On Tue 30 Sep 11:27 PDT 2014, Stephen Boyd wrote:
> On 09/29/14 22:02, Bjorn Andersson wrote:
> > On Mon 29 Sep 15:17 PDT 2014, Stephen Boyd wrote:
> >
> >> On 09/29/14 02:14, Srinivas Kandagatla wrote:
> >>> @@ -246,6 +247,24 @@
> >>> #reset-cells = <1>;
> >>> };
> >>>
> >>> + apcs: syscon@2011000 {
> >>> + compatible = "syscon";
> >>> + reg = <0x2011000 0x1000>;
> >>> + };
> >> This is actually a clock controller block that hw designers decided was
> >> good place to shove the ipc bits (because there's room!). Can we call it
> >>
> >> l2cc: clock-controller@2011000 {
> >> compatible = "syscon";
> >> reg = <0x2011000 0x1000>;
> >> };
> >>
> >> Eventually I'll add the specific krait compatible when we merge krait
> >> clock support:
> >>
> >> l2cc: clock-controller@2011000 {
> >> compatible = "qcom,kpss-gcc", "syscon";
> >> reg = <0x2011000 0x1000>;
> >> clock-output-names = "acpu_l2_aux";
> >> };
> >>
> > As long as we can get hold of the regmap that would be fine. I pressume the
> > idea is to have the kpss-gcc using syscon, just like the rpm. But hopefully the
> > syscon patches that are floating around (merged?) will allow any driver to
> > expose a "syscon regmap".
>
> Agreed. I'm not sure if kpss-gcc will use syscon. With the floating
> patches I don't see a reason why we need to use it.
>
All we need is a regmap for the kpss-gcc, but the restriction that it has to be
a syscon is more of a implementation detail. But you probably don't want to
implement the phandle->regmap lookup code in kpss-gcc (and all the other
places) and if I understand the syscon improvements being suggested they are
supposed to help with exactly that (help any device expose regmaps).
Need to look into that again though.
> >
> >>> +
> >>> + rpm@108000 {
> >>> + compatible = "qcom,rpm-apq8064";
> >>> + reg = <0x108000 0x1000>;
> >>> + qcom,ipc = <&apcs 0x8 2>;
> >> There are actually 3 ipc bits. I guess if we ever have to use the other
> >> two we'll extend this binding to have the other bits specified some
> >> other way?
> > I haven't seen any indications of us using more than this bit. If we want to do
> > that, we could simply make it <&apcs 8 2 &apcs 8 3 &apcs 8 4> (or whatever
> > those indices are). That way this should be easy it keep compatible.
> >
>
> Good to hear we're not forced to use a new property. The indices have
> always been 2, 1, and 0 as far as I know.
>
> We use bit 1 to ack the RPM if it ever crashes. We know that it crashes
> because we receive the error interrupt. Instead of calling panic like we
> do downstream we can gracefully fail any new rpm requests. There really
> isn't any way to recover from this scenario though besides rebooting.
> Looking at the code I see this is all wrong and we use bit 2 to ack the
> error.
>
Hmm, seems I got that wrong, sorry about that.
But do you mean "all wrong" as in that I use the wrong bit or to some greater
extent? Currently all following requests should timeout, but maybe we should
have a faster fail-path when we've hit this point?
>From a practical pov I guess that once the rpm starts returning errors on
updates to regulators, root clocks and bus scaling then most of the system is
becoming useless.
Regards,
Bjorn
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
