compiling 2.6.37 kernel in THUMB2 mode

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

hello all,

I am trying to compile the 2.6.37 kernel in THUMB2 mode and am
experiencing some compilation problems. Some inline assembly in some
files uses commands inappropriate for THUMB2 mode.

I wonder if the kernel is expected to compile cleanly when
CONFIG_THUMB2_KERNEL is enabled (and I am doing something wrong), or
is it know not to compile/work in Thumb2 node for any reason?

Thank you in advance for your help,

Vadim

compiling 2.6.37 kernel in THUMB2 mode

[Nicolas Pitre]

On Tue, 15 Feb 2011, vb at vsbe.com wrote:

> hello all,
> 
> I am trying to compile the 2.6.37 kernel in THUMB2 mode and am
> experiencing some compilation problems. Some inline assembly in some
> files uses commands inappropriate for THUMB2 mode.
> 
> I wonder if the kernel is expected to compile cleanly when
> CONFIG_THUMB2_KERNEL is enabled (and I am doing something wrong), or
> is it know not to compile/work in Thumb2 node for any reason?

It is known not to compile in mainline right now depending on your 
target, but fixes are slowly being merged upstream.

Dave Martin is maintaining a Git tree with all the fixes required for 
Thumb2 here (see the Thumb2 related branches):

http://git.linaro.org/gitweb?p=people/dmart/linux-2.6-arm.git;a=summary

Beware that some patches might not be final yet.


Nicolas

compiling 2.6.37 kernel in THUMB2 mode

[Dave Martin]

Hi,

On Wed, Feb 16, 2011 at 4:03 AM, Nicolas Pitre <nico@fluxnic.net> wrote:
> On Tue, 15 Feb 2011, vb at vsbe.com wrote:
>
>> hello all,
>>
>> I am trying to compile the 2.6.37 kernel in THUMB2 mode and am
>> experiencing some compilation problems. Some inline assembly in some
>> files uses commands inappropriate for THUMB2 mode.
>>
>> I wonder if the kernel is expected to compile cleanly when
>> CONFIG_THUMB2_KERNEL is enabled (and I am doing something wrong), or
>> is it know not to compile/work in Thumb2 node for any reason?
>
> It is known not to compile in mainline right now depending on your
> target, but fixes are slowly being merged upstream.
>
> Dave Martin is maintaining a Git tree with all the fixes required for
> Thumb2 here (see the Thumb2 related branches):
>
> http://git.linaro.org/gitweb?p=people/dmart/linux-2.6-arm.git;a=summary
>
> Beware that some patches might not be final yet.

Note that while the bulk of the kernel is Thumb-2 safe, there are
occasional regressions, and individual platforms and drivers could
still have problems.  For now, I've been focusing on omap--- see

git://git.linaro.org/people/dmart/linux-2.6-arm.git dirty/arm/omap-thumb2+merged

ARM: Add local symbols in relocate_kernel.S to work around gas bugs

If you're trying to support a specific board I'm unlikely to be able
to debug it for you, but feel free to ping me with questions.

Cheers
---Dave

compiling 2.6.37 kernel in THUMB2 mode

[Dave Martin]

On Wed, Feb 16, 2011 at 9:17 AM, Dave Martin <dave.martin@linaro.org> wrote:
> Hi,
>
> On Wed, Feb 16, 2011 at 4:03 AM, Nicolas Pitre <nico@fluxnic.net> wrote:
>> On Tue, 15 Feb 2011, vb at vsbe.com wrote:
>>
>>> hello all,
>>>
>>> I am trying to compile the 2.6.37 kernel in THUMB2 mode and am
>>> experiencing some compilation problems. Some inline assembly in some
>>> files uses commands inappropriate for THUMB2 mode.
>>>
>>> I wonder if the kernel is expected to compile cleanly when
>>> CONFIG_THUMB2_KERNEL is enabled (and I am doing something wrong), or
>>> is it know not to compile/work in Thumb2 node for any reason?
>>
>> It is known not to compile in mainline right now depending on your
>> target, but fixes are slowly being merged upstream.
>>
>> Dave Martin is maintaining a Git tree with all the fixes required for
>> Thumb2 here (see the Thumb2 related branches):
>>
>> http://git.linaro.org/gitweb?p=people/dmart/linux-2.6-arm.git;a=summary
>>
>> Beware that some patches might not be final yet.
>
> Note that while the bulk of the kernel is Thumb-2 safe, there are
> occasional regressions, and individual platforms and drivers could
> still have problems. ?For now, I've been focusing on omap--- see
>
> git://git.linaro.org/people/dmart/linux-2.6-arm.git dirty/arm/omap-thumb2+merged
>
> ARM: Add local symbols in relocate_kernel.S to work around gas bugs

