[webkit-dev] WebKit on OpenBSD

[James Turner]

On Fri, May 30, 2008 at 01:56:33PM -0700, Maciej Stachowiak wrote:
>
> On May 30, 2008, at 1:51 PM, James Turner wrote:
>
>> On Fri, May 30, 2008 at 12:22:54PM -0700, Maciej Stachowiak wrote:
>>>
>>> On May 30, 2008, at 12:15 PM, James Turner wrote:
>>>
>>>> On Fri, May 30, 2008 at 12:13:43PM -0700, Maciej Stachowiak wrote:
>>>>>
>>>>> On May 30, 2008, at 10:37 AM, James Turner wrote:
>>>>>
>>>>>> I've spent the last couple days trying to compile WebKit on OpenBSD
>>>>>> -current.  After adding the signbit function, changing isfinite to
>>>>>> finite and adding the missing pthread_attr_get_np function, I was
>>>>>> finally able to get it to compile.
>>>>>>
>>>>>> However, when I try to run the Gtk test browser or midori (another gtk
>>>>>> browser based on webkit), I receive or core dump whenever I click on 
>>>>>> any
>>>>>> part of a webpage or try to navigate to another.  After inspecting the
>>>>>> core dump with gdb I'm left with this:
>>>>>>
>>>>>> #0  0x0d501dc8 in KJS::Collector::markStackObjectsConservatively () 
>>>>>> from
>>>>>> /usr/local/lib/libwebkit-1.0.so.1.0
>>>>>
>>>>> Sounds like pthread_attr_get_np is not working as expected. It is used 
>>>>> to
>>>>> get the stack base, for purposes of the conservative garbage collector.
>>>>>
>>>>> - Maciej
>>>>
>>>> Thanks, I'll look into my implementation and check it with people who
>>>> actually know what they're doing!
>>>
>>> pthread_attr_get_np is not really all that important, just any way to get
>>> the stack base. If OpenBSD has any function at all that works for that
>>> purpose you could just add another branch to the ifdef. Regrettably there
>>> is no portable way to do it.
>>>
>>> - Maciej
>>
>> So, I looked through OpenBSD's diff stack function and tried using
>> pthread_stackseg_np like:
>>
>> stack_t stack;
>> pthread_stackseg_np(thread, &stack);
>>
>> But I'm still getting a segfault at the same location.  This again could
>> be the wrong function to use, but it's the only thing I can find that
>> looks remotely like it would work :/
>
> It sounds like it should work (looks analogous to the Solaris 
> thr_stksegment). Can you try in the debugger? You should be able to see if 
> it is trying to access an address outside the stack range, or doing an 
> unaligned access, or something.
>
> Regards,
> Maciej

Thanks so much for your help, I was able to get it to work, I wasn't
returning the correct info, below works for OpenBSD:

pthread_t thread = pthread_self();
stack_t stack;
pthread_stackseg_np(thread, &stack);
return stack.ss_sp;

-- 
James Turner
BSD Group Consulting
http://www.bsdgroup.org