Re: [PATCH RFC 2/2] serial: 8250: Add PRUSS UART driver

[Judith Mendez]

Hi Andy

On 5/2/25 4:50 AM, Andy Shevchenko wrote:
> On Wed, Apr 30, 2025 at 07:31:13PM -0500, Judith Mendez wrote:
> > From: Bin Liu <b-liu@xxxxxx>
> >
> > This adds a new serial 8250 driver that supports the UART in PRUSS
> > module.
> >
> > The PRUSS has a UART sub-module which is based on the industry standard
> > TL16C550 UART controller, which has 16-bytes FIFO and supports 16x and
> > 13x over samplings.
>
> > Signed-off-by: Bin Liu <b-liu@xxxxxx>
> > Signed-off-by: Judith Mendez <jm@xxxxxx>
>
> Are you just a committer and not a co-developer of this code?

Commiter only. I am only carrying the driver across kernel version,
fixing merge conflicts, testing the driver, and now up-streaming it.

> ...
>
> > +/*
> > + *  Serial Port driver for PRUSS UART on TI platforms
> > + *
> > + *  Copyright (C) 2020-2021 by Texas Instruments Incorporated - http://www.ti.com/
>
> My calendar shows 2025...
>
> > + *  Author: Bin Liu <b-liu@xxxxxx>
> > + */
>
> ...
>
> The list of the inclusions is semi-random. Please, follow the IWYU principle.
>
> > +#include <linux/clk.h>
> > +#include <linux/module.h>
> > +#include <linux/serial_reg.h>
> > +#include <linux/serial_core.h>
>
> > +#include <linux/of_irq.h>
> > +#include <linux/of_address.h>
> > +#include <linux/of_platform.h>
>
> Please, no of*.h in a new code.

Will only keep linux/of_platform.h for of_device_id struct.

> > +#include <linux/remoteproc.h>
>
> Keep them ordered as well.
>
> + blank line here.
>
> > +#include "8250.h"
>
> ...
>
> > +#define DEFAULT_CLK_SPEED	192000000
>
> Units as a suffix? HZ_PER_MHZ?
> You can also fix 8250_omap in the same way.
>
> ...
>
> > +static inline void uart_writel(struct uart_port *p, u32 offset, int value)
> > +{
> > +	writel(value, p->membase + (offset << p->regshift));
> > +}
>
> Why? Or how does it differ from the ones that serial core provides?
>
> ...
>
> > +static int pruss8250_startup(struct uart_port *port)
> > +{
> > +	int ret;
> > +
> > +	uart_writel(port, PRUSS_UART_PEREMU_MGMT, 0);
> > +
> > +	ret = serial8250_do_startup(port);
> > +	if (!ret)
>
> Please, use usual pattern, check for errors first
>
> 	if (ret)
> 		return ret;
> 	...
> 	return 0;
>
> > +		uart_writel(port, PRUSS_UART_PEREMU_MGMT, PRUSS_UART_TX_EN |
> > +							  PRUSS_UART_RX_EN |
> > +							  PRUSS_UART_FREE_RUN);
> > +	return ret;
> > +}
>
> ...
>
> > +static unsigned int pruss8250_get_divisor(struct uart_port *port,
> > +					  unsigned int baud,
> > +					  unsigned int *frac)
> > +{
> > +	unsigned int uartclk = port->uartclk;
> > +	unsigned int div_13, div_16;
> > +	unsigned int abs_d13, abs_d16;
> > +	u16 quot;
>
> > +	/* Old custom speed handling */
> > +	if (baud == 38400 && (port->flags & UPF_SPD_MASK) == UPF_SPD_CUST) {
> > +		quot = port->custom_divisor & UART_DIV_MAX;
> > +		if (port->custom_divisor & (1 << 16))
> > +			*frac = PRUSS_UART_MDR_13X_MODE;
> > +		else
> > +			*frac = PRUSS_UART_MDR_16X_MODE;
> > +
> > +		return quot;
> > +	}
>
> Why?! Please, try to avoid adding more drivers with this ugly hack.

My understanding is that this is not a hack, for 38400 we need to pass
as custom baud. What is the alternative here?

I see other drivers are doing this as well, will look into this further
but not sure if there is a better solution for this.

> > +	div_13 = DIV_ROUND_CLOSEST(uartclk, 13 * baud);
> > +	div_16 = DIV_ROUND_CLOSEST(uartclk, 16 * baud);
> > +	div_13 = div_13 ? : 1;
> > +	div_16 = div_16 ? : 1;
> > +
> > +	abs_d13 = abs(baud - uartclk / 13 / div_13);
> > +	abs_d16 = abs(baud - uartclk / 16 / div_16);
> > +
> > +	if (abs_d13 >= abs_d16) {
> > +		*frac = PRUSS_UART_MDR_16X_MODE;
> > +		quot = div_16;
> > +	} else {
> > +		*frac = PRUSS_UART_MDR_13X_MODE;
> > +		quot = div_13;
> > +	}
> > +
> > +	return quot;
>
> Are you sure it doesn't repeat existing things from other driver(s)?

core driver does this differently so we cant use that.

