[PATCH v1 1/3] dt-bindings: interrupt-controller: Require trigger type for T-HEAD PLIC

Guo Ren guoren at kernel.org
Mon Jun 27 20:04:22 PDT 2022


On Mon, Jun 27, 2022 at 10:14 PM Samuel Holland <samuel at sholland.org> wrote:
>
> On 6/27/22 2:40 AM, Guo Ren wrote:
> > On Mon, Jun 27, 2022 at 1:13 PM Samuel Holland <samuel at sholland.org> wrote:
> >>
> >> The RISC-V PLIC specification unfortunately allows PLIC implementations
> >> to ignore edges seen while an edge-triggered interrupt is being handled:
> >>
> >>   Depending on the design of the device and the interrupt handler,
> >>   in between sending an interrupt request and receiving notice of its
> >>   handler’s completion, the gateway might either ignore additional
> >>   matching edges or increment a counter of pending interrupts.
> >>
> >> For PLICs with that misfeature, software needs to know the trigger type
> >> of each interrupt. This allows it to work around the issue by completing
> >> edge-triggered interrupts before handling them. Such a workaround is
> >> required to avoid missing any edges.
> >>
> >> The T-HEAD C9xx PLIC is an example of a PLIC with this behavior.
> > Actually, C9xx support pulse signals which configed by
> > pad_plic_int_cfg_x for SoC vendor:
> >
> > https://github.com/T-head-Semi/openc906/blob/main/C906_RTL_FACTORY/gen_rtl/plic/rtl/plic_int_kid.v
> > 104: assign int_new_pending = pad_plic_int_cfg_x ? int_pulse
> > 105:
> >         : level_int_pending;
> >
> > They could put pad_plic_int_cfg_x into the SoC software config
> > registers region or bind them to constant values.
>
> The issue here is the "!int_active" condition for int_new_set_pending,
> regardless of that configuration input.
>
> For interrupt sources that send pulses, those pulses will be lost if they are
> received while int_active is true. So the driver has to make sure int_active is
> false _before_ servicing an interrupt, or it may miss the next one. This means
> the driver needs to know which interrupt _sources_ send pulses and which ones
> assert levels.
You are right, in plus mode, we can't receive any interrupt between
claim & complete, then the irq could be lost.

I think the design follows the sentence written by PLIC spec:
https://github.com/riscv/riscv-plic-spec/blob/master/riscv-plic.adoc

If the global interrupt source was edge-triggered, the gateway will
convert the first matching signal edge into an interrupt request.
Depending on the design of the device and the interrupt handler, in
between sending an interrupt request and receiving notice of its
handler’s completion, the gateway might either **ignore additional
matching edges** or increment a counter of pending interrupts.

Also, the hardware could easily support the feature, but the engineer
follows the first choice and disable the useful function. -_*!

>
> For level interrupts, pad_plic_int_cfg_x had better be 0, but that is a
> hardware/firmware configuration concern. The driver should not have to care
> about that.
>
> (On the Allwinner D1, those inputs are memory mapped, which was the reason for
> the stacked interrupt controller mentioned in my other reply. But while
> pad_plic_int_cfg_x == 1 only works with edge interrupts, the pad_plic_int_cfg_x
> == 0 choice works with either kind of interrupt source, so the stacked driver is
> not really needed.)
>
> Regards,
> Samuel
--
Best Regards
 Guo Ren

ML: https://lore.kernel.org/linux-csky/



More information about the linux-riscv mailing list