uClibc copying kernel headers
I have been trying to work out who copied the kernel headers from toolchain into staging. It appears this is done in uclibc.mk
Looking at the makefile this seems to happen three times, first copying into $(TOOLCHAIN_DIR)/uClibc_dev/usr/include, then into $(STAGING_DIR)/usr/include and finally into $(TARGET_DIR)/usr/include (only if libc.a is on target)
All of these cases have different behaviour depending on whether ifeq ($(LINUX_HEADERS_IS_KERNEL),y) is true. if it is it just copies all th files, otherwise it picks out a specific subset. However search all the buildroot files LINUX_HEADERS_IS_KERNEL is never mentioned except in uclibc.mk
Looking back through earlier buildroot versions, Peter removed the last mention to LINUX_HEADERS_IS_KERNEL with this commit to remove 2.4.x kernel header support
http://lists.busybox.net/pipermail/buildroot/2009-February/026144.html
It appears that before this LINUX_HEADERS_IS_KERNEL was set true for 2.6.x and false for 2.4.x. It looks to me like that change broke the copying of kernel headers for 2.6.x and left code in uclibc.mk that was meant for 2.4.x headers.
I think I can just change uclibc.mk so that it acts as if LINUX_HEADERS_IS_KERNEL is set.
Does this look right?
As an aside there looks to be an error in uclibc.mk at the moment on line 429, although that is code that i think should be removed.
I will attempt to make the changes and if it all works I will submit a patch
Will
Quoted 7 lines · Will Wagner
I have been trying to work out who copied the kernel headers from toolchain into staging. It appears this is done in uclibc.mk
Looking at the makefile this seems to happen three times, first copying into $(TOOLCHAIN_DIR)/uClibc_dev/usr/include, then into $(STAGING_DIR)/usr/include and finally into $(TARGET_DIR)/usr/include (only if libc.a is on target)
Some time ago we had the kernel sources unpacked into one place and acted on this one copy. This was later on changed to have (at least) two places where unpacked kernel tree lives. Just wastes space for no good reason (and complicates stuff along the way). Just related but i guess your users would welcome the time and space savings by just having one.
Quoted 3 lines
All of these cases have different behaviour depending on whether ifeq ($(LINUX_HEADERS_IS_KERNEL),y) is true. if it is it just copies all
As you may remember older (as in pre 2.6.13 or so) kernels had to use manually sanitized headers. LINUX_HEADERS_IS_KERNEL previously ment that you could make headers_install (or however it's called) and get useable headers out of the kernel release tarball -- as opposed to headers for approx. <=2.6.12. ISTR that there was also the scenario where the kernel-headers wouldn't come from the kernel-tarball some were about to build (and run) but distributed by a third-party, and in these cases the above was also false (and potentially asking for trouble due to mismatches).
HTH,
Quoted 33 lines
th files, otherwise it picks out a specific subset. However search all the buildroot files
LINUX_HEADERS_IS_KERNELis never mentioned except in uclibc.mkLooking back through earlier buildroot versions, Peter removed the
last mention toLINUX_HEADERS_IS_KERNELwith this commit to remove
2.4.x kernel header support http://lists.busybox.net/pipermail/buildroot/2009-February/026144.htmlIt appears that before this
LINUX_HEADERS_IS_KERNELwas set true for 2.6.x and false for 2.4.x. It looks to me like that change broke the copying of kernel headers for 2.6.x and left code in uclibc.mk that was meant for 2.4.x headers.I think I can just change uclibc.mk so that it acts as if
LINUX_HEADERS_IS_KERNELis set.Does this look right?
As an aside there looks to be an error in uclibc.mk at the moment on line 429, although that is code that i think should be removed.
I will attempt to make the changes and if it all works I will submit a patch
Will
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot
Have tested out this patch on a couple of configs and it all seems to work fine.
Patch just switches behaviour to 2.6.x kernels back to way it used to be
Signature · 10 lines
------------------------------------------------------------------------ Will Wagner will_wagner at carallon.com Development Manager Office Tel: +44 (0)20 7371 2032 Carallon Ltd, Studio G20, Shepherds Building, Rockley Rd, London W14 0DA ------------------------------------------------------------------------ -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: 0011-Fix-for-improper-copying-of-kernel-headers.patch URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20100128/132983d2/attachment.diff>