Re: [libmicrohttpd] libmicrohttpd, serialized access

[Christian Grothoff]

Hi!

If you have to replace pthread_mutex_* anyway and are using
MHD_USE_SELECT_INTERNALLY, it is indeed safe to replace those
functions with NOP.

Happy hacking!

Christian

On 06/29/2014 11:12 PM, Martin Velek wrote:
> Hi,
> 
> I am using FreeRTOS running on CortexM3. Every mutex takes valuable
> resources (SRAM and a record in the handle table). The NOP should be
> No operation, e.g. the create function returns valid value but not
> takes any resources, lock/unlock/delete does nothing. I have to
> replace the pthread_mutex_* anyway because the FreeRTOS does not
> support POSIX threads.
> 
> I think the logic should not be broken but I do not know if there is
> code starting some thread requiring to lock the mutex even with
> MHD_USE_SELECT_INTERNALLY. I did not notice and find such behaviour so
> far. To convince myself I had to ask :).
> 
> Regards
> Martin
> 
> On Fri, Jun 27, 2014 at 11:44 AM, Christian Grothoff
> <address@hidden> wrote:
>> Hi!
>>
>> If you're on Linux, this is almost the case, as with a single thread
>> you obviously cannot have lock contention, and you're using a Futex
>> (http://en.wikipedia.org/wiki/Futex) -- so the cost is very close to
>> that of a NOP (no system call).  In fact, I would be surprised if
>> replacing the logic with the equivalent of "if (! (daemon->options &
>> MHD_USE_SELECT_INTERNALLY)) lock()" would give you any speed-up (it'd
>> still technically be a branch and not a real NOP).
>>
>> Anyway, if you have a different reason for asking (other than
>> performance concerns) or have real data showing that there is a
>> difference, please explain more and I'll be happy to listen.
>>
>> Happy hacking!
>>
>> Christian
>>
>> On 06/27/2014 11:18 AM, Martin Velek wrote:
>>> Hello,
>>>
>>> I am using libmicrohttpd with the MHD_USE_SELECT_INTERNALLY flag only
>>> (one thread). Every request is processed in the libmicrohttpd thread
>>> and neither MHD_Connection * nor MHD_Response * are processed outside
>>> the thread.
>>>
>>> Is it possible to make all pthread_mutex_* calls as no-ops?
>>>
>>> Martin
>>>
>>
>