RFC: Use mesa gl with sgx #116
Conversation
The SGX side is the GPU userland driver plus TI's libgbm. They both provide the libraries and the header files. So with SGX one needs to use mesa-gl, not mesa. Signed-off-by: Andrey Konovalov <[email protected]>
Using mesa-gl instead of mesa as libgl and mesa provider in the SGX case is not enough. I also had to add the bbappend to remove libgbm and libgbm-dev from mesa-gl's PACKAGES as mesa-gl (with gbm removed from PACKAGECONFIG) creates zero-sized libgbm and libgbm-dev packages which confuse libgbm's do_packagedata. Signed-off-by: Andrey Konovalov <[email protected]>
|
@ndechesne discussed today at the sync. I asked Andrey to send an RFC. The layer where it should go is still TBD. |
| features = d.getVar('MACHINE_FEATURES', True).split() | ||
|
|
||
| mali_libs = d.getVar('MALI_USERLAND_LIBARIES', True); | ||
| sgx_libs = d.getVar('SGX_USERLAND_LIBARIES', True); | ||
|
|
||
| if 'mali450' in features: | ||
| provider = mali_libs.split()[0] | ||
| if target == 'mesa' or target == 'libgl': | ||
| provider = "mesa" |
There was a problem hiding this comment.
what does this specific case correspond to? when would we use mesa if mali450 is in features?
There was a problem hiding this comment.
mali450 driver is only binaries (libMali.so and the lib[gbm|EGL|GLES].so symlinks pointing to the former). So the process is essentially building mesa, and replacing the corresponding mesa binaries with the mali450's ones. So that the include and the pkgconfig files comes from mesa in the mali450 case.
The SGX driver comes with the relevant include and the *.pc files, and mesa-gl is the best fit there.
| @@ -0,0 +1 @@ | |||
| PACKAGES_remove = "${@bb.utils.contains("MACHINE_FEATURES", "sgx", "libgbm libgbm-dev", "", d)}" | |||
There was a problem hiding this comment.
maybe we should fix mesa recipe instead of this hack?
There was a problem hiding this comment.
So producing empty libgbm and libgbm-dev instead of not building them at all is a bug in mesa - not the other's inability to handle empty packages. Then yes, fixing mesa is the most correct approach. Will try this. Thanks!
I am not sure where the change done in mesa-gl_%.bbappend belongs to. Hence the RFC. Thanks