^ Oops, forgot to explain that line: this is the patch required to
work around a specific bug in the assembler.  I'm not currently
proposing to merge this upstream, since I prefer the assembler to get
fixed...
You can grab this patch from the branch.

>
> If you're trying to support a specific board I'm unlikely to be able
> to debug it for you, but feel free to ping me with questions.
>
> Cheers
> ---Dave
>

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

On Wed, Feb 16, 2011 at 1:19 AM, Dave Martin <dave.martin@linaro.org> wrote:
>>
>> Note that while the bulk of the kernel is Thumb-2 safe, there are
>> occasional regressions, and individual platforms and drivers could
>> still have problems. ?For now, I've been focusing on omap--- see
>>
>> git://git.linaro.org/people/dmart/linux-2.6-arm.git dirty/arm/omap-thumb2+merged
>>
>> ARM: Add local symbols in relocate_kernel.S to work around gas bugs
>
> ^ Oops, forgot to explain that line: this is the patch required to
> work around a specific bug in the assembler. ?I'm not currently
> proposing to merge this upstream, since I prefer the assembler to get
> fixed...
> You can grab this patch from the branch.
>

Dave, thank you for this information. I checked out this branch and
would like to build a kernel. I don't know what omap is though, can
you please share a config file one could use to build the thumb
enabled kernel for it?

TIA,
cheers,
Vadim



>>
>> If you're trying to support a specific board I'm unlikely to be able
>> to debug it for you, but feel free to ping me with questions.
>>
>> Cheers
>> ---Dave
>>
>

compiling 2.6.37 kernel in THUMB2 mode

[Dave Martin]

Hi,

On Thu, Feb 17, 2011 at 2:02 AM,  <vb@vsbe.com> wrote:
> On Wed, Feb 16, 2011 at 1:19 AM, Dave Martin <dave.martin@linaro.org> wrote:
>>>
>>> Note that while the bulk of the kernel is Thumb-2 safe, there are
>>> occasional regressions, and individual platforms and drivers could
>>> still have problems. ?For now, I've been focusing on omap--- see
>>>
>>> git://git.linaro.org/people/dmart/linux-2.6-arm.git dirty/arm/omap-thumb2+merged
>>>
>>> ARM: Add local symbols in relocate_kernel.S to work around gas bugs
>>
>> ^ Oops, forgot to explain that line: this is the patch required to
>> work around a specific bug in the assembler. ?I'm not currently
>> proposing to merge this upstream, since I prefer the assembler to get
>> fixed...
>> You can grab this patch from the branch.
>>
>
> Dave, thank you for this information. I checked out this branch and
> would like to build a kernel. I don't know what omap is though, can
> you please share a config file one could use to build the thumb
> enabled kernel for it?

OMAP is one of TI's applications processor families.

