Re: update ARM platform HALs

Re: update ARM platform HALs

[John Dallaway]

Hi Bart

Bart Veer wrote:

> This patch updates all ARM platform HALs to use arm-eabi-gcc instead
> of arm-elf-gcc, and removes -mno-short-load-words from platforms that
> still used it.

Thank you for the check-in.

The eCosCentric arm-eabi pre-built toolchains do not include StrongARM
multi-libs. AFAIK, users who want to build eCos for these targets will
therefore need to either build their own arm-eabi-gcc or revert to
CYGBLD_GLOBAL_COMMAND_PREFIX == "arm-elf" and use the older arm-elf-gcc
3.2.1. Can you please let me know what the intention is here so I can
mention this in the eCos 3.0 release notes?

John Dallaway

Re: update ARM platform HALs

[Jonathan Larmour]

John Dallaway wrote:
> Hi Bart
> 
> Bart Veer wrote:
> 
> 
>>This patch updates all ARM platform HALs to use arm-eabi-gcc instead
>>of arm-elf-gcc, and removes -mno-short-load-words from platforms that
>>still used it.
> 
> 
> Thank you for the check-in.
> 
> The eCosCentric arm-eabi pre-built toolchains do not include StrongARM
> multi-libs. AFAIK, users who want to build eCos for these targets will
> therefore need to either build their own arm-eabi-gcc or revert to
> CYGBLD_GLOBAL_COMMAND_PREFIX == "arm-elf" and use the older arm-elf-gcc
> 3.2.1. Can you please let me know what the intention is here so I can
> mention this in the eCos 3.0 release notes?

This was discussed before IRL, and I believe until I hear to the contrary 
that strongarm should probably still work. It just won't be perfectly 
optimal as it will fall back to the arm7tdmi multilib (the default); but 
now actually looking at GCC sources, I think it may have only a virtually 
perceptible effect even then.

Actually there is one exception that should probably not be arm-eabi which 
is the AEB target, which is arm7di and so can't work with the new tools. 
Sorry I should have mentioned that yesterday Bart. Maybe it's worth 
putting that one back, but it may not be worth even that effort since 
no-one will be using it... I think it hasn't been sold for going on 10 
years now.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["The best things in life aren't things."]------      Opinions==mine

Re: update ARM platform HALs

[Gary Thomas]

Jonathan Larmour wrote:
> John Dallaway wrote:
>> Hi Bart
>>
>> Bart Veer wrote:
>>
>>
>>> This patch updates all ARM platform HALs to use arm-eabi-gcc instead
>>> of arm-elf-gcc, and removes -mno-short-load-words from platforms that
>>> still used it.
>>
>>
>> Thank you for the check-in.
>>
>> The eCosCentric arm-eabi pre-built toolchains do not include StrongARM
>> multi-libs. AFAIK, users who want to build eCos for these targets will
>> therefore need to either build their own arm-eabi-gcc or revert to
>> CYGBLD_GLOBAL_COMMAND_PREFIX == "arm-elf" and use the older arm-elf-gcc
>> 3.2.1. Can you please let me know what the intention is here so I can
>> mention this in the eCos 3.0 release notes?
> 
> This was discussed before IRL, and I believe until I hear to the 
> contrary that strongarm should probably still work. It just won't be 
> perfectly optimal as it will fall back to the arm7tdmi multilib (the 
> default); but now actually looking at GCC sources, I think it may have 
> only a virtually perceptible effect even then.
> 
> Actually there is one exception that should probably not be arm-eabi 
> which is the AEB target, which is arm7di and so can't work with the new 
> tools. Sorry I should have mentioned that yesterday Bart. Maybe it's 
> worth putting that one back, but it may not be worth even that effort 
> since no-one will be using it... I think it hasn't been sold for going 
> on 10 years now.

... but someone was actually asking about it in the last few months.
Better to try and not break things, if possible, no matter how old.

-- 
------------------------------------------------------------
Gary Thomas                 |  Consulting for the
MLB Associates              |    Embedded world
------------------------------------------------------------

Re: update ARM platform HALs

[Jonathan Larmour]

Gary Thomas wrote:
> Jonathan Larmour wrote:
>>
>> Actually there is one exception that should probably not be arm-eabi 
>> which is the AEB target, which is arm7di and so can't work with the 
>> new tools. Sorry I should have mentioned that yesterday Bart. Maybe 
>> it's worth putting that one back, but it may not be worth even that 
>> effort since no-one will be using it... I think it hasn't been sold 
>> for going on 10 years now.
> 
> 
> ... but someone was actually asking about it in the last few months.
> Better to try and not break things, if possible, no matter how old.

There's the answer then :-).

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["The best things in life aren't things."]------      Opinions==mine

Re: update ARM platform HALs

[John Dallaway]

Hi Jifl

Jonathan Larmour wrote:

> John Dallaway wrote:
> 
>> The eCosCentric arm-eabi pre-built toolchains do not include StrongARM
>> multi-libs. AFAIK, users who want to build eCos for these targets will
>> therefore need to either build their own arm-eabi-gcc or revert to
>> CYGBLD_GLOBAL_COMMAND_PREFIX == "arm-elf" and use the older arm-elf-gcc
>> 3.2.1. Can you please let me know what the intention is here so I can
>> mention this in the eCos 3.0 release notes?
> 
> This was discussed before IRL, and I believe until I hear to the
> contrary that strongarm should probably still work. It just won't be
> perfectly optimal as it will fall back to the arm7tdmi multilib (the
> default); but now actually looking at GCC sources, I think it may have
> only a virtually perceptible effect even then.

