Date
1 - 8 of 8
Correct way to set DRIDRIVERS and GALLIUMDRIVERS from mesa bbappend
Mike Crowe
I'm building for a specific chip and therefore don't wish to waste time and
electricity building and disk space on the target installing unwanted mesa drivers. However, mesa.inc contains: GALLIUMDRIVERS = "swrast" GALLIUMDRIVERS:x86-x32 = "" GALLIUMDRIVERS:append:x86:class-target = ",i915,iris,crocus" GALLIUMDRIVERS:append:x86-64:class-target = ",i915,iris,crocus" and mesa_21.3.0.bb contains: DRIDRIVERS ??= "" DRIDRIVERS:append:x86:class-target = ",r100,r200,nouveau,i965" DRIDRIVERS:append:x86-64:class-target = ",r100,r200,nouveau,i965" I'm unable to find a way to override these values. Using (for example): DRIDRIVERS:forcevariable = "" GALLIUMDRIVERS:forcevariable = "swrast" doesn't work because the append still happens after the forcevariable override takes effect. :( Is there a way that I can override GALLIUMDRIVERS and DRIDRIVERS with my own values for x86 and x86-64 without modifying oe-core itself? If not, should the oe-core recipe being using something like: GALLIUMDRIVERS_DEFAULT = "swrast" GALLIUMDRIVERS_DEFAULT:x86-x32 = "" GALLIUMDRIVERS_DEFAULT:append:x86:class-target = ",i915,iris,crocus" GALLIUMDRIVERS_DEFAULT:append:x86-64:class-target = ",i915,iris,crocus" GALLIUMDRIVERS ?= "${GALLIUMDRIVERS_DEFAULT}" and similar for DRIDRIVERS to support this? Thanks. Mike. |
|
Alexander Kanavin
I think you do need to modify oe-core unfortunately, like is done for other drivers: PACKAGECONFIG[etnaviv] = "" GALLIUMDRIVERS:append ="${@bb.utils.contains('PACKAGECONFIG', 'etnaviv', ',etnaviv', '', d)}" Alex I'm building for a specific chip and therefore don't wish to waste time and |
|
Mike Crowe
Hi Alex,
toggle quoted message
Show quoted text
Thanks for responding. Just to make sure I understand, you think that I should add something like: PACKAGECONFIG[i915] = "" PACKAGECONFIG[iris] = "" PACKAGECONFIG[crocus] = "" GALLIUMDRIVERS:append = "${@bb.utils.contains('PACKAGECONFIG', 'i915', ',i915', '', d)}" GALLIUMDRIVERS:append = "${@bb.utils.contains('PACKAGECONFIG', 'iris', ',iris', '', d)}" GALLIUMDRIVERS:append = "${@bb.utils.contains('PACKAGECONFIG', 'crocus', ',crocus', '', d)}" But then I need to find a way to get "i915 iris crocus" into PACKAGECONFIG for x86 and x86-64 whilst still letting others override that. The best I can come up with is to modify the existing default PACKAGECONFIG to be something like: PACKAGECONFIG_DEFAULT_FOR_TARGET:x86 = "i915 iris crocus" PACKAGECONFIG_DEFAULT_FOR_TARGET:x86-64 = "i915 iris crocus" PACKAGECONFIG:class-target ??= "${@bb.utils.filter('DISTRO_FEATURES', 'wayland vulkan', d)} \ ${@bb.utils.contains('DISTRO_FEATURES', 'opengl', 'opengl egl gles gbm dri gallium virgl', '', d)} \ ${@bb.utils.contains('DISTRO_FEATURES', 'x11 opengl', 'x11 dri3', '', d)} \ ${@bb.utils.contains('DISTRO_FEATURES', 'x11 vulkan', 'dri3', '', d)} \ elf-tls \ ${PACKAGECONFIG_DEFAULT_FOR_TARGET} \ " This will probably break anyone who currently sets their own PACKAGECONFIG for x86 and x86-64 until they add the drivers they require. This would seem to be worse than my GALLIUMDRIVERS_DEFAULT suggestion. :( Have I misunderstood you, or is there a better way? Thanks. Mike. On Wednesday 01 December 2021 at 18:21:44 +0100, Alexander Kanavin wrote:
I think you do need to modify oe-core unfortunately, like is done for other |
|
Justin Bronder
On 01/12/21 16:43 +0000, Mike Crowe via lists.openembedded.org wrote:
I'm building for a specific chip and therefore don't wish to waste time andYou can use an anonymous python function which runs after parsing. https://www.yoctoproject.org/docs/current/mega-manual/mega-manual.html#anonymous-python-functions python __anonymous () { d.setVar("DRIDRIVERS", "i965") } -- Justin Bronder |
|
Mike Crowe
On Wednesday 01 December 2021 at 14:18:19 -0500, Justin Bronder wrote:
On 01/12/21 16:43 +0000, Mike Crowe via lists.openembedded.org wrote:Thanks for the suggestion. It solves my problem. However, it seems likeI'm building for a specific chip and therefore don't wish to waste time andYou can use an anonymous python function which runs after parsing. quite a subversive way to do the sort of external customisation that recipes normally support. Thanks. Mike. |
|
Alexander Kanavin
I'd like to see specifics though to understand the use case better. What kind of target you're on, how much space will be saved by only building the driver you need and which driver is that? If the drivers are in separate .so objects, you can simply delete the ones you don't need from do_install_append. The build time savings would be very minimal. Alex On Wed, 1 Dec 2021 at 19:06, Mike Crowe <mac@...> wrote: Hi Alex, |
|
Mike Crowe
Hi Alex,
toggle quoted message
Show quoted text
It's not a case of simply deleting the so files we don't need since the mesa build system stuffs everything into a small number of files and then hard links them together. For example, the current unmodified oe-core build for qemux86-64 yields: $ ls -l tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk -rw-r--r-- 2 mac mac 7993248 Dec 2 09:13 tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk $ dpkg-deb -c tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/ -rwxr-xr-x root/root 17728664 2021-11-17 20:17 ./usr/lib/dri/crocus_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/i915_dri.so link to ./usr/lib/dri/crocus_dri.so -rwxr-xr-x root/root 15903952 2021-11-17 20:17 ./usr/lib/dri/i965_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/iris_dri.so link to ./usr/lib/dri/crocus_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/kms_swrast_dri.so link to ./usr/lib/dri/crocus_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/nouveau_vieux_dri.so link to ./usr/lib/dri/i965_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/r200_dri.so link to ./usr/lib/dri/i965_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/radeon_dri.so link to ./usr/lib/dri/i965_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/swrast_dri.so link to ./usr/lib/dri/crocus_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/virtio_gpu_dri.so link to ./usr/lib/dri/crocus_dri.so drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/share/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/share/drirc.d/ -rw-r--r-- root/root 44273 2021-11-17 20:16 ./usr/share/drirc.d/00-mesa-defaults.conf If I prune the default configuration do_install:append so I only keep swrast_dri.so then I get: $ ls -l tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk -rw-r--r-- 2 mac mac 4247362 Dec 2 15:29 tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk $ dpkg-deb -c tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/ -rwxr-xr-x root/root 17728664 2021-11-17 20:17 ./usr/lib/dri/swrast_dri.so drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/share/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/share/drirc.d/ -rw-r--r-- root/root 44273 2021-11-17 20:16 ./usr/share/drirc.d/00-mesa-defaults.conf If I do no pruning but force: DRIDRIVERS = "" GALLIUMDRIVERS = "swrast" then I get: $ ls -l tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk -rw-r--r-- 2 mac mac 2590320 Dec 2 15:15 tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk $ dpkg-deb -c tmp-glibc/deploy/ipk/core2-64/mesa-megadriver_21.3.0-r0_core2-64.ipk drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/ -rwxr-xr-x root/root 11981240 2021-11-17 20:17 ./usr/lib/dri/kms_swrast_dri.so hrwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/lib/dri/swrast_dri.so link to ./usr/lib/dri/kms_swrast_dri.so drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/share/ drwxr-xr-x root/root 0 2021-11-17 20:17 ./usr/share/drirc.d/ -rw-r--r-- root/root 44273 2021-11-17 20:16 ./usr/share/drirc.d/00-mesa-defaults.conf This is an uncompressed saving of over five megabytes (32%) and a compressed saving of 1.6MiB (39%) (assuming that the ipk compression is indicative of whatever rootfs compression is used.) I tried just crocus and just iris and there were still savings, albeit not quite as large as for swrast. (I have to admit that when I first started investigating this I misread the numbers and thought they were a factor of ten higher. Nevertheless, it's clear that there is a difference.) Justin Bronder's suggestion has given me a workaround if you don't believe that the space saving is more generally useful. Thanks. Mike. On Wednesday 01 December 2021 at 21:33:39 +0100, Alexander Kanavin wrote:
I'd like to see specifics though to understand the use case better. What |
|
Richard Purdie
On Wed, 2021-12-01 at 20:20 +0000, Mike Crowe via lists.openembedded.org wrote:
On Wednesday 01 December 2021 at 14:18:19 -0500, Justin Bronder wrote:I would like to work our how we can improve this situation, I'm just strugglingOn 01/12/21 16:43 +0000, Mike Crowe via lists.openembedded.org wrote:Thanks for the suggestion. It solves my problem. However, it seems likeI'm building for a specific chip and therefore don't wish to waste time andYou can use an anonymous python function which runs after parsing. to see what we should do :/ Cheers, Richard |
|