[systemd-devel] [PATCH] missing: Correct and extend getrandom syscall defines

[Dave Reisner]

On Thu, Oct 30, 2014 at 12:39:53PM +0100, Lennart Poettering wrote:
> On Wed, 29.10.14 21:02, Dave Reisner (d at falconindy.com) wrote:
> 
> > On Wed, Oct 29, 2014 at 11:55:29PM +0100, Lennart Poettering wrote:
> > > On Wed, 29.10.14 14:29, Cristian Rodríguez (crrodriguez at opensuse.org) wrote:
> > > 
> > > > Add syscall numbers for 32 bit x86 and arm and Correct
> > > > the system call number for x86_64 (it is 318 not 278)
> > > 
> > > Hmm? I did my testing on x86_64 3.18rc2, 278 is what worked
> > > here... Did you test 318? Where does that number come from?
> > 
> > I didn't see Cristian's patch and committed this:
> > 
> > http://cgit.freedesktop.org/systemd/systemd/commit/?id=74a550c
> 
> Humm? Did you test it with that? I tested it on my 3.18 with 278, and
> it worked fine. 
> 

Sure did. A value of 278 leads me to a unusuable nspawn binary and an
unbootable system. After my change to 318, nspawn works again, and I can
boot.

Where did you get this value from? How did you test it? My hypothesis:
your kernel API headers are from 3.17 and provide a definition of
__NR_getrandom. Therefore, this fallback was never picked up from
configure.ac in your test build.

> > Is there a reason to avoid the syscall on i386?
> 
> No, I just didn't have any system to test it with. The other numebrs
> should of course be added...

Fair enough.