[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH] linux-aio: fix submit aio as a batch
From: |
Stefan Hajnoczi |
Subject: |
Re: [Qemu-devel] [PATCH] linux-aio: fix submit aio as a batch |
Date: |
Wed, 9 Jul 2014 10:29:51 +0200 |
User-agent: |
Mutt/1.5.23 (2014-03-12) |
On Tue, Jul 08, 2014 at 11:45:10PM +0800, Ming Lei wrote:
> In the enqueue path, we can't complete request, otherwise
> "Co-routine re-entered recursively" may be caused, so this
> patch fixes the issue with below ideas:
Thi probably happens when the caller is in coroutine context and its
completion function invokes qemu_coroutine_enter() on itself. The
solution is to invoke completions from a BH (other places in the block
layer do this too).
> - for -EAGAIN, retry the submission in an introduced event handler
I agree with Paolo that a BH is appropriate.
> - for part of completion, just update the io queue, since it is
> moving on after all
If we do this then we need to guarantee that io_submit() will be called
at some point soon. Otherwise requests could get stuck if the guest
doesn't submit any more I/O requests to push the queue.
Please split this into separate patches. You're trying to do too much.
Overall, I would prefer it if we avoid the extra complexity of deferring
io_submit() on EAGAIN and partial submission. Do you understand why the
kernel is producing this behavior? Can we set the right capacity in
io_setup() so it doesn't happen?
> + if (enqueue)
> + return ret;
Please set up a git hook to run checkpatch.pl. It will alert you when
you violate QEMU coding style:
http://blog.vmsplice.net/2011/03/how-to-automatically-run-checkpatchpl.html
I already mentioned coding style in previous patches, using a git hook
will avoid it happening again.
pgpxjnyzmykrz.pgp
Description: PGP signature