2022-05-24 18:47:17

by Hector Martin

Subject: [PATCH] pinctrl: apple: Use a raw spinlock for the regmap

The irqchip ops are called with a raw spinlock held, so the subsequent
regmap usage cannot use a plain spinlock.

spi-hid-apple-of spi0.0: spihid_apple_of_probe:74

=============================
[ BUG: Invalid wait context ]
5.18.0-asahi-00176-g0fa3ab03bdea #1337 Not tainted
-----------------------------
kworker/u20:3/86 is trying to lock:
ffff8000166b5018 (pinctrl_apple_gpio:462:(&regmap_config)->lock){....}-{3:3}, at: regmap_lock_spinlock+0x18/0x30
other info that might help us debug this:
context-{5:5}
7 locks held by kworker/u20:3/86:
#0: ffff800017725d48 ((wq_completion)events_unbound){+.+.}-{0:0}, at: process_one_work+0x1c8/0x670
#1: ffff80001e33bdd0 (deferred_probe_work){+.+.}-{0:0}, at: process_one_work+0x1c8/0x670
#2: ffff800017d629a0 (&dev->mutex){....}-{4:4}, at: __device_attach+0x30/0x17c
#3: ffff80002414e618 (&ctlr->add_lock){+.+.}-{4:4}, at: spi_add_device+0x40/0x80
#4: ffff800024116990 (&dev->mutex){....}-{4:4}, at: __device_attach+0x30/0x17c
#5: ffff800022d4be58 (request_class){+.+.}-{4:4}, at: __setup_irq+0xa8/0x720
#6: ffff800022d4bcc8 (lock_class){....}-{2:2}, at: __setup_irq+0xcc/0x720

Fixes: a0f160ffcb83 ("pinctrl: add pinctrl/GPIO driver for Apple SoCs")
Signed-off-by: Hector Martin <[email protected]>
---
drivers/pinctrl/pinctrl-apple-gpio.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/drivers/pinctrl/pinctrl-apple-gpio.c b/drivers/pinctrl/pinctrl-apple-gpio.c
index 3a33731b8cdb..b02dad2b2f69 100644
--- a/drivers/pinctrl/pinctrl-apple-gpio.c
+++ b/drivers/pinctrl/pinctrl-apple-gpio.c
@@ -72,6 +72,7 @@ struct regmap_config regmap_config = {
.max_register = 512 * sizeof(u32),
.num_reg_defaults_raw = 512,
.use_relaxed_mmio = true,
+ .use_raw_spinlock = true,
};

/* No locking needed to mask/unmask IRQs as the interrupt mode is per pin-register. */
--
2.35.1

2022-05-25 20:59:24

by Linus Walleij

Subject: Re: [PATCH] pinctrl: apple: Use a raw spinlock for the regmap

On Tue, May 24, 2022 at 4:22 PM Hector Martin <[email protected]> wrote:

> The irqchip ops are called with a raw spinlock held, so the subsequent
> regmap usage cannot use a plain spinlock.
>
> spi-hid-apple-of spi0.0: spihid_apple_of_probe:74
>
> =============================
> [ BUG: Invalid wait context ]
> 5.18.0-asahi-00176-g0fa3ab03bdea #1337 Not tainted
> -----------------------------
> kworker/u20:3/86 is trying to lock:
> ffff8000166b5018 (pinctrl_apple_gpio:462:(&regmap_config)->lock){....}-{3:3}, at: regmap_lock_spinlock+0x18/0x30
> other info that might help us debug this:
> context-{5:5}
> 7 locks held by kworker/u20:3/86:
> #0: ffff800017725d48 ((wq_completion)events_unbound){+.+.}-{0:0}, at: process_one_work+0x1c8/0x670
> #1: ffff80001e33bdd0 (deferred_probe_work){+.+.}-{0:0}, at: process_one_work+0x1c8/0x670
> #2: ffff800017d629a0 (&dev->mutex){....}-{4:4}, at: __device_attach+0x30/0x17c
> #3: ffff80002414e618 (&ctlr->add_lock){+.+.}-{4:4}, at: spi_add_device+0x40/0x80
> #4: ffff800024116990 (&dev->mutex){....}-{4:4}, at: __device_attach+0x30/0x17c
> #5: ffff800022d4be58 (request_class){+.+.}-{4:4}, at: __setup_irq+0xa8/0x720
> #6: ffff800022d4bcc8 (lock_class){....}-{2:2}, at: __setup_irq+0xcc/0x720
>
> Fixes: a0f160ffcb83 ("pinctrl: add pinctrl/GPIO driver for Apple SoCs")
> Signed-off-by: Hector Martin <[email protected]>

Patch applied!

Yours,
Linus Walleij