headers
John Dallaway | 26 Jun 2012 13:08
Picon

eCos arm-eabi GNU tools - test release 4.6.3-20120623

I have generated a new test release of the GNU tools for ARM targets.
The new test release avoids an issue with the length of GDB 'g' packet
replies when working with Cortex-M targets that was seen with the
previous test release. GDB is now built from 7.4.1 sources with an
M-profile patch based on current GDB sources. GCC 4.6.3 is now built
with additional multilib setup for Cortex-A9 processors.

Thank you to Jifl for the GDB work and to Ilija Kocho for driving
everything forward.

These tools are intended for testing within the eCos community. The test
builds can be downloaded from the eCos ftp site and are located under
the "gnutools" directory:

> Cygwin-hosted:
> gnutools/cygwin/test/ecos-gnutools-arm-eabi-20120623.cygwin.tar.bz2
> 
> Linux-hosted:
> gnutools/i386linux/test/ecos-gnutools-arm-eabi-20120623.i386linux.tar.bz2

Please use a mirror site in your own geographical region to access these
files. Sites which mirror the eCos ftp area are listed at:

  http://ecos.sourceware.org/mirror.html

The tools may be extracted into the current working directory from the
downloaded archive files by using one of the following commands at a
bash prompt:

  tar jxvf ecos-gnutools-arm-eabi-20120623.cygwin.tar.bz2
(Continue reading)

Permalink | Reply |
headers
Ilija Kocho | 26 Jun 2012 13:44
Picon

Fwd: [ECOS] eCos arm-eabi GNU tools - test release 4.6.3-20120623

Hi Alex

As announced by John we have new test release of the GNU tools. This
release fixes the issues found in previous test releases in both GCC and
GDB so we can continue with testing.
Therefore I would ask you to put it on eCosCentric test farm. The
details for download and installation are below.

Best regards
Ilija

-------- Original Message --------
Subject: 	[ECOS] eCos arm-eabi GNU tools - test release 4.6.3-20120623
Date: 	Tue, 26 Jun 2012 12:08:28 +0100
From: 	John Dallaway <john <at> dallaway.org.uk>
To: 	eCos Discussion <ecos-discuss <at> ecos.sourceware.org>

I have generated a new test release of the GNU tools for ARM targets.
The new test release avoids an issue with the length of GDB 'g' packet
replies when working with Cortex-M targets that was seen with the
previous test release. GDB is now built from 7.4.1 sources with an
M-profile patch based on current GDB sources. GCC 4.6.3 is now built
with additional multilib setup for Cortex-A9 processors.

Thank you to Jifl for the GDB work and to Ilija Kocho for driving
everything forward.

These tools are intended for testing within the eCos community. The test
builds can be downloaded from the eCos ftp site and are located under
the "gnutools" directory:
(Continue reading)

Permalink | Reply |
headers
Alex Schuilenburg | 27 Jul 2012 13:37
Favicon

Re: Fwd: [ECOS] eCos arm-eabi GNU tools - test release 4.6.3-20120623

Hi Ilija,

On 2012-06-26 12:44, Ilija Kocho wrote:
> Hi Alex
>
> As announced by John we have new test release of the GNU tools. This
> release fixes the issues found in previous test releases in both GCC and
> GDB so we can continue with testing.
> Therefore I would ask you to put it on eCosCentric test farm. The
> details for download and installation are below.

The 4.6.3-20120623 arm tools are now in the farm running on the stm3210e
eval board and lpc2468.  Actually, I had the lpc2468 start in the farm a
while ago but failed to monitor the results - they were all timing out
because the old anoncvs redboot I built in March 2012 is now
incompatible with the current anoncvs. Obviously RedBoot is now updated.

