Re: [Xenomai-help] problem with libanalogy

[Alexis Berlemont <berlemont.hauw@domain.hid>]

Simon Boulay wrote:
> On 01/14/2010 06:07 PM, Alessio Margan @ IIT wrote:
>> Simon Boulay wrote:
>>> Hi,
>>>
>>> On 01/13/2010 06:19 PM, Alessio Margan @ IIT wrote:
>>>
>>>> Hi,
>>>>
>>>> I've got "Segmentation fault" from libanalogy.so.1
>>>> Here is part of code ...
>>>>
>>>> int s526Analogy::Open()
>>>> {
>>>> int ret = 0;
>>>> boardIsOpen = false;
>>>>
>>>> ret = a4l_open(&dsc, dev_name);
>>>> if (ret<   0) {
>>>> fprintf(stderr, "s526Analogy::Open : a4l_open %s failed (ret=%d)\n",
>>>> dev_name, ret);
>>>> return boardIsOpen;
>>>> }
>>>>
>>>> fprintf(stderr, "s526Analogy::Open : name : %s\nsubd %d\nrd subd %d\nwr
>>>> subd %d\nfd %d\nmagic %x\nsize %d\n",
>>>> dsc.board_name,
>>>> dsc.nb_subd,
>>>> dsc.idx_read_subd,
>>>> dsc.idx_write_subd,
>>>> dsc.fd,
>>>> dsc.magic,
>>>> dsc.sbsize);
>>>>
>>>> dsc.sbdata = malloc(dsc.sbsize);
>>>> if (dsc.sbdata == NULL) {
>>>> ret = -ENOMEM;
>>>> fprintf(stderr, "s526Analogy::Open : info buffer allocation failed\n");
>>>> return boardIsOpen;
>>>> }
>>>> ret = a4l_fill_desc(&dsc);
>>>> if (ret<   0) {
>>>> fprintf(stderr, "s526Analogy::Open : a4l_fill_desc failed (ret=%d)\n", ret);
>>>> return boardIsOpen;
>>>> }
>>>>
>>>> boardIsOpen = true;
>>>> printf("Open %s\n", dev_name);
>>>>
>>>> return boardIsOpen;
>>>> }
>>>>
>>>> and here is the back trace from gdb ....
>>>>
>>>> s526Analogy::Open : name : analogy_s526
>>>> subd 4
>>>> rd subd 1
>>>> wr subd 2
>>>> fd 0
>>>> magic 1234abcd
>>>> size 1560
>>>>
>>>> Program received signal SIGSEGV, Segmentation fault.
>>>> [Switching to Thread 0xb7151990 (LWP 6138)]
>>>> 0xb773a64a in a4l_sys_desc () from /usr/xenomai/lib/libanalogy.so.1
>>>> Current language: auto; currently asm
>>>> (gdb) bt
>>>> #0 0xb773a64a in a4l_sys_desc () from /usr/xenomai/lib/libanalogy.so.1
>>>> #1 0xb773a748 in a4l_fill_desc () from /usr/xenomai/lib/libanalogy.so.1
>>>> #2 0x0804ebb2 in s526Analogy::Open (this=0x805e560) at s526Analogy.cpp:55
>>>> #3 0x0804f8c6 in configure_board () at s526Main.cpp:95
>>>> #4 0x0804f96a in main () at s526Main.cpp:115
>>>> (gdb)
>>>>
>>>> dmesg give me ...
>>>>
>>>> app[6454]: segfault at 40 ip b772964a sp bfd372d0 error 4 in
>>>> libanalogy.so.1.0.0[b7728000+3000]
>>>>
>>>>
>>>> Yesterday I pull xenomai-head to get last commits, before I've run the
>>>> application without this segfault.
>>>>
>>> I can't reproduce your problem.
>>> Does the same thing happen with the following command:
>>> insn_read -v -d analogy0?
>>>
>>> Simon.
>>>
>>>   >
>>>
>>>> TIA
>>>>
>>>> Alessio
>>>>
>>>>
>>>>
>> Thanks,
>>
>> insn_read and insn_write works fine, seems that reset the buffer wiht
>> zeroes works ...
>>
>> dsc.sbdata = malloc(dsc.sbsize);
>> bzero(dsc.sbdata, dsc.sbsize);
>> or
>> dsc.sbdata = calloc(dsc.sbsize,1);
>>
>> without I got segfaults.
>>
>> Any idea ?
> It seems more likely that the problem comes from the application this 
> time but I can't confirm without repoducing the problem. You can send me 
> more code if you will. I don't see the scope of the dsc variable for 
> example.
> For my tests, I simply replaced S526Analogy::Open() by a main function 
> and add this at the begining:
> const char *dev_name = "analogy0";
> bool boardIsOpen = false;
> a4l_desc_t dsc = { .sbdata = NULL };
> 
> Does this work for you?

Argh! There was a bug in the initialization procedure of the allocated 
buffer (with malloc(dsc.sbsize)). Some field of an internal analogy 
structure (a4l_root) was not set to 0.

I reproduced the bug by memsetting random values into the allocated buffer.

I committed the fixes and made a pull request.

> 
> Simon.
> 
> 
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help
> 
Alexis.