On Thu, Jun 19, 2014 at 07:40:02PM +0100, Marc Zyngier wrote:

Hi Mark,

On 19/06/14 15:01, Mark Rutland wrote:

quoted

Hi Marc,

On Thu, Jun 19, 2014 at 10:19:27AM +0100, Marc Zyngier wrote:

quoted

Linux has some requirements that must be satisfied in order to boot
on a system built with a GICv3.

Acked-by: Christoffer Dall <redacted>
Signed-off-by: Marc Zyngier <redacted>
---
 Documentation/arm64/booting.txt | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/Documentation/arm64/booting.txt b/Documentation/arm64/booting.txt
index 37fc4f6..e28ccec 100644
--- a/Documentation/arm64/booting.txt
+++ b/Documentation/arm64/booting.txt

@@ -141,6 +141,12 @@ Before jumping into the kernel, the following conditions must be met:
   the kernel image will be entered must be initialised by software at a
   higher exception level to prevent execution in an UNKNOWN state.
 
+  For systems with a GICv3 interrupt controller, it is expected that:
+  - If EL3 is present, it must program ICC_SRE_EL3.Enable (bit 3) to
+    0b1 and ICC_SRE_EL3.SRE (bit 0) to 0b1.
+  - If the kernel is entered at EL1, EL2 must set ICC_SRE_EL2.Enable
+    (bit 3) to 0b1 and ICC_SRE_EL2.SRE (bit 0) to 0b1.

Apologies for spotting this so late, but to me this sounds slightly
ambiguous. The use of "it is expected" doesn't read like a hard
requirement, and in the first point, it's ambiguous as to what "it" is.

I assume that if the GIC is communicated to the kernel as a GICv2 then
these points do not hold?

The first point always holds, specially if the kernel is entered at EL2
(see patch #2 and the way we initialize System Registers in head.S). At
this stage, we haven't looked at DT yet, and must setup EL2
independently of what the platform will describe. The only source of
information we have is whether or not this CPU implements GICv3 System
Registers (id_aa64pfr0_el1).

Assuming EL3 doesn't set these two bits, you will end up trapping back
to EL3. You can hope that EL3 will do the right thing (do what is
described above and restart the offending instruction at EL2). If it
doesn't, you're dead.

quoted

How about:

  For systems with a GICv3 interrupt controller, where the presence of
  GICv3 is communicated to the kernel:
  - If EL3 is present:
    ICC_SRE_EL3.Enable (bit 3) must be initialiased to 0b1.
    ICC_SRE_EL3.SRE (bit 0) must be initialised to 0b1.
  - If the kernel is entered at EL1:
    ICC.SRE_EL2.Enable (bit 3) must be initialised to 0b1
    ICC_SRE_EL2.SRE (bit 0) must be initialised to 0b1.

I'm happy with that change, provided that we get rid of the ", where the
presence of GICv3 is communicated to the kernel".

Sure, given the above that sounds fine by me.

Thanks,
Mark.