One big issue is that the ustl permutation fails with every test on the
lpc2468 for both arm and thumb. The tests fail to start  - they either
dont hit the cyg_test_init breakpoint or they fail with SIGBUS or, if
they do hit cyg_test_start, they crash and burn afterwards with another
SIGBUS.  I have terminated all ustl tests for the lp2468 as a result as
they simply slow things down.  Unfortunately, excluding the ustl
failures, the failure rate is currently 13 out of 484 tests run so far.
I'll post a more complete list when it has run 5000 or more tests.

The stm3210e is faring a lot better. 1113 tests run so far and only 6
failures: ustl tests that have failed so far are bvt05, bvt13, bvt17 and
sprintf2.  Both testintr and kexcept1 are failing in other perms as before.
(Continue reading)

Permalink | Reply |
headers
Ilija Kocho | 30 Jul 2012 11:38
Picon

Re: Fwd: [ECOS] eCos arm-eabi GNU tools - test release 4.6.3-20120623

Hi Alex

Thanks for report.

ustl has been upgraded recently
http://bugs.ecos.sourceware.org/show_bug.cgi?id=1001545
so we need to check if that is ustl or compiler or both.

Unfortunately I have no LPC2468, or any target other than Cortex-M handy.
I'm adding Sergei to the CC, I hope he may have some insight.

Thanks
Ilija

On 27.07.2012 13:37, Alex Schuilenburg wrote:
> Hi Ilija,
>
> On 2012-06-26 12:44, Ilija Kocho wrote:
>> Hi Alex
>>
>> As announced by John we have new test release of the GNU tools. This
>> release fixes the issues found in previous test releases in both GCC and
>> GDB so we can continue with testing.
>> Therefore I would ask you to put it on eCosCentric test farm. The
>> details for download and installation are below.
> The 4.6.3-20120623 arm tools are now in the farm running on the stm3210e
> eval board and lpc2468.  Actually, I had the lpc2468 start in the farm a
> while ago but failed to monitor the results - they were all timing out
> because the old anoncvs redboot I built in March 2012 is now
> incompatible with the current anoncvs. Obviously RedBoot is now updated.
(Continue reading)

Permalink | Reply |
headers
Sergei Gavrikov | 30 Jul 2012 12:08
Picon

Re: Fwd: [ECOS] eCos arm-eabi GNU tools - test release 4.6.3-20120623

On Mon, 30 Jul 2012, Ilija Kocho wrote:

> Hi Alex
> 
> Thanks for report.
> 
> ustl has been upgraded recently
> http://bugs.ecos.sourceware.org/show_bug.cgi?id=1001545
> so we need to check if that is ustl or compiler or both.
> 
> Unfortunately I have no LPC2468, or any target other than Cortex-M handy.
> I'm adding Sergei to the CC, I hope he may have some insight.

Hi

Ilija, unfortunately, I have no LPC2468 too. That is Uwe's Kindler BSP
for the Embedded Artists LCP2468 OEM board. Mine Olimex LPC229X targets.
I'm sorry I cannot assist here too :-( But, I reported that I have no
problems with the previous GNU toolchains builds (4.6 based) on theB
AMT7TDMI-S targets from Olimex (all non-interactive USTL tests passed
smoothly in RedBoot+GDB environment).

And what's about the below (was it applied?)

  packages/hal/arm/lpc24xx/ea2468/current/ChangeLog
  2012-04-23  Nick Garnett  <nickg <at> ecoscentric.com>

  	* include/pkgconf/mlt_arm_lpc24xx_ea2468_rom.ldi:
  	* include/pkgconf/mlt_arm_lpc24xx_ea2468_ram.ldi: Relocate fixed
	vectors to 0x40000020. Previously, the HAL_VSR_GET/SET macros
(Continue reading)

