Hi Marc,

On Mon, Nov 22, 2021 at 5:58 PM Geert Uytterhoeven [off-list ref] wrote:

On Mon, Nov 22, 2021 at 2:54 PM Marc Zyngier [off-list ref] wrote:

quoted

On Mon, 22 Nov 2021 13:10:32 +0000,
Geert Uytterhoeven [off-list ref] wrote:

quoted

On Mon, Nov 22, 2021 at 11:30 AM Marc Zyngier [off-list ref] wrote:

quoted

Since 041284181226 ("of/irq: Allow matching of an interrupt-map local
to an interrupt controller"), a handful of interrupt controllers have
stopped working correctly. This is due to the DT exposing a non-sensical
interrupt-map property, and their drivers relying on the kernel ignoring
this property.

Since we cannot realistically fix this terrible behaviour, add a quirk
for the limited set of devices that have implemented this monster,
and document that this is a pretty bad practice.

quoted

quoted

quoted

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

@@ -76,6 +76,36 @@ struct device_node *of_irq_find_parent(struct device_node *child)
 }
 EXPORT_SYMBOL_GPL(of_irq_find_parent);

+/*
+ * These interrupt controllers abuse interrupt-map for unspeakable
+ * reasons and rely on the core code to *ignore* it (the drivers do
+ * their own parsing of the property).
+ *
+ * If you think of adding to the list for something *new*, think
+ * again. There is a high chance that you will be sent back to the
+ * drawing board.
+ */
+static const char * const of_irq_imap_abusers[] = {
+       "CBEA,platform-spider-pic",
+       "sti,platform-spider-pic",
+       "realtek,rtl-intc",
+       "fsl,ls1021a-extirq",
+       "fsl,ls1043a-extirq",
+       "fsl,ls1088a-extirq",
+       "renesas,rza1-irqc",
+};

Are you sure "renesas,rza1-irqc" handles this wrong? How should it
be handled instead? I read the other thread[1], but didn't became
any wiser: interrupts are mapped one-to-one with the RZ/A1 IRQC.

In both v5.15 and v5.16-rc1, interrupts seem to work fine on RSK+RZA1
and RZA2MEVB, both with gpio-keys and when used as a wake-up interrupt.

Oops, it turned out my "v5.15" tree was not plain v5.15, but v5.15 with
some parts of next, including an older version of commit 041284181226.

quoted

This is odd. 5.16-rc1 should actively breaks the behaviour, as each
interrupt is directly routed to the GIC. Here's an extract of the DT
for r7s9210:

    interrupt-map = <0 0 &gic GIC_SPI 4 IRQ_TYPE_LEVEL_HIGH>,
                    <1 0 &gic GIC_SPI 5 IRQ_TYPE_LEVEL_HIGH>,
                    <2 0 &gic GIC_SPI 6 IRQ_TYPE_LEVEL_HIGH>,
                    <3 0 &gic GIC_SPI 7 IRQ_TYPE_LEVEL_HIGH>,
                    <4 0 &gic GIC_SPI 8 IRQ_TYPE_LEVEL_HIGH>,
                    <5 0 &gic GIC_SPI 9 IRQ_TYPE_LEVEL_HIGH>,
                    <6 0 &gic GIC_SPI 10 IRQ_TYPE_LEVEL_HIGH>,
                    <7 0 &gic GIC_SPI 11 IRQ_TYPE_LEVEL_HIGH>;

I expect v5.16-rc1 to honour the routing described here and not
involve rza1-irqc, because that's what the DT says.

quoted

With this patch applied, I see double keypresses with evtest: when
pressing a key, I get a key-down event, immediately followed by a
key-up event. When releasing the key, I again get two events.

