I sent the following followup report to the bug I filed at
bugme.osdl.org several months ago about my hpt372 ide controller being
slow.
I have determined what causes the dramatic slowdown problem. It is not
drive specific but it may be specific to the hpt controllers. The
problem is due to using automatic dma. If I don't have the following
two options set in my kernel then it runs at full speed when I turn on
dma with hdparm. Otherwise using automatic dma I get somewhere between
a 50% - 700% slowdown on writes, reads seem to not be as badly affected.
This is reproducible on both 2.4.23-rc1-xfs and 2.6.0-test9-bk24 (the
two I tested on).
CONFIG_IDEDMA_PCI_AUTO=3Dy=20
CONFIG_IDEDMA_AUTO=3Dy=20
Chris Cheney
Chris, please post output of 'lspci -vvv -xxx' - it will be very useful.
We can see what registers are programmed differently when autodma is off.
--bart
On Saturday 22 of November 2003 21:48, Chris Cheney wrote:
> I sent the following followup report to the bug I filed at
> bugme.osdl.org several months ago about my hpt372 ide controller being
> slow.
>
> I have determined what causes the dramatic slowdown problem. It is not
> drive specific but it may be specific to the hpt controllers. The
> problem is due to using automatic dma. If I don't have the following
> two options set in my kernel then it runs at full speed when I turn on
> dma with hdparm. Otherwise using automatic dma I get somewhere between
> a 50% - 700% slowdown on writes, reads seem to not be as badly affected.
> This is reproducible on both 2.4.23-rc1-xfs and 2.6.0-test9-bk24 (the
> two I tested on).
>
> CONFIG_IDEDMA_PCI_AUTO=3Dy=20
> CONFIG_IDEDMA_AUTO=3Dy=20
On Sat, Nov 22, 2003 at 11:20:18PM +0100, Bartlomiej Zolnierkiewicz wrote:
>
> Chris, please post output of 'lspci -vvv -xxx' - it will be very useful.
> We can see what registers are programmed differently when autodma is off.
>
> --bart
Ok, I have attached the lspci output. This is the output for when
autodma is off and hdparm -c3 -d1 -u1 has been run.
Chris
On Sunday 23 of November 2003 02:35, Chris Cheney wrote:
> On Sat, Nov 22, 2003 at 11:20:18PM +0100, Bartlomiej Zolnierkiewicz wrote:
> > Chris, please post output of 'lspci -vvv -xxx' - it will be very useful.
> > We can see what registers are programmed differently when autodma is off.
> >
> > --bart
>
> Ok, I have attached the lspci output. This is the output for when
> autodma is off and hdparm -c3 -d1 -u1 has been run.
Thanks, I need also output of lspci when autodma is on.
--bart
On Sun, Nov 23, 2003 at 07:51:06PM +0100, Bartlomiej Zolnierkiewicz wrote:
> On Sunday 23 of November 2003 02:35, Chris Cheney wrote:
> > On Sat, Nov 22, 2003 at 11:20:18PM +0100, Bartlomiej Zolnierkiewicz wrote:
> > > Chris, please post output of 'lspci -vvv -xxx' - it will be very useful.
> > > We can see what registers are programmed differently when autodma is off.
> > >
> > > --bart
> >
> > Ok, I have attached the lspci output. This is the output for when
> > autodma is off and hdparm -c3 -d1 -u1 has been run.
>
> Thanks, I need also output of lspci when autodma is on.
>
> --bart
I have attached the lspci output from the kernel with autodma on. By the
way do I need to have cold booted the machine for both tests? It seems
to give slightly different results for the non autodma kernel today than
what I sent to you before.
Chris
On Mon, Nov 24, 2003 at 02:18:27PM +0100, Bartlomiej Zolnierkiewicz wrote:
>
> Yes, you should cold boot before doing tests.
Ok, here is the updated lspci output.
Chris