[PATCH] i2c: bcm-iproc: Fix bcm_iproc_i2c_isr deadlock issue

[Ray Jui]

Hi Chengfeng,

On 6/24/2023 12:36 PM, YE Chengfeng wrote:
> iproc_i2c_rd_reg and iproc_i2c_wr_reg are called from both
> interrupt context (e.g. bcm_iproc_i2c_isr) and process context
> (e.g. bcm_iproc_i2c_suspend). Therefore, interrupts should be
> disabled to avoid potential deadlock. To prevent this deadlock,
> use spin_lock_irqsave.
> 
> Signed-off-by: Chengfeng Ye <cyeaa at connect.ust.hk>
> ---
>  drivers/i2c/busses/i2c-bcm-iproc.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/i2c/busses/i2c-bcm-iproc.c b/drivers/i2c/busses/i2c-bcm-iproc.c
> index 85d8a6b04885..d02245e4db8c 100644
> --- a/drivers/i2c/busses/i2c-bcm-iproc.c
> +++ b/drivers/i2c/busses/i2c-bcm-iproc.c
> @@ -233,13 +233,14 @@ static inline u32 iproc_i2c_rd_reg(struct bcm_iproc_i2c_dev *iproc_i2c,
>                                    u32 offset)
>  {
>         u32 val;
> +       unsigned long flags;
> 
>         if (iproc_i2c->idm_base) {
> -               spin_lock(&iproc_i2c->idm_lock);
> +               spin_lock_irqsave(&iproc_i2c->idm_lock, flags);
>                 writel(iproc_i2c->ape_addr_mask,
>                        iproc_i2c->idm_base + IDM_CTRL_DIRECT_OFFSET);
>                 val = readl(iproc_i2c->base + offset);
> -               spin_unlock(&iproc_i2c->idm_lock);
> +               spin_unlock_irqrestore(&iproc_i2c->idm_lock, flags);
>         } else {
>                 val = readl(iproc_i2c->base + offset);
>         }
> @@ -250,12 +251,13 @@ static inline u32 iproc_i2c_rd_reg(struct bcm_iproc_i2c_dev *iproc_i2c,
>  static inline void iproc_i2c_wr_reg(struct bcm_iproc_i2c_dev *iproc_i2c,
>                                     u32 offset, u32 val)
>  {
> +       unsigned long flags;
>         if (iproc_i2c->idm_base) {
> -               spin_lock(&iproc_i2c->idm_lock);
> +               spin_lock_irqsave(&iproc_i2c->idm_lock, flags);
>                 writel(iproc_i2c->ape_addr_mask,
>                        iproc_i2c->idm_base + IDM_CTRL_DIRECT_OFFSET);
>                 writel(val, iproc_i2c->base + offset);
> -               spin_unlock(&iproc_i2c->idm_lock);
> +               spin_unlock_irqrestore(&iproc_i2c->idm_lock, flags);
>         } else {
>                 writel(val, iproc_i2c->base + offset);
>         }
> -- 
> 2.17.1

This fix looks good to me. Thanks. Just curious, did you actually see a
race condition issue as a result of this, or the fix is done completely
based on the analysis of the code?

Acked-by: Ray Jui <ray.jui at broadcom.com>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4194 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20230626/f37d970e/attachment-0001.p7s>