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

On Thu, Feb 17, 2011 at 1:27 AM, Dave Martin <dave.martin@linaro.org> wrote:
> Hi,
>
> On Thu, Feb 17, 2011 at 2:02 AM, ?<vb@vsbe.com> wrote:
>> On Wed, Feb 16, 2011 at 1:19 AM, Dave Martin <dave.martin@linaro.org> wrote:
>>>>
>>>> Note that while the bulk of the kernel is Thumb-2 safe, there are
>>>> occasional regressions, and individual platforms and drivers could
>>>> still have problems. ?For now, I've been focusing on omap--- see
>>>>
>>>> git://git.linaro.org/people/dmart/linux-2.6-arm.git dirty/arm/omap-thumb2+merged
>>>>
>>>> ARM: Add local symbols in relocate_kernel.S to work around gas bugs
>>>
>>> ^ Oops, forgot to explain that line: this is the patch required to
>>> work around a specific bug in the assembler. ?I'm not currently
>>> proposing to merge this upstream, since I prefer the assembler to get
>>> fixed...
>>> You can grab this patch from the branch.
>>>
>>
>> Dave, thank you for this information. I checked out this branch and
>> would like to build a kernel. I don't know what omap is though, can
>> you please share a config file one could use to build the thumb
>> enabled kernel for it?
>
> OMAP is one of TI's applications processor families.
>
> From the developer perspective, people are using platforms like the
> Beagle or Panda boards
>
> http://www.beagleboard.org/
> http://www.pandaboard.org/
>
> But there are also many other variants, including boards made by
> third-party companies such as IGEPv2
>
> Are you targeting a specifc platform?
>

Hi Dave,

thank you again for the hints.

I am trying to build an image for a Tegra2 based platform, it s a
2.6.37 based tree with a fair amount of private additions (to be
upstreamed at some point), which does now come up and run in ARM mode.
I am trying to see if it can run in Thumb mode.

It looks like certain files need to be compiled in ARM mode even when
THUMB2_KERNEL is enabled, but I don't see this happening in 2.6.37, I
was wondering how this is done in your tree.

I tried building an OMAP image, but it does not build in Thumb mode,
and Thumb can't even be enabled for it because CPU_V6 configured in.
So I was wondering if this tree has some other working config which
does use Thumb mode. I'll look at the config files you mentioned, but
any suggestions/warnings about converting a kernel to work in Thumb
mode are welcome,

cheers,
/v

>
> There's a config file here, which should build a kernel which works on
> the Beagle or Panda board:
> http://people.linaro.org/~dmart/arm_omap-thumb2+v2_config
>
> Note that you may hit a section discard problem when the final link is
> done to create vmlinux. ?The top patch from this branch should work
> around it, but hopefully it should be fixed upstream soon:
>
> git://git.linaro.org/people/dmart/linux-2.6-arm.git
> arm/smpup-section-discard-fix
>
> Cheers
> ---Dave
>
>>
>> TIA,
>> cheers,
>> Vadim
>>
>>
>>
>>>>
>>>> If you're trying to support a specific board I'm unlikely to be able
>>>> to debug it for you, but feel free to ping me with questions.
>>>>
>>>> Cheers
>>>> ---Dave
>>>>
>>>
>>
>

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

On Thu, Feb 17, 2011 at 9:30 AM,  <vb@vsbe.com> wrote:
>
> It looks like certain files need to be compiled in ARM mode even when
> THUMB2_KERNEL is enabled, but I don't see this happening in 2.6.37, I
> was wondering how this is done in your tree.
>

ah, I think I figured this part - there is an assembler directive .arm
which overrides the command line specified mode (ARM vs THUMB)

cheers,
/v

compiling 2.6.37 kernel in THUMB2 mode

[Dave Martin]

On Thu, Feb 17, 2011 at 5:30 PM,  <vb@vsbe.com> wrote:

[...]

>
> Hi Dave,
>
> thank you again for the hints.
>
> I am trying to build an image for a Tegra2 based platform, it s a
> 2.6.37 based tree with a fair amount of private additions (to be
> upstreamed at some point), which does now come up and run in ARM mode.
> I am trying to see if it can run in Thumb mode.
>
> It looks like certain files need to be compiled in ARM mode even when
> THUMB2_KERNEL is enabled, but I don't see this happening in 2.6.37, I
> was wondering how this is done in your tree.

Generally, this can be kept to a minimum.

On OMAP, the low-level PM code calls into firmware which cannot cope
with interworking between ARM and Thumb.  The PM code on the ARM side
could be made to work with this even if it is mostly built in Thumb,
but that would be complex and not really worth the pain.

You may have this problem if you're trying to communicate with
firmware, but it's best if you port everything else to be buildable in
Thumb-2: the primary reason is that for backwards compatibility
reasons, much of the assembler code in the kernel doesn't support
ARM/Thumb interworking either.  This means that mixing ARM and Thumb
code in the kernel will lead to problems unless you're careful.

