Re: [PATCH 3/4] Consolidate pread/pread64 implementations

[Adhemerval Zanella <adhemerval dot zanella at linaro dot org>]

On 05-01-2016 17:39, Mike Frysinger wrote:
> On 05 Jan 2016 17:19, Adhemerval Zanella wrote:
>> On 05-01-2016 17:11, Mike Frysinger wrote:
>>> On 05 Jan 2016 10:49, Adhemerval Zanella wrote:
>>>> On 04-01-2016 22:53, Mike Frysinger wrote:
>>>>> On 04 Jan 2016 10:42, Adhemerval Zanella wrote:
>>>>>> On 29-12-2015 15:28, Mike Frysinger wrote:
>>>>>>> On 18 Nov 2015 12:02, Adhemerval Zanella wrote:
>>>>>>>> This patch consolidates all the pread/pread64 implementation for Linux
>>>>>>>> in only one (sysdeps/unix/sysv/linux/pread.c).  It also removes the
>>>>>>>> syscall from the auto-generation using assembly macros.
>>>>>>>>
>>>>>>>> For pread{64} offset argument placement the new SYSCALL_LL{64} macro
>>>>>>>> is used.  For pread ports that do not define __NR_pread will use
>>>>>>>> __NR_pread64 and for pread64 ports that dot define __NR_pread64 will
>>>>>>>> use __NR_pread for the syscall.
>>>>>>>
>>>>>>> pretty sure you just broke sh here.  while it doesn't have the 64-bit
>>>>>>> arg alignment issue, it has a wart where it copied the kernel interface
>>>>>>> of one.  you can see it in the files you deleted:
>>>>>>
>>>>>> Indeed, my understanding is SH also requires __ASSUME_ALIGNED_REGISTER_PAIRS.
>>>>>> I will change it for the 1/4 patch part.
>>>>>
>>>>> just to be clear, SuperH (the ABI) does not have these 64-bit register
>>>>> pair requirements like ARM/EABI.  its syscall ABI for pread64/pwrite64
>>>>> has legacy cruft where there's a dummy reg before the 64-bit value.  so
>>>>> you can (and want) to use the same logic just for these two syscalls as
>>>>> the result happens to look the same.
>>>>
>>>> Right, so for SuperH kernel interface the 64-bit pair cruft is only for
>>>> pread64/pwrite64 or is it for all 64-bits arguments?
>>>
>>> just pread64/pwrite64:
>>> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/arch/sh/kernel/sys_sh32.c?h=v4.3
>>
>> sigh... so __ASSUME_ALIGNED_REGISTER_PAIRS is not the way to go on SH. I 
>> would prefer to avoid this, but since kernel interface is what it is I
>> think for SH it would be better to provide specific implementations.
>> I will change that, thanks for the information.
> 
> i think your patches are the right way to go.  what we did in uClibc was
> add a sh-specific pread/write file that had a comment and then defined
> __ASSUME_ALIGNED_REGISTER_PAIRS before including the common one.
> -mike
> 
That is exactly what I just did.