Re: Bug#915333: git-annex: Illegal Instruction on armel (Fujitsu Q700 like QNAP TS-21x/TS-22x)

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: Bug#915333: git-annex: Illegal Instruction on armel (Fujitsu Q700 like QNAP TS-21x/TS-22x)

Adrian Bunk-3
Control: severity -1 serious
Control: reassign -1 ghc 8.4.4+dfsg1-1
Control: affects -1 git-annex

On Thu, Jan 31, 2019 at 08:12:17PM +0100, Bernhard Übelacker wrote:
> Hello Everyone,
> I own a qnap ts-119pII with a similar cpu.
>
> See attached file with several debugging attempts.

Thanks a lot for this.

>...
> If I read [1] right, then the UXTH instruction is just supported
> on ARMv6 or later.

Looking at the code, the bug seems to be in
https://sources.debian.org/src/ghc/8.4.4+dfsg1-2/debian/patches/llvm-arm-unknown-linux-gnueabi.patch/

ARM1136JF-S is ARM11, which is ARMv6.

arm9e would be correct here instead of arm1136jf-s.

Due to the static-only nature of the ghc ecosystem the fix would then
require a complete rebuild of all Haskell packages on armel, but with
the buildds otherwise mostly idle now this should be finished within
2-3 days.

> Kind regards,
> Bernhard
>
> [1] http://infocenter.arm.com/help/topic/com.arm.doc.dui0489i/CIHHJCFE.html

cu
Adrian

--

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

Reply | Threaded
Open this post in threaded view
|

Re: Bug#915333: git-annex: Illegal Instruction on armel (Fujitsu Q700 like QNAP TS-21x/TS-22x)

Ilias Tsitsimpis-2
On Mon, Mar 11, 2019 at 12:05PM, Adrian Bunk wrote:
> On Thu, Jan 31, 2019 at 08:12:17PM +0100, Bernhard Übelacker wrote:
> > See attached file with several debugging attempts.
>
> Looking at the code, the bug seems to be in
> https://sources.debian.org/src/ghc/8.4.4+dfsg1-2/debian/patches/llvm-arm-unknown-linux-gnueabi.patch/

Hey all,

Thank you Bernhard and Adrian for debugging this. I have opened an unblock
request for ghc to see if RT will allow us to fix this for buster.

[#928882] https://bugs.debian.org/928882

Running the `utils/llvm-targets/gen-data-layout.sh` script[1], outputs
the following triplet for arm-unknown-linux-gnueabi:

  ("arm-unknown-linux-gnueabi", ("e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64", "arm7tdmi", "+soft-float -fp-only-sp -d16 -vfp2 -vfp3 -fp16 -vfp4 -fp-armv8 -neon -crypto +strict-align"))

The ARM7TDMI processor core implements the ARM architecture v4T, so it
looks like the correct value for armel, but it would be great if someone
from the debian-arm team could verify that the rest of the flags are OK.

Hopefully, we can fix this for buster.

[1] https://sources.debian.org/src/ghc/8.4.4+dfsg1-2/utils/llvm-targets/gen-data-layout.sh

--
Ilias

Reply | Threaded
Open this post in threaded view
|

Re: Bug#915333: git-annex: Illegal Instruction on armel (Fujitsu Q700 like QNAP TS-21x/TS-22x)

Adrian Bunk-3
On Sun, May 12, 2019 at 03:16:10PM +0300, Ilias Tsitsimpis wrote:

> On Mon, Mar 11, 2019 at 12:05PM, Adrian Bunk wrote:
> > On Thu, Jan 31, 2019 at 08:12:17PM +0100, Bernhard Übelacker wrote:
> > > See attached file with several debugging attempts.
> >
> > Looking at the code, the bug seems to be in
> > https://sources.debian.org/src/ghc/8.4.4+dfsg1-2/debian/patches/llvm-arm-unknown-linux-gnueabi.patch/
>
> Hey all,
>
> Thank you Bernhard and Adrian for debugging this. I have opened an unblock
> request for ghc to see if RT will allow us to fix this for buster.
>
> [#928882] https://bugs.debian.org/928882

Thanks a lot for this.

> Running the `utils/llvm-targets/gen-data-layout.sh` script[1], outputs
> the following triplet for arm-unknown-linux-gnueabi:
>
>   ("arm-unknown-linux-gnueabi", ("e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64", "arm7tdmi", "+soft-float -fp-only-sp -d16 -vfp2 -vfp3 -fp16 -vfp4 -fp-armv8 -neon -crypto +strict-align"))
>
> The ARM7TDMI processor core implements the ARM architecture v4T, so it
> looks like the correct value for armel, but it would be great if someone
> from the debian-arm team could verify that the rest of the flags are OK.
>...

For buster the baseline of the Debian armel port was raised
from v4T to v5TE.

v4T is still fine on the armel port (only slightly slower).

For an upstreamable change arm7tdmi might be better since v4T
is the lowest with Thumb (and Thumb interworking) support.

cu
Adrian

--

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed