On Tue, 6 Feb 2024 at 09:39, Alexander Stein
<alexander.stein@xxxxxxxxxxxxxxx> wrote:
>
> This change uses the appropriate _cansleep or non-sleeping API for
> reading GPIO read-only state. This allows users with GPIOs that
> never sleepbeing called in atomic context.
>
> Implement the same mechanism as in commit 52af318c93e97 ("mmc: Allow
> non-sleeping GPIO cd").
>
> Signed-off-by: Alexander Stein <alexander.stein@xxxxxxxxxxxxxxx>
Applied for fixes and by adding a stable tag, thanks!
Kind regards
Uffe
> ---
> With commit 5d5dfc50e5689 ("gpiolib: remove extra_checks") thw following error
> is showing on my imx8qxp-tqma8xqp-mba8xx.dts platform:
> > BUG: sleeping function called from invalid context at drivers/gpio/gpiolib.c:3740
>
> The reason is that mmc_gpio_get_ro() is called from within atomic context from
> sdhci_check_ro(), which uses spin_lock_irqsave().
>
> The problem was always there, so I assume I didn't actually deadlock cause
> nobody was using SD card with a sleepable read-only GPIO.
>
> drivers/mmc/core/slot-gpio.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/mmc/core/slot-gpio.c b/drivers/mmc/core/slot-gpio.c
> index 2a2d949a9344e..39f45c2b6de8a 100644
> --- a/drivers/mmc/core/slot-gpio.c
> +++ b/drivers/mmc/core/slot-gpio.c
> @@ -75,11 +75,15 @@ EXPORT_SYMBOL(mmc_gpio_set_cd_irq);
> int mmc_gpio_get_ro(struct mmc_host *host)
> {
> struct mmc_gpio *ctx = host->slot.handler_priv;
> + int cansleep;
>
> if (!ctx || !ctx->ro_gpio)
> return -ENOSYS;
>
> - return gpiod_get_value_cansleep(ctx->ro_gpio);
> + cansleep = gpiod_cansleep(ctx->ro_gpio);
> + return cansleep ?
> + gpiod_get_value_cansleep(ctx->ro_gpio) :
> + gpiod_get_value(ctx->ro_gpio);
> }
> EXPORT_SYMBOL(mmc_gpio_get_ro);
>
> --
> 2.34.1
>
