From mboxrd@z Thu Jan  1 00:00:00 1970
From: gangchen@rdamicro.com (Chen Gang)
Date: Tue, 12 May 2015 18:42:47 +0800
Subject: [PATCH 1/1] ARM:r0 is also corrupted after calling __do_div64.
In-Reply-To: <20150512100042.GN2067@n2100.arm.linux.org.uk>
References: <1431412386-1249-1-git-send-email-gangchen@rdamicro.com>
 <20150512100042.GN2067@n2100.arm.linux.org.uk>
Message-ID: <5551D927.7070306@rdamicro.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org



On 05/12/2015 06:00 PM, Russell King - ARM Linux wrote:
> On Tue, May 12, 2015 at 02:33:06PM +0800, Chen Gang wrote:
>> R0 is corrupted after calling __do_div64 and compiler is not informed
>> about this in macro __do_div_asm. If n is to be used again after this
>> macro, r0 is not reloaded and n will contain incorrect value
> This isn't going to work if we're building for big endian (__ARMEB__
> set) because then we end up telling the compiler that __rem and __n
> are both in r0, which is impossible.
We tested your suggestion , it works and it feels better. I have
sent another patch based on your suggestions.
>
>> diff --git a/arch/arm/include/asm/div64.h b/arch/arm/include/asm/div64.h
>> index a66061a..93a9985 100644
>> --- a/arch/arm/include/asm/div64.h
>> +++ b/arch/arm/include/asm/div64.h
>> @@ -34,13 +34,13 @@
>>   	register unsigned long long __n   asm("r0") = n;	\
>>   	register unsigned long long __res asm("r2");		\
>>   	register unsigned int __rem       asm(__xh);		\
> 	register unsigned int __rem_xh    asm(__xh);            \
> 	register unsigned int __rem_xl    asm(__xl);		\
>
>> -	asm(	__asmeq("%0", __xh)				\
>> +	asm(	__asmeq("%R0", __xh)				\
> 	asm(	__asmeq("%0", __xh)				\
> 		__asmeq("%1", __xl)				\
>
> ... and increment each following register %-spec by one...
>
>>   		__asmeq("%1", "r2")				\
>>   		__asmeq("%2", "r0")				\
>>   		__asmeq("%3", "r4")				\
>>   		"bl	__do_div64"				\
>> -		: "=r" (__rem), "=r" (__res)			\
>> -		: "r" (__n), "r" (__base)			\
>> +		: "=r" (__n), "=r" (__res)			\
>> +		: "0"(__n), "r" (__base)			\
> This is obviously wrong because __rem is left unset, and that's one of
> the two values this macro returns.
>
> 		: "=r" (__rem_xh), "=r" (__rem_xl), "=r" (__res) \
> 		: "r" (__n), "r" (__base)			\
>
>>   		: "ip", "lr", "cc");				\
>>   	n = __res;						\
>>   	__rem;							\
> 	__rem_xh;
>
> Basically this renames __rem to __rem_xh, and introduces a new __rem_xl
> to take the other half of the remainder which we don't use.
>