[aarch64-port-dev ] [RFC] [PATCH] replace fork syscall with equivalent clone

[Andrew Haley]

On 10/03/2013 02:55 PM, Edward Nevill wrote:
> 
> On Thu, 2013-10-03 at 16:17 +0300, Riku Voipio wrote:
>> On 3 October 2013 15:56, Andrew Haley <aph at redhat.com> wrote:
>>> On 10/03/2013 01:36 PM, Riku Voipio wrote:
>>>> The following patch is against openjdk-7 25b30 / icedtea 2.1.8 as used
>>>> in linaro openembedded overlay for building openjdk-7. I see openjdk-8
>>>> has taken a more conservative approach by defining the fork system
>>>> call number for Aarch64. While this works, I think it would be more
>>>> cleaner to use clone system call, since that is guaranteed to be
>>>> available on all linux platforms. If you disagree, then we should just
>>>> backport the openjdk-8 patch to openjdk-7.
>>>
>>> I'm trying really hard to figure out why this matters.  BTW, there was
>>> no patch attached.
>>
>> Less lines of code and less arch specific ifdefs, mainly.
>> https://git.linaro.org/gitweb?p=openembedded/meta-linaro.git;a=blob_plain;f=meta-aarch64/recipes-core/openjdk/files/icedtea-openjdk-aarch64.patch
> 
> Thanks for this. Yes I agree that the current sequence of architecture
> specific conditions is a bit of a mess and just using SYS_clone is
> neater.
> 
> However, the tip of OpenJDK8 currently does it this way and it is very
> difficult to get patches pushed into the tip of OpenJDK8.
> 
> Therefore I have just followed what was done before and defined the fork
> nos for aarch64.
> 
> We could change the aarch64 branch to just use SYS_clone and then when
> it comes time to merge with the OpenJDK8 tip review this and see if they
> will accept this. Maybe this piece of code would have already been fixed
> by such time.
> 
> Do you plan to propose a patch against the OpenJDK7 tip for these
> changes. If the changes were accepted in OpenJDK7 then accepting them in
> OpenJDK8 would be pretty automatic.
> 
> Andrew: Do you want me to make these changes (with a suitable comment)
> as an aide memoir when we come to merger with OpenJDK8 tip. Or just
> leave as is?

It's a divergence from upstream that affects all arches.  It's being done
for the sake of tidiness.  I say we revisit this at some time in the future.
I would accept a change surrounded by #ifdef AARCH64 that does not touch
existing code.

Andrew.