> > +}
>
> ...
>
> > +static int pruss8250_probe(struct platform_device *pdev)
> > +{
> > +	struct device_node *np = pdev->dev.of_node;
>
> You don't need this.
>
> > +	struct uart_8250_port port8250;
> > +	struct uart_port *up = &port8250.port;
> > +	struct pruss8250_info *info;
> > +	struct resource resource;
> > +	unsigned int port_type;
> > +	struct clk *clk;
> > +	int ret;
> > +
> > +	port_type = (unsigned long)of_device_get_match_data(&pdev->dev);
> > +	if (port_type == PORT_UNKNOWN)
> > +		return -EINVAL;
> > +
> > +	info = devm_kzalloc(&pdev->dev, sizeof(*info), GFP_KERNEL);
> > +	if (!info)
> > +		return -ENOMEM;
> > +
> > +	memset(&port8250, 0, sizeof(port8250));
> > +
> > +	ret = of_address_to_resource(np, 0, &resource);
>
> Yeah, no modifications from 2021?
>
> > +	if (ret) {
> > +		dev_err(&pdev->dev, "invalid address\n");
> > +		return ret;
> > +	}
>
> > +	ret = of_alias_get_id(np, "serial");
> > +	if (ret > 0)
> > +		up->line = ret;
>
> Use uart_read_port_properties() instead of this and other related code.
>
> > +	clk = devm_clk_get(&pdev->dev, NULL);
> > +	if (IS_ERR(clk)) {
> > +		if (PTR_ERR(clk) == -EPROBE_DEFER)
> > +			return -EPROBE_DEFER;
>
> We have other errors which are effectively being ignored here, why?
>
> > +		up->uartclk = DEFAULT_CLK_SPEED;
> > +	} else {
> > +		up->uartclk = clk_get_rate(clk);
>
> > +		devm_clk_put(&pdev->dev, clk);
>
> Why? Maybe you should not to try devm_ to begin with?
>
> > +	}
> > +
> > +	up->dev = &pdev->dev;
> > +	up->mapbase = resource.start;
> > +	up->mapsize = resource_size(&resource);
> > +	up->type = port_type;
> > +	up->iotype = UPIO_MEM;
> > +	up->regshift = 2;
> > +	up->flags = UPF_SHARE_IRQ | UPF_BOOT_AUTOCONF | UPF_FIXED_PORT |
> > +		    UPF_FIXED_TYPE | UPF_IOREMAP;
> > +	up->irqflags |= IRQF_SHARED;
> > +	up->startup = pruss8250_startup;
> > +	up->rs485_config = serial8250_em485_config;
> > +	up->get_divisor = pruss8250_get_divisor;
> > +	up->set_divisor = pruss8250_set_divisor;
> > +
> > +	ret = of_irq_get(np, 0);
> > +	if (ret < 0) {
> > +		if (ret != -EPROBE_DEFER)
> > +			dev_err(&pdev->dev, "missing irq\n");
> > +		return ret;
> > +	}
>
> A lot of this will be simplified by using the above mentioned API.
>
> > +	up->irq = ret;
> > +	spin_lock_init(&port8250.port.lock);
> > +	port8250.capabilities = UART_CAP_FIFO | UART_CAP_AFE;
> > +
> > +	ret = serial8250_register_8250_port(&port8250);
> > +	if (ret < 0)
> > +		goto err_dispose;
> > +
> > +	info->type = port_type;
> > +	info->line = ret;
> > +	platform_set_drvdata(pdev, info);
> > +
> > +	return 0;
>
> > +err_dispose:
> > +	irq_dispose_mapping(port8250.port.irq);
>
> Why this is needed?

No longer needed will remove

> > +	return ret;
> > +}
>
> ...
>
> > +static const struct of_device_id pruss8250_table[] = {
> > +	{ .compatible = "ti,pruss-uart", .data = (void *)PORT_16550A, },
>
> Inner comma is redundant.
>
> > +	{ /* end of list */ },
>
> No trailing comma in the terminator.
>
> > +};
>
> ...
>
> > +config SERIAL_8250_PRUSS
> > +	tristate "TI PRU-ICSS UART support"
> > +	depends on SERIAL_8250
>
> > +	depends on PRU_REMOTEPROC && TI_PRUSS_INTC
>
> No COMPILE_TEST?
>
> > +	help
> > +	  This driver is to support the UART module in PRU-ICSS which is
> > +	  available in some TI platforms.
> > +	  Say 'Y' here if you wish to use PRU-ICSS UART.
> > +	  Otherwise, say 'N'.
>
> If m, how would the module be called? See many new driver examples in Kconfigs
> (in particular IIO follows this pattern).

All other comments I will fix for v1, thanks for reviewing.

~ Judith