[RFC PATCH 0/5] Add smp booting support for Qualcomm ARMv8 SoCs

[Catalin Marinas]

On Tue, Apr 14, 2015 at 02:49:04PM -0500, Kumar Gala wrote:
> On Apr 14, 2015, at 11:36 AM, Mark Rutland <mark.rutland at arm.com> wrote:
> > On Fri, Apr 10, 2015 at 11:05:29AM +0100, Catalin Marinas wrote:
> >> On Thu, Apr 09, 2015 at 12:37:06PM -0500, Kumar Gala wrote:
> >>> This patch set adds support for SMP boot on the MSM8x16 family of Qualcomm SoCs.
> >>> 
> >>> To support SMP on the MSM8x16 SoCs we need to add ARMv8/64-bit SCM interfaces to
> >>> setup the boot/release addresses for the secondary CPUs.  In addition we need
> >>> a uniquie set of cpu ops.  I'm aware the desired methods for booting secondary
> >>> CPUs is either via spintable or PSCI.  However, these SoCs are shipping with a
> >>> firmware that does not support those methods.
> >> 
> >> And the reason is? Some guesses:
> >> 
> >> a) QC doesn't think boot interface (and cpuidle) standardisation is
> >>   worth the effort (to put it nicely)
> >> b) The hardware was available before we even mentioned PSCI
> >> c) PSCI is not suitable for the QC's SCM interface
> >> d) Any combination of the above
> >> 
> >> I strongly suspect it's point (a). Should we expect future QC hardware
> >> to do the same?
> >> 
> >> You could argue the reason was (b), though we've been discussing PSCI
> >> for at least two years and, according to QC press releases, MSM8916
> >> started sampling in 2014.
> >> 
> >> The only valid reason is (c) and if that's the case, I would expect a
> >> proposal for a new firmware interface protocol (it could be PSCI-based),
> >> well documented, that can be shared with others that may encounter the
> >> same shortcomings.
> > 
> > There's no need to even fork PSCI. The PSCI specification will evolve
> > over time as vendors request changes and we try to accomodate them.
> > 
> > If there's something that PSCI doesn't do that you need it to, contact
> > ARM. Other vendors already have.

Mostly yes but there may be valid reasons for not being able to use
PSCI. The spin-table method is still a firmware interface, though not
necessarily secure (a.k.a. SMC-based). The ACPI parking protocol is
another and, who knows, maybe we define a way to park CPUs back to
firmware without SMC calls (when EL3 is not available).

> But what is someone to do between the period of getting PSCI spec
> updated and needing to ship a product with firmware?
> 
> The take still sounds like if you don’t implement an exact version of
> PSCI you are screwed from being supported in the upstream ARM64
> kernel.

These are silly arguments. There is a big difference between "we
couldn't get the firmware implementing the standard for the early
silicon but we are working on fixing it for future revisions" vs. "we
don't give a s**t about these standards, the kernel must be inclusive".
So please make up your mind on which direction you want to pursue.

-- 
Catalin