>
> I tried building an OMAP image, but it does not build in Thumb mode,
> and Thumb can't even be enabled for it because CPU_V6 configured in.
> So I was wondering if this tree has some other working config which
> does use Thumb mode. I'll look at the config files you mentioned, but
> any suggestions/warnings about converting a kernel to work in Thumb
> mode are welcome,

Assuming you started with omap2plus_defconfig, you need to remove
CONFIG_ARCH_OMAP2 from the config.  It's not possible to incorporate
support for omap2 with CONFIG_THUMB2_KERNEL enabled because ARMv6
doesn't support Thumb-2.

Here's the config I was using:
http://people.linaro.org/~dmart/arm_omap-thumb2+v2_config (based on
the linaro config-- note: you don't need to build all the modules!
Just make zImage or uImage)

Hope that helps!
---Dave

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

Hello Dave,

I had to stop working on trying to run my Tegra2 platform in Thumb-2
mode, but I am back to it.

Basically, I got it compilable, it starts up and goes through
initialization all the way to mounting the file system of the USB
stick. And this is where it gets stuck. The kernel build from the same
set of sources less my thumb related compilation fixes does not stop
there and continues to the login prompt.

I see that interrupts work (by adding some print statements showing
incrementing jiffies and basic interrupt counts), which shows that
normal<->thumb mode transitions happen just fine.

The last thing the thumb kernel prints is

Waiting for root device /dev/sda3...

but the normal kernel proceeds to mount the FS after that.

One thing which seems really weird is that the two versions (thumb and
normal) report different USB serial numbers:

boot.ok:usb 3-1.4: SerialNumber: 07A80C01174E078E
boot.th:usb 3-1.4: SerialNumber: 07AA08011D51D473

I wonder if this is an indication of some USB problems (which would
explain why the file system fails to mount), but find it hard to come
up with a theory.

I wonder if this rings a bell with anyone,

thank you in advance,

Vadim


On Fri, Feb 18, 2011 at 1:26 AM, Dave Martin <dave.martin@linaro.org> wrote:
> On Thu, Feb 17, 2011 at 5:30 PM, ?<vb@vsbe.com> wrote:
>
> [...]
>
>>
>> Hi Dave,
>>
>> thank you again for the hints.
>>
>> I am trying to build an image for a Tegra2 based platform, it s a
>> 2.6.37 based tree with a fair amount of private additions (to be
>> upstreamed at some point), which does now come up and run in ARM mode.
>> I am trying to see if it can run in Thumb mode.
>>
>> It looks like certain files need to be compiled in ARM mode even when
>> THUMB2_KERNEL is enabled, but I don't see this happening in 2.6.37, I
>> was wondering how this is done in your tree.
>
> Generally, this can be kept to a minimum.
>
> On OMAP, the low-level PM code calls into firmware which cannot cope
> with interworking between ARM and Thumb. ?The PM code on the ARM side
> could be made to work with this even if it is mostly built in Thumb,
> but that would be complex and not really worth the pain.
>
> You may have this problem if you're trying to communicate with
> firmware, but it's best if you port everything else to be buildable in
> Thumb-2: the primary reason is that for backwards compatibility
> reasons, much of the assembler code in the kernel doesn't support
> ARM/Thumb interworking either. ?This means that mixing ARM and Thumb
> code in the kernel will lead to problems unless you're careful.
>
>>
>> I tried building an OMAP image, but it does not build in Thumb mode,
>> and Thumb can't even be enabled for it because CPU_V6 configured in.
>> So I was wondering if this tree has some other working config which
>> does use Thumb mode. I'll look at the config files you mentioned, but
>> any suggestions/warnings about converting a kernel to work in Thumb
>> mode are welcome,
>
> Assuming you started with omap2plus_defconfig, you need to remove
> CONFIG_ARCH_OMAP2 from the config. ?It's not possible to incorporate
> support for omap2 with CONFIG_THUMB2_KERNEL enabled because ARMv6
> doesn't support Thumb-2.
>
> Here's the config I was using:
> http://people.linaro.org/~dmart/arm_omap-thumb2+v2_config (based on
> the linaro config-- note: you don't need to build all the modules!
> Just make zImage or uImage)
>
> Hope that helps!
> ---Dave
>

