Re: ide-cd turning off DMA when verifying DVD-R

Re: ide-cd turning off DMA when verifying DVD-R

[Bartlomiej Zolnierkiewicz]

Hi,

On 1/13/06, Volker Kuhlmann <list0570@paradise.net.nz> wrote:
> On Fri 13 Jan 2006 17:40:33 NZDT +1300, Robert Hancock wrote:
>
> > I'm thinking the IDE code is too aggressive in assuming that the failure
> >  is because of a DMA problem and disabling it.. Most likely all that's
> > happening is the drive is taking a long time to complete the current
> > command.

What actually happened is that normal command timed out
and because of that driver reset the device which caused
it to loose DMA:

->ide_atapi_error()
    ->ide_do_reset()
      ->pre_reset()
        ->check_dma_crc()
          ->__ide_dma_off()

Somebody needs to investigate why __ide_dma_off() is called
et all and if we need to restore DMA after reset (don't count ATM
on me, I'm buried by bugreports).  Ondrej, could you fill the bug at
http://bugzilla.kernel.org so we don't lose it?

> Yes! Each time when inserting a faulty CD/DVD, or whenever the drive
> gives read errors for whatever reason, the kernel decides to turn DMA
> off and try again, fail (again) and leave DMA off. And this after having
> successfully used DMA before - so it's not that the device is
> DMA-incapable.

This is a separate issue and is related to retrying failed DMA
commands.  Driver doesn't know what was the cause of the timeout
and it shouldn't disable DMA for the device (only the failed request
should be retried in PIO mode).

Any takers? :)

PS1 please don't trim cc: list
PS2 please use linux-ide@vger.kernel.org for ATA problem

Thanks,
Bartlomiej

Re: ide-cd turning off DMA when verifying DVD-R

[Ondrej Zary]

Bartlomiej Zolnierkiewicz wrote:
> Ondrej, could you fill the bug at
> http://bugzilla.kernel.org so we don't lose it?
> 
Here it is http://bugzilla.kernel.org/show_bug.cgi?id=5882

-- 
Ondrej Zary

Re: Fwd: ide-cd turning off DMA when verifying DVD-R

[Bartlomiej Zolnierkiewicz]

I'm adding missing cc:, I hope I didn't forget about anybody this time.

There is now bugzilla entry for this bug (thanks Ondrej):
http://bugzilla.kernel.org/show_bug.cgi?id=5882

Jens, could you take a look at the part related to ide-cd timeouts?

On 1/13/06, Ville Syrjälä <syrjala@sci.fi> wrote:
> On Fri, Jan 13, 2006 at 10:49:22AM +0100, Bartlomiej Zolnierkiewicz wrote:
> > On 1/13/06, Ville Syrjälä <syrjala@sci.fi> wrote:
> > > On Fri, Jan 13, 2006 at 10:21:44AM +0100, Bartlomiej Zolnierkiewicz wrote:
> > > > On 1/13/06, Volker Kuhlmann <list0570@paradise.net.nz> wrote:
> > > > > On Fri 13 Jan 2006 17:40:33 NZDT +1300, Robert Hancock wrote:
> > > > >
> > > > > > I'm thinking the IDE code is too aggressive in assuming that the failure
> > > > > >  is because of a DMA problem and disabling it.. Most likely all that's
> > > > > > happening is the drive is taking a long time to complete the current
> > > > > > command.
> > > >
> > > > What actually happened is that normal command timed out
> > > > and because of that driver reset the device which caused
> > > > it to loose DMA:
> > > >
> > > > ->ide_atapi_error()
> > > >     ->ide_do_reset()
> > > >       ->pre_reset()
> > > >         ->check_dma_crc()
> > > >           ->__ide_dma_off()
> > > >
> > > > Somebody needs to investigate why __ide_dma_off() is called
> > > > et all and if we need to restore DMA after reset (don't count ATM
> > > > on me, I'm buried by bugreports).  Ondrej, could you fill the bug at
> > > > http://bugzilla.kernel.org so we don't lose it?
> > >
> > > It looks like this has been fixed recently. In 2.6.14-rc4-mm1
> > > check_dma_crc() used to call __ide_dma_off() when no CRC errors were
> > > registered. The bug has been fixed before 2.6.15-rc5-mm1. I'm still
> > > running 2.6.14-rc4-mm1 on my burning system so I haven't actually tested
> > > this on a recent kernel.
> >
> > I don't see it being fixed, could you point me at the patch?
>
> It seems the patch was carried along during the 2.6.15-rc-mm series but
> was later dropped.
>
> Here it is:
> http://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.15-rc5/2.6.15-rc5-mm3/broken-out/ide-promise-flushing-hang-fix.patch
>
> There's a bunch of other stuff in that patch besides the __ide_dma_off()
> change. Maybe that approach isn't correct and just increasing some
> timeout value would work. LG burners are just really slow recognizing
> discs.

The patch was NACK-ed by Alan Cox and I agree with him (this should be
done differently).  This __ide_dma_off() chunk looks sensible but does it fix
the issue?  I was under impression that after a reset drive looses its DMA
xfer mode and needs to be reprogrammed (ATA spec has the answer).

> I just took some random discs from my colletion and measured the time it
> took for the drive to recognize them.
>
> ~12 seconds for CD-R
> ~15 seconds for some DVD+R
> ~27 seconds for some DVD+R
>
> I'm not sure why it takes so long for some discs. I know they aren't
> bad burns because my laptop DVD-ROM drive, which is more sensitive to
> disc quality than the LG, can read them without any problems and it
> recognizes them in a few seconds.

I'll let Jens comment on this one. :)

Bartlomiej

Re: ide-cd turning off DMA when verifying DVD-R

[Rashkae]

On Fri, Jan 13, 2006 at 10:19:17AM +0100, Bartlomiej Zolnierkiewicz wrote:


> On 1/13/06, Volker Kuhlmann <list0570@paradise.net.nz> wrote:
> > On Fri 13 Jan 2006 17:40:33 NZDT +1300, Robert Hancock wrote:
> >
> > > I'm thinking the IDE code is too aggressive in assuming that the failure
> > >  is because of a DMA problem and disabling it.. Most likely all that's
> > > happening is the drive is taking a long time to complete the current
> > > command.
> 
> What actually happened is that normal command timed out
> and because of that driver reset the device which caused
> it to loose DMA:
> 
> ->ide_atapi_error()
>     ->ide_do_reset()
>       ->pre_reset()
>         ->check_dma_crc()
>           ->__ide_dma_off()
> 
> Somebody needs to investigate why __ide_dma_off() is called
> et all and if we need to restore DMA after reset (don't count ATM
> on me, I'm buried by bugreports).  Ondrej, could you fill the bug at
> http://bugzilla.kernel.org so we don't lose it?

Shouldn't whether DMA is restored after reset be governed by keepsettings flag?

From the hdparm man page:

      -k     Get/set  the  keep_settings_over_reset flag for the
              drive.  When this flag is set, the driver will pre­
              serve  the -dmu options over a soft reset, (as done
              during the error  recovery  sequence).

I was a little disapointed that this didn't work as advertised.

Re: ide-cd turning off DMA when verifying DVD-R

[Bartlomiej Zolnierkiewicz]

On 1/13/06, Rashkae <rashkae@tigershaunt.com> wrote:
> On Fri, Jan 13, 2006 at 10:19:17AM +0100, Bartlomiej Zolnierkiewicz wrote:
>
>
> > On 1/13/06, Volker Kuhlmann <list0570@paradise.net.nz> wrote:
> > > On Fri 13 Jan 2006 17:40:33 NZDT +1300, Robert Hancock wrote:
> > >
> > > > I'm thinking the IDE code is too aggressive in assuming that the failure
> > > >  is because of a DMA problem and disabling it.. Most likely all that's
> > > > happening is the drive is taking a long time to complete the current
> > > > command.
> >
> > What actually happened is that normal command timed out
> > and because of that driver reset the device which caused
> > it to loose DMA:
> >
> > ->ide_atapi_error()
> >     ->ide_do_reset()
> >       ->pre_reset()
> >         ->check_dma_crc()
> >           ->__ide_dma_off()
> >
> > Somebody needs to investigate why __ide_dma_off() is called
> > et all and if we need to restore DMA after reset (don't count ATM
> > on me, I'm buried by bugreports).  Ondrej, could you fill the bug at
> > http://bugzilla.kernel.org so we don't lose it?
>
> Shouldn't whether DMA is restored after reset be governed by keepsettings flag?

settings should always be preserved and keepsettings flag should die

> From the hdparm man page:
>
>       -k     Get/set  the  keep_settings_over_reset flag for the
>               drive.  When this flag is set, the driver will pre­
>               serve  the -dmu options over a soft reset, (as done
>               during the error  recovery  sequence).
>
> I was a little disapointed that this didn't work as advertised.

AFAIR (2.5/2.6 and probably 2.4 also) it never worked w.r.t. to DMA
as documented in hdparm manual.

Bartlomiej

Re: Fwd: ide-cd turning off DMA when verifying DVD-R

[Alan Cox]

On Gwe, 2006-01-13 at 14:33 +0100, Bartlomiej Zolnierkiewicz wrote:
> The patch was NACK-ed by Alan Cox and I agree with him (this should be
> done differently).  This __ide_dma_off() chunk looks sensible but does it fix
> the issue?  I was under impression that after a reset drive looses its DMA
> xfer mode and needs to be reprogrammed (ATA spec has the answer).

Yes and the behaviour is determined by hdparm -k/-K. Its all a bit
random after a CRC error however as the recovery code does a polling
speed change and the locking versus timers/irq's is totally broken.

I did some minimal fixes in the older -ac patches but even with all the
other locking cleanup I did back then this one remained. Essentially the
old IDE layer needs to switch from polled speed changes in task context
to issuing speed changes as state machine sequences.

Thats non trivial

Alan

Re: Fwd: ide-cd turning off DMA when verifying DVD-R

[Rashkae]

On Fri, Jan 13, 2006 at 02:24:25PM +0000, Alan Cox wrote:

> On Gwe, 2006-01-13 at 14:33 +0100, Bartlomiej Zolnierkiewicz wrote:
> > The patch was NACK-ed by Alan Cox and I agree with him (this should be
> > done differently).  This __ide_dma_off() chunk looks sensible but does it fix
> > the issue?  I was under impression that after a reset drive looses its DMA
> > xfer mode and needs to be reprogrammed (ATA spec has the answer).
> 
> Yes and the behaviour is determined by hdparm -k/-K. Its all a bit

Great theory, but unfortunately, as things are now, -k does not preserve DMA setting, contrary to it's documented function.

Quote from Bartlomiej Zolnierkiewicz:

AFAIR (2.5/2.6 and probably 2.4 also) it never worked w.r.t. to DMA
as documented in hdparm manual.