Good (v5.15 or v5.16-rc1):

    Event: time 1637585631.288990, type 1 (EV_KEY), code 2 (KEY_1), value 1
    Event: time 1637585631.288990, -------------- SYN_REPORT ------------
    Event: time 1637585631.499924, type 1 (EV_KEY), code 2 (KEY_1), value 0
    Event: time 1637585631.499924, -------------- SYN_REPORT ------------

Bad (v5.16-rc1 + this patch):

    Event: time 1637585341.946647, type 1 (EV_KEY), code 2 (KEY_1), value 1
    Event: time 1637585341.946647, -------------- SYN_REPORT ------------
    Event: time 1637585341.960256, type 1 (EV_KEY), code 2 (KEY_1), value 0
    Event: time 1637585341.960256, -------------- SYN_REPORT ------------
    Event: time 1637585342.146775, type 1 (EV_KEY), code 2 (KEY_1), value 1
    Event: time 1637585342.146775, -------------- SYN_REPORT ------------
    Event: time 1637585342.160092, type 1 (EV_KEY), code 2 (KEY_1), value 0
    Event: time 1637585342.160092, -------------- SYN_REPORT ------------

Is there any chance you could trace whether rza1-irqc gets called at
all when setting up and handling the interrupt?

I reran my tests ([A] pristine v5.15, [B] my current tree based on v5.16-rc1,
[C] my tree plus your patch).

[A] and [B] behave the same:

Sorry, [A] and [C]:

  Boot:

    rza1_irqc_translate:152: domain :soc:interrupt-controller@fcfef800
hwirq 3 type 3
    rza1_irqc_alloc:115: domain :soc:interrupt-controller@fcfef800
virq 41 nr_irqs 1
    rza1_irqc_alloc:127: param[0] = 0
    rza1_irqc_alloc:127: param[1] = 3
    rza1_irqc_alloc:127: param[2] = 4
    rza1_irqc_translate:152: domain :soc:interrupt-controller@fcfef800
hwirq 2 type 3
    rza1_irqc_alloc:115: domain :soc:interrupt-controller@fcfef800
virq 42 nr_irqs 1
    rza1_irqc_alloc:127: param[0] = 0
    rza1_irqc_alloc:127: param[1] = 2
    rza1_irqc_alloc:127: param[2] = 4
    rza1_irqc_translate:152: domain :soc:interrupt-controller@fcfef800
hwirq 5 type 3
    rza1_irqc_alloc:115: domain :soc:interrupt-controller@fcfef800
virq 43 nr_irqs 1
    rza1_irqc_alloc:127: param[0] = 0
    rza1_irqc_alloc:127: param[1] = 5
    rza1_irqc_alloc:127: param[2] = 4
    rza1_irqc_set_type:76: hwirq 3 type 3
    rza1_irqc_set_type:76: hwirq 2 type 3
    rza1_irqc_set_type:76: hwirq 5 type 3

  Pressing all 3 keys on RSK+RZA1:

    rza1_irqc_eoi:62: hw_irq 3 IRQRR 0x8
    rza1_irqc_eoi:62: hw_irq 3 IRQRR 0x8
    rza1_irqc_eoi:62: hw_irq 2 IRQRR 0x4
    rza1_irqc_eoi:62: hw_irq 2 IRQRR 0x4
    rza1_irqc_eoi:62: hw_irq 5 IRQRR 0x20
    rza1_irqc_eoi:62: hw_irq 5 IRQRR 0x20

  /proc/interrupts:

    41:          2  rza1-irqc   3 Edge      SW1
    42:          2  rza1-irqc   2 Edge      SW2
    43:          2  rza1-irqc   5 Edge      SW3

  evtest:

