RE: [PATCH 1/1] ARM: compressed/head.S: fix __nop macro rept number for ARMv7M

[Vikas MANOCHA]

Hi Patrice,

> -----Original Message-----
> From: Patrice CHOTARD
> Sent: Friday, May 19, 2017 6:41 AM
> To: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
> Cc: linux@xxxxxxxxxxxxxxx; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Christophe PRIOUZEAU
> <christophe.priouzeau@xxxxxx>; Christophe KERELLO <christophe.kerello@xxxxxx>; Vikas MANOCHA <vikas.manocha@xxxxxx>;
> Patrick DELAUNAY <patrick.delaunay@xxxxxx>
> Subject: Re: [PATCH 1/1] ARM: compressed/head.S: fix __nop macro rept number for ARMv7M
> 
> Hi Ard
> 
> On 05/18/2017 01:16 PM, Ard Biesheuvel wrote:
> > On 12 May 2017 at 08:22, Patrice CHOTARD <patrice.chotard@xxxxxx> wrote:
> >> Hi all
> >>
> >> It's a gentle reminder to not forgot this patch.
> >>
> >> Thanks
> >>
> >> Patrice
> >>
> >> On 04/25/2017 11:01 AM, patrice.chotard@xxxxxx wrote:
> >>> From: Patrice Chotard <patrice.chotard@xxxxxx>
> >>>
> >>> In zImage, _magic_sig is located at offset 0x24.
> >>> But for ARMv7M architecture which enable CPU_THUMBONLY and !EFI_STUB
> >>> kernel flag, _magic_sig is located at offset 0x14
> >>>
> >>> 00000000  00 46 00 46 00 46 00 46  00 46 00 46 00 46 0f f2
> >>> |.F.F.F.F.F.F.F.Ã|
> >>> 00000010  15 0c 60 47 18 28 6f 01  00 00 00 00 28 5c 0e 00
> >>> |..`G.(o.....(\..|
> >>> 00000020  01 02 03 04 0f 46 90 46  00 00 00 00 00 00 00 00
> >>> |.....F.F........|
> >>>
> >>> Reproduced with stm32_defconfig with XIP_KERNEL flag disabled.
> >>>
> >>> To restore correct _magic_sig offset at 0x24, we add 8 additional
> >>> "mov r0, r0" instruction.
> >>>
> >>> Signed-off-by: Patrice Chotard <patrice.chotard@xxxxxx>
> >>> ---
> >>>   arch/arm/boot/compressed/head.S | 3 ++-
> >>>   1 file changed, 2 insertions(+), 1 deletion(-)
> >>>
> >>> diff --git a/arch/arm/boot/compressed/head.S
> >>> b/arch/arm/boot/compressed/head.S index 9150f97..5d9cda3 100644
> >>> --- a/arch/arm/boot/compressed/head.S
> >>> +++ b/arch/arm/boot/compressed/head.S
> >>> @@ -127,7 +127,8 @@
> >>>    AR_CLASS(   .arm    )
> >>>   start:
> >>>                .type   start,#function
> >>> -             .rept   7
> >>> +   ARM(              .rept   7)
> >>> + THUMB(              .rept   15)
> >>>
> >
> > Could you try using W(mov) instead of mov in the definition of __nop ?
> 
> If i use W(mov) instead of mov in __nop macro, i need to add one additional "mov r0,r0" in THUMB case as following :
> 
> 
>   start:
>   		.type	start,#function
>   		.rept	7
>   		__nop
>   		.endr
>      ARM(		mov	r0, r0		)
>      ARM(		b	1f		)
> + THUMB(		mov	r0, r0		)

No need of extra THUMB instruction, compiling previous mov r0,r0 should be good for both arm & thumb :

   		__nop
   		.endr
      			mov	r0, r0		/* 32 bit for arm & 16 bit for thumb */
      ARM(		b	1f		)
     THUMB(		badr	r12, 1f		)

Cheers,
Vikas

>    THUMB(		badr	r12, 1f		)
>    THUMB(		bx	r12		)
> 
>   		.word	_magic_sig	@ Magic numbers to help the loader
>   		.word	_magic_start	@ absolute load/run zImage address
>   		.word	_magic_end	@ zImage end address
>   		.word	0x04030201	@ endianness flag
> 
> 
> Which implementation do you prefer ?
> 
> Thanks
> 
> Patrice