[PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[James Ausmus]

Some IDE -> Compact Flash media adapters are not capable of supporting
UDMA, which can cause very slow boot times when the CF media itself
reports as capable of UDMA transfer speeds. Create Kconfig option to
turn off the UDMA capability bit when media is identified as Compact
Flash.

Signed-off-by: James Ausmus <james.ausmus@gmail.com>

---

diff -uprN -X linux-2.6.16.5/Documentation/dontdiff
linux-2.6.16.5.orig/drivers/ide/Kconfig
linux-2.6.16.5/drivers/ide/Kconfig
--- linux-2.6.16.5.orig/drivers/ide/Kconfig     2006-04-17
13:37:47.000000000 -0700
+++ linux-2.6.16.5/drivers/ide/Kconfig  2006-04-17 13:43:37.000000000 -0700
@@ -447,6 +447,18 @@ config IDEDMA_ONLYDISK

          Generally say N here.

+config IDE_COMPACT_FLASH_NO_UDMA
+       bool "Disable UDMA for Compact Flash Devices"
+       help
+         This turns off the UDMA capability flag bit for any IDE device
+         that identifies as Compact Flash - This is a workaround for
+         cheap Compact Flash -> IDE adapters that don't support UDMA
+         transfer speeds even if the CF card does.
+
+         Enable this if you are using a Compact Flash card as an IDE
+         device and you see re-occuring 0x41 DMA Timeout errors from
+         your IDE driver.
+
 config BLK_DEV_AEC62XX
        tristate "AEC62XX chipset support"
        help
diff -uprN -X linux-2.6.16.5/Documentation/dontdiff
linux-2.6.16.5.orig/drivers/ide/ide-probe.c
linux-2.6.16.5/drivers/ide/ide-probe.c
--- linux-2.6.16.5.orig/drivers/ide/ide-probe.c 2006-04-17
13:37:48.000000000 -0700
+++ linux-2.6.16.5/drivers/ide/ide-probe.c      2006-04-17
13:44:45.000000000 -0700
@@ -248,6 +248,11 @@ static inline void do_identify (ide_driv
        if ((id->config != 0x848a) && (id->config & (1<<7)))
                drive->removable = 1;

+#if defined(CONFIG_IDE_COMPACT_FLASH_NO_UDMA)
+       if (id->config == 0x848a)
+               id->capability &= ~1;
+#endif
+
        drive->media = ide_disk;
        printk("%s DISK drive\n", (id->config == 0x848a) ? "CFA" : "ATA" );
        QUIRK_LIST(drive);

Re: [PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[Mark Lord]

James Ausmus wrote:
>..
> +config IDE_COMPACT_FLASH_NO_UDMA
> +       bool "Disable UDMA for Compact Flash Devices"
> +       help
> +         This turns off the UDMA capability flag bit for any IDE device
> +         that identifies as Compact Flash - This is a workaround for
> +         cheap Compact Flash -> IDE adapters that don't support UDMA
> +         transfer speeds even if the CF card does.

I dunno.  This might be better done as a module/boot option
than as a rebuild-your-kernel configuration choice.

??

Re: [PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[Alan Cox]

On Llu, 2006-04-17 at 16:11 -0700, James Ausmus wrote:
> Some IDE -> Compact Flash media adapters are not capable of supporting
> UDMA, which can cause very slow boot times when the CF media itself
> reports as capable of UDMA transfer speeds. Create Kconfig option to
> turn off the UDMA capability bit when media is identified as Compact
> Flash.

This would be far better if it was a boot time option than a kernel
configuration or set using sysfs/procfs controls as users will not want
to recompile their kernel in such cases.

Re: [PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[Andreas Mohr]

Hi,

On Tue, Apr 18, 2006 at 12:14:18PM +0100, Alan Cox wrote:
> On Llu, 2006-04-17 at 16:11 -0700, James Ausmus wrote:
> > Some IDE -> Compact Flash media adapters are not capable of supporting
> > UDMA, which can cause very slow boot times when the CF media itself
> > reports as capable of UDMA transfer speeds. Create Kconfig option to
> > turn off the UDMA capability bit when media is identified as Compact
> > Flash.
> 
> This would be far better if it was a boot time option than a kernel
> configuration or set using sysfs/procfs controls as users will not want
> to recompile their kernel in such cases.

Count me in.
However, while this is much better than a compile-time setting, it's still
not fully satisfying since many users won't realize that they're hitting this
problem and thus won't search for and find this obscure boot parameter.
Is there any way at all to get this condition detected automatically?
I think one should try hard to find a way to detect it, if at all possible.

Andreas Mohr

Re: [PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[Alan Cox]

On Maw, 2006-04-18 at 13:34 +0200, Andreas Mohr wrote:
> However, while this is much better than a compile-time setting, it's still
> not fully satisfying since many users won't realize that they're hitting this
> problem and thus won't search for and find this obscure boot parameter.
> Is there any way at all to get this condition detected automatically?

Not that I can think off immediately. The controller and the drive both
report the modes they support. If that is wrong then we either need to
be able to identify the specific device (as libata does with the Palmax
systems) or try it and see (which we indeed do but takes time to error
out).

