This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH v3 2/2] posix: Implement preadv2 and pwritev2


On Sat, Jun 3, 2017 at 6:48 AM, Zack Weinberg <zackw@panix.com> wrote:
> On 06/03/2017 07:27 AM, H.J. Lu wrote:
>> On Sat, Jun 3, 2017 at 4:23 AM, Florian Weimer <fweimer@redhat.com> wrote:
>>> On 06/03/2017 01:04 PM, H.J. Lu wrote:
>>>> On Sat, Jun 3, 2017 at 1:22 AM, Andreas Schwab <schwab@linux-m68k.org> wrote:
>>>>> On Jun 02 2017, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>>>>>
>>>>>> The x86-64 LO_HI_LONG can't be used for p{readv,writev}{64}v2.  Add a
>>>>>> new macro, LO_HI_LONG_FLAGS, to pass the off{64}_t and flags arguments.
>>>>>
>>>>> Why can't LO_HI_LONG just pass the padding unconditionally on x86_64?
>>>>>
>>>>
>>>> To avoid the unnecessary (long) (((uint64_t) (val)) >> 32).
>>>
>>> I think the question is why you can't define it like this:
>>>
>>>    (val), 0
>>>
>>> ?  Are you concerned about the additional overhead of passing that
>>> unnecessary zero at the end of the parameter list for other system
>>> calls?  Or would this result in an observable kernel interface
>>> difference and break stuff?
>>
>> My patch has
>>
>> ndex 7b8bd79..a3fe2fa 100644
>> --- a/sysdeps/unix/sysv/linux/x86_64/sysdep.h
>> +++ b/sysdeps/unix/sysv/linux/x86_64/sysdep.h
>> @@ -389,4 +389,9 @@
>>  #undef LO_HI_LONG
>>  #define LO_HI_LONG(val) (val)
>>
>> +/* Provide a macro to pass the off{64}_t and flags arguments on
>> +   p{readv,writev}{64}v2.  */
>> +#undef LO_HI_LONG_FLAGS
>> +#define LO_HI_LONG_FLAGS(val, flags) (val), 0, (flags)
>> +
>>  #endif /* linux/x86_64/sysdep.h */
>>
>> For LO_HI_LONG, it doesn't mater what the second one is.  It makes
>> no difference if -1 is passed.  Why bother with 0?
>
> If nothing else, always passing the unused argument as 0 will reduce
> confusion when people inspect these syscalls, e.g. through the debugger
> interface or the seccomp filter interface.  I think that's a sufficient
> reason to do it.  These are all I/O syscalls that can block, it's the
> wrong place to be shaving cycles.

LO_HI_LONG_FLAGS is still needed for x32, regardless what we
do with LO_HI_LONG.

-- 
H.J.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]