On 4/2/24 16:07, Matti Vaittinen wrote:
Another "oddity" is that the PMIC has two physical IRQ lines. When I
last wrote this patch in 2021 I had some naming collison in debugfs for
the IRQ domains. Back then I used:
irq_domain_update_bus_token(intb_domain, DOMAIN_BUS_WIRED);
to work-around the issue. Now, when rebasing to v6.9-rc1 the naming
collision was gone and things seemed to work. However, it'd be great if
the IRQ code in MFD driver was reviewed by greater minds :)
It appears my statement "things seemed to work" is a bit too optimistic.
I am afraid my approach of having two separate IRQ domains for one
device (and DT-node) is just somehow fundamentally wrong. It'd be great
to learn what's the correct "ideology" here.
It appears the naming collision is still there. My config just had the
CONFIG_GENERIC_IRQ_DEBUGFS disabled. Enabling it shows the same naming
collison:
debugfs: File
':ocp:interconnect@48000000:segment@100000:target-module@9c000:i2c@0:pmic@60'
in directory 'domains' already present!
If I'm not mistaken the debugfs file name is generated from the
device-tree node path+name. This is a subtle hint that it is not
expected there are more than 1 IRQ-domain / device. I guess this kind of
makes sense if we can have more than 1 HWIRQ handled by a single domain
(I don't recall having to ever write such domain/IRQ-controller before,
but I think it should be possible).
I have now 3 new questions =)
1. Should we be able to have more than 1 IRQ domain / device?
2. Should regmap_irq support having more than 1 HWIRQ
3. If answer to 1 is "no" - should we protect against this somehow? (see
why below).
When CONFIG_GENERIC_IRQ_DEBUGFS is disabled, adding the two IRQ
controllers with own IRQ domains (intb and errb here) to a single device
is seemingly successful. I see no complaints / errors. Also, most of the
IRQs seem to work - but not all. In my case trying to issue:
cat /proc/interrupts
will oops. Also, looking in the /sys/kernel/irq/ lists folders for all
the "intb" and "errb" IRQs - but reading the files contained in these
directories will cause an oops for all "errb" interrupts except for the
first 16.
Finally, if I use the
irq_domain_update_bus_token(intb_domain, DOMAIN_BUS_WIRED);
to add "-1" at the end of the "intb" - domain name resulting domains:
:ocp:interconnect@48000000:segment@100000:target-module@9c000:i2c@0:pmic@60
:ocp:interconnect@48000000:segment@100000:target-module@9c000:i2c@0:pmic@60-1
then it seems that reading the IRQ information from the /proc/interrupts
works as expected. Here I am making a wild guess that the name of the
domain is used as a key for some data-lookups, and having two domains
with a same name will either overwrite something or cause wrong domain
data to be fetched. (This is just guessing for now).
Any tips, hints or thoughts on this?
Yours,
-- Matti
--
Matti Vaittinen
Linux kernel developer at ROHM Semiconductors
Oulu Finland
~~ When things go utterly wrong vim users can always type :help! ~~
