From mboxrd@z Thu Jan  1 00:00:00 1970
From: Dave Kleikamp <dave.kleikamp@oracle.com>
Date: Mon, 25 Apr 2016 21:57:17 +0000
Subject: Re: [PATCH V3 2/3] Implement byte-by-byte memory access facilitators
Message-Id: <571E92BD.5050407@oracle.com>
List-Id: <sparclinux.vger.kernel.org>
References: <1461608888-5063-3-git-send-email-dave.kleikamp@oracle.com>
In-Reply-To: <1461608888-5063-3-git-send-email-dave.kleikamp@oracle.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: sparclinux@vger.kernel.org

On 04/25/2016 03:48 PM, Sam Ravnborg wrote:
> Hi Dave.
> 
>> Sparc64 requires type-aligned memory access. Since data buffers may not
>> be properly aligned, implement a safe copy from memory per data type.
>>
>> Signed-off-by: Dave Kleikamp <dave.kleikamp@oracle.com>
>> ---
>>  defs.h |   41 +++++++++++++++++++++++++++++++++++++++++
>>  1 files changed, 41 insertions(+), 0 deletions(-)
>>
>> diff --git a/defs.h b/defs.h
>> index e3afc58..d9d4559 100644
>> --- a/defs.h
>> +++ b/defs.h
>> @@ -2187,6 +2187,45 @@ struct builtin_debug_table {
>>   *  Facilitators for pulling correctly-sized data out of a buffer at a
>>   *  known address. 
>>   */
>> +
>> +#ifdef NEED_ALIGNED_MEM_ACCESS
>> +
>> +#define DEF_LOADER(TYPE)			\
>> +static inline TYPE				\
>> +load_##TYPE (char *addr)			\
>> +{						\
>> +	TYPE ret;				\
>> +	size_t i = sizeof(TYPE);		\
>> +	while (i--)				\
>> +		((char *)&ret)[i] = addr[i];	\
>> +	return ret;				\
>> +}
>> +
>> +DEF_LOADER(int);
>> +DEF_LOADER(uint);
>> +DEF_LOADER(long);
>> +DEF_LOADER(ulong);
>> +DEF_LOADER(ulonglong);
>> +DEF_LOADER(ushort);
>> +DEF_LOADER(short);
>> +typedef void *pointer_t;
>> +DEF_LOADER(pointer_t);
>> +
>> +#define LOADER(TYPE) load_##TYPE
> A simpler model would be
> #define LOADER(TYPE, addr)                         \
> ({                                                 \
> 	TYPE ret;                                  \
> 	memcpy(&ret, (void *)addr, sizeof(TYPE));  \
> 	ret;                                       \
> })

This does simplify things a bit.

> 
> I would expect the compiler to optimize this away for archs
> that do not require aligned access.

I hesitate to to replace the macros for every architecture with this.
I'll let Dave A. make that call, but I think it can improve the
NEED_ALIGNED_MEM_ACCESS case.

> 
> Quick test:
> 
> #include <string.h>
> #include <stdio.h>
> 
> #define LOADER(TYPE, addr)                       \
> ({                                               \
>         TYPE ret;                                \
>         memcpy(&ret, (void*)addr, sizeof(TYPE)); \
>         ret;                                     \
> })
> 
> #define INT(ADDR)       LOADER(int, ADDR)
> 
> void main(void)
> {
> 	int i = 37;
> 	int j = -17;
> 
>         printf("i=%d j=%d\n", INT(&i), INT(&j));
> }
> 
> 
> 	Sam
>