On Mon, Oct 15, 2012 at 11:33:08PM +0800, Fei Yang wrote: > 2012/10/15 Mikael Pettersson <mikpe@xxxxxxxx>: > > Yangfei (Felix) writes: > > > Hi all, > > > > > > I found that hardcoded instruction in inline asm can cause certains certain features fail to work on ARM platform due to endianness. > > > As an example, consider the following code snippet of platform_do_lowpower function from arch/arm/mach-realview/hotplug.c: > > > / * > > > * here's the WFI > > > */ > > > asm(".word 0xe320f003\n" > > > : > > > : > > > : "memory", "cc"); > > > > > > The instruction generated from this inline asm will not work on big-endian ARM platform, such as ARM BE-8 format. Instead, an exception will be generated. > > > > > > Here the code should be: > > > / * > > > * here's the WFI > > > */ > > > asm("WFI\n" > > > : > > > : > > > : "memory", "cc"); > > > > > > Seems the kernel doesn't support ARM BE-8 well. I don't know why this problem happens. > > > Can anyone tell me who owns this part? I can prepare a patch then. > > > Thanks. > > > > Questions regarding the ARM kernel should go to the linux-arm-kernel mailing list > > (see the MAINTAINERS file), with an optional cc: to the regular LKML. > > > > BE-8 is, if I recall correctly, ARMv7's broken format where code and data have > > different endianess. GAS supports an ".inst" directive which is like ".word" > > except the data is assumed to be code. This matters for disassembly, and may > > also be required for BE-8. > > > > That is, just s/.word/.inst/g above and report back if that works or not. > > -- > > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > > the body of a message to majordomo@xxxxxxxxxxxxxxx > > More majordomo info at http://vger.kernel.org/majordomo-info.html > > Please read the FAQ at http://www.tux.org/lkml/ > > > > > > Hi Mikael, > > Thanks for the reply. I modified the code as suggested and rebuilt the > kernel, cpu-hotplug feature now works on big-endian(BE-8) ARM > platform. > Since the ARM core can be configured by system software to work in > big-endian mode, it's necessary to fix this problem. And here is a > small patch : > > diff -urN linux-3.6.2/arch/arm/mach-exynos/hotplug.c > linux/arch/arm/mach-exynos/hotplug.c > --- linux-3.6.2/arch/arm/mach-exynos/hotplug.c 2012-10-13 > 04:50:59.000000000 +0800 > +++ linux/arch/arm/mach-exynos/hotplug.c 2012-10-15 23:05:44.000000000 +0800 > @@ -72,7 +72,7 @@ > /* > * here's the WFI > */ > - asm(".word 0xe320f003\n" > + asm(".inst 0xe320f003\n" The cleanest fix would simply be to build these files with appropriate modified CFLAGS (-march=armv6k or -march=armv7-a), and use the proper "wfi" mnemonic. Failing that, you could use the facilities in <asm/opcodes.h> to declare a wrapper macro for injecting this opcode (see <asm/opcodes-virt.h> for an example). However, putting custom opcodes into the assembler should only be done if it's really necessary. Nowadays, I think we can consider tools which don't understand the WFI mnemonic to be obsolete, at least for platforms which only build for v7 and above. The relevant board maintainers would need to sign off on such a change, so we don't end up breaking their builds. If any of these boards needs to build for v6K, the custom opcode might be worth it -- some people might just possibly be relying on older tools for such platforms. Cheers ---Dave > : > : > : "memory", "cc"); > diff -urN linux-3.6.2/arch/arm/mach-realview/hotplug.c > linux/arch/arm/mach-realview/hotplug.c > --- linux-3.6.2/arch/arm/mach-realview/hotplug.c 2012-10-13 > 04:50:59.000000000 +0800 > +++ linux/arch/arm/mach-realview/hotplug.c 2012-10-15 23:05:00.000000000 +0800 > @@ -66,7 +66,7 @@ > /* > * here's the WFI > */ > - asm(".word 0xe320f003\n" > + asm(".inst 0xe320f003\n" > : > : > : "memory", "cc"); > diff -urN linux-3.6.2/arch/arm/mach-shmobile/hotplug.c > linux/arch/arm/mach-shmobile/hotplug.c > --- linux-3.6.2/arch/arm/mach-shmobile/hotplug.c 2012-10-13 > 04:50:59.000000000 +0800 > +++ linux/arch/arm/mach-shmobile/hotplug.c 2012-10-15 23:05:25.000000000 +0800 > @@ -53,7 +53,7 @@ > /* > * here's the WFI > */ > - asm(".word 0xe320f003\n" > + asm(".inst 0xe320f003\n" > : > : > : "memory", "cc"); > > _______________________________________________ > linux-arm-kernel mailing list > linux-arm-kernel@xxxxxxxxxxxxxxxxxxx > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel -- To unsubscribe from this list: send the line "unsubscribe linux-assembly" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
- Follow-Ups:
- References:
- Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- From: Yangfei (Felix)
- Re: Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- From: Mikael Pettersson
- [PATCH] Re: Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- From: Fei Yang
- Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- Prev by Date: [PATCH] Re: Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- Next by Date: Re: [PATCH] Re: Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- Previous by thread: [PATCH] Re: Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- Next by thread: Re: [PATCH] Re: Hardcoded instruction causes certain features to fail on ARM platfrom due to endianness
- Index(es):