible.

Thanks,

-- 
Alex Bennée
Virtualisation Tech Lead @ Linaro