[NUMA x86_64] problem accessing global Node List pgdat_list

[NUMA x86_64] problem accessing global Node List pgdat_list

[devesh sharma]

Hi all,
On an dual opteron machine and 2.6.9 kernel, I am accessing the global
node list pgdat_list but I am not getting the desired results

#include<linux/module.h>
#include<linux/config.h>
#include<linux/kernel.h>
#include<linux/mmzone.h>

struct pglist_data *pgdat_list ;

int init_module( void )
{

  pg_data_t *pg_dat ;

  printk ("<1>****Module initialized to retrive the information of
pgdat_list list in the Kernel****\n" ) ;


  for_each_pgdat(pg_dat)
  {
    printk ("<1>The number of zones on this node are %x\n", pg_dat ->
nr_zones ) ;

    printk ("<1>The Node Id of this node is %d\n", pg_dat -> node_id ) ;

  }

  return 0 ;
}

void cleanup_module ( void )
{
    printk ("<1>********Module Exiting***********\n" ) ;
}

MODULE_LICENSE("GPL") ;

How I can access this list any body tell me the solution.

Re: [NUMA x86_64] problem accessing global Node List pgdat_list

[Eric Dumazet]

Hi Devesh

devesh sharma a écrit :
> Hi all,
> On an dual opteron machine and 2.6.9 kernel, I am accessing the global
> node list pgdat_list but I am not getting the desired results
> 
> #include<linux/module.h>
> #include<linux/config.h>
> #include<linux/kernel.h>
> #include<linux/mmzone.h>
> 
> struct pglist_data *pgdat_list ;

What are you doing here ? You declare a local variable on this module.
You should instead write :

extern struct pglist_data *pgdat_list ;
(But it seems already declared in mmzone.h)

But pgdat_list is an exported symbol of linux kernel : a module cannot access it.

So I suspect you will have to add in mm/page_alloc.c  (and recompile your kernel)

EXPORT_SYMBOL(pgdat_list);


And please use a recent kernel (2.6.13 at least) or few people will answer you.


> 
> int init_module( void )
> {
> 
>   pg_data_t *pg_dat ;
> 
>   printk ("<1>****Module initialized to retrive the information of
> pgdat_list list in the Kernel****\n" ) ;
> 
> 
>   for_each_pgdat(pg_dat)
>   {
>     printk ("<1>The number of zones on this node are %x\n", pg_dat ->
> nr_zones ) ;
> 
>     printk ("<1>The Node Id of this node is %d\n", pg_dat -> node_id ) ;
> 
>   }
> 
>   return 0 ;
> }
> 
> void cleanup_module ( void )
> {
>     printk ("<1>********Module Exiting***********\n" ) ;
> }
> 
> MODULE_LICENSE("GPL") ;
> 
> How I can access this list any body tell me the solution.

Eric

Re: [NUMA x86_64] problem accessing global Node List pgdat_list

[devesh sharma]

Hi Eric,
I have tried numa=fake=4 on intel xeon with 2.6.13 kernel it is
working fine and by adding EXPORT_SYMBOL(pgdat_list)  in
mm/page_alloc.c now I am able to access pgdat_list also. But on
opteron machine there is some problem in kernel compilation
at make install stage I am getting following warning

WARNING: No module mptbase found for kernel 2.6.13, continuing anyway
WARNING: No module mptscsih found for kernel 2.6.13, continuing anyway

now when I boot my kernel, panic is received

Booting the kernel.
Red Hat nash version 4.1.18 starting
mkrootdev: lable / not found
mount: error 2 mounting ext3
mount: error 2 mounting none
switchroot: mount failed : 22
umount : /initrd/dev failed : 22
kernel panic - not syncing : Attempted to kill init

On the other hand when I complie same source code on XEON machine this
works fine.

what could be the problem?

Re: [NUMA x86_64] problem accessing global Node List pgdat_list

[Eric Dumazet]

devesh sharma a écrit :
> Hi Eric,
> I have tried numa=fake=4 on intel xeon with 2.6.13 kernel it is
> working fine and by adding EXPORT_SYMBOL(pgdat_list)  in
> mm/page_alloc.c now I am able to access pgdat_list also. But on
> opteron machine there is some problem in kernel compilation
> at make install stage I am getting following warning
> 
> WARNING: No module mptbase found for kernel 2.6.13, continuing anyway
> WARNING: No module mptscsih found for kernel 2.6.13, continuing anyway
> 
> now when I boot my kernel, panic is received
> 
> Booting the kernel.
> Red Hat nash version 4.1.18 starting
> mkrootdev: lable / not found
> mount: error 2 mounting ext3
> mount: error 2 mounting none
> switchroot: mount failed : 22
> umount : /initrd/dev failed : 22
> kernel panic - not syncing : Attempted to kill init
> 
> On the other hand when I complie same source code on XEON machine this
> works fine.
> 
> what could be the problem?

Hi Devesh

I believe you have basic kernel compilation problem :)

Your kernel cannot mount / because the boot loader was configured to find a 
root=LABEL=/
As this is a RedHat kernel extension, your self compiled 2.6.13 kernel cannot 
understand this.
You should change your lilo/grub config to match your root device as 
root=/dev/hda2  (if hda2 is your root device)

Also check that your kernel (or initrd) contains the disk driver you need.

Eric