Re: [PATCH] ARM: cacheflush: Fix exynos build breakage on ARMv6 by using macros for ISB/DSB

[Krzysztof Kozlowski]

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<k.kozlowski@xxxxxxxxxxx>
> > > Reported-by: Mark Brown<broonie@xxxxxxxxxx>
> > > 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

--
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/