    Event: time 1637597938.224621, type 1 (EV_KEY), code 2 (KEY_1), value 1
    Event: time 1637597938.224621, -------------- SYN_REPORT ------------
    Event: time 1637597938.232198, type 1 (EV_KEY), code 2 (KEY_1), value 0
    Event: time 1637597938.232198, -------------- SYN_REPORT ------------
    Event: time 1637597938.532939, type 1 (EV_KEY), code 2 (KEY_1), value 1
    Event: time 1637597938.532939, -------------- SYN_REPORT ------------
    Event: time 1637597938.542304, type 1 (EV_KEY), code 2 (KEY_1), value 0
    Event: time 1637597938.542304, -------------- SYN_REPORT ------------
    Event: time 1637597941.772467, type 1 (EV_KEY), code 3 (KEY_2), value 1
    Event: time 1637597941.772467, -------------- SYN_REPORT ------------
    Event: time 1637597941.782309, type 1 (EV_KEY), code 3 (KEY_2), value 0
    Event: time 1637597941.782309, -------------- SYN_REPORT ------------
    Event: time 1637597942.110321, type 1 (EV_KEY), code 3 (KEY_2), value 1
    Event: time 1637597942.110321, -------------- SYN_REPORT ------------
    Event: time 1637597942.122303, type 1 (EV_KEY), code 3 (KEY_2), value 0
    Event: time 1637597942.122303, -------------- SYN_REPORT ------------
    Event: time 1637597945.256109, type 1 (EV_KEY), code 4 (KEY_3), value 1
    Event: time 1637597945.256109, -------------- SYN_REPORT ------------
    Event: time 1637597945.262132, type 1 (EV_KEY), code 4 (KEY_3), value 0
    Event: time 1637597945.262132, -------------- SYN_REPORT ------------
    Event: time 1637597945.630469, type 1 (EV_KEY), code 4 (KEY_3), value 1
    Event: time 1637597945.630469, -------------- SYN_REPORT ------------
    Event: time 1637597945.642299, type 1 (EV_KEY), code 4 (KEY_3), value 0
    Event: time 1637597945.642299, -------------- SYN_REPORT ------------

So despite seeing only 2 interrupts per key, gpio-keys generates
4 events per key.

With my v5.16-rc1-based tree, rza1_irqc_translate(), rza1_irqc_alloc(),
rza1_irqc_set_type(), and rza1_irqc_eoi() are indeed not called.

Hence this is [B], i.e. after the bad commit:

  /proc/interrupts:

    41:     242419     GIC-0  35 Level     SW1
    42:     142771     GIC-0  34 Level     SW2
    43:     136355     GIC-0  37 Level     SW3
            ^^^^^^
            Oops

  evtest:

    Event: time 1637598499.076306, type 1 (EV_KEY), code 2 (KEY_1), value 1
    Event: time 1637598499.076306, -------------- SYN_REPORT ------------
    Event: time 1637598499.350985, type 1 (EV_KEY), code 2 (KEY_1), value 0
    Event: time 1637598499.350985, -------------- SYN_REPORT ------------
    Event: time 1637598501.979770, type 1 (EV_KEY), code 3 (KEY_2), value 1
    Event: time 1637598501.979770, -------------- SYN_REPORT ------------
    Event: time 1637598502.370948, type 1 (EV_KEY), code 3 (KEY_2), value 0
    Event: time 1637598502.370948, -------------- SYN_REPORT ------------
    Event: time 1637598504.660146, type 1 (EV_KEY), code 4 (KEY_3), value 1
    Event: time 1637598504.660146, -------------- SYN_REPORT ------------
    Event: time 1637598505.030947, type 1 (EV_KEY), code 4 (KEY_3), value 0
    Event: time 1637598505.030947, -------------- SYN_REPORT ------------

So despite receiving an interrupt storm, gpio-keys behaves as expected.

I will retest tomorrow with an old kernel, as I do not remember seeing such
behavior when I wrote the rza1-irqc driver.

Summarized:
  - Before the bad commit, and after your fix, irqc-rza1 is invoked,
    and the number of interrupts seen is correct, but input events
    are doubled.
  - After the bad commit, irqc-rza1 is not invoked, and there is an
    interrupt storm, but input events are OK.

Sorry for the confusion.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds