Limiting DMA speeds for individual IDE drives

Limiting DMA speeds for individual IDE drives

[Alan Stern]

Is there any simple way to force the old IDE driver to limit the DMA
speed for a particular device?

I've got a situation where a drive claims to be capable of supporting
UDMA/100, but it's in a noisy environment and gets lots of errors at
that speed.  I'd like to limit it to UDMA/66 or even UDMA/33.

The hdparm command should be able to do this but I can't run it until 
the system has booted, by which time a bunch of CRC and possibly other 
errors have already occurred.  Ideally it should be possible to limit 
the speed starting as early as device detection, but I can't find any 
way to do it.  Is there support for such a thing or will I have to hack 
it in?

Thanks,

Alan Stern

Re: Limiting DMA speeds for individual IDE drives

[Frederik Deweerdt]

On Tue, Sep 08, 2009 at 12:03:48PM -0400, Alan Stern wrote:
> Is there any simple way to force the old IDE driver to limit the DMA
> speed for a particular device?
> 
> I've got a situation where a drive claims to be capable of supporting
> UDMA/100, but it's in a noisy environment and gets lots of errors at
> that speed.  I'd like to limit it to UDMA/66 or even UDMA/33.
> 
> The hdparm command should be able to do this but I can't run it until 
> the system has booted, by which time a bunch of CRC and possibly other 
> errors have already occurred.  Ideally it should be possible to limit 
> the speed starting as early as device detection, but I can't find any 
> way to do it.  Is there support for such a thing or will I have to hack 
> it in?
Does passing ide=nodma at bootime, and then having init set the DMA at
the right speed, would work?

Regards,
Frederik

Re: Limiting DMA speeds for individual IDE drives

[Alan Cox]

> I've got a situation where a drive claims to be capable of supporting
> UDMA/100, but it's in a noisy environment and gets lots of errors at
> that speed.  I'd like to limit it to UDMA/66 or even UDMA/33.

That should never occur with a proper cable and I would be concerned the
fault might be something more problematic such as speed misconfiguration
or an incompatibility. Which driver is in use ?

> The hdparm command should be able to do this but I can't run it until 
> the system has booted, by which time a bunch of CRC and possibly other 
> errors have already occurred.  Ideally it should be possible to limit 

Only the data transfers are CRC protected and at high speed, but noise at
low speed would be a real concern as the commands are sent low speed but
without protection on PATA devices - so a bit flip can send a DMA to the
wrong sector.

> the speed starting as early as device detection, but I can't find any 
> way to do it.  Is there support for such a thing or will I have to hack 
> it in?

You can disallow DMA but not clip DMA to UDMA33 with the old driver. You
could disallow DMA at boot and reallow it with a speed set by hdparm in
your boot scripts...

Re: Limiting DMA speeds for individual IDE drives

[Sergei Shtylyov]

Hello.

Frederik Deweerdt wrote:

>>Is there any simple way to force the old IDE driver to limit the DMA
>>speed for a particular device?

    No, you can only disable DMA totally, and then set the needed speed via 
hdparm.

>>I've got a situation where a drive claims to be capable of supporting
>>UDMA/100, but it's in a noisy environment and gets lots of errors at
>>that speed.  I'd like to limit it to UDMA/66 or even UDMA/33.

    Are you sure that it's all because of the noise and not a cable type 
misdetection?

>>The hdparm command should be able to do this but I can't run it until 
>>the system has booted, by which time a bunch of CRC and possibly other 
>>errors have already occurred.  Ideally it should be possible to limit 
>>the speed starting as early as device detection, but I can't find any 
>>way to do it.  Is there support for such a thing or will I have to hack 
>>it in?

