2.6.24 vs. 2.6.27/29 (was RE: [Linux-Xtensa] shmat and
sigpausefailed using LTP on Linux 2.6.24)
marc at tensilica.com
Wed Apr 22 16:38:01 PDT 2009
I can't speak for any agreements you have with Tensilica
(hadn't heard of committing to backporting to 2.6.24).
Porting everything down to 2.6.24 would take some time, it's not
something you could get, say, tomorrow. It also adds some risk,
because the extensive testing in question has been done on later
kernels. We also don't know offhand exactly which patches you
need to backport to address your concerns.
Is there any reason you need 2.6.24 specifically? Do you have
driver dependencies or other kernel changes that you made to
that version that would need to be re-applied to a newer kernel?
New files (such as a new platform port) are simply copied to the
new kernel. For modified files, you have to re-apply the diff
onto (or merge with) the new kernel. For a platform port, I
would hope you'd have very few such modifications, so that moving
to a new kernel is much quicker than backporting our changes
(which touch just about everything for the Xtensa architecture).
The easiest is to use the xtensa-2.6.27-smp tree (in the 2.6.29
tree, include directories have moved around).
There are two things we know of that need to be addressed.
One is to make sure you configure the kernel for single processor.
The other is that the system call that provides atomic operations
may need to be re-enabled, if your toolchain is based on the last
snapshot rather than from a more recent pull from the buildroot git tree.
We're looking into re-enabling it right now (it's a trivial fix,
just needs testing).
-Marc (again, I speak for myself, not for Tensilica)
Shimon Edelhaus wrote:
> Greg told me you guys will port down to 2.6.24. We have no
> time to mess with new kernels. We have a chip coming
> tomorrow. Literally tomorrow...
> I am going to talk to Edy and Greg. This is completely unacceptable.
> Simon Edelhaus
> Centillium 2008
> ----- Original Message -----
> From: linux-xtensa-bounces at linux-xtensa.org
> <linux-xtensa-bounces at linux-xtensa.org>
> To: Piet Delaney <Piet.Delaney at mail.tensilica.com>;
> hitheshm at tataelxsi.co.in <hitheshm at tataelxsi.co.in>
> Cc: linux-xtensa at linux-xtensa.org
> <linux-xtensa at linux-xtensa.org>; Piet at linux-xtensa.org
> <Piet at linux-xtensa.org>; Delaney <piet.delaney at gmail.com>
> Sent: Wed Apr 22 10:51:46 2009
> Subject: 2.6.24 vs. 2.6.27/29 (was RE: [Linux-Xtensa] shmat
> and sigpausefailed using LTP on Linux 2.6.24)
> Just a note to provide some background on 2.6.24 vs. 2.6.27+
> kernels for Xtensa cores. You have probably noticed that most
> Wiki pages on linux-xtensa.org still refer to the 2.6.24 kernel
> as being the latest stable one. It is also the latest for which
> we've created a "snapshot" tarball.
> At some point early last year, work began in earnest on adding
> SMP support for Xtensa. This development would completely
> destabilize the kernel for a while, so was done in a completely
> separate tree. SMP also required much more extensive testing to
> find and debug subtle (and not so subtle) issues, and stabilize
> to a point where we have reasonable confidence in the port.
> Piet recently got it to that point, where it runs for days under
> constant LTP testing, without problems. However, the code base
> for it has moved beyond 2.6.24, and is still progressing quickly
> as the focus is now on merging all these changes with the latest
> kernel and submitting them to kernel.org . Once this is done and
> more testing has been done, we expect to create another 'snapshot'.
> Which is why, officially, the latest stable tree may still be
> considered to be the 2.6.24 one. Meanwhile you can get to the
> latest 2.6.27 or 2.6.29 SMP trees using git. If you look at the
> home page on www.linux-xtensa.org and follow the GIT link, you'll
> see these two trees listed (xtensa-2.6.27-smp and xtensa-2.6.29-smp).
> You should be able to clone and pull these trees in the usual
> manner with git, using these names instead of xtensa-2.6.24 .
> Note that the existing buildroot build instructions might not
> match exactly with it. For example, currently the default kernel
> config might be setup for SMP instead of single-processor,
> requiring you to change that parameter when doing make menuconfig
> on the kernel (hopefully the default gets fixed soon).
> There might other differences. You'll have to try and find out.
> Step-by-step instructions will only get updated when things
> stabilize a bit more and we create a new snapshot.
> Hope that helps,
> P.S. Anything opinions expressed by a Tensilica employee mailing to
> linux-xtensa.org is their own, not necessarily that of Tensilica.
> linux-xtensa mailing list
> linux-xtensa at linux-xtensa.org
More information about the linux-xtensa