Re: [RFD] New driver system (was Adding new syscalls via modules)

Khimenko Victor (khim@sch57.msk.ru)
Tue, 6 Jul 1999 22:05:52 +0400 (MSD)


In <000e01bec6ed$196bda40$28032ed4@sasha.storactive.ru> Alexander Maryanchick (Sasha@GDev.msk.ru) wrote:
> I'm afraid to hurt somebody, but it's must be said:
> Linux extensibility is unacceptable poor. :-(

> Some examples:
> 1. ext2compr, Undelete Tools, KGI,... are patches.
> User must rebulid his kernel to install them.
> Linus must spend his time to think about these patches.
> Developers do not write cool features
> (patches are hard to distribute and require coordination
> with other patches)
> Such patches MUST be distributed as modules.

Why ? It's MUCH more easy to write and support such patches then to write and
support windows-style "separate module" for Windows which will patch kernel
in memory on the fly :-))

> 2. See 'albods', 'forks' & related discussion.
> Too long discussion for one open() flag.
> It would be cool to write a little module and see :-)

You can write patch with easy... And you can write module as well with
technique borrowed from windows toys: kernel patching in memory...
It's just non-practical in case of Linux since there are just a few
Windows kernel but A LOT OF linux kernels (there are more Windows kernels
then official Windows releases but still not as many as different Linux
kernels -- if you'll count all possibilities including usage of different
compilers, different compiler options, etc :-)

> 3. See winfiles.com, download.com, etc.
> A lot of virtual drivers: CD-ROM cache, HDD encryption,
> web acceleration, ...
> Most of these features can not be in full volume
> implemented in Linux without the evil patching the kernel.

And you know what ? Thay can not be implemented in Windows without evil
patching the kernel as well :-)) Just there you must patch binary-only kernel
without sources on the fly in memory (it's not theory -- I seen code for
some of this toys). And this way is not practical for Linux: there are contless
combinations of compilers/compiler flags...

> While we spend time rewritting patches, M$ distributes drivers :-(.

Yes. And when our patches are finally fixed M$ drivers still do not work
as expected... And I STILL can not use Delphi4 due crash in video driver :-((

> I think, the lack of extensibility is a MOST important weakness of Linux
> kernel.
> We need to fix this problem. Do you agree?

No. Adding new feature to Linux is more easy then addign new feature to Windows.
We have completely other problem: most nice toys from winfiles.com are not
interested to hackers with enough skills to implement them and users who want
them do not have enough skills...

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/