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

On Fri, Mar 11, 2011 at 2:44 PM,  <vb@vsbe.com> wrote:
>
> boot.ok:usb 3-1.4: SerialNumber: 07A80C01174E078E
> boot.th:usb 3-1.4: SerialNumber: 07AA08011D51D473
>

oh, geez, this is embarrassing - please disregard this 'clue', but th
rest of mystery remains - linux does not boot all the way...

cheers,
/vb


> I wonder if this is an indication of some USB problems (which would
> explain why the file system fails to mount), but find it hard to come
> up with a theory.
>
> I wonder if this rings a bell with anyone,
>
> thank you in advance,
>
> Vadim
>
>
> On Fri, Feb 18, 2011 at 1:26 AM, Dave Martin <dave.martin@linaro.org> wrote:
>> On Thu, Feb 17, 2011 at 5:30 PM, ?<vb@vsbe.com> wrote:
>>
>> [...]
>>
>>>
>>> Hi Dave,
>>>
>>> thank you again for the hints.
>>>
>>> I am trying to build an image for a Tegra2 based platform, it s a
>>> 2.6.37 based tree with a fair amount of private additions (to be
>>> upstreamed at some point), which does now come up and run in ARM mode.
>>> I am trying to see if it can run in Thumb mode.
>>>
>>> It looks like certain files need to be compiled in ARM mode even when
>>> THUMB2_KERNEL is enabled, but I don't see this happening in 2.6.37, I
>>> was wondering how this is done in your tree.
>>
>> Generally, this can be kept to a minimum.
>>
>> On OMAP, the low-level PM code calls into firmware which cannot cope
>> with interworking between ARM and Thumb. ?The PM code on the ARM side
>> could be made to work with this even if it is mostly built in Thumb,
>> but that would be complex and not really worth the pain.
>>
>> You may have this problem if you're trying to communicate with
>> firmware, but it's best if you port everything else to be buildable in
>> Thumb-2: the primary reason is that for backwards compatibility
>> reasons, much of the assembler code in the kernel doesn't support
>> ARM/Thumb interworking either. ?This means that mixing ARM and Thumb
>> code in the kernel will lead to problems unless you're careful.
>>
>>>
>>> I tried building an OMAP image, but it does not build in Thumb mode,
>>> and Thumb can't even be enabled for it because CPU_V6 configured in.
>>> So I was wondering if this tree has some other working config which
>>> does use Thumb mode. I'll look at the config files you mentioned, but
>>> any suggestions/warnings about converting a kernel to work in Thumb
>>> mode are welcome,
>>
>> Assuming you started with omap2plus_defconfig, you need to remove
>> CONFIG_ARCH_OMAP2 from the config. ?It's not possible to incorporate
>> support for omap2 with CONFIG_THUMB2_KERNEL enabled because ARMv6
>> doesn't support Thumb-2.
>>
>> Here's the config I was using:
>> http://people.linaro.org/~dmart/arm_omap-thumb2+v2_config (based on
>> the linaro config-- note: you don't need to build all the modules!
>> Just make zImage or uImage)
>>
>> Hope that helps!
>> ---Dave
>>
>

compiling 2.6.37 kernel in THUMB2 mode

[Dave Martin]

On Sat, Mar 12, 2011 at 12:14 AM,  <vb@vsbe.com> wrote:
> On Fri, Mar 11, 2011 at 2:44 PM, ?<vb@vsbe.com> wrote:
>>
>> boot.ok:usb 3-1.4: SerialNumber: 07A80C01174E078E
>> boot.th:usb 3-1.4: SerialNumber: 07AA08011D51D473
>>
>
> oh, geez, this is embarrassing - please disregard this 'clue', but th
> rest of mystery remains - linux does not boot all the way...

All C code "should just work", so I'd be surprised if generic driver
stacks like USB got broken (for example). The most likely cause of
problems is that there is some mach/ or plat/ code containing some
assembler which is not Thumb-compatible.

