Re: how quickly do PA-RISC patches make it to mainstream kernel?
On Thu, 4 Dec 2003, Matthew Wilcox wrote:
> On Thu, Dec 04, 2003 at 04:03:21PM +0200, Martin-Éric Racine wrote:
> > On Thu, 4 Dec 2003, Matthew Wilcox wrote:
> > > That's not the case. Marcelo is willing to take patches, but nobody is
> > > interested in generating them and sending them to him.
> >
> > Why?
>
> Because we're interested in 2.6. Why do you care whether the parisc
> patches are in Marcelo's tree or not? 2.4 is already hopelessly forked.
Because I need to be able to apply the exact same patches e.g. grsecurity, etc.
to all kernels accross the board, regardless of architecture, and there is no
way this can work if all non-x86 architectures are hopelessly out of sync with
Marcelo's releases.
My most frequent example is grsecurity. This one fails on precisely every one of
the non-x86 architectures I run: sometimes, one can get lucky and the previous
grsecurity release will apply to the CVS or rsync tree of parisc or powermac,
but most of the time it's hopelessly out of sync.
As such, if nobody working on non-x86 bothers with submitting their diffs to
Marcello, or only does it once or twice a year, then, yes indeed, 2.4 becomes
hopelessly forked and out-of-sync and it makes it impossible to use any of the
common patches on non-x86 development trees of matching major versions.
This means that the only two options are either:
1) bend grsecurity until it fits 2.4-xx.paNN (not practical for people who are
not kernel developers and whose job is to batch-patch a customer's servers)
2) apply the current grsecurity to vanilla, but end up with completely outdated
hardware support and missing bugfixes on non-x86 platforms.
Neither options are acceptable, from a user or administrator's point of view.
Bear in mind that while 2.6 is a very promising kernel, people in real-life
situations tend to be extremely conservative in their choices. Nowadays, it
means trusting kernel 2.4 for regular deployments.
That's why frequent commits of non-x86 code to Marcelo's 2.4 tree is important.
--
Martin-Éric Racine, ICT Consultant
http://www.pp.fishpool.fi/~q-funk/
Reply to: