ACPI stack overflow

ACPI stack overflow

[Stuart_Hayes]

Hello...

I think I am getting a stack overflow when Linux is parsing the ACPI tables (initializing all the devices and running all the _STA methods).  I am using the x86_64 architecture.  I would like to try increasing the kernel stack size, but I'm not sure how to go about doing this.  Could someone tell me how to increase the kernel stack size?  (And, has anyone else seen a problem with stack overflows with ACPI?)

Thanks!
Stuart
stuart_hayes@dell.com

Re: ACPI stack overflow

[Richard B. Johnson]

On Fri, 5 Mar 2004 Stuart_Hayes@Dell.com wrote:

>
> Hello...
>
> I think I am getting a stack overflow when Linux is parsing the ACPI
> tables (initializing all the devices and running all the _STA methods).
> I am using the x86_64 architecture.  I would like to try increasing the
> kernel stack size, but I'm not sure how to go about doing this.
> Could someone tell me how to increase the kernel stack size?
> (And, has anyone else seen a problem with stack overflows with ACPI?)
>

Please fix your mailer. In Unix/Linux, we put in a [Enter] ('\n')
every once in awhile, usually every 79 charcters so that a line
of text does not exceed 80 characters. We do not let some indefinite
screen "auto-wrap".

> Thanks!
> Stuart
> stuart_hayes@dell.com

There have been continual changes over the years to reduce the
amount of kernel stack that the kernel uses because kernel stack-
space is "expensive". It needs to be changed in pages.  I think
that if you have a stack-overflow, then you are writing poor
kernel code. In the kernel, do not put arrays on the stack, i.e,
in "local" space. Use kmalloc()/kfree() instead.

Basically, do not increase the stack size. It just masks problems.
It does not make them go away. If you need more stack, you
are doing something wrong.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.24 on an i686 machine (797.90 BogoMips).
            Note 96.31% of all statistics are fiction.

Re: ACPI stack overflow

[Chris Wedgwood]

On Fri, Mar 05, 2004 at 03:01:34PM -0600, Stuart_Hayes@Dell.com wrote:

> I think I am getting a stack overflow when Linux is parsing the ACPI
> tables (initializing all the devices and running all the _STA
> methods).  I am using the x86_64 architecture.  I would like to try
> increasing the kernel stack size, but I'm not sure how to go about
> doing this.  Could someone tell me how to increase the kernel stack
> size?  (And, has anyone else seen a problem with stack overflows
> with ACPI?)

Please find the source of the stack overflow and fix that.  Increasing
the kernel stack size is not a suitable solution.


  --cw

Re: ACPI stack overflow

[Andi Kleen]

Stuart_Hayes@Dell.com writes:

> Hello...
> 
> I think I am getting a stack overflow when Linux is parsing the ACPI tables (initializing all the devices and running all the _STA methods).  I am using the x86_64 architecture.  I would like to try increasing the kernel stack size, but I'm not sure how to go about doing this.  Could someone tell me how to increase the kernel stack size?  (And, has anyone else seen a problem with stack overflows with ACPI?)

Increasing THREAD_ORDER to 2 in include/asm-x86_64/page.h should do 
the trick in theory (not tested). There is also an old 2.4 exact stack
overflow checking patch at ftp://ftp.x86-64.org/pub/linux/debug/stackcheck-1
that could be probably ported to newer kernels.

I haven't heard of ACPI stack overflows before.

-Andi

RE: ACPI stack overflow

[Stuart_Hayes]

Richard B. Johnson wrote:
> On Fri, 5 Mar 2004 Stuart_Hayes@Dell.com wrote:
> 
>> 
>> Hello...
>> 
>> I think I am getting a stack overflow when Linux is parsing the ACPI
>> tables (initializing all the devices and running all the _STA
>> methods). I am using the x86_64 architecture.  I would like to try
>> increasing the kernel stack size, but I'm not sure how to go about
>> doing this. 
>> Could someone tell me how to increase the kernel stack size?
>> (And, has anyone else seen a problem with stack overflows with ACPI?)
>> 
> 
> Please fix your mailer. In Unix/Linux, we put in a [Enter] ('\n')
> every once in awhile, usually every 79 charcters so that a line
> of text does not exceed 80 characters. We do not let some indefinite
> screen "auto-wrap".
> 
>> Thanks!
>> Stuart
>> stuart_hayes@dell.com
> 
> There have been continual changes over the years to reduce the
> amount of kernel stack that the kernel uses because kernel stack-
> space is "expensive". It needs to be changed in pages.  I think
> that if you have a stack-overflow, then you are writing poor
> kernel code. In the kernel, do not put arrays on the stack, i.e,
> in "local" space. Use kmalloc()/kfree() instead.
> 
> Basically, do not increase the stack size. It just masks problems.
> It does not make them go away. If you need more stack, you
> are doing something wrong.
> 
> Cheers,
> Dick Johnson
> Penguin : Linux version 2.4.24 on an i686 machine (797.90 BogoMips).
>             Note 96.31% of all statistics are fiction.

Thanks.  Unfortunately, it isn't my code that's using so much stack 
space--it's the ACPI "driver".  It has many deep nested calls, and 
I believe each call in x86_64 is taking more stack space than it would
 in i386 because of the larger registers.  But, what I really wanted
 to do is increase the kernel size to verify that the problem is a 
stack overflow (my other tests have led me to believe so).  I wasn't 
going to propose that as a solution.

Thanks!
Stuart