Re: [PATCH v4 6/7] serial: exar: add CTI specific setup code

From: Parker Newman
Date: Thu Apr 18 2024 - 13:04:45 EST


On Thu, 18 Apr 2024 19:29:44 +0300 (EEST)
Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx> wrote:

> On Thu, 18 Apr 2024, Parker Newman wrote:
> > On Thu, 18 Apr 2024 16:20:15 +0300 (EEST)
> > Ilpo Järvinen <ilpo.jarvinen@xxxxxxxxxxxxxxx> wrote:
> >
> > > On Wed, 17 Apr 2024, Parker Newman wrote:
> > > > From: Parker Newman <pnewman@xxxxxxxxxxxxxxx>
> > > >
> > > > This is a large patch but is only additions. All changes and removals
> > > > are made in previous patches in this series.
> > > >
> > > > - Add CTI board_init and port setup functions for each UART type
> > > > - Add CTI_EXAR_DEVICE() and CTI_PCI_DEVICE() macros
> > > > - Add support for reading a word from the Exar EEPROM.
> > > > - Add support for configuring and setting a single MPIO
> > > > - Add various helper functions for CTI boards.
> > > > - Add osc_freq to struct exar8250
> > > >
> > > > Signed-off-by: Parker Newman <pnewman@xxxxxxxxxxxxxxx>
>
> > > > @@ -192,11 +252,201 @@ struct exar8250_board {
> > > >
> > > > struct exar8250 {
> > > > unsigned int nr;
> > > > + unsigned int osc_freq;
> > > > struct exar8250_board *board;
> > > > void __iomem *virt;
> > > > int line[];
> > > > };
> > > >
> > > > +static inline void exar_write_reg(struct exar8250 *priv,
> > > > + unsigned int reg, u8 value)
> > > > +{
> > > > + writeb(value, priv->virt + reg);
> > > > +}
> > > > +
> > > > +static inline u8 exar_read_reg(struct exar8250 *priv, unsigned int reg)
> > > > +{
> > > > + return readb(priv->virt + reg);
> > > > +}
> > >
> > > I tried to understand what is going on with this priv->virt in 8250_exar
> > > in general and why it exists but I failed. It seems to BAR0 is mapped
> > > there but also serial8250_pci_setup_port() does map the same BAR and
> > > sets it up into the usual place in membase.
> > >
> >
> > Exar PCI/PCIe UARTs have global configuration registers from 0x80-0x9B.
> > These registers are for reading the EEPROM, configuring the MPIO, etc.
> > As these registers are only at 0x80, and not port specific, the driver maps
> > BAR0 to priv->virt for accessing them.
>
> Okay, thanks for explaining. The naming & lack of comments wasn't exactly
> making it easy to follow this bit (this is not your fault in anyway but
> a pre-existing problem in the driver's code).
>
> I've a follow up question now that it's confirmed they're different,
> see below...
>
> > > > + exar_write_reg(priv, (offset + UART_EXAR_8XMODE), 0x00);
> > > > + exar_write_reg(priv, (offset + UART_EXAR_FCTR), UART_FCTR_EXAR_TRGD);
> > > > + exar_write_reg(priv, (offset + UART_EXAR_TXTRG), 128);
> > > > + exar_write_reg(priv, (offset + UART_EXAR_RXTRG), 128);
> > >
> > > Unnecessary parenthesis.
>
> > > > + exar_write_reg(priv, (offset + UART_EXAR_8XMODE), 0x00);
> > > > + exar_write_reg(priv, (offset + UART_EXAR_FCTR), UART_FCTR_EXAR_TRGD);
> > > > + exar_write_reg(priv, (offset + UART_EXAR_TXTRG), 32);
> > > > + exar_write_reg(priv, (offset + UART_EXAR_RXTRG), 32);
> > >
> > > Unnecessary parenthesis.
> > >
> >
> > I will fix these in my cleanup patches.
>
> Based on the wording in your response, I'm not sure you got this right. It
> is code you're adding in this patch so the parenthesis should be removed
> from this change so they never appear in the commit history.
>

Greg has already merged this series into his testing branch. I assumed any
changes would need to be made in a separate patch series? Sorry if I
misunderstood. I already sent these fixes in a new mini-series.

> > > I recommend you add a helper for this as it is repeated twice. Are the
> > > values 32 and 128 literal or do they have some specific meaning? If the
> > > latter case, they should be using named defines (this likely applies to
> > > the existing trigger code in the driver too).
> > >
> > >
> >
> > They are the FIFO trigger levels so they are literally 128 and 32.
>
> Okay, no problem then if its 128 characters and 32 characters.
>
> > These 4 writes come from Exar's out-of-tree driver and are in
> > pci_xr17v35x_setup() and some other vendor specific functions.
> >
> > I am not sure why/if these are needed.
>
> ...So the follow-up question. I see the existing code in
> pci_fastcom335_setup() and pci_xr17v35x_setup() writes into membase
> based address but your code uses exar_write_reg() which is priv->virt
> based. Is this difference intentional?
>

Both methods are effectively the same thing. I used exar_write_reg() to be
consistent with my other code and it is a bit cleaner than:

u8 __iomem *p;

p = port->port.membase;

writeb(0x00, p + UART_EXAR_8XMODE);