[Angstrom-devel] ppoll() syscall patch, was Re: [RFT] udev 141 for the unstable feed

[pHilipp Zabel]

On Fri, May 22, 2009 at 2:38 PM, Koen Kooi <k.kooi at student.utwente.nl> wrote:
> On 18-05-09 15:32, Koen Kooi wrote:
>>
>> Hi,
>>
>> More and more projects are starting to use libudev for their magic
>> (devicekit-*, connman, hal and more), but the udev we are currently
>> using (124) is too old for that.
>>
>> After a lot of fiddling and reading debian/rules and .spec files I have
>> udev 141 working well on my beagleboard.
>
> Udev >= 137 is using ppoll() which isn't implemented in the arm kernel,
> which may lead to a race condition on boot.
>
> I haven't seen such a hang on sheevaplug or beagle, but the testapp[1] is
> able to hang after only 5 iterations on my beagle.
>
> I've taken the liberty to update the proposed patch against 2.6.20 to
> 2.6.30rc, but it should apply any post-header more kernels (>=2.6.28).
> Please give this patch a try, I haven't even compile-tested it yet.

There also seems to be a bit of activity about this in the Ubuntu bug
tracker, but I can't find any discussion on the LAKML? Shouldn't this
patch be pushed upstream if it is correct?

> If this works we should consider patching all kernels with this. I'm unsure
> if we should make the glibc stubs go away to force a compile-time failure,
> any opinions on that?

I've already got the compile-time failure on magician and hx4700
(because of the age-old kernel the currently use in OE? __NR_ppoll not
defined in kernel headers --> no ppoll emulation in glibc?).

regards
Philipp