Re: [PATCH v1] i3c: master: svc: Fix npcm845 FIFO_EMPTY quirk

[Frank Li <Frank.li@nxp.com>]

On Fri, Jul 25, 2025 at 10:50:34PM +0800, Stanley Chu wrote:
> Hi Frank,
>
> On Fri, Jul 25, 2025 at 9:45 PM Frank Li <Frank.li@nxp.com> wrote:
> >
> > On Fri, Jul 25, 2025 at 03:03:10PM +0800, Stanley Chu wrote:
> > > From: Stanley Chu <yschu@nuvoton.com>
> > >
> > > Prefilling in private write transfers is only necessary when the FIFO
> > > is empty. Otherwise, if the transfer is NACKed due to IBIWON and retries
> > > continue, data may be prefilled again but could be lost because the FIFO
> > > is not empty.
> >
> > why "maybe prefilled", please use certain words.
> The original code didn't consider the retry case, data lost happen in
> the retry case only when total write bytes is larger than FIFO size,
> not always happen.
> In the example of writing 17 bytes (FIFO size is 16), the last byte is lost.
> 1. Emit S+Addr/W
> 2. Write 16 bytes to FIFO (FIFO is full)
> 3. NACKed due to IBIWON
> 4. (retry) Emit Sr+Addr/W
> 5. Write last byte to FIFO (data lost)

You need descript clearly at commit message about this.

>
> I just need to prefill FIFO in the beginning to make sure FIFO is not
> empty when HW starts transmitting.
> So, this patch adds a condition that prefill FIFO only when FIFO is empty.
>
> >
> > >
> > > Fixes: 4008a74e0f9b ("i3c: master: svc: Fix npcm845 FIFO empty issue")
> > > Signed-off-by: Stanley Chu <yschu@nuvoton.com>
> > > ---
> > >  drivers/i3c/master/svc-i3c-master.c | 11 ++++++++++-
> > >  1 file changed, 10 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/i3c/master/svc-i3c-master.c b/drivers/i3c/master/svc-i3c-master.c
> > > index 7e1a7cb94b43..34b6e125b18a 100644
> > > --- a/drivers/i3c/master/svc-i3c-master.c
> > > +++ b/drivers/i3c/master/svc-i3c-master.c
> > > @@ -104,6 +104,7 @@
> > >  #define   SVC_I3C_MDATACTRL_TXTRIG_FIFO_NOT_FULL GENMASK(5, 4)
> > >  #define   SVC_I3C_MDATACTRL_RXTRIG_FIFO_NOT_EMPTY 0
> > >  #define   SVC_I3C_MDATACTRL_RXCOUNT(x) FIELD_GET(GENMASK(28, 24), (x))
> > > +#define   SVC_I3C_MDATACTRL_TXCOUNT(x) FIELD_GET(GENMASK(20, 16), (x))
> > >  #define   SVC_I3C_MDATACTRL_TXFULL BIT(30)
> > >  #define   SVC_I3C_MDATACTRL_RXEMPTY BIT(31)
> > >
> > > @@ -280,6 +281,13 @@ static inline bool is_events_enabled(struct svc_i3c_master *master, u32 mask)
> > >       return !!(master->enabled_events & mask);
> > >  }
> > >
> > > +static inline bool svc_i3c_master_tx_empty(struct svc_i3c_master *master)
> > > +{
> > > +     u32 reg = readl(master->regs + SVC_I3C_MDATACTRL);
> > > +
> > > +     return (SVC_I3C_MDATACTRL_TXCOUNT(reg) == 0);
> > > +}
> > > +
> > >  static bool svc_i3c_master_error(struct svc_i3c_master *master)
> > >  {
> > >       u32 mstatus, merrwarn;
> > > @@ -1303,7 +1311,8 @@ static int svc_i3c_master_xfer(struct svc_i3c_master *master,
> > >                * The only way to work around this hardware issue is to let the
> > >                * FIFO start filling as soon as possible after EmitStartAddr.
> > >                */
> > > -             if (svc_has_quirk(master, SVC_I3C_QUIRK_FIFO_EMPTY) && !rnw && xfer_len) {
> > > +             if (svc_has_quirk(master, SVC_I3C_QUIRK_FIFO_EMPTY) && !rnw && xfer_len &&
> > > +                 svc_i3c_master_tx_empty(master)) {
> > >                       u32 end = xfer_len > SVC_I3C_FIFO_SIZE ? 0 : SVC_I3C_MWDATAB_END;
> > >                       u32 len = min_t(u32, xfer_len, SVC_I3C_FIFO_SIZE);
> >
> > if prefill to FIFO SIZE each time, such as
> >
> > replace SVC_I3C_FIFO_SIZE with
> >
> > SVC_I3C_FIFO_SIZE - SVC_I3C_MDATACTRL_TXCOUNT(readl(master->regs + SVC_I3C_MDATACTRL)).
> >
> The free space is SVC_I3C_FIFO_SIZE when FIFO is empty, no need to
> check TX count again.

My means is that if use above logic, needn't check helper function
svc_i3c_master_tx_empty(master).

Does this method work?

Frank

>
> --
> Stanley
>
> > Frank
> > >
> > > --
> > > 2.34.1
> > >
>
> --
> linux-i3c mailing list
> linux-i3c@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-i3c