add support for i586-unknown-linux-gnu #157

gnzlbg · 2017-11-14T13:27:31Z

Closes #156 .

gnzlbg · 2017-11-14T13:32:08Z

I don't have a linux machine, is there a way to get the tests to run? Or do I have to test it manually (e.g. by forking cargo and trying to compile it with cross for i586 ?)

malbarbo

It's LGTM.

malbarbo · 2017-11-14T13:35:59Z

docker/i586-unknown-linux-gnu/Dockerfile

+    gcc \
+    libc6-dev \
+    make \
+    file \


Is file dependency required?

No, this is a c&p-error from the i586 file in the stdsimd crate. EDIT: fixed.

malbarbo · 2017-11-14T13:39:54Z

docker/i586-unknown-linux-gnu/Dockerfile

@@ -0,0 +1,27 @@
+FROM ubuntu:17.04


This file is based on i686-unknown-linux-gnu, right? If the ubuntu version is kept in 12.04, the docker layers can be shared between i686-unknown-linux-gnu and i586-unknown-linux-gnu.

I don't really know my way around docker. Should I downgrade the version in this file, upgrade the version in the i686 docker file, or do I need to do something else to reuse the docker layers?

(FWIW in the stdsimd crate we use 17.04 for everything and I am trying to replace our testing infrastructure there with cross now that you are adding support here for qemu-system).

Per @japaric 's comment below I've downgraded this to 12.04.

japaric · 2017-11-14T13:45:04Z

README.md

@@ -198,6 +198,7 @@ worst, "hang" (never terminate).
 | `armv7-unknown-linux-gnueabihf`      | 2.15   | 4.6.2   | 1.0.2m  | ✓   | 2.8.0 |   ✓    |
 | `armv7-unknown-linux-musleabihf`     | 1.1.15 | 5.3.1   | N/A     |     | 2.8.0 |   ✓    |
 | `asmjs-unknown-emscripten` [4]       | 1.1.15 | 1.37.13 | N/A     | ✓   | N/A   |   ✓    |
+| `i586-unknown-linux-gnu`             | 2.15   | 4.6.2   | 1.0.2m  | ✓   | N/A   |   ✓    |


the glibc version is wrong. Ubuntu 17.04 ships with glibc 2.24.

I've downgraded the ubuntu version to 12.04, so that this should now be correct.

japaric · 2017-11-14T13:46:35Z

docker/i586-unknown-linux-gnu/Dockerfile

@@ -0,0 +1,27 @@
+FROM ubuntu:17.04


This should be as low as possible, i.e. 12.04 as the i686 target, to make the binaries produced with Cross compatible with legacy distributions like centos.

Fixed. This should make the version in the readme file correct right?

malbarbo · 2017-11-14T13:48:25Z

I don't known how to run the test in a non Linux machine... Maybe you can install Linux in a virtual machine?

Instructions to run the tests in a Linux machine:

First you need to build the docker image:

./build-docker-image.sh i586-unknown-linux-gnu

Then you can run the tests:

TRAVIS_OS_NAME=linux TARGET=i586-unknown-linux-gnu CPP=1 DYLIB=1 STD=1 OPENSSL=0.5.5 RUN=1 bash ci/script.sh

japaric · 2017-11-14T13:48:27Z

I don't have a linux machine, is there a way to get the tests to run?

I can throw this into homu as a try build but test times will be in the order of hours due to the large number of travis builders. If you temporarily comment all the other travis builds I can start a try build.

gnzlbg · 2017-11-14T13:50:28Z

@japaric I have another PR with an new ARM target. Maybe after that one passes review I can merge both PRs and then we can do a homu build of both together?

japaric · 2017-11-14T13:55:49Z

@gnzlbg Fine by me.

gnzlbg · 2017-11-14T14:38:55Z

@japaric I've merged both PR here and disabled the other builds so that homu can be run.

japaric

I just remembered:

apt-get install -y --no-install-recommends g++-arm-linux-gnueabihf

This is wrong for arm-unknown-linux-gnueabihf because that targets supports both ARMv6 and ARMv7 but g++-arm-linux-gnueabihf targets ARMv7 thus with this combination the binary will contain ARMv7 instructions and will crash when executed on a ARMv6 device. You won't detect this problem when testing under QEMU because qemu-arm emulates an ARMv7 processor by default. But quite a few people, including me, have run into this runtime crash (SIGILL) when using the Ubuntu toolchain for cross compilation.

