Regression: starting X locks box with git master

Regression: starting X locks box with git master

[James Cloos]

3dc5063786b273f1aee545844f6bd4e9651ebffe was current when I compiled.

With that kernel, my laptop locks up as soon as I startx.

The Xorg.0.log showed that the radeon driver got as far as starting
the i2c calls to the M7 video card.

After that the box would reply to ICMP pings and SysRq-B, but was
otherwise non-responsive.  Even SysRq-[KSU] seemed to do nothing.

v2.6.25 works fine.

There are 3562 commits (including merges) between those two, and
it takes an hour or more for each kernel compile.  Bisecting will
be quite painful.  And I only have the one box to work on.  But I
will see what I can do.

Posting before bisecting in case it helps anyone else....

-JimC
-- 
James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6

Re: Regression: starting X locks box with git master

[Justin Mattock]

On Thu, Apr 24, 2008 at 7:59 PM, James Cloos <cloos@jhcloos.com> wrote:
> 3dc5063786b273f1aee545844f6bd4e9651ebffe was current when I compiled.
>
>  With that kernel, my laptop locks up as soon as I startx.
>
>  The Xorg.0.log showed that the radeon driver got as far as starting
>  the i2c calls to the M7 video card.
>
>  After that the box would reply to ICMP pings and SysRq-B, but was
>  otherwise non-responsive.  Even SysRq-[KSU] seemed to do nothing.
>
>  v2.6.25 works fine.
>
>  There are 3562 commits (including merges) between those two, and
>  it takes an hour or more for each kernel compile.  Bisecting will
>  be quite painful.  And I only have the one box to work on.  But I
>  will see what I can do.
>
>  Posting before bisecting in case it helps anyone else....
>
>  -JimC
>  --
>  James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6
>
>
>  --
>  To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>  the body of a message to majordomo@vger.kernel.org
>  More majordomo info at  http://vger.kernel.org/majordomo-info.html
>  Please read the FAQ at  http://www.tux.org/lkml/
>

I seem to expiriencing lockups as well, vary annoying, "I'm getting
freezes before X is even started", "and after as well." : (
 using nmi_watchdog=1, or nohz=off, or nolapic_timer, or idle=poll
seems to help me with having no freezes, but using powertop
I'm seeing (100%) under AVG residency, which is a bit alarming to me.
If I dont give any boot options and leave my system as it was, I need
to boot and as fast as possible use glxgears to keep the processor at
a certain level, from freezing, but is hard to do.
overall I think this has to do with nohz, right now I have nohz=off
with highress on and the system seems to be behaving properly.
regards;

-- 
Justin P. Mattock

Re: Regression: starting X locks box with git master

[Yinghai Lu]

On Thu, Apr 24, 2008 at 12:59 PM, James Cloos <cloos@jhcloos.com> wrote:
> 3dc5063786b273f1aee545844f6bd4e9651ebffe was current when I compiled.
>
>  With that kernel, my laptop locks up as soon as I startx.
>
>  The Xorg.0.log showed that the radeon driver got as far as starting
>  the i2c calls to the M7 video card.
>
>  After that the box would reply to ICMP pings and SysRq-B, but was
>  otherwise non-responsive.  Even SysRq-[KSU] seemed to do nothing.
>
>  v2.6.25 works fine.
>
>  There are 3562 commits (including merges) between those two, and
>  it takes an hour or more for each kernel compile.  Bisecting will
>  be quite painful.  And I only have the one box to work on.  But I
>  will see what I can do.
>
>  Posting before bisecting in case it helps anyone else....

Do you have CONFIG_X86_PAT in your .config?

YH

Re: Regression: starting X locks box with git master

[James Cloos]

>>>>> "Justin" == Justin Mattock <justinmattock@gmail.com> writes:

Justin> I seem to expiriencing lockups as well, vary annoying, ...
Justin> overall I think this has to do with nohz, right now I have
Justin> nohz=off with highress on and the system seems to be behaving
Justin> properly.

I hadn't thought of trying with nohz=off.

I do use nohz and (non-smp) preempt.

I'll have to give nohz=off a test before bisecting.

Thanks for the note.

-JimC
-- 
James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6

Re: Regression: starting X locks box with git master

[Justin Mattock]

On Thu, Apr 24, 2008 at 8:52 PM, Yinghai Lu <yhlu.kernel@gmail.com> wrote:
> On Thu, Apr 24, 2008 at 12:59 PM, James Cloos <cloos@jhcloos.com> wrote:
>  > 3dc5063786b273f1aee545844f6bd4e9651ebffe was current when I compiled.
>  >
>  >  With that kernel, my laptop locks up as soon as I startx.
>  >
>  >  The Xorg.0.log showed that the radeon driver got as far as starting
>  >  the i2c calls to the M7 video card.
>  >
>  >  After that the box would reply to ICMP pings and SysRq-B, but was
>  >  otherwise non-responsive.  Even SysRq-[KSU] seemed to do nothing.
>  >
>  >  v2.6.25 works fine.
>  >
>  >  There are 3562 commits (including merges) between those two, and
>  >  it takes an hour or more for each kernel compile.  Bisecting will
>  >  be quite painful.  And I only have the one box to work on.  But I
>  >  will see what I can do.
>  >
>  >  Posting before bisecting in case it helps anyone else....
>
>  Do you have CONFIG_X86_PAT in your .config?
>
>  YH
>
>
> --
>  To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>  the body of a message to majordomo@vger.kernel.org
>  More majordomo info at  http://vger.kernel.org/majordomo-info.html
>  Please read the FAQ at  http://www.tux.org/lkml/
>
I cant seem to find X86_PAT,
I have mtrr to "y", but for the other dependencie NONPROMISC_DEVMEM I
cant seem to locate.
regards;


-- 
Justin P. Mattock

Re: Regression: starting X locks box with git master

[James Cloos]

>>>>> "Yinghai" == Yinghai Lu <yhlu.kernel@gmail.com> writes:

Yinghai> Do you have CONFIG_X86_PAT in your .config?

I do not find that in .config at all, set or not:

:; grep X86_PAT /usr/src/linux-2.6/.config

returns nothing.

But the kernel which dies does generate this:

Apr 24 04:28:09 [kernel] [    0.000000] x86: PAT support disabled.

-JimC
-- 
James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6

Re: Regression: starting X locks box with git master

[Rafael J. Wysocki]

On Friday, 25 of April 2008, James Cloos wrote:
> >>>>> "Yinghai" == Yinghai Lu <yhlu.kernel@gmail.com> writes:
> 
> Yinghai> Do you have CONFIG_X86_PAT in your .config?
> 
> I do not find that in .config at all, set or not:
> 
> :; grep X86_PAT /usr/src/linux-2.6/.config
> 
> returns nothing.
> 
> But the kernel which dies does generate this:
> 
> Apr 24 04:28:09 [kernel] [    0.000000] x86: PAT support disabled.

Has the problem been fixed in 2.6.26-rc1 or in the current -git?

Rafael

Re: Regression: starting X locks box with git master

[Justin Mattock]

On Wed, May 7, 2008 at 9:53 PM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
> On Friday, 25 of April 2008, James Cloos wrote:
>  > >>>>> "Yinghai" == Yinghai Lu <yhlu.kernel@gmail.com> writes:
>  >
>  > Yinghai> Do you have CONFIG_X86_PAT in your .config?
>  >
>  > I do not find that in .config at all, set or not:
>  >
>  > :; grep X86_PAT /usr/src/linux-2.6/.config
>  >
>  > returns nothing.
>  >
>  > But the kernel which dies does generate this:
>  >
>  > Apr 24 04:28:09 [kernel] [    0.000000] x86: PAT support disabled.
>
>  Has the problem been fixed in 2.6.26-rc1 or in the current -git?
>
>  Rafael
>

X86PAT was enabled in 2.6.26-rc1, the problem is when loading fglrx
there is a conflict, you have to disable X86pat for fglrx PAT to be
enabled.
regards;

-- 
Justin P. Mattock

Re: Regression: starting X locks box with git master

[James Cloos]

>>>>> "Rafael" == Rafael J Wysocki <rjw@sisk.pl> writes:

>> But the kernel which dies does generate this:

>> Apr 24 04:28:09 [kernel] [    0.000000] x86: PAT support disabled.

