From mboxrd@z Thu Jan 1 00:00:00 1970 From: k.kozlowski@samsung.com (Krzysztof Kozlowski) Date: Wed, 24 Sep 2014 09:51:41 +0200 Subject: [PATCH] ARM: cacheflush: Fix exynos build breakage on ARMv6 by using macros for ISB/DSB In-Reply-To: References: <1410871979-12470-1-git-send-email-k.kozlowski@samsung.com> <54219AEC.2060204@samsung.com> Message-ID: <1411545101.4861.1.camel@AMDC1943> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On wto, 2014-09-23 at 12:54 -0400, Nicolas Pitre wrote: > On Wed, 24 Sep 2014, Kukjin Kim wrote: > > > On 09/16/14 21:52, Krzysztof Kozlowski wrote: > > > This fixes build breakage of platsmp.c if ARMv6 was chosen for compile > > > time options (e.g. by building allmodconfig): > > > > > > $ make allmodconfig > > > $ make > > > CC arch/arm/mach-exynos/platsmp.o > > > /tmp/ccdQM0Eg.s: Assembler messages: > > > /tmp/ccdQM0Eg.s:432: Error: selected processor does not support ARM mode > > > `isb ' > > > /tmp/ccdQM0Eg.s:437: Error: selected processor does not support ARM mode > > > `isb ' > > > /tmp/ccdQM0Eg.s:438: Error: selected processor does not support ARM mode > > > `dsb ' > > > make[1]: *** [arch/arm/mach-exynos/platsmp.o] Error 1 > > > > > > The error was introduced in commit "ARM: EXYNOS: Move code from > > > hotplug.c to platsmp.c". Previously code using > > > v7_exit_coherency_flush() macro was built with '-march=armv7-a' flag but > > > this flag dissapeared during the movement. > > > > > > Use isb() and dsb() macros in v7_exit_coherency_flush() so the proper > > > code will be generated for ARMv6. > > > > > > Signed-off-by: Krzysztof Kozlowski > > > Reported-by: Mark Brown > > > Link: http://www.spinics.net/lists/linux-samsung-soc/msg36790.html > > > --- > > > arch/arm/include/asm/cacheflush.h | 17 +++++++++++------ > > > 1 file changed, 11 insertions(+), 6 deletions(-) > > > > > > diff --git a/arch/arm/include/asm/cacheflush.h > > > b/arch/arm/include/asm/cacheflush.h > > > index 79ecb4f34ffb..b74eeec971c0 100644 > > > --- a/arch/arm/include/asm/cacheflush.h > > > +++ b/arch/arm/include/asm/cacheflush.h > > > @@ -464,22 +464,27 @@ static inline void __sync_cache_range_r(volatile void > > > *p, size_t size) > > > * CONFIG_FRAME_POINTER=y. ip is saved as well if ever r12-clobbering > > > * trampoline are inserted by the linker and to keep sp 64-bit aligned. > > > */ > > > -#define v7_exit_coherency_flush(level) \ > > > +#define v7_exit_coherency_flush(level) do { \ > > > asm volatile( \ > > > "stmfd sp!, {fp, ip} \n\t" \ > > > "mrc p15, 0, r0, c1, c0, 0 @ get SCTLR \n\t" \ > > > "bic r0, r0, #"__stringify(CR_C)" \n\t" \ > > > "mcr p15, 0, r0, c1, c0, 0 @ set SCTLR \n\t" \ > > > - "isb \n\t" \ > > > + : : : "r0","r1","r2","r3","r4","r5","r6","r7", \ > > > + "r9","r10","lr","memory" ); \ > > > + isb(); \ > > > + asm volatile( \ > > > "bl v7_flush_dcache_"__stringify(level)" \n\t" \ > > > "mrc p15, 0, r0, c1, c0, 1 @ get ACTLR \n\t" \ > > > "bic r0, r0, #(1<< 6) @ disable local coherency \n\t" \ > > > "mcr p15, 0, r0, c1, c0, 1 @ set ACTLR \n\t" \ > > > - "isb \n\t" \ > > > - "dsb \n\t" \ > > > - "ldmfd sp!, {fp, ip}" \ > > > : : : "r0","r1","r2","r3","r4","r5","r6","r7", \ > > > - "r9","r10","lr","memory" ) > > > + "r9","r10","lr","memory" ); \ > > > + isb(); \ > > > + dsb(); \ > > > + asm volatile( \ > > > + "ldmfd sp!, {fp, ip}" ); \ > > > + } while (0) > > > > > > int set_memory_ro(unsigned long addr, int numpages); > > > int set_memory_rw(unsigned long addr, int numpages); > > > > Hi Russell, > > > > Can you please check this patch? Unfortunately I'm not sure about this, this > > should resolve the build error though... > > I don't like this patch at all. > > The original assembly sequence was carefully written in a single segment > because it is important that the compiler does not attempt to schedule > anything in the middle. Some definitions of isb() and dsb() need a > temporary register which needs to be allocated and initialized and who > knows what the compiler might do (such as spilling to the stack which > might be fatal here). > > Instead, you should insert ".arch armv7-a" in the assembly sequence to > tell the assembler the isb and dsb instructions are going to be executed > on an ARMv7 capable CPU. > Thank you for pointing this out. I'll send fixed patch. Best regards, Krzysztof > > Nicolas