Re: [RFT PATCH v2 08/10] drivers: qcom: rpmh-rsc: spin_lock_irqsave() for tcs_invalidate()

[Doug Anderson]

Hi,

On Wed, Mar 11, 2020 at 4:14 PM Douglas Anderson <dianders@xxxxxxxxxxxx> wrote:
>
> Auditing tcs_invalidate() made me worried.  Specifically I saw that it
> used spin_lock(), not spin_lock_irqsave().  That always worries me.
>
> As I understand it, using spin_lock() is only valid in these
> situations:
>
> a) You know you are running in the interrupt handler (and all other
>    users of the lock use the "irqsave" variant).
> b) You know that nobody using the lock is ever running in the
>    interrupt handler.
> c) You know that someone else has always disabled interrupts before
>    your code runs and thus the "irqsave" variant is pointless.
>
> From auditing the driver we look OK.  ...except that there is one
> further corner case.  If sometimes your code is called with IRQs
> disabled and sometimes it's not you will get in trouble if someone
> ever boots your board with "nosmp" (AKA in uniprocessor mode).  In
> such a case if someone else has the lock (without disabling
> interrupts) and they get swapped out then your code (with interrupts
> disabled) might loop forever waiting for the spinlock.
>
> It's just safer to use the irqsave version, so let's do that.  In
> future patches I believe tcs_invalidate() will always be called with
> interrupts off anyway.
>
> Signed-off-by: Douglas Anderson <dianders@xxxxxxxxxxxx>
> ---
>
> Changes in v2: None
>
>  drivers/soc/qcom/rpmh-rsc.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/soc/qcom/rpmh-rsc.c b/drivers/soc/qcom/rpmh-rsc.c
> index ba489d18c20e..c82c734788b1 100644
> --- a/drivers/soc/qcom/rpmh-rsc.c
> +++ b/drivers/soc/qcom/rpmh-rsc.c
> @@ -218,9 +218,10 @@ static bool tcs_is_free(struct rsc_drv *drv, int tcs_id)
>  static int tcs_invalidate(struct rsc_drv *drv, int type)
>  {
>         int m;
> +       unsigned long flags;
>         struct tcs_group *tcs = &drv->tcs[type];
>
> -       spin_lock(&tcs->lock);
> +       spin_lock_irqsave(&tcs->lock, flags);
>         if (bitmap_empty(tcs->slots, MAX_TCS_SLOTS)) {
>                 spin_unlock(&tcs->lock);

Noticed a bug while doing a code review of:

https://lkml.kernel.org/r/1585244270-637-7-git-send-email-mkshah@xxxxxxxxxxxxxx

...specifically my patch forgets to change the error case to
spin_unlock_irqrestore().  ...but perhaps if that other patch lands
when we can just remove the spinlocks from this function...  I'll post
more in my reply to that other patch.


-Doug