Hi Michael,

On Wed, 03 Jul 2024 12:30:38 +0100,
Michael Ellerman [off-list ref] wrote:

Rob Herring [off-list ref] writes:

quoted

On Tue, Jul 2, 2024 at 10:54 AM Marc Zyngier [off-list ref] wrote:

quoted

On Sun, 30 Jun 2024 11:21:55 +0100,
Christian Zigotzky [off-list ref] wrote:

quoted

Hello,

There is an issue with the identification of ATA drives with our
P.A. Semi Nemo boards [1] after the
commit "of/irq: Factor out parsing of interrupt-map parent
phandle+args from of_irq_parse_raw()" [2].

[snip]

My earlier request for valuable debug information still stands. But
while you're at it, can you please give the following hack a go?

        M.

--- a/drivers/of/irq.c
+++ b/drivers/of/irq.c

@@ -282,8 +282,10 @@ int of_irq_parse_raw(const __be32 *addr, struct of_phandle_args *out_irq)

                        oldimap = imap;
                        imap = of_irq_parse_imap_parent(oldimap, imaplen, out_irq);
-                       if (!imap)
-                               goto fail;
+                       if (!imap) {
+                               match = 0;
+                               break;
+                       }

AFAICT reading the DT, I don't think this would fix it. imap should
only be null if malformed. This case to me looks like interrupt-map
has the correct cell sizes, but just never matches to do the mapping.
So maybe imaplen is off and that causes us to end up here, but if
there's an error I don't see it. A boot with DEBUG enabled in
drivers/of/irq.c would help.

quoted

                        match &= of_device_is_available(out_irq->np);
                        if (match)

This may not be the final workaround even if it solves your boot
problem, but will at least give us a hint at what is going wrong.

I have the fuzzy feeling that we may be able to lob this broken system
as part of the of_irq_imap_abusers[] array, which would solve things
pretty "neatly".

I think this would work and would consolidate the work-arounds. It
would need either "pasemi,rootbus" or "pa-pxp" added to the list.

Not sure if it helps, but there's already some code in arch/powerpc to
"fixup" the nemo device tree at boot.

I'm not sure if it's actually the problem here, but it might be, it does
renumber some interrupts. Or possibly it could be tweaked to fix
whatever the issue is.

The code is in fixup_device_tree_pasemi():

  https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/powerpc/kernel/prom_init.c?h=v6.10-rc5#n3114

Ah, that's quite interesting, thanks for the pointer.

I can see two possibilities here:

- either we remove the interrupt-map from the DT (no idea if that is
  possible)

- or we patch the interrupt-map to be slightly more useful and
  actually match its input

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.