Re: [PATCH v2 1/3] power: reset: read priority from device tree

From: Guenter Roeck
Date: Mon Dec 01 2014 - 13:14:20 EST


On Mon, Dec 01, 2014 at 05:59:08PM +0000, Mark Rutland wrote:
> On Mon, Dec 01, 2014 at 05:51:34PM +0000, Guenter Roeck wrote:
> > On Mon, Dec 01, 2014 at 05:41:00PM +0000, Mark Rutland wrote:
> > > On Mon, Dec 01, 2014 at 05:24:37PM +0000, Stefan Agner wrote:
> > > > On 2014-12-01 18:11, Mark Rutland wrote:
> > > > > On Mon, Dec 01, 2014 at 05:03:07PM +0000, Stefan Agner wrote:
> > > > >> This patch adds an optional property which allows to specify the
> > > > >> reset source priority. This priority is used by the kernel restart
> > > > >> handler call chain to sort out the proper reset/restart method.
> > > > >> Depending on the power design of a board or other machine/board
> > > > >> specific peculiarity, it is not possible to pick a generic priority.
> > > > >>
> > > > >> Signed-off-by: Stefan Agner <stefan@xxxxxxxx>
> > > > >> ---
> > > > >> Documentation/devicetree/bindings/power/reset/syscon-reboot.txt | 3 +++
> > > > >> drivers/power/reset/syscon-reboot.c | 5 ++++-
> > > > >> 2 files changed, 7 insertions(+), 1 deletion(-)
> > > > >>
> > > > >> diff --git a/Documentation/devicetree/bindings/power/reset/syscon-reboot.txt b/Documentation/devicetree/bindings/power/reset/syscon-reboot.txt
> > > > >> index 1190631..ee41d9c 100644
> > > > >> --- a/Documentation/devicetree/bindings/power/reset/syscon-reboot.txt
> > > > >> +++ b/Documentation/devicetree/bindings/power/reset/syscon-reboot.txt
> > > > >> @@ -11,6 +11,9 @@ Required properties:
> > > > >> - offset: offset in the register map for the reboot register (in bytes)
> > > > >> - mask: the reset value written to the reboot register (32 bit access)
> > > > >>
> > > > >> +Optional properties:
> > > > >> +- priority: define the priority of the reset (0-255, defaults to 128)
> > > > >> +
> > > > >
> > > > > NAK. This is a leak of Linux-internal details.
> > > > >
> > > > > What is this necessary for?
> > > > >
> > > > > Mark.
> > > >
> > > > Hi Mark,
> > > >
> > > > In my case, it is necessary to be called ahead of the watchdog, which
> > > > has a priority of 128. This syscon-reboot driver currently has a default
> > > > priority of 128 too. I could live with a higher default priority for the
> > > > syscon-reboot driver, in fact I proposed that in the discussion of v1 of
> > > > that patch:
> > > > https://lkml.org/lkml/2014/11/28/484
> > >
> > > Thanks for the link.
> > >
> > > > IMHO, this priority might make sense for most cases, I guess that
> > > > dedicated "syscon" capabilities are usually better suited as a reboot
> > > > source than watchdog.
> > >
> > > I would think likewise.
> > >
> > > > If dt, then the question which arises: If there are different
> > > > capabilities to reset/reboot a whole system, how do we reflect which is
> > > > the best suited one in dt?
> > >
> > > I'm not sure, but I don't think that exposing a priority variable in
> > > this way is the best, because it implicitly relies on what the kernel
> > > may have selected for other devices and/or FW interfaces, which may not
> > > have been described in DT.
> > >
> > > So if we can get away with a fixed priority for now, then I would prefer
> > > that.
> > >
> > > Otherwise, I would imagine that most systems have a single preferred
> > > mechanism with some possible fallback(s), for which a single
> > > preferred-poweroff property might suffice, and has better interaction
> > > w.r.t. priority (in that it should _always_ be tried first). Even that's
> > > difficult to reconcile with FW bindings though, especially EFI (which we
> > > sometimes must use in preference for variable storage and capsule
> > > updates).
> > >
> > Hi mark,
> >
> > reboot, not poweroff, but I like that idea; it is system independent,
> > and I agree that there should be only one "preferred-reboot" mechanism.
> > If we need more we can always add something like "fallback-reboot".
>
> Ah, sorry, I got myself confused regarding reboot/poweroff. Goot to hear
> we agree on the preferred-$X (where X is reboot in this case).
>
In this context, note that the poweroff handler patchset is abandoned/dead.
Neither the power maintainers nor Linus liked the idea of having prioritized
poweroff handlers.

> > Not sure I understand the reference to EFI. Does EFI install a means
> > to restart the system ?
>
> Through EFI runtime services there's a mechanism to reset the machine.
> In certain cases this must be called so as to allow EFI to set things up
> to run after reboot (e.g. capsules for FW updates), while in other cases
> it's perfectly fine to call other mechanisms directly.
>
Sounds like its priority depends on the context. That might be a bit tricky,
but it might be possible if EFI uses the API to register its restart handler
and re-registers itself if needed. Would that make sense ?

Thanks,
Guenter
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/