Re: [PATCH 1/3] mailbox: sprd: Introduce refcnt when clients requests/free channels

[Orson Zhai <orsonzhai@gmail.com>]

On Mon, Feb 08, 2021 at 10:06:47PM +0800, Baolin Wang wrote:
> Hi Orson,
> 
> On Mon, Feb 8, 2021 at 7:52 PM Orson Zhai <orsonzhai@gmail.com> wrote:
> >
> > From: Orson Zhai <orson.zhai@unisoc.com>
> >
> > Unisoc mailbox has no way to be enabled/disabled for any single channel.
> > They can only be set to startup or shutdown as a whole device at same time.
> >
> > Add a variable to count references to avoid mailbox FIFO being reset
> > unexpectedly when clients are requesting or freeing channels.
> >
> > Also add a lock to dismiss possible conflicts from register r/w in
> > different startup or shutdown threads.
> >
> > Fixes: ca27fc26cd22 ("mailbox: sprd: Add Spreadtrum mailbox driver")
> > Signed-off-by: Orson Zhai <orson.zhai@unisoc.com>
> > ---
> >  drivers/mailbox/sprd-mailbox.c | 38 +++++++++++++++++++++++++-------------
> >  1 file changed, 25 insertions(+), 13 deletions(-)
> >
> > diff --git a/drivers/mailbox/sprd-mailbox.c b/drivers/mailbox/sprd-mailbox.c
> > index f6fab24..e606f52 100644
> > --- a/drivers/mailbox/sprd-mailbox.c
> > +++ b/drivers/mailbox/sprd-mailbox.c
> > @@ -60,6 +60,8 @@ struct sprd_mbox_priv {
> >         struct clk              *clk;
> >         u32                     outbox_fifo_depth;
> >
> > +       struct mutex            lock;
> > +       u32                     refcnt;
> >         struct mbox_chan        chan[SPRD_MBOX_CHAN_MAX];
> >  };
> >
> > @@ -215,18 +217,22 @@ static int sprd_mbox_startup(struct mbox_chan *chan)
> >         struct sprd_mbox_priv *priv = to_sprd_mbox_priv(chan->mbox);
> >         u32 val;
> >
> > -       /* Select outbox FIFO mode and reset the outbox FIFO status */
> > -       writel(0x0, priv->outbox_base + SPRD_MBOX_FIFO_RST);
> > +       mutex_lock(&priv->lock);
> > +       if (priv->refcnt++ == 0) {
> > +               /* Select outbox FIFO mode and reset the outbox FIFO status */
> > +               writel(0x0, priv->outbox_base + SPRD_MBOX_FIFO_RST);
> >
> > -       /* Enable inbox FIFO overflow and delivery interrupt */
> > -       val = readl(priv->inbox_base + SPRD_MBOX_IRQ_MSK);
> > -       val &= ~(SPRD_INBOX_FIFO_OVERFLOW_IRQ | SPRD_INBOX_FIFO_DELIVER_IRQ);
> > -       writel(val, priv->inbox_base + SPRD_MBOX_IRQ_MSK);
> > +               /* Enable inbox FIFO overflow and delivery interrupt */
> > +               val = readl(priv->inbox_base + SPRD_MBOX_IRQ_MSK);
> > +               val &= ~(SPRD_INBOX_FIFO_OVERFLOW_IRQ | SPRD_INBOX_FIFO_DELIVER_IRQ);
> > +               writel(val, priv->inbox_base + SPRD_MBOX_IRQ_MSK);
> >
> > -       /* Enable outbox FIFO not empty interrupt */
> > -       val = readl(priv->outbox_base + SPRD_MBOX_IRQ_MSK);
> > -       val &= ~SPRD_OUTBOX_FIFO_NOT_EMPTY_IRQ;
> > -       writel(val, priv->outbox_base + SPRD_MBOX_IRQ_MSK);
> > +               /* Enable outbox FIFO not empty interrupt */
> > +               val = readl(priv->outbox_base + SPRD_MBOX_IRQ_MSK);
> > +               val &= ~SPRD_OUTBOX_FIFO_NOT_EMPTY_IRQ;
> > +               writel(val, priv->outbox_base + SPRD_MBOX_IRQ_MSK);
> > +       }
> > +       mutex_unlock(&priv->lock);
> 
> I think using the atomic_add/sub_and_test() related APIs can remove
> the mutex lock.

Yes, atomic could make refcnt itself safe. But mutex lock is to make whole processing of
reading/writing registers safe.

Consider case like this:

  channel #1             channel #2
-------------------------------------
   startup
   .....
   shutdown               startup
     |-refcnt==0            |
     |                      |-retcnt+1
     |                      |-enable mailbox
     |-disable mailbox 

Mailbox will be wrongly disabled after client requests channel #2's startup.

> 
> >
> >         return 0;
> >  }
> > @@ -235,9 +241,13 @@ static void sprd_mbox_shutdown(struct mbox_chan *chan)
> >  {
> >         struct sprd_mbox_priv *priv = to_sprd_mbox_priv(chan->mbox);
> >
> > -       /* Disable inbox & outbox interrupt */
> > -       writel(SPRD_INBOX_FIFO_IRQ_MASK, priv->inbox_base + SPRD_MBOX_IRQ_MSK);
> > -       writel(SPRD_OUTBOX_FIFO_IRQ_MASK, priv->outbox_base + SPRD_MBOX_IRQ_MSK);
> > +       mutex_lock(&priv->lock);
> > +       if (--priv->refcnt == 0) {
> > +               /* Disable inbox & outbox interrupt */
> > +               writel(SPRD_INBOX_FIFO_IRQ_MASK, priv->inbox_base + SPRD_MBOX_IRQ_MSK);
> > +               writel(SPRD_OUTBOX_FIFO_IRQ_MASK, priv->outbox_base + SPRD_MBOX_IRQ_MSK);
> > +       }
> > +       mutex_unlock(&priv->lock);
> >  }
> >
> >  static const struct mbox_chan_ops sprd_mbox_ops = {
> > @@ -266,6 +276,8 @@ static int sprd_mbox_probe(struct platform_device *pdev)
> >                 return -ENOMEM;
> >
> >         priv->dev = dev;
> > +       priv->refcnt = 0;
> 
> No need to do this, the priv structure is already cleared to 0.

Right, will remove at next version.

Thanks for reviewing!

-Orson

> 
> > +       mutex_init(&priv->lock);
> >
> >         /*
> >          * The Spreadtrum mailbox uses an inbox to send messages to the target
> > --
> > 2.7.4
> >
> 
> 
> -- 
> Baolin Wang