Re: [PATCH v5 4/6] i2c: npcm: Modify the client address assignment

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Fri, Oct 04, 2024 at 10:29:10AM +0800, Tyrone Ting wrote:
> Hi Andy:
> 
> Thank you for your comments.
> 
> After a second thought, I'll explain why slave_addr << 1 is given here.
> 
> Tyrone Ting <warp5tw@gmail.com> 於 2024年10月4日 週五 上午9:49寫道：
> >
> > Hi Andy:
> >
> > Thank you for your comments and they'll be addressed.
> >
> > Andy Shevchenko <andriy.shevchenko@linux.intel.com> 於 2024年10月1日 週二 下午9:17寫道：
> > >
> > > On Tue, Oct 01, 2024 at 02:28:53PM +0800, Tyrone Ting wrote:
> > > > From: Tyrone Ting <kfting@nuvoton.com>
> > > >
> > > > Store the client address earlier since it might get called in
> > > > the i2c_recover_bus() logic flow at the early stage of
> > > > npcm_i2c_master_xfer().
> > >
> > > ...
> > >
> > > > +     /*
> > > > +      * Previously, the address was stored w/o left-shift by one bit and
> > > > +      * with that shift in the following call to npcm_i2c_master_start_xmit().
> > > > +      *
> > > > +      * Since there are cases that the i2c_recover_bus() gets called at the
> > > > +      * early stage of npcm_i2c_master_xfer(), the address is stored with
> > > > +      * the shift and used in the i2c_recover_bus().
> > > > +      *
> > > > +      * The address is stored from bit 1 to bit 7 in the register for
> > > > +      * sending the i2c address later so it's left-shifted by 1 bit.
> > > > +      */
> > > > +     bus->dest_addr = slave_addr << 1;
> > >
> > > I'm wondering if it's better to use i2c_8bit_addr_from_msg() here?
> > >
> 
> The current implementation of i2c_8bit_addr_from_msg() (ref link:
> https://github.com/torvalds/linux/blob/master/include/linux/i2c.h#L947)
> is
> "return (msg->addr << 1) | (msg->flags & I2C_M_RD);" and it takes
> extra consideration about the read flag when retrieving the i2c
> address.
> IOW, if there is a read event, the i2c address contains a read
> indication (bit 0 of the i2c address is 1).
> 
> The patch code "bus->dest_addr = slave_addr << 1;" might get used in
> i2c_recover_bus() later. (ref link:
> https://github.com/torvalds/linux/blob/master/drivers/i2c/busses/i2c-npcm7xx.c#L1691)
> 
> Suppose there is a read event and the i2c address is 0x60.
> 
> With i2c_8bit_addr_from_msg(), bus->dest_addr will be 0xc1.
> With the original patch, bus->dest_addr will be 0xc0.
> 
> If some error condition occurs and it requires i2c_recover_bus() to
> recover the bus, according to the description at
> https://github.com/torvalds/linux/blob/master/drivers/i2c/busses/i2c-npcm7xx.c#L1742,
> the address "0xc1" is used
> as a parameter to npcm_i2c_wr_byte() which is used to send the address
> in the write direction.
> 
> If i2c_8bit_addr_from_msg() is applied, it might not fit the scenario
> described at
> https://github.com/torvalds/linux/blob/master/drivers/i2c/busses/i2c-npcm7xx.c#L1742,
> which is about to send
> an address in a write direction since the address from
> i2c_8bit_addr_from_msg() contains a read indication.

Okay, then I would do the i2c_8bit_addr_from_msg() call here as AFAICS
this is the real event where you save the address *of the event*.

And in the respective user update the comment to summarize above and do
rather ->dest_addr & ~I2C_M_RD there.

-- 
With Best Regards,
Andy Shevchenko