Permalink | Reply |
headers
Alex Schuilenburg | 31 Jul 2012 13:48
Favicon

Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

On 2012-07-30 11:08, Sergei Gavrikov wrote:
> [...]
>
> And what's about the below (was it applied?)
>
>   packages/hal/arm/lpc24xx/ea2468/current/ChangeLog
>   2012-04-23  Nick Garnett  <nickg <at> ecoscentric.com>
>   
>   	* include/pkgconf/mlt_arm_lpc24xx_ea2468_rom.ldi:
>   	* include/pkgconf/mlt_arm_lpc24xx_ea2468_ram.ldi: Relocate fixed
> 	vectors to 0x40000020. Previously, the HAL_VSR_GET/SET macros
> 	accessed the wrong locations and didn't have any effect; now
> 	both hardware and software have the same idea of where the VSR
> 	table is.
> 	NOTE: that this renders RAM applications built with this layout
> 	incompatible with RedBoot built with the old layout. A new
> 	RedBoot needs to be installed.

Yes, it was applied and RedBoot upgraded.

So far 4900 tests have been run on the lpc2468 (we have 2 boards) with
4035 passes and 669 failures - the rest are N/A.

The stm3210e has run 2905 tests, 2803 of which were passes and 20
failures.  Almost all the failures are the same tests as before, just in
different eCos configurations. I'll give a breakdown of these when the
run is complete.

-- Alex
(Continue reading)

Permalink | Reply |
headers
David Fernandez | 26 Jun 2012 23:50

Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

Hi John,

On 26/06/12 12:08, John Dallaway wrote:
> I have generated a new test release of the GNU tools for ARM targets.
> The new test release avoids an issue with the length of GDB 'g' packet
> replies when working with Cortex-M targets that was seen with the
> previous test release. GDB is now built from 7.4.1 sources with an
> M-profile patch based on current GDB sources.
Could you provide a link to the patch tarball? I usually build the
tool-chain, and like to try that.
>  GCC 4.6.3 is now built
> with additional multilib setup for Cortex-A9 processors.
Now that you mention multilib, I noticed that, when building gcc for
arm-eabi, cpus like cortex-m3 require you (perhaps that is my mistake)
to build it --with-cpu=cortex-m3 --with-mode=thumb (at least), so that
newlib's crt0.S gets compiled in thumb-2 mode, so that the right code is
generated when using the options -mcpu=cortex-m3 -mthumb.

I wonder why crt0.S is not compiled multiple times with all the options
required for the cpus and modes supported by the arm-eabi target... I
thought that multilib was to ensure that (although I may be wrong about
that).

Do you know something about this? The crossgcc list seems to be very
quiet, and nobody there seems to be willing to give an answer on that.

Cheers
David Fernandez

--

--
(Continue reading)

Permalink | Reply |
headers
John Dallaway | 27 Jun 2012 09:00
Picon

Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

Hi David

On 26/06/12 22:50, David Fernandez wrote:

> Could you provide a link to the patch tarball? I usually build the
> tool-chain, and like to try that.

You should find the patch tarball on your local eCos mirror site at:

  gnutools/src/test/ecos-gnutools-20120623-patches.tar.bz2

Many of the individual patches within the tarball are unchanged from
previous eCos arm-eabi GNU tools test release builds and their filenames
reflect this fact.

John Dallaway
eCos maintainer
http://www.dallaway.org.uk/john

--

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
Permalink | Reply |
headers
David Fernandez | 27 Jun 2012 13:28

Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

>> Could you provide a link to the patch tarball? I usually build the
>> tool-chain, and like to try that.
>
> You should find the patch tarball on your local eCos mirror site at:
>
>  gnutools/src/test/ecos-gnutools-20120623-patches.tar.bz2
>
Yes, it works. Great!

Do you know anything on the multilib question? My experience is that,
if you don't set the default cpu and mode appropriately, gcc will
compile newlib's crt0.S using the wrong default options, and you end
up with a crt0.o with ARM32 code where you would expect thumb or
thumb-2... I don't know if the gcc-4.6.3 sources have this problem,
but surely the gcc-4.7.1 has it, unless you know of some other
configuration option that makes this work fine.