For example, the power management code in plat-omap and mach-omap2 has
some issues of this type.  This prevented successful bootup, as well
as preventing power management operations from working properly until
the code was fixed.

For a list of all the patches which could be relevant, try grepping
the history of the linaro 2.6.38 tree
(git://git.linaro.org/kernel/linux-linaro-2.6.38.git master) for
anything matching "thumb".  The non-omap specific patches are relevant
to you -- pretty much everything is upstream or in Russell's tree now.
 The omap-specific patches will be a useful illustration of the kind
of problems you might encounter with the BSP.

Hope that helps
---Dave

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

On Sun, Mar 13, 2011 at 4:08 AM, Dave Martin <dave.martin@linaro.org> wrote:
 >
> For a list of all the patches which could be relevant, try grepping
> the history of the linaro 2.6.38 tree
> (git://git.linaro.org/kernel/linux-linaro-2.6.38.git master) for
> anything matching "thumb". ?The non-omap specific patches are relevant
> to you -- pretty much everything is upstream or in Russell's tree now.
> ?The omap-specific patches will be a useful illustration of the kind
> of problems you might encounter with the BSP.
>
> Hope that helps

Thank upi, it certainly does! Just one more question: whats Russell's tree?

cheers,,
/vb
> ---Dave
>

compiling 2.6.37 kernel in THUMB2 mode

[Vincent Palatin]

> Just one more question: whats Russell's tree?

AFAIK: That one :
http://ftp.arm.linux.org.uk/pub/armlinux/kernel/git-cur/linux-2.6-arm.git

-- 
Vincent

compiling 2.6.37 kernel in THUMB2 mode

[vb at vsbe.com]

On Sun, Mar 13, 2011 at 9:17 AM, Vincent Palatin <
vincent.palatin_lkml@m4x.org> wrote:

> > Just one more question: whats Russell's tree?
>
> AFAIK: That one :
> http://ftp.arm.linux.org.uk/pub/armlinux/kernel/git-cur/linux-2.6-arm.git
>
>
this does not seem to be a pointer to a valid git repo, or am I missing
something?

cheers,
/vb




> --
> Vincent
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20110316/49eeb978/attachment-0001.html>

compiling 2.6.37 kernel in THUMB2 mode

[Vincent Palatin]

>> > Just one more question: whats Russell's tree?
>>
>> AFAIK: That one :
>> http://ftp.arm.linux.org.uk/pub/armlinux/kernel/git-cur/linux-2.6-arm.git
>>
>
> this does not seem to be a pointer to a valid git repo, or am I missing
> something?

The URL is unusual but
git clone http://ftp.arm.linux.org.uk/pub/armlinux/kernel/git-cur/linux-2.6-arm.git
works for me.

by the way this server connectivity has limited bandwidth, you would
rather clone the linus Tree and add this one as another remote :

git remote add arm
http://ftp.arm.linux.org.uk/pub/armlinux/kernel/git-cur/linux-2.6-arm.git
git remote update
git checkout -b arm-devel arm/devel

-- 
Vincent

compiling 2.6.37 kernel in THUMB2 mode

[Russell King - ARM Linux]

On Fri, Mar 18, 2011 at 07:03:06AM -0400, Vincent Palatin wrote:
> >> > Just one more question: whats Russell's tree?
> >>
> >> AFAIK: That one :
> >> http://ftp.arm.linux.org.uk/pub/armlinux/kernel/git-cur/linux-2.6-arm.git
> >>
> >
> > this does not seem to be a pointer to a valid git repo, or am I missing
> > something?
> 
> The URL is unusual but
> git clone http://ftp.arm.linux.org.uk/pub/armlinux/kernel/git-cur/linux-2.6-arm.git
> works for me.
> 
> by the way this server connectivity has limited bandwidth, you would
> rather clone the linus Tree and add this one as another remote :

No it doesn't.  It's very close to the UK backbone.

http://www.arm.linux.org.uk/developer/git-arm.php

That page includes an explaination of the issues, which has more to do
with your internet connectivity and your monthly data allowance than
the server.  Then again, maybe I'm old fashioned and the idea of pulling
450MB to gain a few KB of updates is the in-thing today.