Re: [Jackit-devel] Re: little NPTL SCHED_FIFO test program

Re: [Jackit-devel] Re: little NPTL SCHED_FIFO test program

[Lee Revell]

On Fri, 2004-08-20 at 05:20, martin rumori wrote:
> On Fri, Aug 20, 2004 at 03:38:44AM -0400, Lee Revell wrote:
> > > I have built a glibc with the patch you (Lee) sent to the bug-report.
> > 
> > This fixes the problem, jackd now works correctly.  I cannot take credit
> > for finding the problem though, Florian did most of the work in tracking
> > this down.
> 
> same here, really great (debian unstable).
> 
> wanted to try the volunteer preempt patch for the first time, (until
> now running 2.6.7-cko5 atm).
> 
> tried to compile 2.6.8.1 with volunteer-preempt-P5 (latest), but
> wasn't able to boot into the new kernel.  shows some lines of ACPI....
> directly after unpacking, then stops for about 1 second (without an
> error message, last line is something with a specific processor flag),
> then reboots.  not long enough to see something meaningful...
> 

I think you have an off-by-one error.  The latest version is -P4.

Try the 'noacpi' boot option.  Also, try disabling all power
management.  I know this is not realistic for a laptop, it's just for
debugging purposes.

> ACPI is disabled in favor of APM, changed that, same behavior: there
> is some lines starting with ACPI on the screen, either way.  applied
> latest acpi.sf.net patch, same thing. would like to keep APM since
> ACPI doesn't work that well with my machine.  stock 2.6.8.1-acpi boots
> fine (with the usual acpi oddities, with or without acpi patch), apm
> with 2.6.8 not yet tried (don't expect problems).
> 
> thinkpad t40p, PM-1600, 512 MB, latest bios/embeddecontroller.
> 
> is there something like that known with volunteer-preempt?  should i
> try an older version (== P4) first?  are there some config options
> which are especially meaningful?
> 

I am adding LKML and Ingo to the cc: list as he has requested any
reports of ACPI problems with the VP patches.

Lee

Re: problems with volunteer preempt patch WAS: little NPTL SCHED_FIFO test program

[martin rumori]

On Fri, Aug 20, 2004 at 05:42:59AM -0400, Lee Revell wrote:
> On Fri, 2004-08-20 at 05:20, martin rumori wrote:
> > tried to compile 2.6.8.1 with volunteer-preempt-P5 (latest), but
> > wasn't able to boot into the new kernel.  shows some lines of ACPI....

> I think you have an off-by-one error.  The latest version is -P4.

o.k., you're right.  this time it's really -P5 :-)

> Try the 'noacpi' boot option.  Also, try disabling all power
> management.  I know this is not realistic for a laptop, it's just for
> debugging purposes.

tried it with the options, no success.  built a kernel without
powermanagement and without cpu speedstep at all.  exactly the same
behavior, directly after unpacking the kernel at the beginning of
startup:

0 MB highmem
511 mb lowmem available

ACPI RSDP .....
.....
ACPI build 1 zone list
Detected Processor...
Memory ...
...
Checking if processor honors the WP bit even in supervisor mode... ok.

---
that's it.  freezes for 1-2 seconds after this last message, then
reboots.  tried to write down these contents of the screen, as far as
i could recognize them during the short period of time.

i am wondering especially about the ACPI messages, as ACPI is
completely switched off now.

second (minor) issue: when having enabled CONFIG_PREEMPT_VOLUNTARY(=y)
but not CONFIG_PREEMPT_TIMING(=n), i get the following when linking:

  LD      init/built-in.o
  LD      .tmp_vmlinux1
arch/i386/kernel/built-in.o(.text+0x2da1): In function `do_nmi':
: undefined reference to `__trace'
arch/i386/kernel/built-in.o(.text+0x342c): In function `do_IRQ':
: undefined reference to `__trace'
arch/i386/mm/built-in.o(.text+0x705): In function `do_page_fault':
: undefined reference to `__trace'
make: *** [.tmp_vmlinux1] Error 1

maybe latency.o is not included when linking without
CONFIG_PREEMPT_TIMING?

bests,

martin

Re: problems with volunteer preempt patch WAS: little NPTL SCHED_FIFO test program

[Ingo Molnar]

* martin rumori <lists@rumori.de> wrote:

> ACPI RSDP .....
> .....
> ACPI build 1 zone list
> Detected Processor...
> Memory ...
> ...
> Checking if processor honors the WP bit even in supervisor mode... ok.
> 
> ---
> that's it.  freezes for 1-2 seconds after this last message, then
> reboots.  tried to write down these contents of the screen, as far as
> i could recognize them during the short period of time.
> 
> i am wondering especially about the ACPI messages, as ACPI is
> completely switched off now.

is ACPI switched off in the .config too?

> second (minor) issue: when having enabled CONFIG_PREEMPT_VOLUNTARY(=y)
> but not CONFIG_PREEMPT_TIMING(=n), i get the following when linking:
> 
>   LD      init/built-in.o
>   LD      .tmp_vmlinux1
> arch/i386/kernel/built-in.o(.text+0x2da1): In function `do_nmi':
> : undefined reference to `__trace'

> maybe latency.o is not included when linking without
> CONFIG_PREEMPT_TIMING?

right. We dont want to link it in so i've added a NOP define for
__trace() to sched.h. This fix will show up in -P6.

	Ingo

Re: [Jackit-devel] Re: problems with volunteer preempt patch WAS: little NPTL SCHED_FIFO test program

[martin rumori]

On Fri, Aug 20, 2004 at 08:35:59PM +0200, Ingo Molnar wrote:
> > i am wondering especially about the ACPI messages, as ACPI is
> > completely switched off now.
> 
> is ACPI switched off in the .config too?

basically yes, but there is one strange thing:

#
# Power management options (ACPI, APM)
#
# CONFIG_PM is not set

#
# ACPI (Advanced Configuration and Power Interface) Support
#
# CONFIG_ACPI is not set
CONFIG_ACPI_BOOT=y

i don't know where to find the last option in the config programs
(menuconfig, xconfig).  if i comment it manually, it'll get
uncommented as soon as the next make command is issued, regardless
whether make xconfig again, make oldconfig, or make.  it gets reverted
to the above shown state.

i figured out that in stock 2.6.8.1 and 2.6.7-cko5 this option is
enabled, too, (even with the rest of acpi disabled and apm enabled)
maybe i can't see the messages at boot time since they are going to
fast.

> > maybe latency.o is not included when linking without
> > CONFIG_PREEMPT_TIMING?
> 
> right. We dont want to link it in so i've added a NOP define for
> __trace() to sched.h. This fix will show up in -P6.

great, many thanks.

bests,

martin

Re: [Jackit-devel] Re: problems with volunteer preempt patch WAS: little NPTL SCHED_FIFO test program

[martin rumori]

On Fri, Aug 20, 2004 at 08:35:59PM +0200, Ingo Molnar wrote:
> * martin rumori <lists@rumori.de> wrote:
> > Checking if processor honors the WP bit even in supervisor mode... ok.
> > 
> > that's it.  freezes for 1-2 seconds after this last message, then
> > reboots.  tried to write down these contents of the screen, as far as
> > i could recognize them during the short period of time.

works now with -P6 and apm support in the kernel.  great.  thanks a
lot.

bests,

martin