Skip to contentSkip to conversation

Stripping kernel modules for ARC

Buildroot Alexey Brodkin 5 participants 12 messages Sep 14, 2016 – Sep 15, 2016 1 message not in the archive

lore
Share & tools
lore
#0 lore raw

Hello,

This is a follow-up of discussion started here
https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/issues/86

The problem in hand is kernel stack unwinding on ARC historically was done with use of debug_frame instead of more common eh_frame. And if stripping of target binaries is enabled in Buildroot invocation of strip with "--strip-unneeded" gets rid of debug_frame section effectively making kernel unwinder helpless.

I'm wondering what would be a better way to strip modules on ARC in Buildroot?

I may see at least following solutions:
?1. Empty STRIP_STRIP_UNNEEDED or even KSTRIPCMD for ARC
? ? This together with?http://lists.infradead.org/pipermail/linux-snps-arc/2016-September/001483.html ? ? will still produce pretty compact .ko files on target
?2. Use objdump instead of strip with explicitly mentioned sections to keep or remove

Any thoughts are much appreciated.

-Alexey

#1 lore raw
Quoted 11 lines · Alexey Brodkin

Hello,

This is a follow-up of discussion started here
https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/issues/86

The problem in hand is kernel stack unwinding on ARC historically was done with use of debug_frame instead of more common eh_frame. And if stripping of target binaries is enabled in Buildroot invocation of strip with "--strip-unneeded" gets rid of debug_frame section effectively making kernel unwinder helpless.

Quoted 7 lines

I'm wondering what would be a better way to strip modules on ARC in Buildroot?

I may see at least following solutions:
1. Empty STRIP_STRIP_UNNEEDED or even KSTRIPCMD for ARC This together with
http://lists.infradead.org/pipermail/linux-snps-arc/2016-September/001483.html
will still produce pretty compact .ko files on target

How about using --keep-symbol to keep the debug_frame? Or is it not a symbol?

Quoted 2 lines

2. Use objdump instead of strip with explicitly mentioned sections to keep or
remove

strip also has --remove-section.

Regards,
Arnout

Quoted 9 lines

Any thoughts are much appreciated.

-Alexey
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot

Signature · 6 lines
Arnout Vandecappelle arnout at mind be
Senior Embedded Software Architect +32-16-286500
Essensium/Mind http://www.mind.be
G.Geenslaan 9, 3001 Leuven, Belgium BE 872 984 063 RPR Leuven
LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle
GPG fingerprint: 7493 020B C7E3 8618 8DEC 222C 82EB F404 F9AC 0DDF
#2 lore raw
In reply to Alexey Brodkin
Quoted 7 lines · Alexey Brodkin

Hello,

This is a follow-up of discussion started here
https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/issues/86

The problem in hand is kernel stack unwinding on ARC historically was done with use of debug_frame instead of more common eh_frame.

Just for the record, I will be switching kernel unwinder to .eh_frame anyways - it's just taking longer than I hoped to. So there is no need to do anything special for .debug_frame - not as of now.

The question is if we enable BR2_STRIP_NONE=y in buildroot, what strip is run on kernel / modules. And also does it strip build and target/install modules differently such that target ones are stripped, but not those for debugging etc.

-Vineet

Quoted 21 lines

And if stripping
of target binaries is enabled in Buildroot invocation of strip with "--strip-unneeded" gets rid of debug_frame section effectively making kernel unwinder helpless.

I'm wondering what would be a better way to strip modules on ARC in Buildroot?

I may see at least following solutions:
1. Empty STRIP_STRIP_UNNEEDED or even KSTRIPCMD for ARC
This together with http://lists.infradead.org/pipermail/linux-snps-arc/2016-September/001483.html
will still produce pretty compact .ko files on target
2. Use objdump instead of strip with explicitly mentioned sections to keep or remove

Any thoughts are much appreciated.

-Alexey
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot

#3 lore raw
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Vineet" == Vineet Gupta <[email protected]> writes:

Quoted 5 lines · Alexey Brodkin

Hello,

This is a follow-up of discussion started here
https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/issues/86

Quoted 6 lines
Quoted 2 lines

The problem in hand is kernel stack unwinding on ARC historically was done with use of debug_frame instead of more common eh_frame.

