Re: [PATCH v2 2/2] gpio: tegra: Use raw_spinlock

[Bartosz Golaszewski]

On Wed, Nov 4, 2020 at 6:04 PM Dmitry Osipenko <digetx@xxxxxxxxx> wrote:
>
> Use raw_spinlock in order to fix spurious messages about invalid context
> when spinlock debugging is enabled. This happens because there is a legit
> nested raw_spinlock->spinlock locking usage within IRQ-related code. IRQ
> core uses raw spinlock and then Tegra GPIO driver uses a nested spinlock.
> The debug code can't recognize and handle this case, hence we need to use
> raw spinlock in the GPIO driver.
>
>  [ BUG: Invalid wait context ]
>  ...
>   (dump_stack) from (__lock_acquire)
>   (__lock_acquire) from (lock_acquire)
>   (lock_acquire) from (_raw_spin_lock_irqsave)
>   (_raw_spin_lock_irqsave) from (tegra_gpio_irq_set_type)
>   (tegra_gpio_irq_set_type) from (__irq_set_trigger)
>   (__irq_set_trigger) from (__setup_irq)
>   (__setup_irq) from (request_threaded_irq)
>   (request_threaded_irq) from (devm_request_threaded_irq)
>   (devm_request_threaded_irq) from (elants_i2c_probe)
>   (elants_i2c_probe) from (i2c_device_probe)
>  ...
>
> Tested-by: Peter Geis <pgwipeout@xxxxxxxxx>
> Signed-off-by: Dmitry Osipenko <digetx@xxxxxxxxx>
> ---
>
> Changelog:
>
> v2: - Only lvl_lock is converted to raw_spinlock. The dbc_lock doesn't
>       relate to IRQ and doesn't need the conversion.
>
>     - Improved commit message by clarifying that IRQ core uses raw
>       spinlock.
>
>     - Added clarifying comment to the code for the lvl_lock.
>
>  drivers/gpio/gpio-tegra.c | 18 +++++++++++++-----
>  1 file changed, 13 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpio/gpio-tegra.c b/drivers/gpio/gpio-tegra.c
> index 98fc78739ebf..e19ebff6018c 100644
> --- a/drivers/gpio/gpio-tegra.c
> +++ b/drivers/gpio/gpio-tegra.c
> @@ -61,8 +61,16 @@ struct tegra_gpio_info;
>  struct tegra_gpio_bank {
>         unsigned int bank;
>         unsigned int irq;
> -       spinlock_t lvl_lock[4];
> -       spinlock_t dbc_lock[4]; /* Lock for updating debounce count register */
> +
> +       /*
> +        * IRQ-core code uses raw locking, and thus, nested locking also
> +        * should be raw in order not to trip spinlock debug warnings.
> +        */
> +       raw_spinlock_t lvl_lock[4];
> +
> +       /* Lock for updating debounce count register */
> +       spinlock_t dbc_lock[4];
> +
>  #ifdef CONFIG_PM_SLEEP
>         u32 cnf[4];
>         u32 out[4];
> @@ -334,14 +342,14 @@ static int tegra_gpio_irq_set_type(struct irq_data *d, unsigned int type)
>                 return -EINVAL;
>         }
>
> -       spin_lock_irqsave(&bank->lvl_lock[port], flags);
> +       raw_spin_lock_irqsave(&bank->lvl_lock[port], flags);
>
>         val = tegra_gpio_readl(tgi, GPIO_INT_LVL(tgi, gpio));
>         val &= ~(GPIO_INT_LVL_MASK << GPIO_BIT(gpio));
>         val |= lvl_type << GPIO_BIT(gpio);
>         tegra_gpio_writel(tgi, val, GPIO_INT_LVL(tgi, gpio));
>
> -       spin_unlock_irqrestore(&bank->lvl_lock[port], flags);
> +       raw_spin_unlock_irqrestore(&bank->lvl_lock[port], flags);
>
>         tegra_gpio_mask_write(tgi, GPIO_MSK_OE(tgi, gpio), gpio, 0);
>         tegra_gpio_enable(tgi, gpio);
> @@ -675,7 +683,7 @@ static int tegra_gpio_probe(struct platform_device *pdev)
>                                                  tegra_gpio_irq_handler, bank);
>
>                 for (j = 0; j < 4; j++) {
> -                       spin_lock_init(&bank->lvl_lock[j]);
> +                       raw_spin_lock_init(&bank->lvl_lock[j]);
>                         spin_lock_init(&bank->dbc_lock[j]);
>                 }
>         }
> --
> 2.27.0
>

Patch applied, thanks!

Bartosz