Rafael> Has the problem been fixed in 2.6.26-rc1 or in the current -git?

I plan on giving current master a test tonight.

-JimC
-- 
James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6

Re: Regression: starting X locks box with git master

[Justin Mattock]

On Wed, May 7, 2008 at 10:06 PM, James Cloos <cloos@jhcloos.com> wrote:
> >>>>> "Rafael" == Rafael J Wysocki <rjw@sisk.pl> writes:
>
>  >> But the kernel which dies does generate this:
>
>  >> Apr 24 04:28:09 [kernel] [    0.000000] x86: PAT support disabled.
>
>  Rafael> Has the problem been fixed in 2.6.26-rc1 or in the current -git?
>
>  I plan on giving current master a test tonight.
>
>  -JimC
>  --
>  James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6
>

Cool;
>From what I've experienced out of 2.6.25-rc* to the latest git  with a
macbook pro; 2.6.25-rc7,8,or 9 runs the best i.g.
not as hot, fans don't trip as often, when listening to music, and
then compiling you don't have a skip., suspend works
lircd works, isight works.  As for 2.6.26-rc1 lircd doesnt work for
some reason, isight now works, atheros now works, fglrx does'nt due to
something deep into the kernel was changed(who knows what that might be)
did find the kernel running kind of quieter  but noticed the fans
would rev up every so often (estimation would be around 35 min to 60
min + or - )
probably a new mechanism to cool things down. Sound is working, but
every so often a song sounds as if you have blown out speakers.
I probably will pull the latest git, but after spending countless
hours, trying to get fglrx up and running,  and realizing something
was drastically changed in that area, I've decided to go get some
liquid propulsion and call it a day.
regards;

-- 
Justin P. Mattock

Re: Regression: starting X locks box with git master

[James Cloos]

Git master as of 3de2403e allows me to run X again.

After installing the new kernel I also re-compiled all of the X server
stack (libdrm, libpciaccess, mesa, the X server and the video driver).  
I don't know how much of a difference that made.

I don't know whether the problem was sched, x86, agp or dri related,
but AFAICT it is now fixed.

Thanks,

-JimC
-- 
James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6

Re: Regression: starting X locks box with git master

[Justin Mattock]

On Thu, May 8, 2008 at 10:57 PM, James Cloos <cloos@jhcloos.com> wrote:
> Git master as of 3de2403e allows me to run X again.
>
> After installing the new kernel I also re-compiled all of the X server
> stack (libdrm, libpciaccess, mesa, the X server and the video driver).
> I don't know how much of a difference that made.
>
> I don't know whether the problem was sched, x86, agp or dri related,
> but AFAICT it is now fixed.
>
> Thanks,
>
> -JimC
> --
> James Cloos <cloos@jhcloos.com>         OpenPGP: 1024D/ED7DAEA6
>

Well; After reverting to 2.6.25-rc* then contemplating(beer) for a
while I decided to pull the latest git today, the problem now was with
atapiix, So with that in mind I decided to load 2.6.26-rc1 everything
seems to be O.K., the problem I have is: I receive a blank screen when
I load fglrx,
then startx. I think it has to do with /proc/mtrr and pat almost as if
I'm on the wrong screen. With that said I wouldn't be such a pain
about wanting fglrx, hell I don't even play video games, well a long
time ago at least(Myst, Riven, schism)or use 3D applications, but this
module was the only module that was at least working with the suspend
mechanism, This way  I can test X11 and see if appletouch can survive,
scroll function on the mighty mouse can survive,and DVD playback can
survive properly after a wake-up. With that in mind has anybody gotten
suspend to work with radeonhd, or any other module for the macbook
pro. That way I can at least have that function working,and out of the
way, so I can move on to other things, besides having my eyes go
crosseyed trying to find code that isn't there.
Anyways I'll let you know if I come upon anything interesting;
regards;


-- 
Justin P. Mattock