--

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
Permalink | Reply |
headers
Ilija Kocho | 27 Jun 2012 14:08
Picon

Re: Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

On 27.06.2012 13:28, David Fernandez wrote:
>>> Could you provide a link to the patch tarball? I usually build the
>>> tool-chain, and like to try that.
>> You should find the patch tarball on your local eCos mirror site at:
>>
>>  gnutools/src/test/ecos-gnutools-20120623-patches.tar.bz2
>>
> Yes, it works. Great!
>
> Do you know anything on the multilib question? My experience is that,
> if you don't set the default cpu and mode appropriately, gcc will
> compile newlib's crt0.S using the wrong default options, and you end
> up with a crt0.o with ARM32 code where you would expect thumb or
> thumb-2... I don't know if the gcc-4.6.3 sources have this problem,
> but surely the gcc-4.7.1 has it, unless you know of some other
> configuration option that makes this work fine.
>

You will find multilibs for all architectures of interest for eCos under
gnutools/arm-eabi/lib/gcc/arm-eabi/4.6.3/ directory. This multilib
configuration is produced by gcc-4.6.3-ecos-120524_t-arm-elf.diff.

Ilija

--

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
Permalink | Reply |
headers
David Fernandez | 27 Jun 2012 18:22

Re: Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

On Wed, Jun 27, 2012 at 1:08 PM, Ilija Kocho <ilijak <at> siva.com.mk> wrote:
>> Do you know anything on the multilib question? My experience is that,
>> if you don't set the default cpu and mode appropriately, gcc will
>> compile newlib's crt0.S using the wrong default options, and you end
>> up with a crt0.o with ARM32 code where you would expect thumb or
>> thumb-2... I don't know if the gcc-4.6.3 sources have this problem,
>> but surely the gcc-4.7.1 has it, unless you know of some other
>> configuration option that makes this work fine.
>>
>
> You will find multilibs for all architectures of interest for eCos under
> gnutools/arm-eabi/lib/gcc/arm-eabi/4.6.3/ directory. This multilib
> configuration is produced by gcc-4.6.3-ecos-120524_t-arm-elf.diff.
>
>

Ilija, are you then saying that gcc does not do it right as-is, and we
need to patch it?

Does this happen only with newlib, or does it happen also with the
standard libraries?

--

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
Permalink | Reply |
headers
Stanislav Meduna | 27 Jun 2012 18:38
Favicon

Re: Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

On 27.06.2012 18:22, David Fernandez wrote:

>> You will find multilibs for all architectures of interest for eCos under
>> gnutools/arm-eabi/lib/gcc/arm-eabi/4.6.3/ directory. This multilib
>> configuration is produced by gcc-4.6.3-ecos-120524_t-arm-elf.diff.
> 
> Ilija, are you then saying that gcc does not do it right as-is, and we
> need to patch it?

FWIW I do not remember such problems - I am compiling for an ARM9
and a thumb2 with the same toolchain and it works. I did not test
4.6.3 specifically, but it worked for 4.5.something compiled
under MinGW (although with other problems) and works for 4.7
canadian cross (built on linux targeting MinGW and compiling
for eCos).

As far as I remember the abovementioned patch only tweaks what
multilibs from the zillion possible actually get build.

I can't rule out some other patch affecting it, but in case
of such problems I'd suspect some bash/autotools/make/sed/whatever
incompatibility first. Trying in a from-scratch environment
is the first step.

-- 
                                       Stano

--

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
(Continue reading)

Permalink | Reply |
headers
Ilija Kocho | 27 Jun 2012 18:49
Picon

Re: Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

On 27.06.2012 18:38, Stanislav Meduna wrote:
> On 27.06.2012 18:22, David Fernandez wrote:
>
>>> You will find multilibs for all architectures of interest for eCos under
>>> gnutools/arm-eabi/lib/gcc/arm-eabi/4.6.3/ directory. This multilib
>>> configuration is produced by gcc-4.6.3-ecos-120524_t-arm-elf.diff.
>> Ilija, are you then saying that gcc does not do it right as-is, and we
>> need to patch it?
> FWIW I do not remember such problems - I am compiling for an ARM9
> and a thumb2 with the same toolchain and it works. I did not test
> 4.6.3 specifically, but it worked for 4.5.something compiled
> under MinGW (although with other problems) and works for 4.7
> canadian cross (built on linux targeting MinGW and compiling
> for eCos).
>
> As far as I remember the abovementioned patch only tweaks what
> multilibs from the zillion possible actually get build.

