| Message ID | 1312141497-13854-1-git-send-email-lacombar@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Sun, 2011-07-31 at 15:44 -0400, Aunt Tillie wrote: > This change make Kbuild honor the ARCH setting of the existing configuration, if > any .config is present so that it does not get reset if there is a conflict > with the default ARCH for the platform and the current setting. I think this is the wrong approach. There are plans to make the architecture a *real* configuration option, and that will be lovely when it happens (and changing the 32-vs-64 distinction so that it's no longer a different ARCH but just a config option is one step in that direction). But I don't think this patch is really a helpful step towards that goal; we should do that *properly* or not at all, and not give some temporary half-baked behaviour in the meantime, that people will come to depend on (and others will hate because it means they can't just take an ARM config and build it on their x86 box and have it automatically converted). And the other problem is that this is not really addressing the underlying issue, which is that we are still clinging to the legacy ARCH values for directories which don't exist in the arch/ directory any more, and which are *redundant* with the setting of CONFIG_64BIT in the config. The only problem here is that we're not using the merged ARCH=x86 by default, as most of the other 64-bit capable architectures in the kernel do — and it works just fine for all of them. If the x86 merge still hasn't been completed, four years after we deleted arch/{i386,x86_64}, then we need to complete it. Thanks for helping to root out the few esoteric things that still don't quite work right on x86; it's very useful to find them and fix them.
Hi, On Sun, Jul 31, 2011 at 4:20 PM, David Woodhouse <dwmw2@infradead.org> wrote: > On Sun, 2011-07-31 at 15:44 -0400, Aunt Tillie wrote: >> This change make Kbuild honor the ARCH setting of the existing configuration, if >> any .config is present so that it does not get reset if there is a conflict >> with the default ARCH for the platform and the current setting. > > I think this is the wrong approach. > > There are plans to make the architecture a *real* configuration option, > and that will be lovely when it happens (and changing the 32-vs-64 > distinction so that it's no longer a different ARCH but just a config > option is one step in that direction). > > But I don't think this patch is really a helpful step towards that goal; > we should do that *properly* or not at all, and not give some temporary > half-baked behaviour in the meantime, that people will come to depend on > (and others will hate because it means they can't just take an ARM > config and build it on their x86 box and have it automatically > converted). > Irrelevant. This can still be done explicitly: % make ARCH=arm defconfig [... time passes ...] % make ARCH=i386 oldconfig > And the other problem is that this is not really addressing the > underlying issue, which is that we are still clinging to the legacy ARCH > values for directories which don't exist in the arch/ directory any > more, and which are *redundant* with the setting of CONFIG_64BIT in the > config. > no. The relation between the target architecture and bitness' width is not a bijection, and treating it as such is short-sighted. > The only problem here is that we're not using the merged ARCH=x86 by > default, as most of the other 64-bit capable architectures in the kernel > do — and it works just fine for all of them. > > If the x86 merge still hasn't been completed, four years after we > deleted arch/{i386,x86_64}, then we need to complete it. Thanks for > helping to root out the few esoteric things that still don't quite work > right on x86; it's very useful to find them and fix them. > I think you totally miss the point of the patch as you keep being self-centered on x86. I am working with configuration for mips, sh, powerpc, arm and x86. Some of them are for real board, some of them are to regress-test compilers, binutils and kernel builds. Each of those config hardcode the CROSS_COMPILER string and have their own build directory. In each case, I want to be able to just run "make O=/src/obj/v3.0-arm oldnoconfig all" without having to worry about anything else. -- To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 2011-07-31 at 18:07 -0400, Arnaud Lacombe wrote: > I think you totally miss the point of the patch as you keep being > self-centered on x86. I'm focusing on x86 at the moment because x86 is the only architecture that's *broken* in this respect. The other architectures that you mention below, where they support 64-bit kernels, all work fine with only *one* ARCH= setting covering all configurations. It's only x86 where we haven't actually *finished* the merge. > I am working with configuration for mips, sh, powerpc, arm and x86. > Some of them are for real board, some of them are to regress-test > compilers, binutils and kernel builds. Each of those config hardcode > the CROSS_COMPILER string and have their own build directory. In each > case, I want to be able to just run "make O=/src/obj/v3.0-arm > oldnoconfig all" without having to worry about anything else. Yes, that's a valid but *separate* problem. FWIW I usually solve this problem with a two-line GNUmakefile: ARCH := arm include Makefile I haven't checked whether it works for out-of-source-tree builds; I bet it could be made to. I would love to see $ARCH turned into a proper configuration option.
