| Message ID | 1354557403-20386-2-git-send-email-will.deacon@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 12/03/2012 11:56 AM, Will Deacon wrote: > Recent ARMv7 toolchains assume that unaligned memory accesses will not > fault and will instead be handled by the processor. > > For the nommu case (without an MPU), memory will be treated as > strongly-ordered and therefore unaligned accesses may fault regardless > of the SCTLR.A setting. > > This patch passes -mno-unaligned-access to GCC when compiling for nommu > targets, preventing the generation of unaligned memory access in the > kernel. BTW, this option will not fully prevent unaligned accesses. This code will still generate unaligned accesses: void func(void) { char array[] = { 1, 2, 3 }; } u-boot has been fixing these up by making the arrays static. Rob > > Signed-off-by: Will Deacon <will.deacon@arm.com> > --- > arch/arm/Makefile | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/arch/arm/Makefile b/arch/arm/Makefile > index 5f914fc..03363e2 100644 > --- a/arch/arm/Makefile > +++ b/arch/arm/Makefile > @@ -32,6 +32,7 @@ KBUILD_DEFCONFIG := versatile_defconfig > # defines filename extension depending memory management type. > ifeq ($(CONFIG_MMU),) > MMUEXT := -nommu > +KBUILD_CFLAGS += $(call cc-option,-mno-unaligned-access) > endif > > ifeq ($(CONFIG_FRAME_POINTER),y) >
On Mon, 3 Dec 2012, Rob Herring wrote: > On 12/03/2012 11:56 AM, Will Deacon wrote: > > Recent ARMv7 toolchains assume that unaligned memory accesses will not > > fault and will instead be handled by the processor. > > > > For the nommu case (without an MPU), memory will be treated as > > strongly-ordered and therefore unaligned accesses may fault regardless > > of the SCTLR.A setting. > > > > This patch passes -mno-unaligned-access to GCC when compiling for nommu > > targets, preventing the generation of unaligned memory access in the > > kernel. > > BTW, this option will not fully prevent unaligned accesses. This code > will still generate unaligned accesses: > > void func(void) > { > char array[] = { 1, 2, 3 }; > } Isn't that a gcc bug? Nicolas
diff --git a/arch/arm/Makefile b/arch/arm/Makefile index 5f914fc..03363e2 100644 --- a/arch/arm/Makefile +++ b/arch/arm/Makefile @@ -32,6 +32,7 @@ KBUILD_DEFCONFIG := versatile_defconfig # defines filename extension depending memory management type. ifeq ($(CONFIG_MMU),) MMUEXT := -nommu +KBUILD_CFLAGS += $(call cc-option,-mno-unaligned-access) endif ifeq ($(CONFIG_FRAME_POINTER),y)
Recent ARMv7 toolchains assume that unaligned memory accesses will not fault and will instead be handled by the processor. For the nommu case (without an MPU), memory will be treated as strongly-ordered and therefore unaligned accesses may fault regardless of the SCTLR.A setting. This patch passes -mno-unaligned-access to GCC when compiling for nommu targets, preventing the generation of unaligned memory access in the kernel. Signed-off-by: Will Deacon <will.deacon@arm.com> --- arch/arm/Makefile | 1 + 1 file changed, 1 insertion(+)