I've been busy and have been a bit late responding to this. In short,
have a problem with disk performance on a HP Pavilion laptop in the
2.6.1[34] kernels (it works fine under 2.6.12.x). I've summarized much
of the tests I've run on in a post to linux-kernel and linux-ide (when
we thought it was a problem in the ATIIXP module) at:
http://marc.theaimsgroup.com/?l=linux-kernel&m=112802950411697&w=2
After working with Stefan for a bit, we determined that the problem is
in the ACPI modules (in fact setting ACPI=ht at boot fixes it, although
other things seem to break, as described below). Stefan suggested I
repost the bug to the linux-kernel list, and so here it is.
David
On Fri, 2005-11-04 at 00:24 +0100, Stefan wrote:
> David Ronis wrote:
>
> >Hi Stefan,
> >
> >Did you receive my e-mail with the results of your suggestion?
> >
> >
> Ah sometimes yahoo marks some mails from the list as spam.
>
> Anyways, I just looked through the changelog from 2.6.12 to 2.6.13, lots
> of acpi changes.
> It looks like the problem is within the acpi code.
>
> Just send another mail to linux-kernel kernel list instead of linux-ide
> and tell them about your problem.
>
> I'd send along the hdparm tests 'hdparm /dev/hda' + 'hdparm -i /dev/hda'
> with both kernels and the 'dmesg' output of both + 'lspci -vv'
>
> To my believe it is unlikely, that any changes in the drivers for ide or
> atiixp are responsible for your performance loss.
>
> >I've been playing around a bit more; specifically, I tried booting with
> >apci=ht (without noapic) as was suggested by someone from my earlier
> >post on the linux-kernel list. This alone seems to fix the problem (as
> >did your suggestion). Now however, X doesn't work because
> >
> >"The /dev/input/event* device nodes seem to be missing"
> >
> >I don't know if your suggestion gives the same problem, but my bet is
> >that it does.
> >
> >Anyhow, if you didn't receive my last e-mail, I'll resend it (it has the
> >output of dmesg as well as some timings).
> >
> >Thanks again.
> >
> >David
> >
> >
> >On Tue, 2005-11-01 at 11:05 +0100, Stefan wrote:
> >
> >
> >>Lets try something different,
> >>
> >>boot your machine with those arguments:
> >>
> >>noapic acpi=off
> >>
> >>If you use grub, just press e to edit the boot command and add these
> >>flags to the existing line which boots your kernel.
> >>Lets see if this makes any performance difference
> >>
> >>
> >>
> >>
> >>
> >>___________________________________________________________
> >>Gesendet von Yahoo! Mail - Jetzt mit 1GB Speicher kostenlos - Hier anmelden: http://mail.yahoo.de
> >>
> >>
> >>
> >
> >
> >
>
>
>
>
>
>
> ___________________________________________________________
> Gesendet von Yahoo! Mail - Jetzt mit 1GB Speicher kostenlos - Hier anmelden: http://mail.yahoo.de
>
David Ronis <[email protected]> wrote:
>
> I've been busy and have been a bit late responding to this. In short,
> have a problem with disk performance on a HP Pavilion laptop in the
> 2.6.1[34] kernels (it works fine under 2.6.12.x). I've summarized much
> of the tests I've run on in a post to linux-kernel and linux-ide (when
> we thought it was a problem in the ATIIXP module) at:
>
> http://marc.theaimsgroup.com/?l=linux-kernel&m=112802950411697&w=2
>
> After working with Stefan for a bit, we determined that the problem is
> in the ACPI modules (in fact setting ACPI=ht at boot fixes it, although
> other things seem to break, as described below). Stefan suggested I
> repost the bug to the linux-kernel list, and so here it is.
The acpi team are very bugzilla-oriented. Could you please open a bug
report for this, against acpi at bugzilla.kernel.org?
Thanks.
David Ronis wrote:
> I've been busy and have been a bit late responding to this. In short,
> have a problem with disk performance on a HP Pavilion laptop in the
> 2.6.1[34] kernels (it works fine under 2.6.12.x). I've summarized much
> of the tests I've run on in a post to linux-kernel and linux-ide (when
> we thought it was a problem in the ATIIXP module) at:
>
> http://marc.theaimsgroup.com/?l=linux-kernel&m=112802950411697&w=2
>
> After working with Stefan for a bit, we determined that the problem is
> in the ACPI modules (in fact setting ACPI=ht at boot fixes it, although
> other things seem to break, as described below). Stefan suggested I
> repost the bug to the linux-kernel list, and so here it is.
Is this read performance, write performance, or both? And have you
checked the output in dmesg? See anything with blockdev? Detailed lspci
output and /proc/interrupts?
I don't have anything in mand, just looking for information which might
in some universe get diddled by ACPI.
--
-bill davidsen ([email protected])
"The secret to procrastination is to put things off until the
last possible moment - but no longer" -me
Thanks for the reply. As per Andrew Morton's suggestion, I've opened a
bug at bugzilla.kernel.org
(http://bugzilla.kernel.org/show_bug.cgi?id=5594), it has all the
relevant information (basically copies of what I've posted on
linux-kernel and linux-ide). So far it's still sitting there as "NEW".
David
On Mon, 2005-11-14 at 16:17 -0500, Bill Davidsen wrote:
> David Ronis wrote:
> > I've been busy and have been a bit late responding to this. In short,
> > have a problem with disk performance on a HP Pavilion laptop in the
> > 2.6.1[34] kernels (it works fine under 2.6.12.x). I've summarized much
> > of the tests I've run on in a post to linux-kernel and linux-ide (when
> > we thought it was a problem in the ATIIXP module) at:
> >
> > http://marc.theaimsgroup.com/?l=linux-kernel&m=112802950411697&w=2
> >
> > After working with Stefan for a bit, we determined that the problem is
> > in the ACPI modules (in fact setting ACPI=ht at boot fixes it, although
> > other things seem to break, as described below). Stefan suggested I
> > repost the bug to the linux-kernel list, and so here it is.
>
> Is this read performance, write performance, or both? And have you
> checked the output in dmesg? See anything with blockdev? Detailed lspci
> output and /proc/interrupts?
>
> I don't have anything in mand, just looking for information which might
> in some universe get diddled by ACPI.