Re: [PATCH] markers: modpost

[David Smith <dsmith at redhat dot com>]

Mathieu Desnoyers wrote:
> * David Smith (dsmith@redhat.com) wrote:
>> Mathieu Desnoyers wrote:
>>> * Roland McGrath (roland@redhat.com) wrote:
>>>>> If we want to do it safely, I think we should iterate from
>>>>> __start___markers to __stop___markers symbols of vmlinux and get the
>>>>> pointers to the name/format string pairs.
>>>>>
>>>>> The same can then be done with modules using the __markers section.
>>>>>
>>>>> Or maybe is there some reason not to do that ?
>>>> It's just rather a pain in the ass, a whole lot more fiddly work.
>>>> cf "somewhat crude" and "foreseeable future" in my patch's log entry.
>>>> Knock yourself out if you're looking for more tedious hacking to do in
>>>> modpost.c, but I say fix it when it breaks.
>>>>
>>> Hmmmm, I have rarely seen code go into mainline without addressing valid
>>> technical criticism first. Please fix.
>>>
>>> I'll look into it if I find the time.
>>>
>>> Mathieu
>> Mathieu,
>>
>> Here's an updated patch, written by Roland (that I tested for him), that
>> looks for all marker symbols in the __markers_strings section.  It doesn't
>> get the pointers from the __markers section because that is very difficult
>> to do in modpost (having to handle the architecture-dependent relocations
>> applied to those pointers).
>>
> 
> Hrm, what would happen if a gcc optimization eventually decides to mix
> the memory layout of the strings ? Is there something that specifies
> that they won't ?

I don't believe there is anything in gcc that specifies that the strings
won't get mixed around.  But, I believe this code is good for the
foreseeable future.  We could fix this code if the future breakage does
happen.

-- 
David Smith
dsmith@redhat.com
Red Hat
http://www.redhat.com
256.217.0141 (direct)
256.837.0057 (fax)