Created attachment 463942 [details] build.log My installation of boinc is currently failing at the src_install stage with lots of the following errors: X server found. dri2 connection failed! * ACCESS DENIED: open_wr: /dev/dri/renderD128 open("/dev/dri/card0", O_RDWR) failed: Permission denied Device open failed, aborting... F: open_wr S: deny P: /dev/dri/renderD128 A: /dev/dri/renderD128 R: /dev/dri/renderD128 C: convert /var/tmp/portage/sci-misc/boinc-7.6.33-r3/distdir/boinc.tif -resize 16x16 /var/tmp/portage/sci-misc/boinc-7.6.33-r3/work/boinc_16.png This is simply because my installation of media-gfx/imagemagick is built with the 'opencl' useflag, and of course 'convert' is trying to open the DRM devices causing the access violations.
So, I've tried patching it myself with the following: --- /usr/portage/sci-misc/boinc/boinc-7.6.33-r3.ebuild 2017-02-11 19:45:50.000000000 +0000 +++ boinc-7.6.33-r3.ebuild 2017-02-16 12:50:27.296737072 +0000 @@ -92,6 +92,25 @@ eautoreconf use X && need-wxwidgets unicode + + # For convert, when built with opencl + if use X; then + shopt -s nullglob + mesa_cards=$(echo -n /dev/dri/card* | sed 's/ /:/g') + if test -n "${mesa_cards}"; then + addpredict "${mesa_cards}" + fi + intel_cards=$(echo -n /dev/dri/render* | sed 's/ /:/g') + if test -n "${intel_cards}"; then + addpredict "${intel_cards}" + fi + ati_cards=$(echo -n /dev/ati/card* | sed 's/ /:/g') + if test -n "${ati_cards}"; then + addpredict "${ati_cards}" + fi + shopt -u nullglob + addpredict /dev/nvidiactl + fi } src_configure() { I based that on what's already used in the imagemagick ebuild, but I had to add the Intel Beignet lines (as that's what is actually used on my system). This worked and the installation has completed successfully, but I still had the following repeated messages shown: X server found. dri2 connection failed! open("/dev/dri/card0", O_RDWR) failed: Permission denied Device open failed, aborting... X server found. dri2 connection failed! open("/dev/dri/card0", O_RDWR) failed: Permission denied Device open failed, aborting... cl_get_gt_device(): error, unknown device: ffffffff I suppose actually I should have sued addwrite instead of addpredict? Would that be acceptable in an ebuild?
(In reply to Russell Knighton from comment #0) > This is simply because my installation of media-gfx/imagemagick is built > with the 'opencl' useflag, and of course 'convert' is trying to open the DRM > devices causing the access violations. Sorry for the late reply. I have had a rough private live in spring and forgot about that bug. I'll see what happens if I add the opencl USE flag to imagemagick, and how to deal with it.
I can reproduce this and am working on a solution.
Actually this is not inside the responsibility of the ebuild. At least not as far as I understand the usage of convert[opencl] in a sandboxed environment. The imagemagick ebuild generates /etc/sandbox.d/99imagemagick with: -------- ~ # cat /etc/sandbox.d/99imagemagick SANDBOX_PREDICT="/dev/nvidiactl:/dev/ati/card:/dev/dri/card" -------- If I enable my nvidia card (optimus laptop using bumblebee), then I can reproduce the error, although it is /dev/nvidia-uvm that gets complained about. If I change the sandbox prediction file to -------- ~ # cat /etc/sandbox.d/99imagemagick SANDBOX_PREDICT="/dev/nvidiactl:/dev/nvidia-uvm:/dev/ati/card:/dev/dri/card" -------- The install goes fine. I reckon the imagemagick ebuild should add the (at least two) missing devices there. According to your error, the sandbox prediction written by the imagemagick ebuild should (or might) better read: -------- ~ # cat /etc/sandbox.d/99imagemagick SANDBOX_PREDICT="/dev/nvidiactl:/dev/nvidia-uvm:/dev/ati/card:/dev/dri/card:/dev/dri/renderD128" -------- I'll open a bug for that and reference this one. The question will be, whether it is always /dev/dri/renderD128, or whether other devices will produce other device nodes?
Interestingly portage fills up the sandbox prediction to: -------- ~ # grep SANDBOX_PREDICT /home/portage/sci-misc/boinc-7.6.33-r4/temp/environment declare -x SANDBOX_PREDICT="/dev/ati/card:/dev/crypto:/dev/dri/card: /dev/dri/render:/dev/nvidia:/dev/nvidia-uvm:/dev/nvidiactl:/dev/random: /home/portage/sci-misc/boinc-7.6.33-r4/homedir:/proc/self/coredump_filter: /var/cache/fontconfig" -------- I have a /dev/dri/renderD128 device, too. But I do not get any error when trying to install using the HD chipset.
It seems the issue is already known. See bug 580208
https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=ac141b1953a532e8553b099ac341db25f45d74b8 should have fixed the issue.
*** Bug 632350 has been marked as a duplicate of this bug. ***