IIRC, Bart observed a SIGTRAP in libgcc (udivsi3?) when running up eCos
built with arm-eabi-gcc 4.3.2 on the ipaq target. So perhaps users would
be better off sticking with arm-elf-gcc 3.2.1 for StrongARM? I'm not
proposing that we spend any further time investigating the problem, just
seeking clarification as to what we should recommend to users.

John Dallaway

Re: update ARM platform HALs

[Bart Veer]

>>>>> "Jifl" == Jonathan Larmour <jifl@eCosCentric.com> writes:

    Jifl> Actually there is one exception that should probably not be
    Jifl> arm-eabi which is the AEB target, which is arm7di and so
    Jifl> can't work with the new tools. Sorry I should have mentioned
    Jifl> that yesterday Bart. Maybe it's worth putting that one back,
    Jifl> but it may not be worth even that effort since no-one will
    Jifl> be using it... I think it hasn't been sold for going on 10
    Jifl> years now.

I'll revert the AEB to use arm-elf-gcc.    

>>>>> "John" == John Dallaway <john@dallaway.org.uk> writes:

    John> IIRC, Bart observed a SIGTRAP in libgcc (udivsi3?) when
    John> running up eCos built with arm-eabi-gcc 4.3.2 on the ipaq
    John> target. So perhaps users would be better off sticking with
    John> arm-elf-gcc 3.2.1 for StrongARM? I'm not proposing that we
    John> spend any further time investigating the problem, just
    John> seeking clarification as to what we should recommend to
    John> users.

That was back in October, the toolchain has been respun several times
since then. This is something to look at once the checkins are
complete and we have moved on to testing.

Also note that there are several ARM targets which completely fail to
build at the moment because of bitrot. And no doubt some number of
non-ARM targets will fail to build. Again, this should be looked at
after the checkins.

Bart

-- 
Bart Veer                                   eCos Configuration Architect
eCosCentric Limited    The eCos experts      http://www.ecoscentric.com/
Barnwell House, Barnwell Drive, Cambridge, UK.      Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.

Re: update ARM platform HALs

[Jonathan Larmour]

John Dallaway wrote:
> Hi Jifl
> 
> Jonathan Larmour wrote:
> 
> 
>>John Dallaway wrote:
>>
>>
>>>The eCosCentric arm-eabi pre-built toolchains do not include StrongARM
>>>multi-libs. AFAIK, users who want to build eCos for these targets will
>>>therefore need to either build their own arm-eabi-gcc or revert to
>>>CYGBLD_GLOBAL_COMMAND_PREFIX == "arm-elf" and use the older arm-elf-gcc
>>>3.2.1. Can you please let me know what the intention is here so I can
>>>mention this in the eCos 3.0 release notes?
>>
>>This was discussed before IRL, and I believe until I hear to the
>>contrary that strongarm should probably still work. It just won't be
>>perfectly optimal as it will fall back to the arm7tdmi multilib (the
>>default); but now actually looking at GCC sources, I think it may have
>>only a virtually perceptible effect even then.
> 
> 
> IIRC, Bart observed a SIGTRAP in libgcc (udivsi3?) when running up eCos
> built with arm-eabi-gcc 4.3.2 on the ipaq target. So perhaps users would
> be better off sticking with arm-elf-gcc 3.2.1 for StrongARM? I'm not
> proposing that we spend any further time investigating the problem, just
> seeking clarification as to what we should recommend to users.

Well, I've had a closer look, and indeed the prebuilt tools won't just 
work after all, because, even if thumb interworking is disabled, gcc uses 
ARMv4t insns like 'bx', which aren't available on strongarm's ARMv4.

If people want to use strongarm, they can indeed either use the older 
tools, or rebuild the current tools with strongarm multilibs added.

Of course, I don't expect anyone to be using strongarm in new designs as 
they've been discontinued.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["The best things in life aren't things."]------      Opinions==mine

Re: update ARM platform HALs

[John Dallaway]

Jifl

Jonathan Larmour wrote:

> John Dallaway wrote:
>
>> Jonathan Larmour wrote:
>>
>>> John Dallaway wrote:
>>>
>>>> The eCosCentric arm-eabi pre-built toolchains do not include StrongARM
>>>> multi-libs. AFAIK, users who want to build eCos for these targets will
>>>> therefore need to either build their own arm-eabi-gcc or revert to
>>>> CYGBLD_GLOBAL_COMMAND_PREFIX == "arm-elf" and use the older arm-elf-gcc
>>>> 3.2.1. Can you please let me know what the intention is here so I can
>>>> mention this in the eCos 3.0 release notes?
>>>
>>> This was discussed before IRL, and I believe until I hear to the
>>> contrary that strongarm should probably still work. It just won't be
>>> perfectly optimal as it will fall back to the arm7tdmi multilib (the
>>> default); but now actually looking at GCC sources, I think it may have
>>> only a virtually perceptible effect even then.
>>
>> IIRC, Bart observed a SIGTRAP in libgcc (udivsi3?) when running up eCos
>> built with arm-eabi-gcc 4.3.2 on the ipaq target. So perhaps users would
>> be better off sticking with arm-elf-gcc 3.2.1 for StrongARM? I'm not
>> proposing that we spend any further time investigating the problem, just
>> seeking clarification as to what we should recommend to users.
> 
> Well, I've had a closer look, and indeed the prebuilt tools won't just
> work after all, because, even if thumb interworking is disabled, gcc
> uses ARMv4t insns like 'bx', which aren't available on strongarm's ARMv4.
> 
> If people want to use strongarm, they can indeed either use the older
> tools, or rebuild the current tools with strongarm multilibs added.

OK. Thanks for the clarification.

> Of course, I don't expect anyone to be using strongarm in new designs as
> they've been discontinued.

Yes, I appreciate that this is unlikely to affect many users. The set of
multi-libs in the pre-built arm-eabi tools seems perfectly reasonable
for eCos in 2009.

John Dallaway