Troubles with cuda binaries on dunfell-l4t-r32.4.3

[chhg82]

Hi, I cloned the dunfell branche from git://git.yoctoproject.org/poky.git and the dunfell-l4t-r32.4.3 branche from here on ubuntu 16.04 and 18.04 and also on openSUSE 15.1. With the nvidia sdk I downloaded the cuda target drivers 16.04 and 18.04 and tried to bitbake with following build configuration:

local.conf:

MACHINE = "jetson-xavier" INHERIT += "rm_work" IMAGE_CLASSES += "image_types_tegra" IMAGE_FSTYPES = "tegraflash tar.gz" EXTRA_IMAGE_FEATURES += "package-management" NVIDIA_DEVNET_MIRROR = "/home/jupidu/Downloads/nvidia/sdkm_downloads_ubuntu1804"

CUDA_VERSION="10.2"

CUDA_BINARIES = "cuda-binaries-ubuntu1804" GCCVERSION_aarch64 = "8.%"

bblayer.conf:

POKY_BBLAYERS_CONF_VERSION = "2" BBPATH = "${TOPDIR}" BBFILES ?= "" BBLAYERS ?= "\ /home/jupidu/yocto/poky/meta \ /home/jupidu/yocto/poky/meta-poky \ /home/jupidu/yocto/poky/meta-yocto-bsp \ \ /home/jupidu/yocto/poky/meta-tegra \ /home/jupidu/yocto/poky/meta-tegra/contrib \ "

I get following errors in all my operation systems:

ERROR: /home/jupidu/yocto/poky/meta-tegra/recipes-devtools/cuda/cuda-binaries-ubuntu1804_10.2.89-1.bb: Error executing a python function in \<code>: The stack trace of python calls that resulted in this exception/failure was: File: '\<code>', lineno: 3, function: \<module> 0001:anon_20home_jupidu_yocto_poky_meta_conf_machine_include_arm_feature_arm_thumb_inc(d) 0002:anon_32home_jupidu_yocto_poky_meta_classes_patch_bbclass(d) 0003:anon_707__home_jupidu_yocto_poky_meta_classes_base_bbclass(d) 0004:anon_202home_jupidu_yocto_poky_meta_classes_reproducible_build_bbclass(d) 0005:anon_1130home_jupidu_yocto_poky_meta_classes_insane_bbclass(d) 0006:__anon_1370home_jupidu_yocto_poky_meta_classes_insane_bbclass(d) 0007:anon_252__home_jupidu_yocto_poky_meta_classes_package_bbclass(d) File: '/home/jupidu/yocto/poky/meta/classes/base.bbclass', lineno: 590, function: anon_707__home_jupidu_yocto_poky_meta_classes_base_bbclass 0586: 0587: needsrcrev = False 0588: srcuri = d.getVar('SRC_URI') 0589: for uri in srcuri.split(): 0590: (scheme, _ , path) = bb.fetch.decodeurl(uri)[:3] 0591: 0592: # HTTP/FTP use the wget fetcher 0593: if scheme in ("http", "https", "ftp"): 0594: d.appendVarFlag('do_fetch', 'depends', ' wget-native:do_populate_sysroot') File: '/home/jupidu/yocto/poky/bitbake/lib/bb/fetch2/init.py', lineno: 357, function: decodeurl 0353: """ 0354: 0355: m = re.compile('(?Ptype>[^:]*)://((?P<user>[^/;]+)@)?(?P<location[^;]+)(;(?P.*))?').match(url) 0356: if not m: *** 0357: raise MalformedUrl(url) 0358: 0359: type = m.group('type') 0360: location = m.group('location') 0361: if not location: Exception: bb.fetch2.MalformedUrl: The URL: '/home/jupidu/Downloads/nvidia/sdkm_downloads_ubuntu1604/cuda-repo-ubuntu1804-10-2-local-10.2.89-440.40_1.0-1_amd64.deb' is invalid and cannot be interpreted ERROR: Failed to parse recipe: /home/jupidu/yocto/poky/meta-tegra/recipes-devtools/cuda/cuda-binaries-ubuntu1804_10.2.89-1.bb

This error stays even with a wrong NVIDIA_DEVNET_MIRROR, but everything works fine without cuda drivers.

Any ideas? Is it a python problem? Thanks!

[madisongh]

Your NVIDIA_DEVNET_MIRROR is missing the file:// URL prefix. Try adding that, as in:

NVIDIA_DEVNET_MIRROR = "file:///home/jupidu/Downloads/nvidia/sdkm_downloads_ubuntu1804"

and the fetcher should stop complaining.

[chhg82]

Great, thank you!!! :-)