[PATCH] s390: Use long branches across object boundaries (jgh instead of jh)

Stefan Liebler stli@linux.ibm.com
Wed Nov 10 13:57:48 GMT 2021


On 09/11/2021 20:06, Carlos O'Donell wrote:
> On 11/9/21 12:50, Florian Weimer wrote:
>> Depending on the layout chosen by the linker, the 16-bit displacement
>> of the jh instruction is insufficient to reach the target label.
>>
>> Analysis of the linker failure was carried out by Nick Clifton.
>>
>> Tested on a z13 and z15, s390x-linux-gnu only.
> 
> Looks correct to me. Converting from BRC to BRCL doubles the available offset bits.
> I tested assembling a few variants and they look good to me.
> 
> Reviewed-by: Carlos O'Donell <carlos@redhat.com>
> 
>> ---
>>  sysdeps/s390/memmem-arch13.S | 2 +-
>>  sysdeps/s390/strstr-arch13.S | 2 +-
>>  2 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/sysdeps/s390/memmem-arch13.S b/sysdeps/s390/memmem-arch13.S
>> index c5c8d8c97e..58df8cdb14 100644
>> --- a/sysdeps/s390/memmem-arch13.S
>> +++ b/sysdeps/s390/memmem-arch13.S
>> @@ -41,7 +41,7 @@ ENTRY(MEMMEM_ARCH13)
>>  #  error The arch13 variant of memmem needs the z13 variant of memmem!
>>  # endif
>>  	clgfi	%r5,9
>> -	jh	MEMMEM_Z13
> 
> OK. jh is BRC (branch relative on condition, A7-M-4-RI) with a 16-bit offset.
> 
>> +	jgh	MEMMEM_Z13
> 
> OK. jgh is BRCL (branch relative on condition long, C0-M-4-RI) with a 32-bit offset.
> 
>>  
>>  	aghik	%r0,%r5,-1		/* vll needs highest index.  */
>>  	bc	4,0(%r14)		/* cc==1: return if needle-len == 0.  */
>> diff --git a/sysdeps/s390/strstr-arch13.S b/sysdeps/s390/strstr-arch13.S
>> index c7183e627c..222a6de91a 100644
>> --- a/sysdeps/s390/strstr-arch13.S
>> +++ b/sysdeps/s390/strstr-arch13.S
>> @@ -49,7 +49,7 @@ ENTRY(STRSTR_ARCH13)
>>  #  error The arch13 variant of strstr needs the z13 variant of strstr!
>>  # endif
>>  	clgfi	%r4,9
>> -	jh	STRSTR_Z13
>> +	jgh	STRSTR_Z13
> 
> Likewise.
> 
>>  
>>  	/* In case of a partial match, the vstrs instruction returns the index
>>  	   of the partial match in a vector-register.  Then we have to
>>
> 
> 

This patch is okay. Thanks for catching this.
I've also had a look into the wcsmbs implementations like
sysdeps/s390/wcscpy-vx.S where we jump to c-implementation as fallback.
There we also use jg which is brcl with all bits set in the condition mask.

Thanks,
Stefan


More information about the Libc-alpha mailing list