fully bisectable real time series

fully bisectable real time series

[Daniel Walker]

I've re-worked the -rt series, and some of the core patches to make it
all bisectable. I've done bisect updates in the past which really only
scratched the surface.

The main problems where around the spinlock, semaphore, and rw-semaphore
changes. I made a number of updates to how the bottom level spinlocks
get defined to ease the transition per-patch and architecture. For
semaphores I removed some constraints on how many changes need to be
made per architecture.

So what's missing? Well I dropped every architecture port except x86,
since I broke them all. I also dropped a number of non-essential patches
(mostly from Peter) not for any technical reason but just to reduce the
amount of code. 

I didn't fix all the architectures since I want to make sure x86 is
solid before I start moving the other architectures to the new format.
All the dropped patches are still part of the archive, just not in the
series.

x86 32-bit is compile and boot bisectable, and 64-bit is compile
bisectable (should boot too, but I haven't tested it.)

There's tons of back merging to do, but I keep it to a minimum for the
time being. I'm sure there are some un-bisectable areas that I just
haven't found yet, however if you report it to me I'll be happy to fix
them.

I'm only providing a broken out series and you can download in from,

ftp://source.mvista.com/pub/dwalker/rt/patch-2.6.24.4-rt4-dw1-broken-out.tar.bz2

Enjoy!

Daniel

Re: fully bisectable real time series

[Steven Rostedt]

On Mon, 7 Apr 2008, Daniel Walker wrote:

>
> I've re-worked the -rt series, and some of the core patches to make it
> all bisectable. I've done bisect updates in the past which really only
> scratched the surface.
>

Daniel,

Thanks, for doing this. I'll try to look at it when I get a chance. If you
can make a bisectable series that still gives a byte for byte identical
kernel with a make allyesconfig, I'll incorporate it.

-- Steve

Re: fully bisectable real time series

[Sven-Thorsten Dietrich]

On Mon, 2008-04-07 at 16:34 -0700, Daniel Walker wrote:
> I've re-worked the -rt series, and some of the core patches to make it
> all bisectable. I've done bisect updates in the past which really only
> scratched the surface.

I had started on a forward-port to .25, but I got completely
side-tracked with a lot of other stuff.

The bi-sectable series is vary valuable for such porting efforts, and
the current series needs major cleanup, since there are a lot of fixes,
and code-backouts (e.g. softirq from hardirq thread context), which
interfere with testing of a back-port at each level.

So this is a valuable effort, and I hope that at least some of what you
have done will make it into .25's RT port.

If I have time to work on the port in the next couple of days, before
Steven gets to it, I will see what if any of your stuff can be used.

Thanks,

Sven

Re: fully bisectable real time series

[Daniel Walker]

On Mon, 2008-04-07 at 20:33 -0400, Steven Rostedt wrote:
> On Mon, 7 Apr 2008, Daniel Walker wrote:
> 
> >
> > I've re-worked the -rt series, and some of the core patches to make it
> > all bisectable. I've done bisect updates in the past which really only
> > scratched the surface.
> >
> 
> Daniel,
> 
> Thanks, for doing this. I'll try to look at it when I get a chance. If you
> can make a bisectable series that still gives a byte for byte identical
> kernel with a make allyesconfig, I'll incorporate it.

I haven't actually tested if it's byte for byte identical, I can or you
can review it an decide what you think of it.. The executable code parts
should be identical. I just moved patches around, patch split-ups, did
name changes, and added macros ..

Daniel