Re: [Xen-devel] [PATCH] ARM: refine compiler target architecture

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Wed, 2013-12-18 at 08:48 +0100, Andre Przywara wrote:
> But actually the ABI is orthogonal to the -march setting, right? AFAIK 
> -march just controls which instruction the compiler emits and which 
> instructions the assembler allows.

I thought they were subtly intertwined, even if technically they are
orthogonal.

> $ arm-linux-gnueabi-gcc -v
> Target: arm-linux-gnueabi
> ... --with-arch=armv4t ...
> 
> This is pretty conservative, but the compiler is fine and with 
> -march=armv7-a -mfloat-abi=hard it even generates modern code.

I think we can't force float-abi=hard unless the rest of the libraries
on the system use it? Which won't be the case on armel.

> Are there really people out there still using OABI?

I thought the Debian armel port was OABI, it seems I was mistaken,
sorry. From the native Debian armel compiler:

$ gcc -v
[...]
Target: arm-linux-gnueabi
Configured with: [...]
 --with-arch=armv4t --with-float=soft --enable-checking=release
--build=arm-linux-gnueabi --host=arm-linux-gnueabi
--target=arm-linux-gnueabi
[...]
gcc version 4.8.2 (Debian 4.8.2-7) 

https://wiki.debian.org/ArmEabiPort confirms that I was wrong.

The armhf port is EABI and v7:
$ gcc -v
[...]
Target: arm-linux-gnueabihf
Configured with: [...]
 --with-arch=armv7-a --with-fpu=vfpv3-d16 --with-float=hard
--with-mode=thumb --enable-checking=release --build=arm-linux-gnueabihf
--host=arm-linux-gnueabihf --target=arm-linux-gnueabihf
[...]
gcc version 4.6.3 (Debian 4.6.3-14) 

> Do we care?

I care about people not getting confused into think that things will
work on such systems when it will not. The bugs resulting from an OABI
vs EABI mismatch will likely be quite subtle.

I guess this means the original Debian arm (not -el nor -hf) port which
I suppose people are less likely to have these days.

> Both compilers I tested (Slackware and Debian cross) are purely EABI, 
> but have less than armv7 as their default.

I wonder if we can either add some option which will force EABI and
therefore fail to build on an OABI system (maybe march v7 is that test
given the age of OABI?) or add a configure check for struct alignment?

In any case looking back to the original patch I am now convinced that
it would be OK. I'm not sure about for 4.4 though, I'm a bit wary of
changing compiler defaults at this stage.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel