Re: [PATCH] serial: 8250_bcm7271: improve bcm7271 8250 port

[Andy Shevchenko]

On Mon, Aug 14, 2023 at 11:09:46AM -0700, Doug Berger wrote:
> On 8/14/2023 9:28 AM, Justin Chen wrote:
> > On 8/14/23 8:12 AM, Andy Shevchenko wrote:
> > > On Sat, Aug 12, 2023 at 09:24:21PM -0700, Justin Chen wrote:
> > > > On Sat, Aug 12, 2023 at 3:50 AM Greg Kroah-Hartman
> > > > <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > > > > On Fri, Aug 11, 2023 at 03:14:01PM -0700, Justin Chen wrote:

...

> > > > > > +     [PORT_BCM7271] = {
> > > > > > +             .name           = "bcm7271_uart",
> > > 
> > > This is badly named port type.
> > > 
> This may be true, but it does mirror the PORT_BCM63XX naming and I do value
> consistency so it is acceptable to me. However, I will happily yield to a
> better name if one can be determined by popular consensus.
> 
> > 
> > Would "Brcmstb 7271 UART" suffice?
> > 
> Perhaps, "Broadcom BCM7271 UART" but it seems excessively "chatty" to me, so
> as I said I am OK with the original submission.

I'm not okay, sorry. But your variant seems the best from all proposed.

> > > > > > +             .fifo_size      = 32,
> > > > > > +             .tx_loadsz      = 32,
> > > > > > +             .fcr            = UART_FCR_ENABLE_FIFO |
> > > > > > UART_FCR_R_TRIG_01,
> > > > > > +             .rxtrig_bytes   = {1, 8, 16, 30},
> > > > > > +             .flags          = UART_CAP_FIFO | UART_CAP_AFE
> > > > > > +     },
> > > > > >   };
> > > 
> > > This is almost a dup of PORT_ALTR_16550_F32. Use it if you wish.
> > > You can always rename it if it feels the right thing to do.
> > > 
> > 
> > There is some other PORT_ALTR logic that I would like to avoid. I would
> > also like to avoid future changes to PORT_ALTR that wouldn't be
> > applicable to us.
> I too am reluctant to introduce yet another port type, but Justin is correct
> in pointing out that the PORT_ALTR_16550_* port types include Tx FIFO
> threshold programming that is incompatible with the BCM7271 UART hardware.
> This port type does appear necessary to address fundamental differences in
> the hardware unless we are willing to scrap the uart_config[] array and have
> the individual drivers manage these differences (which I would also be OK
> with, but I am just a tail on this dog).
> 
> The BCM7271 UART IP does support programmable Tx FIFO thresholds in a
> different way, so if I (or someone else) decided to enable support for that
> it would appear that this new port type would be necessary at that time as
> well.

All these details are missing in the initial submission. How should we know all
that? Please, amend the commit message accordingly.

> > > But why 8 and not 16 is the default rxtrig?
> > 
> > We were seeing some latency issues on our chips where 16 would cause
> > overflows. Trying to kill 2 birds with one stone. If creating another
> > port type is avoidable then alternatively I can change the default in
> > userspace.

Also choose the number less than 124, IIRC we have gaps that may be filled.

-- 
With Best Regards,
Andy Shevchenko