Skip to contentSkip to conversation

Some Modules fail w/ unresolved cpu_info__per_cpu

Linux-ia64 Lee, Jung-Ik 3 participants 5 messages Oct 17, 2002 – Feb 19, 2003

lore
Share & tools
lore
#0 lore raw

On 2.5.39,
it fails to insmod some modules that refers to "cpu_info__per_cpu". That global IS DEFINE_PER_CPU'd, DECLARE_PER_CPU'd and is in System.map but for some reason it fails to resolve.
Have you guys seen this ?

thanks
J.I.

-----Original Message-----
From: Luck, Tony [mailto:[email protected]]
Sent: Wednesday, October 16, 2002 10:56 AM
To: Luck, Tony; linux ia64 kernel list
Cc: [email protected]
Subject: RE: [Linux-ia64] [patch 2.5.39] allow kernel to be virtually mapp ed from any physi cal address

Quoted 4 lines · Yesterday I

This patch provides just the code needed to virtually map the kernel to a fixed virtual address from whatever physical address it happened to be loaded at (it is assumed that the bootloader handled the issue of finding a suitably aligned piece of memory).

Elilo already knows how to find memory, as long as you either use the "relocatable" keyword in elilo.conf, or the "-r" command-line option to let it know that it is OK to relocate.

Using this elilo option means that the changes I provided for vmlinux.ld.S can be very slightly simplified, it isn't necessary to define BASE_KVADDR as "KERNEL_START + KERNEL_TR_PAGE_SIZE", you can just use:

#define BASE_KVADDR KERNEL_START

-Tony

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64

#1 lore raw
Quoted 10 lines · Jung-Ik

On 2.5.39,
it fails to insmod some modules that refers to "cpu_info__per_cpu". That global IS DEFINE_PER_CPU'd, DECLARE_PER_CPU'd and is in System.map but for some reason it fails to resolve.
Have you guys seen this ?

thanks
J.I.

Yes I have seen something similar with the ipmi driver ("imb") on Tiger but with the 2.5.35 release.
I found that the exported symbol name via /proc/ksyms is in fact:
cpu_info__per_cpu_R__ver_cpu_info__per_cpu

Then I did not ask much more questions to me trying to find the root cause of the problem within the kernel sources and I just added the following trivial modification in the driver's code:
#define cpu_info__per_cpu cpu_info__per_cpu_R__ver_cpu_info__per_cpu

With this workaround insmod didn't complain anymore about "cpu_info__per_cpu" and the driver works correctly.

But I am agree it is just a workaround and we shouldn't have to modify drivers's code for such problem.

Quoted 36 lines

-----Original Message-----
From: Luck, Tony [mailto:[email protected]]
Sent: Wednesday, October 16, 2002 10:56 AM
To: Luck, Tony; linux ia64 kernel list
Cc: [email protected]
Subject: RE: [Linux-ia64] [patch 2.5.39] allow kernel to be virtually mapp ed from any physi cal address

Quoted 4 lines · Yesterday I

This patch provides just the code needed to virtually map the kernel to a fixed virtual address from whatever physical address it happened to be loaded at (it is assumed that the bootloader handled the issue of finding a suitably aligned piece of memory).

Elilo already knows how to find memory, as long as you either use the "relocatable" keyword in elilo.conf, or the "-r" command-line option to let it know that it is OK to relocate.

Using this elilo option means that the changes I provided for vmlinux.ld.S can be very slightly simplified, it isn't necessary to define BASE_KVADDR as "KERNEL_START + KERNEL_TR_PAGE_SIZE", you can just use:

#define BASE_KVADDR KERNEL_START

-Tony

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64

#2 lore raw
In reply to Lee, Jung-Ik
Quoted 8 lines · Christian Cotte-Barrot

I found that the exported symbol name via /proc/ksyms is in fact:
cpu_info__per_cpu_R__ver_cpu_info__per_cpu

Then I did not ask much more questions to me trying to find the root cause of the problem within the kernel sources and I just added the following trivial modification in the driver's code:
#define cpu_info__per_cpu cpu_info__per_cpu_R__ver_cpu_info__per_cpu

Quoted 5 lines

With this workaround insmod didn't complain anymore about "cpu_info__per_cpu" and the driver works correctly.

But I am agree it is just a workaround and we shouldn't have to modify drivers's code for such problem.

Arrghh, don't do that! FAQ: http://www.tux.org/lkml/#s8-8.

#3 lore raw
In reply to Lee, Jung-Ik
Quoted 11 lines · Keith Owens
Quoted 8 lines · Christian Cotte-Barrot

I found that the exported symbol name via /proc/ksyms is in fact:
cpu_info__per_cpu_R__ver_cpu_info__per_cpu

Then I did not ask much more questions to me trying to find the root cause of the problem within the kernel sources and I just added the following trivial modification in the driver's code:
#define cpu_info__per_cpu cpu_info__per_cpu_R__ver_cpu_info__per_cpu

Quoted 7 lines
Quoted 5 lines

With this workaround insmod didn't complain anymore about "cpu_info__per_cpu" and the driver works correctly.

But I am agree it is just a workaround and we shouldn't have to modify drivers's code for such problem.

Arrghh, don't do that! FAQ: http://www.tux.org/lkml/#s8-8.

Agree, agree, agree, ...

In my case I had just the need to verify that the ipmi driver still worked on Tiger platform and I applied a very temporary workaround in order that the module would be successfully loaded.
I agree it is a nasty huge correction and I won't pretend it is a clean one. I just wanted to answer I hit the problem and give an easy and simple way to bybass it.

#4 lore raw
In reply to Lee, Jung-Ik
Quoted 10 lines · Jung-Ik

On 2.5.39,
it fails to insmod some modules that refers to "cpu_info__per_cpu". That global IS DEFINE_PER_CPU'd, DECLARE_PER_CPU'd and is in System.map but for some reason it fails to resolve.
Have you guys seen this ?

thanks
J.I.

That's an old story but we met the "unresolved symbol cpu_info__per_cpu" problem again with e100.o and e1000.o modules on a 2.4.45 release. Looks like the EXPORT_SYMBOL macro has problem to generate the checksum for symbols defined with *PER_CPU* macros.
The kernel exports the wrong cpu_info__per_cpu_R__ver_cpu_info__per_cpu symbol where "__ver_cpu_info__per_cpu" should be a checksum ? Found a cleaner patch than in the past, just replaced

    EXPORT_SYMBOL(cpu_info__per_cpu);
  with
    EXPORT_SYMBOL_NOVERS(cpu_info__per_cpu);
  in
    ./arch/ia64/kernel/ia64_ksyms.c

CCB.