Correct, eCos' patch instructs GCC builder what libs (i.e. for which
targets) are of interest for eCos.

Ilija.

--

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
Permalink | Reply |
headers
David Fernandez | 28 Jun 2012 22:04

Re: Re: Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

On 27/06/12 17:49, Ilija Kocho wrote:
> On 27.06.2012 18:38, Stanislav Meduna wrote:
>> On 27.06.2012 18:22, David Fernandez wrote:
>>
>>>> You will find multilibs for all architectures of interest for eCos under
>>>> gnutools/arm-eabi/lib/gcc/arm-eabi/4.6.3/ directory. This multilib
>>>> configuration is produced by gcc-4.6.3-ecos-120524_t-arm-elf.diff.
>>> Ilija, are you then saying that gcc does not do it right as-is, and we
>>> need to patch it?
>> FWIW I do not remember such problems - I am compiling for an ARM9
>> and a thumb2 with the same toolchain and it works. I did not test
>> 4.6.3 specifically, but it worked for 4.5.something compiled
>> under MinGW (although with other problems) and works for 4.7
>> canadian cross (built on linux targeting MinGW and compiling
>> for eCos).
>>
>> As far as I remember the abovementioned patch only tweaks what
>> multilibs from the zillion possible actually get build.
> Correct, eCos' patch instructs GCC builder what libs (i.e. for which
> targets) are of interest for eCos.
>
> Ilija.
>
Then, is it usually necessary to patch gcc or prepare its source files
in a special way (other than just using the configure command) to get
the C libraries compiled for several architectures, cpus, and modes?

--

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss
(Continue reading)

Permalink | Reply |
headers
David Fernandez | 28 Jun 2012 22:00

Re: Re: Re: eCos arm-eabi GNU tools - test release 4.6.3-20120623

On 27/06/12 17:38, Stanislav Meduna wrote:
> On 27.06.2012 18:22, David Fernandez wrote:
>
>>> You will find multilibs for all architectures of interest for eCos under
>>> gnutools/arm-eabi/lib/gcc/arm-eabi/4.6.3/ directory. This multilib
>>> configuration is produced by gcc-4.6.3-ecos-120524_t-arm-elf.diff.
>> Ilija, are you then saying that gcc does not do it right as-is, and we
>> need to patch it?
> FWIW I do not remember such problems - I am compiling for an ARM9
> and a thumb2 with the same toolchain and it works. I did not test
> 4.6.3 specifically, but it worked for 4.5.something compiled
> under MinGW (although with other problems) and works for 4.7
> canadian cross (built on linux targeting MinGW and compiling
> for eCos).

It will work with ARM9, because the default options will generate ARM32
code. But, if you compiled that gcc for newlib, then if you try to
compile code for Cortex-M3, which runs only in thumb-2 mode, or just
with -mthumb, then you will get in trouble, as your crt0.o will have
ARM32 code, unless you disable the startup files, as eCos does in
(almost) all cases.

The fact that you have to change the default options explicitly
(--with-cpu=cortex-m3 --with-mode=thumb --with-arch=armv7-m) to get
crt0.S compiled in thumb-mode, means that newlib is not getting compiled
multiple times to support all the cpus that arm-eabi-gcc allows you to
compile for.

So I wonder if that is a problem of newlib not being compliant with gcc
multilib, or if it is the gcc build system that does not work well with
(Continue reading)

Permalink | Reply |

Gmane