Just for the record, I will be switching kernel unwinder to .eh_frame anyways - it's just taking longer than I hoped to. So there is no need to do anything special for .debug_frame - not as of now.

Ok, what kind of time frame are we looking at? Within the 2016.11 cycle or longer?

Quoted 3 lines

The question is if we enable BR2_STRIP_NONE=y in buildroot, what strip is run on kernel / modules. And also does it strip build and target/install modules differently such that target ones are stripped, but not those for debugging etc.

git grep -A3 STRIP_none package/Makefile.in
package/Makefile.in:ifeq ($(BR2_STRIP_none),y)
package/Makefile.in-TARGET_STRIP = true
package/Makefile.in-STRIPCMD = $(TARGET_STRIP)
package/Makefile.in-KSTRIPCMD = $(TARGET_STRIP)

So the stripping in target-finalize becomes a NOP when BR2_STRIP_none is selected.

Signature · 1 line
Bye, Peter Korsgaard
#4 lore raw

Hi Vineet,

Quoted 28 lines · Peter Korsgaard
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

"Vineet" == Vineet Gupta <[email protected]> writes:

?> On 09/14/2016 04:29 AM, Alexey Brodkin wrote:
?>> Hello,
?>>?
?>> This is a follow-up of discussion started here
?>> https://github.com/foss-for-synopsys-dwc-arc-processors/toolchain/issues/86
?>>?
?>> The problem in hand is kernel stack unwinding on ARC historically was done ?>> with use of debug_frame instead of more common eh_frame.?

?> Just for the record, I will be switching kernel unwinder to .eh_frame anyways - ?> it's just taking longer than I hoped to. So there is no need to do anything ?> special for .debug_frame - not as of now.

Ok, what kind of time frame are we looking at? Within the 2016.11 cycle or longer?

?> The question is if we enable BR2_STRIP_NONE=y in buildroot, what strip is run ?> on kernel / modules. And also does it strip build and target/install modules ?> differently such that target ones are stripped, but not those for debugging etc.

Quoted 8 lines

git grep -A3 STRIP_none package/Makefile.in
package/Makefile.in:ifeq ($(BR2_STRIP_none),y)
package/Makefile.in-TARGET_STRIP = true
package/Makefile.in-STRIPCMD = $(TARGET_STRIP)
package/Makefile.in-KSTRIPCMD = $(TARGET_STRIP)

So the stripping in target-finalize becomes a NOP when BR2_STRIP_none is selected.

Just to add buildroot doesn't do any stripping in "output/build/XXX" at all. And depending on BR2_STRIP_xxx binaries in "output/target" folder could be stripped. And different binaries could be stripped differently, this in particular is appled to kernel modules (thus KSTRIPCMD is used along with normal STRIPCMD).

-Alexey

#5 lore raw
In reply to Peter Korsgaard
Quoted 9 lines · Peter Korsgaard
Quoted 6 lines
Quoted 2 lines

The problem in hand is kernel stack unwinding on ARC historically was done with use of debug_frame instead of more common eh_frame.

Just for the record, I will be switching kernel unwinder to .eh_frame anyways - it's just taking longer than I hoped to. So there is no need to do anything special for .debug_frame - not as of now.

Ok, what kind of time frame are we looking at? Within the 2016.11 cycle or longer?

What kernel will this BR release be. My changes will go into 4.9.

-Vineet

#6 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Vineet" == Vineet Gupta <[email protected]> writes:

Hi,

Quoted 4 lines
Quoted 2 lines

Ok, what kind of time frame are we looking at? Within the 2016.11 cycle or longer?

What kernel will this BR release be. My changes will go into 4.9.

Ok. Whatever is latest stable by the time 2016.11-rc1 gets released - E.G. presumably 4.8.