> Does passing ide=nodma at bootime, and then having init set the DMA at
> the right speed, would work?

    ide=nodma is now obsolete -- use ide_core.nodma=<interface>,<device> 
instead. Read Documentation/ide/ide.txt before advising. ;-)

> Regards,
> Frederik

WBR, Sergei

Re: Limiting DMA speeds for individual IDE drives

[Alan Stern]

On Tue, 8 Sep 2009, Alan Cox wrote:

> > I've got a situation where a drive claims to be capable of supporting
> > UDMA/100, but it's in a noisy environment and gets lots of errors at
> > that speed.  I'd like to limit it to UDMA/66 or even UDMA/33.
> 
> That should never occur with a proper cable and I would be concerned the
> fault might be something more problematic such as speed misconfiguration
> or an incompatibility. Which driver is in use ?

The cable indeed is likely to be at fault.  The same drive worked okay
at the higher speed with a different cable (which unfortunately is
unavailable for use in the final deployment).  This is using the old
IDE driver.  Here's an extract from the log, with
ide-core.ignore_cable=0 specified on the command line:

Linux version 2.6.27-gentoo-r10 (root@raise) (gcc version 4.1.2 (Gentoo 4.1.2 p1.0.2)) #1 SMP PREEMPT Tue Apr 21 15:06:03 UTC 2009
...
Uniform Multi-Platform E-IDE driver
piix 0000:00:1f.1: IDE controller (0x8086:0x24cb rev 0x02)
pci 0000:00:1f.1: PCI INT A -> GSI 18 (level, low) -> IRQ 18
piix 0000:00:1f.1: IDE port disabled
piix 0000:00:1f.1: not 100% native mode: will probe irqs later
ide: ignoring cable detection for ide0
    ide0: BM-DMA at 0xf000-0xf007
Probing IDE interface ide0...
hda: STEC MACH-8 SSD, ATA DISK drive
hda: host max PIO4 wanted PIO255(auto-tune) selected PIO4
hda: UDMA/100 mode selected
ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
ide_generic: please use "probe_mask=0x3f" module parameter for probing all legacy ISA IDE ports
ide_generic: I/O resource 0x1F0-0x1F7 not free.
ide_generic: I/O resource 0x170-0x177 not free.
hda: max request size: 512KiB
hda: 60789456 sectors (31124 MB), CHS=16383/255/63
hda: cache flushes not supported
 hda: hda1 hda2 hda3
...
hda: dma_intr: status=0x51 { DriveReady SeekComplete Error }
hda: dma_intr: error=0xc4 { DriveStatusError BadCRC UncorrectableError }, LBAsect=14823116, sector=14823116
ide: failed opcode was: unknown
hda: dma_intr: status=0x51 { DriveReady SeekComplete Error }
hda: dma_intr: error=0xc4 { DriveStatusError BadCRC UncorrectableError }, LBAsect=15133492, sector=15133492
ide: failed opcode was: unknown
hda: dma_intr: status=0x51 { DriveReady SeekComplete Error }
hda: dma_intr: error=0xc4 { DriveStatusError BadCRC UncorrectableError }, LBAsect=9478100, sector=9478100
ide: failed opcode was: unknown
Etc.; you get the idea...


> > The hdparm command should be able to do this but I can't run it until 
> > the system has booted, by which time a bunch of CRC and possibly other 
> > errors have already occurred.  Ideally it should be possible to limit 
> 
> Only the data transfers are CRC protected and at high speed, but noise at
> low speed would be a real concern as the commands are sent low speed but
> without protection on PATA devices - so a bit flip can send a DMA to the
> wrong sector.
> 
> > the speed starting as early as device detection, but I can't find any 
> > way to do it.  Is there support for such a thing or will I have to hack 
> > it in?
> 
> You can disallow DMA but not clip DMA to UDMA33 with the old driver. You
> could disallow DMA at boot and reallow it with a speed set by hdparm in
> your boot scripts...


On Tue, 8 Sep 2009, Frederik Deweerdt wrote:

> Does passing ide=nodma at bootime, and then having init set the DMA at
> the right speed, would work?

I'll recommend trying that out.  Thanks to both of you for the advice.

Alan Stern