On 2020/12/7 2:50, Marc Zyngier wrote:
On 2020-12-06 15:02, Linus Walleij wrote:
On Sat, Dec 5, 2020 at 11:15 PM Serge Semin <fancer.lancer@xxxxxxxxx>
wrote:
Hmm, that sounds like a problem, but the explanation is a bit unclear
to me. AFAICS you are saying that the only callbacks which are
called during the IRQ request/release are the irq_enable(), right? If
so then the only reason why we haven't got a problem reported due to
that so far is that the IRQs actually unmasked by default.
What we usually do in cases like that (and I have discussed this
with tglx in the past I think) is to simply mask off all IRQs in
probe().
Then they will be unmasked when requested by drivers.
See e.g. gpio-pl061 that has this line in probe():
writeb(0, pl061->base + GPIOIE); /* disable irqs */
This should definitely be the default behaviour. The code code
expects all interrupt sources to be masked until actively enabled,
usually with the IRQ being requested.
I think this patch is used for that purpose. I do two things in
irq_enable(): unmask irq and then enable IRQ;
and for irq_disable(), it's similar; mask IRQ then disable it.
Thanks
Jiaxing
Thanks,
M.
