Re: [PATCH] Drivemap module

[Marco Gerards <mgerards@xs4all.nl>]

Robert Millan <rmh@aybabtu.com> writes:

[...]

>> > This is a lot of code being added to kernel, and space in kernel is highly
>> > valuable.
>> > 
>> > Would the same functionality work if put inside a module?
>> For the reasons discussed above in the loader.h snippet, I don't think
>> so: the only "lighter" solution would be to just put a drivemap_hook
>> variable that would be called before boot, but as I mentioned before,
>> this solution can be employed by other modules as well.
>> 
>> Besides (and I realize this is not a great defense) it's not _that_ much
>> code: just a simple linked-list implementation with add and delete
>> operations, and the iteration of it on loader_boot. I did not check how
>> many bytes does this patch add by itself, but I can run some simulations
>> (I totally _had_ to say that ^^) if you want.
>
> Having a small kernel is highly desireable for most users.  If the kernel is
> too big, it won't fit and then either we have to use blocklists (which are
> unreliable), or we have to abort the install.
>
> Please, try to find a way that doesn't increase kernel size significantly.
>
> If the kernel interfaces are not extensible enough, you could try to readjust
> them for your needs.  This approach works well most of the time (although I
> haven't studied this particular problem in detail).

Like discussed before.  Bring up such modifications like hooks up in a
*separate* thread.  I already said that not everyone reads this
discussion.  I will not accept a patch that changes the kernel if it
is part of a bigger patch that not many people read.

Please don't discuss this over with Robert and me, you know that it
was pointed out that this has to be a patch in a separate thread.
Furthermore, this is a way to get some feedback from Bean who wants
something similar, IIRC.

--
Marco