For PCMCIA CF adapters we are ok because they are ISA bus so PIO 0
cycles are all that are supported. For other controllers it will depend
whether the CF adapter is integrated into a PCI card with unique
subvendor/dev identifiers which can be blacklisted, or a motherboard
with DMI entries that can be used.

If it's just some poorly engineered 'shove a cable in one end and a CF
card the other' device which is therefore not directly detectable I
think you lose.

Alan

Re: [PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[James Ausmus]

On 4/18/06, Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
> On Maw, 2006-04-18 at 13:34 +0200, Andreas Mohr wrote:
> > However, while this is much better than a compile-time setting, it's still
> > not fully satisfying since many users won't realize that they're hitting this
> > problem and thus won't search for and find this obscure boot parameter.
> > Is there any way at all to get this condition detected automatically?
>
> Not that I can think off immediately. The controller and the drive both
> report the modes they support. If that is wrong then we either need to
> be able to identify the specific device (as libata does with the Palmax
> systems) or try it and see (which we indeed do but takes time to error
> out).
>
> For PCMCIA CF adapters we are ok because they are ISA bus so PIO 0
> cycles are all that are supported. For other controllers it will depend
> whether the CF adapter is integrated into a PCI card with unique
> subvendor/dev identifiers which can be blacklisted, or a motherboard
> with DMI entries that can be used.
>
> If it's just some poorly engineered 'shove a cable in one end and a CF
> card the other' device which is therefore not directly detectable I
> think you lose.
>

This is exactly the situation that I have with 2 separate "dumb" (just
physical interfaces, essentially - not at all detectable) IDE->CF
adapters - both the IDE controller and the CF media support several
UDMA modes, so the IDE driver throws the CF device into UDMA mode on
bootup. However, as the physical interface between the IDE cable and
the CF socket is poorly engineered, it cannot handle the higher
speeds, causing the timeout errors. For some people, this can just be
fixed with an ide=nodma boot option, but as I also have a (quite
large) rotating media device on the controller, this is not an option,
as, if a fsck is performed on a boot, the boot-up time is upwards of
30 minutes.

I would be happy to look into making this a boot option - but my
project schedule won't allow for the kind of time for me to figure out
how to do that :), so this is just a quick hack that works for my
situation.

-James



> Alan
>
>

Re: [PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[Robert Hancock]

James Ausmus wrote:
> This is exactly the situation that I have with 2 separate "dumb" (just
> physical interfaces, essentially - not at all detectable) IDE->CF
> adapters - both the IDE controller and the CF media support several
> UDMA modes, so the IDE driver throws the CF device into UDMA mode on
> bootup. However, as the physical interface between the IDE cable and
> the CF socket is poorly engineered, it cannot handle the higher
> speeds, causing the timeout errors. For some people, this can just be
> fixed with an ide=nodma boot option, but as I also have a (quite
> large) rotating media device on the controller, this is not an option,
> as, if a fsck is performed on a boot, the boot-up time is upwards of
> 30 minutes.

If it's like some of the CF-IDE adapters I've seen, the DMA request/ack 
lines likely aren't even wired up between the card and the cable. 
There's no way the kernel can detect that DMA is not actually possible 
on such a card without trying and waiting for it to timeout. (I've seen 
a few which have jumpers which select whether to connect these or not - 
haven't a clue why you wouldn't want to hook those up unconditionally..)

Isn't there an option to disable DMA for a specific IDE channel 
(ide2=nodma or something like that)?

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/

Re: [PATCH 1/1] ide: Allow disabling of UDMA for Compact Flash devices

[James Ausmus]

On 4/18/06, Robert Hancock <hancockr@shaw.ca> wrote:
> James Ausmus wrote:
> > This is exactly the situation that I have with 2 separate "dumb" (just
> > physical interfaces, essentially - not at all detectable) IDE->CF
> > adapters - both the IDE controller and the CF media support several
> > UDMA modes, so the IDE driver throws the CF device into UDMA mode on
> > bootup. However, as the physical interface between the IDE cable and
> > the CF socket is poorly engineered, it cannot handle the higher
> > speeds, causing the timeout errors. For some people, this can just be
> > fixed with an ide=nodma boot option, but as I also have a (quite
> > large) rotating media device on the controller, this is not an option,
> > as, if a fsck is performed on a boot, the boot-up time is upwards of
> > 30 minutes.
>
> If it's like some of the CF-IDE adapters I've seen, the DMA request/ack
> lines likely aren't even wired up between the card and the cable.
> There's no way the kernel can detect that DMA is not actually possible
> on such a card without trying and waiting for it to timeout. (I've seen
> a few which have jumpers which select whether to connect these or not -
> haven't a clue why you wouldn't want to hook those up unconditionally..)
>
> Isn't there an option to disable DMA for a specific IDE channel
> (ide2=nodma or something like that)?
>

Yes, there is, but that unfortunately does not solve the issue for
configurations where there is also a UDMA capable device on the same
channel. If there was something like ide2=0:nodma,1:dma or some such,
in order to enable/disable dma on a per device basis, instead of a
per-channel basis, then that would also solve the issue.

-James


> --
> Robert Hancock      Saskatoon, SK, Canada
> To email, remove "nospam" from hancockr@nospamshaw.ca
> Home Page: http://www.roberthancock.com/
>
>