TL;DR The C toolchain (gcc + libc) must be compiled from scratch to target ARMv6

gnzlbg · 2017-11-14T15:01:45Z

@japaric is there a way to force qemu-arm to emulate an armv6 processor ?

gnzlbg · 2017-11-14T15:47:35Z

@japaric I've removed the arm target since I don't really know how to do that without a machine where I can test it, the i586 parts should be good to go if you want to try with homu.

japaric · 2017-11-14T19:08:03Z

is there a way to force qemu-arm to emulate an armv6 processor ?

I think I tried setting a specific CPU before using QEMU_CPU but I think I didn't manage to get a SIGILL when supossedly emulating ARMv6 and running a binary that contained ARMv7-only instructions.

I don't really know how to do that without a machine where I can test it

You can probably copy paste some of the Docker build stuff from rust-lang/rust. There they use crosstool-ng to build an ARMv6 toolchain.

@homunkulus try

homunkulus · 2017-11-14T19:08:10Z

⌛ Trying commit 81f03cd with merge 7e225bd...

add support for i586-unknown-linux-gnu Closes #156 .

homunkulus · 2017-11-14T19:21:47Z

💔 Test failed - status-travis

gnzlbg · 2017-11-14T19:52:17Z

It fails to link cargo: https://travis-ci.org/japaric/cross/builds/302095949#L8670

japaric · 2017-11-14T20:01:04Z

That looks slightly familiar. IIRC that error is caused because you used a relatively new binutils / gcc when compiling the std facade but used an old binutils (ld) when linking a Rust binary against the facade. This has happened before with 32-bit targets and it was fixed upstream, perhaps this target was not fixed? I believe the solution was to downgrade the binutils version used in the docker image (in rust-lang/rust) where std is built.

gnzlbg · 2017-11-14T20:09:19Z

@japaric you mean this here, and the comment above?

https://travis-ci.org/japaric/cross/builds/302095949#L8670

It seems like that is done for i686 but not for i586.

japaric · 2017-11-14T20:12:00Z

I was referring to this linker error:

= note: /usr/bin/ld: /rust/lib/rustlib/i586-unknown-linux-gnu/lib/liballoc_jemalloc-208c05c6a7233cbb.rlib(jemalloc.pic.o): unrecognized relocation (0x2b) in section `.text.malloc_conf_init'

Specifically the "unrecognized relocation" you can probably find issues with that phrase in rust-lang/rust.

gnzlbg · 2017-11-14T20:18:00Z

Did you take a look at the link? It seems to exactly be addressing that error, but it seems to be only doing so for i686 and not i586. I'll go through the issues and see what Ican find.

gnzlbg · 2017-11-14T20:33:19Z

Ok I c&p the wrong link, this is the right one: https://github.com/rust-lang/rust/blob/master/src/ci/docker/dist-i586-gnu-i686-musl/Dockerfile#L36

japaric · 2017-11-14T20:37:26Z

Yes, that's the issue. The problem is that std is built in Ubuntu 16.04 and we are using a linker from 12.04.

ENV CFLAGS_i686_unknown_linux_musl=-Wa,-mrelax-relocations=no

This flag should also be applied to i586-gnu otherwise only i686-musl would be "supercompatible". This has to be fixed upstream.

gnzlbg · 2017-11-14T20:40:44Z

rust-lang/rust#45991

fix linking error on i586 Try to fix this linking error on i586 in cross: https://travis-ci.org/japaric/cross/builds/302095949#L8670 The problem is that `std` is built in Ubuntu 16.04 and `cross` uses a linker from 12.04. Currently this fix solves the problem for `i686-musl` making it "supercompatible", this PR applies the fix to `i586` as well. The cross PR is here: cross-rs/cross#157

gnzlbg · 2017-11-18T11:48:08Z

@homunkulus try

gnzlbg · 2017-11-18T11:52:21Z

@japaric might be worth to retry this today or tomorrow, the fix was merged yesterday in rustc :)

japaric · 2017-11-18T14:10:48Z

@homunkulus retry

homunkulus · 2017-11-18T14:10:55Z

⌛ Trying commit 81f03cd with merge 6a7bee8...

add support for i586-unknown-linux-gnu Closes #156 .

homunkulus · 2017-11-18T14:18:59Z

💔 Test failed - status-travis

japaric · 2017-11-18T14:22:17Z

linker error:

/checkout/src/liballoc_jemalloc/../jemalloc/src/jemalloc.c:797: undefined reference to `secure_getenv'

that sounds like the std facade is being built against a newer libc than the one we are using. The easiest fix would be to bump our Ubuntu version to match the one that's being used in rust-lang/rust. The fix that would make i586 binaries more compatible would be to lower the libc version used in rust-lang/rust.

gnzlbg · 2017-11-18T14:32:23Z

I've bumped the ubuntu version to 16.04 (the one used in rust-lang/rust). If that works we could try lowering it to 14.04 and see if that works.

japaric · 2017-11-18T15:05:28Z

@homunkulus try

homunkulus · 2017-11-18T15:05:34Z

⌛ Trying commit 7f3bfb6 with merge 04e7fec...

add support for i586-unknown-linux-gnu Closes #156 .

japaric · 2017-11-18T15:07:20Z

@gnzlbg you can try this PR, and any other PR, by the way. It didn't work before because the branch was in a try (failed) stage. In that stage you have to retry. I think you should be able to retry try-ed PRs but I've never tested that before.

homunkulus · 2017-11-18T15:21:10Z

☀️ Test successful - status-travis
State: approved= try=True

japaric · 2017-11-18T15:41:07Z

We shouldn't use a libc older than the one used by rust-lang/rust because there's the risk that a new libc symbol will be used in the std facade in the future and that will cause linker error due to us having an older libc without that symbol.

If you uncomment the other build jobs I'll r+ this.

gnzlbg · 2017-11-19T16:51:24Z

@japaric I've rebased but I think the versions of libc in the readme are still wrong.

Closes cross-rs#156 .

gnzlbg · 2017-11-19T16:55:36Z

@japaric so i've updated the version of glibc in the readme to 2.23 and gcc to 5.3.1 (those seem to be the values for ubuntu 16.04). IIUC the openssl version is 1.0.2m for all right?

gnzlbg · 2017-11-19T17:57:02Z

@homunkulus try

homunkulus · 2017-11-19T17:57:09Z

⌛ Trying commit ae1f3d5 with merge 999cc89...

add support for i586-unknown-linux-gnu Closes #156 .

homunkulus · 2017-11-19T20:27:19Z

☀️ Test successful - status-travis
State: approved= try=True

japaric · 2017-11-19T21:36:02Z

IIUC the openssl version is 1.0.2m for all right?

Yes. The OpenSSL is the same on all the images.

Thank you, @gnzlbg.

gnzlbg · 2017-11-21T15:17:44Z

@japaric could you do a release of this? I'd like to move stdsimd to cross asap, even if that means loosing arm-unknown-linux-gnueabihf support temporarily for the time being.

japaric · 2017-11-21T23:23:42Z

@gnzlbg v0.1.14 has been released. You'll have to wait like 1 hour until propely tagged images make it to dockerhub and the release works properly. When this build is done the new release should work.

gnzlbg · 2017-11-21T23:48:44Z

Thanks, I'll send a PR to stdsimd to start using cross probably tomorrow. I'll ping you in there.

malbarbo reviewed Nov 14, 2017

View reviewed changes

japaric suggested changes Nov 14, 2017

View reviewed changes

gnzlbg force-pushed the i586 branch from 0120b62 to a56e314 Compare November 14, 2017 13:47

japaric approved these changes Nov 14, 2017

View reviewed changes

gnzlbg force-pushed the i586 branch from a56e314 to b0cadd1 Compare November 14, 2017 14:32

gnzlbg mentioned this pull request Nov 14, 2017

add support for arm-unknown-linux-gnueabihf #158

Closed

gnzlbg force-pushed the i586 branch from b0cadd1 to 3267e22 Compare November 14, 2017 14:38

japaric suggested changes Nov 14, 2017

View reviewed changes

gnzlbg force-pushed the i586 branch from 3267e22 to 81f03cd Compare November 14, 2017 15:46

japaric pushed a commit that referenced this pull request Nov 14, 2017

Auto merge of #157 - gnzlbg:i586, r=<try>

7e225bd

add support for i586-unknown-linux-gnu Closes #156 .

gnzlbg mentioned this pull request Nov 14, 2017

fix linking error on i586 rust-lang/rust#45991

Merged

japaric pushed a commit that referenced this pull request Nov 18, 2017

Auto merge of #157 - gnzlbg:i586, r=<try>

6a7bee8

add support for i586-unknown-linux-gnu Closes #156 .

japaric pushed a commit that referenced this pull request Nov 18, 2017

Auto merge of #157 - gnzlbg:i586, r=<try>

04e7fec

add support for i586-unknown-linux-gnu Closes #156 .

gnzlbg force-pushed the i586 branch from 7f3bfb6 to ba8d878 Compare November 19, 2017 16:50

add i586-unknown-linux-gnu target

ae1f3d5

Closes cross-rs#156 .

gnzlbg force-pushed the i586 branch from ba8d878 to ae1f3d5 Compare November 19, 2017 16:54

japaric pushed a commit that referenced this pull request Nov 19, 2017

Auto merge of #157 - gnzlbg:i586, r=<try>

999cc89

add support for i586-unknown-linux-gnu Closes #156 .

japaric merged commit 91637e1 into cross-rs:master Nov 19, 2017

homunkulus mentioned this pull request Nov 19, 2017

Cross with managed volumes #131

Closed

add support for i586-unknown-linux-gnu #157

add support for i586-unknown-linux-gnu #157

Conversation

gnzlbg commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

malbarbo left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gnzlbg Nov 14, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

malbarbo commented Nov 14, 2017

japaric commented Nov 14, 2017

gnzlbg commented Nov 14, 2017 • edited Loading

japaric commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

japaric left a comment

Choose a reason for hiding this comment

gnzlbg commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

japaric commented Nov 14, 2017

homunkulus commented Nov 14, 2017

homunkulus commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

japaric commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

japaric commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

japaric commented Nov 14, 2017

gnzlbg commented Nov 14, 2017

gnzlbg commented Nov 18, 2017

gnzlbg commented Nov 18, 2017

japaric commented Nov 18, 2017

homunkulus commented Nov 18, 2017

homunkulus commented Nov 18, 2017

japaric commented Nov 18, 2017

gnzlbg commented Nov 18, 2017

japaric commented Nov 18, 2017

homunkulus commented Nov 18, 2017

japaric commented Nov 18, 2017

homunkulus commented Nov 18, 2017

japaric commented Nov 18, 2017

gnzlbg commented Nov 19, 2017

gnzlbg commented Nov 19, 2017

gnzlbg commented Nov 19, 2017

homunkulus commented Nov 19, 2017

homunkulus commented Nov 19, 2017

japaric commented Nov 19, 2017

gnzlbg commented Nov 21, 2017

japaric commented Nov 21, 2017

gnzlbg commented Nov 21, 2017

gnzlbg Nov 14, 2017 •

edited

Loading

gnzlbg commented Nov 14, 2017 •

edited

Loading