Re: [PATCH 1/5] ArmPkg/ArmLib: add support for reading the max physical address space size

[Ard Biesheuvel <ard.biesheuvel@linaro.org>]

On Fri, 23 Nov 2018 at 14:20, Ard Biesheuvel <ard.biesheuvel@linaro.org> wrote:
>
> On Fri, 23 Nov 2018 at 14:16, Andrew Jones <drjones@redhat.com> wrote:
> >
> > On Fri, Nov 23, 2018 at 01:14:27PM +0100, Ard Biesheuvel wrote:
> > > Add a helper function that returns the maximum physical address space
> > > size as supported by the current CPU.
> > >
> > > Contributed-under: TianoCore Contribution Agreement 1.1
> > > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> > > ---
> > >  ArmPkg/Include/Library/ArmLib.h               |  6 ++++++
> > >  ArmPkg/Library/ArmLib/AArch64/ArmLibSupport.S | 16 ++++++++++++++++
> > >  ArmPkg/Library/ArmLib/Arm/ArmLibSupport.S     |  8 ++++++++
> > >  3 files changed, 30 insertions(+)
> > >
> > > diff --git a/ArmPkg/Include/Library/ArmLib.h b/ArmPkg/Include/Library/ArmLib.h
> > > index ffda50e9d767..9a804c15fdb6 100644
> > > --- a/ArmPkg/Include/Library/ArmLib.h
> > > +++ b/ArmPkg/Include/Library/ArmLib.h
> > > @@ -733,4 +733,10 @@ ArmWriteCntvOff (
> > >    UINT64   Val
> > >    );
> > >
> > > +UINTN
> > > +EFIAPI
> > > +ArmGetPhysicalAddressBits (
> > > +  VOID
> > > +  );
> > > +
> > >  #endif // __ARM_LIB__
> > > diff --git a/ArmPkg/Library/ArmLib/AArch64/ArmLibSupport.S b/ArmPkg/Library/ArmLib/AArch64/ArmLibSupport.S
> > > index 1ef2f61f5979..75ab8dade485 100644
> > > --- a/ArmPkg/Library/ArmLib/AArch64/ArmLibSupport.S
> > > +++ b/ArmPkg/Library/ArmLib/AArch64/ArmLibSupport.S
> > > @@ -196,4 +196,20 @@ ASM_FUNC(ArmWriteSctlr)
> > >  3:msr   sctlr_el3, x0
> > >  4:ret
> > >
> > > +ASM_FUNC(ArmGetPhysicalAddressBits)
> > > +  mrs   x0, id_aa64mmfr0_el1
> > > +  adr   x1, .LPARanges
> > > +  and   x0, x0, #7
> > > +  ldrb  w0, [x1, x0]
> > > +  ret
> > > +
> > > +//
> > > +// Bits 0..2 of the AA64MFR0_EL1 system register encode the size of the
> > > +// physical address space support on this CPU:
> > > +// 0 == 32 bits, 1 == 36 bits, etc etc
> > > +// 6 and 7 are reserved
> > > +//
> > > +.LPARanges:
> > > +  .byte 32, 36, 40, 42, 44, 48, -1, -1
> >
> > Hi Ard,
> >
> > One of the things I was wondering is how much it matters what the
> > firmware's opinion of highest physical address is vs. the guest
> > kernel. Do they need to match? This patch series implies they do,
> > or at least that 40-bits won't always be sufficient for firmware.
>
> Yes. The size of the GCD space limits how much memory we can report as
> present to the OS. So it only matters if there is DRAM there.
>
> > However, guests using 64k pages running on supporting hardware can
> > use 52-bits. Considering ArmVirtPkg only uses 4k pages, that's not
> > an option for it, and that justifies not defining index 6 == 52 in
> > the above array, but will that also restrict the guest?
> >
>
> At the moment, yes. UEFI support for 52-bit/64k pages is still under
> discussion, and is presently not supported.
>

... which btw doesn't mean we can't report that much memory in the GCD
memory map, we just can't map it in UEFI.

There were some discussions about how to proceed here, since there
appear to be some SoC vendors that want to use bit 51 to distinguish
between DRAM and MMIO regions, which implies that we have to support
it to be able to boot such systems (and the architecture does not
forbid or discourage the practice)