Signature · 2 lines
Venlig hilsen,
Peter Korsgaard 

A message that is not in the archive (sent off-list or not yet received): CAE2F3rCBwysRFrMf4CD6uLevvfw0TNaZGyptOj6MLDQvGPPwFg@mail.gmail.com

#8 lore raw

Hello,

Quoted 7 lines · Daniel Mentz

I believe there might be a third option where we pass INSTALL_MOD_STRIP=1 to the kernel Makefile when running "make modules_install". It would then require a change in the Linux kernel Makefile to run a different strip/objcopy command for ARC. This would contain the details of stripping .ko files to the Linux kernel build system. Otherwise, we'll have this cross-dependency between the kernel and buildroot. As soon as the kernel switches to .eh_frame, buildroot would have to change again.

Exactly my thinking. I discovered INSTALL_MOD_STRIP only recently, and I believe we should use it for the kernel instead of having our own logic. This would remove the need for special logic in Buildroot to strip kernel modules, and would leave it to the kernel build system to know how modules should be stripped in an architecture-specific way.

Best regards,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
#9 lore raw
Quoted 13 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Thomas" == Thomas Petazzoni <[email protected]> writes:

Hello,

Quoted 7 lines · Daniel Mentz

I believe there might be a third option where we pass INSTALL_MOD_STRIP=1 to the kernel Makefile when running "make modules_install". It would then require a change in the Linux kernel Makefile to run a different strip/objcopy command for ARC. This would contain the details of stripping .ko files to the Linux kernel build system. Otherwise, we'll have this cross-dependency between the kernel and buildroot. As soon as the kernel switches to .eh_frame, buildroot would have to change again.

Quoted 5 lines

Exactly my thinking. I discovered INSTALL_MOD_STRIP only recently, and I believe we should use it for the kernel instead of having our own logic. This would remove the need for special logic in Buildroot to strip kernel modules, and would leave it to the kernel build system to know how modules should be stripped in an architecture-specific way.

Sounds good. When was this introduced and what about external kernel modules?

Signature · 2 lines
Venlig hilsen,
Peter Korsgaard 
#10 lore raw

Hello,

Quoted 1 line · Peter Korsgaard

Sounds good. When was this introduced and what about external kernel modules?

Introduced in commit ac031f26e89cc04fc7504f31ae137857eb83a051, appeared first in 2.6.18.

External kernel modules: the one using kbuild will do fine, as long as we use the modules_install target of the kbuild build system and not manually install/copy the .ko files.

External kernel modules not using kbuild would obviously no longer be stripped. Not sure how many of these we have today, and how to handle them. They could be problematic :-/

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
#11 lore raw
Quoted 9 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Thomas" == Thomas Petazzoni <[email protected]> writes:

Hello,

Quoted 1 line · Peter Korsgaard

Sounds good. When was this introduced and what about external kernel modules?

Introduced in commit ac031f26e89cc04fc7504f31ae137857eb83a051, appeared first in 2.6.18.

Ok, good!

Quoted 7 lines

External kernel modules: the one using kbuild will do fine, as long as we use the modules_install target of the kbuild build system and not manually install/copy the .ko files.

External kernel modules not using kbuild would obviously no longer be stripped. Not sure how many of these we have today, and how to handle them. They could be problematic :-/

I guess we'll just have to handle those explictly whenever we find them.

Signature · 2 lines
Venlig hilsen,
Peter Korsgaard 
#12 lore raw

Hi all,

Quoted 21 lines · Peter Korsgaard
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

"Thomas" == Thomas Petazzoni <[email protected]> writes:

?> Hello,
?> On Thu, 15 Sep 2016 09:18:58 +0200, Peter Korsgaard wrote:

?>> Sounds good. When was this introduced and what about external kernel modules?

?> Introduced in commit ac031f26e89cc04fc7504f31ae137857eb83a051, appeared ?> first in 2.6.18.

Ok, good!

?> External kernel modules: the one using kbuild will do fine, as long as ?> we use the modules_install target of the kbuild build system and not ?> manually install/copy the .ko files.

Quoted 5 lines

?> External kernel modules not using kbuild would obviously no longer be ?> stripped. Not sure how many of these we have today, and how to handle ?> them. They could be problematic :-/

I guess we'll just have to handle those explictly whenever we find them.

Let me cook up a patch which uses?INSTALL_MOD_STRIP=1 then.

-Alexey