qemu-block
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-block] [PATCH] macio: Use blk_drain instead of blk_drain_all


From: Stefan Hajnoczi
Subject: Re: [Qemu-block] [PATCH] macio: Use blk_drain instead of blk_drain_all
Date: Tue, 14 Jun 2016 10:08:21 +0100
User-agent: Mutt/1.6.1 (2016-04-27)

On Mon, Jun 13, 2016 at 11:39:48AM +0200, Kevin Wolf wrote:
> Am 12.06.2016 um 08:56 hat Fam Zheng geschrieben:
> > We only care about the associated backend, so blk_drain is more
> > appropriate here.
> > 
> > Signed-off-by: Fam Zheng <address@hidden>
> 
> [ Cc: John ]
> 
> > ---
> >  hw/ide/macio.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/hw/ide/macio.c b/hw/ide/macio.c
> > index 78c10a0..a8c7321 100644
> > --- a/hw/ide/macio.c
> > +++ b/hw/ide/macio.c
> > @@ -400,7 +400,7 @@ static void pmac_ide_flush(DBDMA_io *io)
> >      IDEState *s = idebus_active_if(&m->bus);
> >  
> >      if (s->bus->dma->aiocb) {
> > -        blk_drain_all();
> > +        blk_drain(s->blk);
> >      }
> >  }
> 
> Looks good to me:
> 
> Reviewed-by: Kevin Wolf <address@hidden>
> 
> However, even this is still doing too much. We only need to drain the
> requests that come from this device and can ignore e.g. block job
> requests.
> 
> Now the part that I'm not completely sure about is whether the problem
> is here in the IDE emulation and it should track its own requests or
> whether it is blk_drain() that actually shouldn't drain the BDS but just
> all requests that came in through this specific BB.
> 
> I'm leaning towards the latter, but I'm unsure whether we have cases
> where we actually need to drain the whole root BDS. Any opinions?

There seem to be two distinct drain use cases:

1. Complete all requests that have been submitted via this BB.

2. Quiesce the BDS sub-graph so operations can be performed without
   in-flight I/O intefering.

Device emulation usually needs #1.  Live migration and monitor commands
need #2.

We could distinguish the two by making blk_drain() have #1 semantics
only.  Users that need #2 must use new
blk_drained_begin()/blk_drained_end() functions so the quiescent region
is explicit.

Stefan

Attachment: signature.asc
Description: PGP signature


reply via email to

[Prev in Thread] Current Thread [Next in Thread]