On Thu, Sep 04, 2025 at 02:05:57PM +0200, Andrew Lunn wrote:
> On Thu, Sep 04, 2025 at 11:19:48AM +0800, Yibo Dong wrote:
> > On Thu, Sep 04, 2025 at 12:24:17AM +0200, Andrew Lunn wrote:
> > > > struct mucse_mbx_info {
> > > > + struct mucse_mbx_stats stats;
> > > > + u32 timeout;
> > > > + u32 usec_delay;
> > > > + u16 size;
> > > > + u16 fw_req;
> > > > + u16 fw_ack;
> > > > + /* lock for only one use mbx */
> > > > + struct mutex lock;
> > > > /* fw <--> pf mbx */
> > > > u32 fw_pf_shm_base;
> > > > u32 pf2fw_mbox_ctrl;
> > >
> > > > +/**
> > > > + * mucse_obtain_mbx_lock_pf - Obtain mailbox lock
> > > > + * @hw: pointer to the HW structure
> > > > + *
> > > > + * This function maybe used in an irq handler.
> > > > + *
> > > > + * Return: 0 if we obtained the mailbox lock or else -EIO
> > > > + **/
> > > > +static int mucse_obtain_mbx_lock_pf(struct mucse_hw *hw)
> > > > +{
> > > > + struct mucse_mbx_info *mbx = &hw->mbx;
> > > > + int try_cnt = 5000;
> > > > + u32 reg;
> > > > +
> > > > + reg = PF2FW_MBOX_CTRL(mbx);
> > > > + while (try_cnt-- > 0) {
> > > > + mbx_ctrl_wr32(mbx, reg, MBOX_PF_HOLD);
> > > > + /* force write back before check */
> > > > + wmb();
> > > > + if (mbx_ctrl_rd32(mbx, reg) & MBOX_PF_HOLD)
> > > > + return 0;
> > > > + udelay(100);
> > > > + }
> > > > + return -EIO;
> > > > +}
> > >
> > > If there is a function which obtains a lock, there is normally a
> > > function which releases a lock. But i don't see it.
> > >
> >
> > The lock is relased when send MBOX_CTRL_REQ in mucse_write_mbx_pf:
> >
> > mbx_ctrl_wr32(mbx, ctrl_reg, MBOX_CTRL_REQ);
> >
> > Set MBOX_PF_HOLD(bit3) to hold the lock, clear bit3 to release, and set
> > MBOX_CTRL_REQ(bit0) to send the req. req and lock are different bits in
> > one register. So we send the req along with releasing lock (set bit0 and
> > clear bit3).
> > Maybe I should add comment like this?
> >
> > /* send the req along with releasing the lock */
> > mbx_ctrl_wr32(mbx, ctrl_reg, MBOX_CTRL_REQ);
>
> As i said, functions like this come in pairs. obtain/release,
> lock/unlock. When reading code, you want to be able to see both of the
> pair in a function, to know the unlock is not missing. The kernel even
> has tools which will validate all paths through a function releasing
> locks. Often error paths get this wrong.
>
> So please make this a function, give it a name which makes it obvious
> it is the opposite of mucse_obtain_mbx_lock_pf().
>
> Andrew
>
Got it, I will update this.
Thanks for your feedback.
