[Buildroot] [PATCH v2, 1/1] package/libcamera: link with atomic when needed

Fabrice Fontaine fontaine.fabrice at gmail.com
Mon Sep 16 11:54:39 UTC 2019


Le lun. 16 sept. 2019 à 13:40, Kieran Bingham
<kieran.bingham at ideasonboard.com> a écrit :
>
> Hi Fabrice,
>
> Thank you for the references, at least it wasn't me doing something
> distinctly wrong!
>
> On 16/09/2019 12:05, Fabrice Fontaine wrote:
> > Hi Kieran,
> > Le dim. 15 sept. 2019 à 22:45, Kieran Bingham
> > <kieran.bingham at ideasonboard.com> a écrit :
> >>
> >> Hi Thomas,
> >>
> >> On 15/09/2019 21:18, Thomas Petazzoni wrote:
> >>> On Thu,  5 Sep 2019 18:53:06 +0200
> >>> Fabrice Fontaine <fontaine.fabrice at gmail.com> wrote:
> >>>
> >>>> Fixes:
> >>>>  - http://autobuild.buildroot.org/results/1f0b8338f5f39aa86b9d432598dae2f53c5f7c84
> >>>>
> >>>> Signed-off-by: Fabrice Fontaine <fontaine.fabrice at gmail.com>
> >>>> ---
> >>>> Changes v1 -> v2 (after review of Kieran Bingham):
> >>>>  - Use an upstreamable solution
> >>>
> >>> Applied to master, thanks.
> >>
> >> Great, thanks for collecting this.
> >>
> >>> Could you send the patch upstream, after
> >>> taking into account the comments from Kieran ?
> >>
> >> This fix has already been integrated into libcamera master :
> >>
> >> https://git.linuxtv.org/libcamera.git/commit/?id=5d05418d9b53e1838692f687a6dc373dad45355c
> >>
> >>
> >> I haven't sent a version-bump patch yet, because we've got usages of
> >> O_TMPFILE in our tests, which even with making our build depend on
> >> kernel headers > 3.11 - I still saw failures in some toolchains.
> >>
> >> Do you know of a 'failsafe' way to build on all (or skip) toolchains
> >> when we make use of O_TMPFILE?
> > There is a pretty long thread about O_TMPFILE in the context of runc
> > here: https://patchwork.ozlabs.org/patch/1044933.
> > You can also find the commit log for runc's patch here:
> > https://git.buildroot.net/buildroot/commit/package/runc/?id=905e976a6af224b3ed015c46fcea2d717c155f55
>
>
> I don't really see a resolution in there except for adding a whole bunch
> of architecture specific definitions (and a hacky workaround for
> non-support) of O_TMPFILE into libcamera.
>
> Is that the expected resolution here?
>
> I really doubt libcamera could be used on old toolchains without
> O_TMPFILE support, as it requires media specific features from new
> kernels anyway.
>
> Is there a clean way to mark libcamera as not supported on /really/ old
> toolchains?
>From my understanding, there is no clean way to check for the
availability of O_TMPFILE.
However, O_TMPFILE is only used by tests and tests can be disabled
through -Dtest=false.
>
>
> >> I've tried adding:
> >>
> >> +       depends on BR2_TOOLCHAIN_HEADERS_AT_LEAST_4_10
> >> +       depends on BR2_TOOLCHAIN_HEADERS_AT_LEAST_4_0 # Still 6 failures
> >> +       depends on BR2_TOOLCHAIN_HEADERS_AT_LEAST_3_11
> >>
> >> But I still get failures related to the O_TMPFILE usage.
> >>
> >> (Yes, I know each HEADERS_AT_LEAST brings in the older dependencies as
> >> well, but I started at 3_11)
> >>
> >> Cheers
> >>
> >> Kieran
> >>
> >>
> >>> Thanks!
> >>>
> >>> Thomas
> >>>
> >>
> >> --
> >> Regards
> >> --
> >> Kieran
> > Best Regards,
> >
> > Fabrice
> >
>
> --
> Regards
> --
> Kieran
Best Regards,

Fabrice



More information about the buildroot mailing list