Hi, On Sun, Jul 31, 2011 at 6:34 PM, David Woodhouse <dwmw2@infradead.org> wrote: >> I am working with configuration for mips, sh, powerpc, arm and x86. >> Some of them are for real board, some of them are to regress-test >> compilers, binutils and kernel builds. Each of those config hardcode >> the CROSS_COMPILER string and have their own build directory. In each >> case, I want to be able to just run "make O=/src/obj/v3.0-arm >> oldnoconfig all" without having to worry about anything else. > > Yes, that's a valid but *separate* problem. yes, but I never really looked at it so far, I had the occasion now, and did the patch. > FWIW I usually solve this problem with a two-line GNUmakefile: > ARCH := arm > include Makefile > this would not work as I might ends up being multiple build in parallel and I am an aficionados of `git clean -fdx' so the extra file would go away. > I haven't checked whether it works for out-of-source-tree builds; I bet > it could be made to. > > I would love to see $ARCH turned into a proper configuration option. > I do not think this would be that hard, the main issue is that it would heavily be cross-tree. - Arnaud -- To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, July 31, 2011 3:34 pm, David Woodhouse wrote: > On Sun, 2011-07-31 at 18:07 -0400, Arnaud Lacombe wrote: >> I think you totally miss the point of the patch as you keep being >> self-centered on x86. > > I'm focusing on x86 at the moment because x86 is the only architecture > that's *broken* in this respect. The other architectures that you > mention below, where they support 64-bit kernels, all work fine with > only *one* ARCH= setting covering all configurations. It's only x86 > where we haven't actually *finished* the merge. > >> I am working with configuration for mips, sh, powerpc, arm and x86. >> Some of them are for real board, some of them are to regress-test >> compilers, binutils and kernel builds. Each of those config hardcode >> the CROSS_COMPILER string and have their own build directory. In each >> case, I want to be able to just run "make O=/src/obj/v3.0-arm >> oldnoconfig all" without having to worry about anything else. > > Yes, that's a valid but *separate* problem. FWIW I usually solve this > problem with a two-line GNUmakefile: > ARCH := arm > include Makefile > > I haven't checked whether it works for out-of-source-tree builds; I bet > it could be made to. > > I would love to see $ARCH turned into a proper configuration option. I don't object in principle, but I'm expecting to have to modify/update multiple *working* scripts as a result. But of course this won't be the first time and I don't expect it to be the last time either. [signed: eeyore]
On Sun, 2011-07-31 at 19:05 -0700, Randy Dunlap wrote: > > > I would love to see $ARCH turned into a proper configuration option. > > I don't object in principle, but I'm expecting to have to modify/update > multiple *working* scripts as a result. But of course this won't be > the first time and I don't expect it to be the last time either. > [signed: eeyore] :) Yeah, that's why I'd kind of like to see it done *once*, *properly*. Arnaud's hack to read it from .config is cute, but might give behaviour that people will come to depend on in their scripts and then we take it away again.
Hi, On Mon, Aug 1, 2011 at 4:42 AM, David Woodhouse <dwmw2@infradead.org> wrote: > On Sun, 2011-07-31 at 19:05 -0700, Randy Dunlap wrote: >> >> > I would love to see $ARCH turned into a proper configuration option. >> >> I don't object in principle, but I'm expecting to have to modify/update >> multiple *working* scripts as a result. But of course this won't be >> the first time and I don't expect it to be the last time either. >> [signed: eeyore] > > :) > > Yeah, that's why I'd kind of like to see it done *once*, *properly*. > > Arnaud's hack to read it from .config is cute, but might give behaviour > that people will come to depend on in their scripts and then we take it > away again. > Instead of talking of irrelevant implementation details, what _behavior_ do you expect ? The priority I expect is (from most to least): 1. command line 2. existing configuration (.config) 3. Makefile's default 4. Kconfig default - Arnaud -- To unsubscribe from this list: send the line "unsubscribe linux-kbuild" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/Makefile b/Makefile index f676d15..bb2f1fc 100644 --- a/Makefile +++ b/Makefile @@ -192,7 +192,12 @@ SUBARCH := $(shell uname -m | sed -e s/i.86/i386/ -e s/sun4u/sparc64/ \ # Default value for CROSS_COMPILE is not to prefix executables # Note: Some architectures assign CROSS_COMPILE in their arch/*/Makefile export KBUILD_BUILDHOST := $(SUBARCH) -ARCH ?= $(SUBARCH) +ifeq ($(shell test -e .config && echo y),y) +ARCH ?= $(shell sed '/^\# Linux\/\(.*\) .* Kernel Configuration/!d; s//\1/' .config) +endif +ifeq ($(ARCH),) +ARCH := $(SUBARCH) +endif CROSS_COMPILE ?= $(CONFIG_CROSS_COMPILE:"%"=%) # Architecture as present in compile.h
This change make Kbuild honor the ARCH setting of the existing configuration, if any .config is present so that it does not get reset if there is a conflict with the default ARCH for the platform and the current setting. Signed-off-by: Arnaud Lacombe aka. Aunt Tillie <lacombar@gmail.com> --- Makefile | 7 ++++++- 1 files changed, 6 insertions(+), 1 deletions(-)