[PATCH v3 2/5] ARM: pm: add generic CPU suspend/resume support

[linux@arm.linux.org.uk (Russell King - ARM Linux)]

On Wed, Feb 09, 2011 at 07:15:25PM -0800, Colin Cross wrote:
> The diagnostic register also needs to be saved to keep the errata bits
> set in __v7_setup.

Saving I've no problem with.  Restoring gets hairy with kernels running
in non-secure mode, as we can't just write the register - we don't know
whether we are running in secure or non-secure mode.  A write to the
register in NS mode will crash.

Santosh: is the diagnostic register on OMAP4 re-initialized by the secure
code on OMAP?

> > + ? ? ? stmia ? r0, {r4 - r11}
> > + ? ? ? ldmfd ? sp!, {r4 - r11, pc}
> > +ENDPROC(cpu_v7_do_suspend)
> > +
> > +ENTRY(cpu_v7_do_resume)
> > + ? ? ? mov ? ? ip, #0
> > + ? ? ? mcr ? ? p15, 0, ip, c8, c7, 0 ? @ invalidate TLBs
> > + ? ? ? mcr ? ? p15, 0, ip, c7, c5, 0 ? @ invalidate I cache
> 
> Does this need the same ALT_SMP/ALT_UP combo as v7_flush_icache_all?

That depends whether you the CPU which is resuming is part of a coherent
SMP system at that point.  This instruction will invalidate the I-cache
for the local CPU only, whereas the c7, c1 variant will invalidate the
instruction caches of all CPUs within the inner sharable domain.

Has anything changed in the other CPUs as a result of this CPU resuming
at this point?  I don't think so, so I think we just need to ensure that
the local CPU instruction cache is invalidated at this point.

> Tegra2 suspend and cpuidle works on top of this patch and the patch
> that adds SMP support to sleep_save_sp.  Tegra seems to need to
> invalidate the entire l1 data cache before enabling it,

As it's undefined what state the data cache is in on resume, I'm surprised
the s5pv210 code doesn't also need a D-cache invalidate too.  Maybe Samsung
folk can answer that.

> so I'm using a
> custom reset vector that branches to cpu_resume, and I'm handling the
> TLB invalidate in the function cpu_resume returns to.
> 
> Tested-by: Colin Cross <ccross@android.com>
> 
> Are you targeting 2.6.39 with these patches?  They replace a few
> hundred lines of code in the Tegra2 suspend, hotplug, and idle
> patches, so I'd like to wait until this is in before pushing mine.

Undecided at the moment.  It's great that you've tested it, and I've
also tested it on Assabet, but PXA and Samsung stuff hasn't been
tested yet.  I guess I could just push the generic and sa1100 bits for
2.6.39, unless the remainder gets tested.

Once the above issues have answers, I'll see about posting a new set of
patches.