On 13/01/16 01:52, Hanjun Guo wrote:

On 2016/1/13 0:45, Tomasz Nowicki wrote:

quoted

On 12.01.2016 13:03, Marc Zyngier wrote:

 > +    typer = readq_relaxed(redist_base + GICR_TYPER);
 > +    /* don't map reserved page as it's buggy to access it */
 > +    size = (typer & GICR_TYPER_VLPIS) ? SZ_64K * 3 : SZ_64K * 2;

[...]

quoted

Also, please map the whole region for the
redistributor as we have on the DT side (4 64kB pages for VLPIS capable
redistributors).

Hanjun, is it something you had problem with?

Yes, for D02, it use 3 64kB pages for GICv4, it removed the reserved
page. In DT case, it has "stride" binding so it can support no standard
cases, seems that we can introduce "stride" for ACPI too.

I don't think there is any need for that, assuming this system only
exposes its redistributors via the ACPI GICC structure. In which case, I
don't think it hurts to have overlapping mappings: For redistributors A
and B, we never touch A[3] (the reserved page), but we will access B[0].

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...