Skip to contentSkip to conversation

_end symbol missing from Symbol.map

REGRESSION Hannes Reinecke 4 participants 5 messages Aug 13, 2009 – Aug 20, 2009 x86

lore
Share & tools
lore
#0 lore raw

Hi all,

with 2.6.31 'crash' on x86_64 falls flat on its face as the '_end' symbol is missing from the System.map file.

The culprit is commit 091e52c3551d3031343df24b573b770b4c6c72b6, which moved the '_end' symbol into it's own section. Apparently this causes kallsyms to not reference it properly.

So either we'd need to revert part of the patch to not include _end in it's own section:

arch/x86/kernel/vmlinux.lds.S +1 -3
index 59f31d2..1422df5 100644
@@ -376,9 +376,7 @@ SECTIONS
                __brk_limit = .;
        }
 
-       .end : AT(ADDR(.end) - LOAD_OFFSET) {
-               _end = .;
-       }
+       _end = .;
 
        /* Sections to be discarded */
        /DISCARD/ : {

or someone has to fixup kallsyms. But this is far beyond my comfort zone.

Cheers,

Hannes
---
Dr. Hannes Reinecke zSeries & Storage
[email protected] +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Markus Rex, HRB 16746 (AG Nürnberg)

#1 lore raw
Quoted 28 lines · On Thu, 13 Aug 2009 08:45:20 +0200 Hannes Reinecke <[email protected]> wrote:

Hi all,

with 2.6.31 'crash' on x86_64 falls flat on its face as the '_end' symbol is missing from the System.map file.

The culprit is commit 091e52c3551d3031343df24b573b770b4c6c72b6, which moved the '_end' symbol into it's own section. Apparently this causes kallsyms to not reference it properly.

So either we'd need to revert part of the patch to not include _end in it's own section:

arch/x86/kernel/vmlinux.lds.S +1 -3
index 59f31d2..1422df5 100644
@@ -376,9 +376,7 @@ SECTIONS
                __brk_limit = .;
        }
 
-       .end : AT(ADDR(.end) - LOAD_OFFSET) {
-               _end = .;
-       }
+       _end = .;
 
        /* Sections to be discarded */
        /DISCARD/ : {
Quoted 2 lines

or someone has to fixup kallsyms. But this is far beyond my comfort zone.

So System.map is part of the kernel API too? Sigh.

Your email client replaces tabs with spaces.

The patch didn't have a signed-off-by:

I queued it up, and tagged it for -stable backporting. Unless we come up with something better.

We might not get this into 2.6.31, in which case this fix or its replacement will need backporting to 2.6.30.x and 2.6.31.x (IMO).

#2 lore raw
Quoted 30 lines · Andrew Morton
Quoted 28 lines · On Thu, 13 Aug 2009 08:45:20 +0200 Hannes Reinecke <[email protected]> wrote:

Hi all,

with 2.6.31 'crash' on x86_64 falls flat on its face as the '_end' symbol is missing from the System.map file.

The culprit is commit 091e52c3551d3031343df24b573b770b4c6c72b6, which moved the '_end' symbol into it's own section. Apparently this causes kallsyms to not reference it properly.

So either we'd need to revert part of the patch to not include _end in it's own section:

arch/x86/kernel/vmlinux.lds.S +1 -3
index 59f31d2..1422df5 100644
@@ -376,9 +376,7 @@ SECTIONS
                __brk_limit = .;
        }
 
-       .end : AT(ADDR(.end) - LOAD_OFFSET) {
-               _end = .;
-       }
+       _end = .;
 
        /* Sections to be discarded */
        /DISCARD/ : {
Quoted 7 lines
Quoted 2 lines

or someone has to fixup kallsyms. But this is far beyond my comfort zone.

So System.map is part of the kernel API too? Sigh.

Your email client replaces tabs with spaces.

I know.

Quoted 2 lines

The patch didn't have a signed-off-by:

I know, too.
However, this is clearly a band-aid, and as such I reported it more as an RFC.
One (Sam?) should really fix up kallsyms to extract the _end symbol. Hence I didn't warrant it with a Signed-off line.

Quoted 3 lines

I queued it up, and tagged it for -stable backporting. Unless we come up with something better.

I was sort of hoping Sam would speak up and present some better approach ...

Quoted 3 lines

We might not get this into 2.6.31, in which case this fix or its replacement will need backporting to 2.6.30.x and 2.6.31.x (IMO).

Thanks for the feedback.

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke		      zSeries & Storage
[email protected]			      +49 911 74053 688
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Markus Rex, HRB 16746 (AG Nürnberg)
#3 lore raw

At Thu, 20 Aug 2009 07:59:26 +0200,

Quoted 49 lines · Hannes Reinecke
Quoted 30 lines · Andrew Morton
Quoted 28 lines · On Thu, 13 Aug 2009 08:45:20 +0200 Hannes Reinecke <[email protected]> wrote:

Hi all,

with 2.6.31 'crash' on x86_64 falls flat on its face as the '_end' symbol is missing from the System.map file.

The culprit is commit 091e52c3551d3031343df24b573b770b4c6c72b6, which moved the '_end' symbol into it's own section. Apparently this causes kallsyms to not reference it properly.

So either we'd need to revert part of the patch to not include _end in it's own section:

arch/x86/kernel/vmlinux.lds.S +1 -3
index 59f31d2..1422df5 100644
@@ -376,9 +376,7 @@ SECTIONS
                __brk_limit = .;
        }
 
-       .end : AT(ADDR(.end) - LOAD_OFFSET) {
-               _end = .;
-       }
+       _end = .;
 
        /* Sections to be discarded */
        /DISCARD/ : {
Quoted 7 lines
Quoted 2 lines

or someone has to fixup kallsyms. But this is far beyond my comfort zone.

So System.map is part of the kernel API too? Sigh.

Your email client replaces tabs with spaces.

I know.

Quoted 2 lines

The patch didn't have a signed-off-by:

I know, too.
However, this is clearly a band-aid, and as such I reported it more as an RFC.
One (Sam?) should really fix up kallsyms to extract the _end symbol. Hence I didn't warrant it with a Signed-off line.

Quoted 7 lines
Quoted 3 lines

I queued it up, and tagged it for -stable backporting. Unless we come up with something better.

I was sort of hoping Sam would speak up and present some better approach ...

Quoted 2 lines

We might not get this into 2.6.31, in which case this fix or its replacement will need backporting to 2.6.30.x and 2.6.31.x (IMO).

Yep, x86-32 had already the own end section in 2.6.30 kernel, which was introduced by the following commit:

commit 0a699af8e613a670be50245366fa18cb19ac5172

Author: H. Peter Anvin <[email protected]>
Date:   Tue Mar 17 14:14:31 2009 -0700

x86-32: move _end to a dummy section

Impact: build fix with CONFIG_RELOCATABLE

Move _end into a dummy section, so that relocs.c will know it is a relocatable symbol.

Cc Jeremy Fitzhardinge <[email protected]>
Cc Jeremy Fitzhardinge <[email protected]>

So, judging from the commit log, simply reverting it seems to have some drawback...?

thanks,

Takashi

#4 lore raw
Quoted 4 lines · Takashi Iwai

So, judging from the commit log, simply reverting it seems to have some drawback...?

Yes. I'm starting to think we should have a list of "known relative symbols" in relocs.c just as we have a list of "known absolute symbols" already.

-hpa