Re: [Xenomai-help] select(2) generates EFAULT

[Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>]

Steve wrote:
> On 18/05/2009, Jan Kiszka <jan.kiszka@domain.hid> wrote:
>> Gilles Chanteperdrix wrote:
>>> Jan Kiszka wrote:
>>>> Gilles Chanteperdrix wrote:
>>>>> Steve wrote:
>>>>>> First, thank you for the help on my previous query.
>>>>>>
>>>>>> I am still working on porting a c++ application from RTAI to Xenomai,
>>>>>> and have encountered a problem with sockets, specifically the select
>>>>>> function.
>>>>>>
>>>>>> When Xenomai is not being used the code runs as expected.  Once the
>>>>>> Xenomai headers, compile options, etc, are used the select function
>>>>>> fails with the error EFAULT.  I have reproduced the error (works
>>>>>> without Xenomai, EFAULTs with Xenomai) using the following program:
>>>>>>
>>>>>> #include <stdio.h>
>>>>>> #include <arpa/inet.h>
>>>>>> #include <errno.h>
>>>>>>
>>>>>> int handleError(char *msg, int err);
>>>>>>
>>>>>> int main(void)
>>>>>> {
>>>>>>    int sfd;
>>>>>>    struct sockaddr_in sai;
>>>>>>    fd_set SockSet;
>>>>>>    struct timeval tv;
>>>>>>    tv.tv_sec = 2;
>>>>>>    tv.tv_usec = 0;
>>>>>>
>>>>>>    sai.sin_family = AF_INET;
>>>>>>    sai.sin_port = htons(1025);
>>>>>>    sai.sin_addr.s_addr = htonl(INADDR_ANY);
>>>>>>
>>>>>>    if((sfd = socket(AF_INET, SOCK_STREAM, 0)) == -1)
>>>>>>    { return handleError("creation error", errno); }
>>>>>>
>>>>>>    if(bind(sfd, (struct sockaddr*)&sai, sizeof(sai)) == -1)
>>>>>>    { return handleError("bind error", errno); }
>>>>>>
>>>>>>    if(listen(sfd, 5) == -1)
>>>>>>    { return handleError("listen error", errno); }
>>>>>>
>>>>>>    FD_ZERO(&SockSet);
>>>>>>    FD_SET(sfd, &SockSet);
>>>>>>
>>>>>>    if(select(sfd+1, &SockSet, 0, 0, &tv) == -1)
>>>>>>    { return handleError("select error", errno); }
>>>>>>
>>>>>>    printf("success\n");
>>>>>>
>>>>>>    return 0;
>>>>>> }
>>>>>>
>>>>>> I am using Xenomai 2.4.6.1 on a vanilla 2.6.24.6 Linux kernel, and
>>>>>> running on Ubuntu 8.04.
>>>>> Could you try 2.4.7, to see if you are not having a bug which has
>>>>> already been fixed?
>>>>>
>>>>> What you are trying should work, even though you are not really using
>>>>> Xenomai's select but Linux plain select.
>>>> This should have been fixed in 2.4.x.git - if my patch were applied. See
>>>> git://git.xenomai.org/xenomai-jki.git for-2.4.x
>>>>
>>>> Did I missed to post some correction of that version, or what delays
>>>> this? Sorry, lost track of it.
>>> Sorry, I do not see what patch you are refering to. Is it:
>>> http://git.xenomai.org/?p=xenomai-jki.git;a=commitdiff;h=86333965a4362f8439822ec9d22b0a84b5701141;hp=a219c4a71b381f4662df45bac5508c3ae64ab924
>>>
>>> In this commit, I only see things which should return more -EFAULT. Or
>>> did I miss something?
>> I mean
>> http://git.xenomai.org/?p=xenomai-jki.git;a=commitdiff;h=8542b9dc56b9edd16dab69e930699bcee17fa6ac
>>
>> That should be...
>> http://permalink.gmane.org/gmane.linux.real-time.xenomai.devel/6237
>>
>> Jan
>>
>> --
>> Siemens AG, Corporate Technology, CT SE 2
>> Corporate Competence Center Embedded Linux
>>
> 
> I am in the process of re-compiling the kernel with the 2.4.7, and I
> have just checked the fault number of my example; it is definitely
> giving +14, not -14 (so EFAULT, not -EFAULT).  Might this be an issue
> with my kernel config options rather than anthing else?

No, that is the unix standard behaviour, select should return -1 and the
14 be set in errno.
Newer calls belonging to the pthread_*, clock_* families return their
errors directly as positive codes too, and not through errno.

-- 
                                                 Gilles.