On 11/21/2013 09:59 AM, Marc Zyngier wrote: > PSCI is an ARM standard that provides a generic interface that > ... Thanks again for posting this, I like the idea of adding PSCI handlers to u-boot for several platforms very much. > This patch series contains 3 parts: > - the first four patches are just bug fixes Those are fine, I already acked those patches. > - the next three contain the generic PSCI code and build infrastructure As I heard you will rework these anyway, I will refrain from commenting in detail, just some generic comments on the approach: * Is the creation of a top-level psci directory for holding the PSCI binaries really necessary? Should this mimic the spl approach? Can you consider to move this at least into the arch/arm directory, as PSCI is ARM specific? Or add it to the SPL directory, as this serves a similar purpose? But maybe your new approach renders this all moot. * Can you keep the SMP bringup code in place and re-use it from the PSCI handlers instead of "#ifndef PSCI"ing it? So maybe rename arch/arm/cpu/armv7/sunxi/psci.S to .../sunxi/smp.S? My idea here is to make PSCI an option in addition to the current SMP HYP mode code. So that for instance on VExpress (or better: Arndale) you could either use the existing code using the kernel's platform SMP code or enable PSCI in u-boot and let the kernel use that, too. I maybe ask too much for the first incarnation of the code, but that is how I would like to eventually see it. AFAIK Linux prefers PSCI over platform-defined SMP code, so this should work out of the box. * Is the use of TPIDRPRW & Co. really safe? It looks like as we seem to be the only secure user (and they are banked), but I am just curious whether there is any "prior art" in using those registers temporarily. > ... > The kernel now boots in HYP mode, finds its secondary CPU without any > additional SMP code, and runs KVM out of the box. Hopefully, the > Xen/ARM guys can do the same fairly easily. BTW: Yesterday my PSCI host patches for Xen have been committed, so Xen should be able to use that feature just like the kernel does. Thanks! Andre. _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/cucslists/listinfo/kvmarm
