Re: Question on memory commit during MR finalize()

From: Paolo Bonzini <pbonzini@redhat.com>
To: Peter Xu <peterx@redhat.com>
Cc: Markus Armbruster <armbru@redhat.com>,
	QEMU Devel Mailing List <qemu-devel@nongnu.org>
Subject: Re: Question on memory commit during MR finalize()
Date: Mon, 20 Apr 2020 23:44:11 +0200	[thread overview]
Message-ID: <fe73a74b-2324-68a5-a37f-530a6bb03ebc@redhat.com> (raw)
In-Reply-To: <20200420210049.GA420399@xz-x1>

On 20/04/20 23:00, Peter Xu wrote:
> 
> I'm still uncertain how the dirty ring branch can easily trigger this, however
> the backtrace looks really odd to me in that we're going to do memory commit
> and even sending KVM ioctls during finalize(), especially in the RCU thread...
> I never expected that.

Short answer: it is really hard to not trigger finalize() from an RCU
callback, and it's the reason why the RCU thread takes the big QEMU lock.

However, instead of memory_region_transaction_commit,
memory_region_finalize probably should do something like

    --memory_region_transaction_depth;
    assert (memory_region_transaction_depth ||
	    (!memory_region_update_pending &&
             !ioeventfd_update_pending));

Paolo