Re: Problems with registers save/restore on context switch?

[David Brownlee <abs%netbsd.org@localhost>]

On 26 December 2013 20:15, Donald Lee <MacPPC2%c.icompute.com@localhost> wrote:
> I have run the same test (with trivial printf's added to show progress)
> on a G4, running NetBSD 5.2, AND NetBSD 6.0.2, and it comes through clean.
>
> I have included the logs from these two machines. mercy fails. Candor
> run the test cleanly.
>
> I'm not sure what this means, but it seems to imply that the poblem is
> CPU specific because it works cleanly on the Beige G3, on a certain G4, but
> not on another (faster, newer) G4

Very interesting find!

(My powerpc skills could probably be listed on the back of a 601 die,
but some random comments below)

There are not many MPC7400 vs MPC7447A differences in
sys/arch/powerpc/oea/cpu_subr.c

There is one 7447A section with the comment /* Enable the 7450 branch
caches */ which also has /* Disable BTIC on 7450 Rev 2.0 or earlier */
- I'm pretty sure its irrelevant, but disabling it would be easy to
test.

The "oea_startup: failed to allocate DEAD ZONE" code has a comment
which indicates it is harmless, but another difference between the two
is the amount of memory - would it be possible to test them both with
the same (lower) amount of memory, such as 512M?

The 7440 cache is reported as write back, while the 7447A is not,
though I suspect that may just be a reporting inconsistency.

The 7440 L2 cache is supposed to be 2-way set associative vs 8-way set
associative for the 7447A - I don't see any differences in the
handling code, though I don't know if that is significant...