[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH] ide-test: fix failure for test_flush
From: |
Kevin Wolf |
Subject: |
Re: [Qemu-devel] [PATCH] ide-test: fix failure for test_flush |
Date: |
Tue, 11 Jun 2013 14:13:37 +0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
Am 11.06.2013 um 12:05 hat Andreas Färber geschrieben:
> Am 10.06.2013 20:23, schrieb Michael Roth:
> > bd07684aacfb61668ae2c25b7dd00b64f3d7c7f3 added a test to ensure BSY
> > flag is set when a flush request is in flight. It does this by setting
> > a blkdebug breakpoint on flush_to_os before issuing a CMD_FLUSH_CACHE.
> > It then resumes CMD_FLUSH_CACHE operation and checks that BSY is unset.
> >
> > The actual unsetting of BSY does not occur until ide_flush_cb gets
> > called in a bh, however, so in some cases this check will race with
> > the actual completion.
> >
> > Fix this by polling the ide status register until BSY flag gets unset
> > before we do our final sanity checks. According to
> > f68ec8379e88502b4841a110c070e9b118d3151c this is in line with how a guest
> > would determine whether or not the device is still busy.
> >
> > Signed-off-by: Michael Roth <address@hidden>
> > ---
> > tests/ide-test.c | 5 ++++-
> > 1 file changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/tests/ide-test.c b/tests/ide-test.c
> > index 828e71a..7e2eb94 100644
> > --- a/tests/ide-test.c
> > +++ b/tests/ide-test.c
> > @@ -455,7 +455,10 @@ static void test_flush(void)
> > data = inb(IDE_BASE + reg_device);
> > g_assert_cmpint(data & DEV, ==, 0);
> >
> > - data = inb(IDE_BASE + reg_status);
> > + do {
> > + data = inb(IDE_BASE + reg_status);
> > + } while (data & BSY);
>
> Is a busy loop really a good idea for a qtest? CC'ing Anthony.
> For the theoretical case that BSY is not cleared it might be better to
> terminate the loop with some timeout to get an assertion failure or at
> least use some form of sleep() to yield the thread while waiting?
FWIW, the floppy test already has a busy wait for IRQs, which results in
the same failure mode. I think it's okay for a test case, but if someone
felt like implementing timeouts, that would be great and we could apply
them on top.
Kevin