Re: [PATCH 2/5] serial: s5p: Add Apple M1 support

[Simon Glass <sjg@chromium.org>]

Hi Mark,

On Sat, 2 Oct 2021 at 16:16, Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
>
> > From: Simon Glass <sjg@chromium.org>
> > Date: Sun, 19 Sep 2021 21:15:59 -0600
> >
> > Hi Mark,
> >
> > On Sat, 18 Sept 2021 at 07:55, Mark Kettenis <kettenis@openbsd.org> wrote:
> > >
> > > Apple M1 SoCs include an S5L UART which is a variant of the S5P
> > > UART.  Add support for this variant and enable it by default
> > > on Apple SoCs.
> > >
> > > Signed-off-by: Mark Kettenis <kettenis@openbsd.org>
> > > ---
> > >  arch/arm/include/asm/arch-m1/clk.h  | 11 ++++++++
> > >  arch/arm/include/asm/arch-m1/uart.h | 41 +++++++++++++++++++++++++++++
> > >  arch/arm/mach-apple/board.c         |  5 ++++
> > >  drivers/serial/Kconfig              |  2 +-
> > >  drivers/serial/serial_s5p.c         | 22 ++++++++++++++++
> > >  5 files changed, 80 insertions(+), 1 deletion(-)
> > >  create mode 100644 arch/arm/include/asm/arch-m1/clk.h
> > >  create mode 100644 arch/arm/include/asm/arch-m1/uart.h
> > >
> > > diff --git a/arch/arm/include/asm/arch-m1/clk.h b/arch/arm/include/asm/arch-m1/clk.h
> > > new file mode 100644
> > > index 0000000000..f4326d0c0f
> > > --- /dev/null
> > > +++ b/arch/arm/include/asm/arch-m1/clk.h
> > > @@ -0,0 +1,11 @@
> > > +// SPDX-License-Identifier: GPL-2.0+
> > > +/*
> > > + * Copyright (C) 2021 Mark Kettenis <kettenis@openbsd.org>
> > > + */
> > > +
> > > +#ifndef __ASM_ARM_ARCH_CLK_H_
> > > +#define __ASM_ARM_ARCH_CLK_H_
> > > +
> > > +unsigned long get_uart_clk(int dev_index);
> >
> > comment? But this should be in a clock driver, unless you are trying
> > to get a debug UART up ASAP with minimal code?
> >
> > > +
> > > +#endif
> > > diff --git a/arch/arm/include/asm/arch-m1/uart.h b/arch/arm/include/asm/arch-m1/uart.h
> > > new file mode 100644
> > > index 0000000000..d2a17a221e
> > > --- /dev/null
> > > +++ b/arch/arm/include/asm/arch-m1/uart.h
> > > @@ -0,0 +1,41 @@
> > > +/* SPDX-License-Identifier: GPL-2.0+ */
> > > +/*
> > > + * (C) Copyright 2009 Samsung Electronics
> > > + * Minkyu Kang <mk7.kang@samsung.com>
> > > + * Heungjun Kim <riverful.kim@samsung.com>
> > > + */
> > > +
> > > +#ifndef __ASM_ARCH_UART_H_
> > > +#define __ASM_ARCH_UART_H_
> > > +
> > > +#ifndef __ASSEMBLY__
> > > +/* baudrate rest value */
> > > +union br_rest {
> > > +       unsigned short  slot;           /* udivslot */
> > > +       unsigned char   value;          /* ufracval */
> > > +};
> > > +
> > > +struct s5p_uart {
> > > +       unsigned int    ulcon;
> > > +       unsigned int    ucon;
> > > +       unsigned int    ufcon;
> > > +       unsigned int    umcon;
> > > +       unsigned int    utrstat;
> > > +       unsigned int    uerstat;
> > > +       unsigned int    ufstat;
> > > +       unsigned int    umstat;
> > > +       unsigned int    utxh;
> > > +       unsigned int    urxh;
> > > +       unsigned int    ubrdiv;
> > > +       union br_rest   rest;
> > > +       unsigned char   res3[0x3fd0];
> > > +};
> > > +
> > > +static inline int s5p_uart_divslot(void)
> > > +{
> > > +       return 0;
> > > +}
> > > +
> > > +#endif /* __ASSEMBLY__ */
> > > +
> > > +#endif
> > > diff --git a/arch/arm/mach-apple/board.c b/arch/arm/mach-apple/board.c
> > > index 0c8b35292e..8bc5c2f69e 100644
> > > --- a/arch/arm/mach-apple/board.c
> > > +++ b/arch/arm/mach-apple/board.c
> > > @@ -156,3 +156,8 @@ ulong board_get_usable_ram_top(ulong total_size)
> > >          */
> > >         return 0x980000000;
> > >  }
> > > +
> > > +unsigned long get_uart_clk(int dev_index)
> > > +{
> > > +       return 24000000;
> >
> > Should add to devicetree for the driver
> >
> > > +}
> > > diff --git a/drivers/serial/Kconfig b/drivers/serial/Kconfig
> > > index 93348c0929..c3ac773929 100644
> > > --- a/drivers/serial/Kconfig
> > > +++ b/drivers/serial/Kconfig
> > > @@ -719,7 +719,7 @@ config ROCKCHIP_SERIAL
> > >
> > >  config S5P_SERIAL
> > >         bool "Support for Samsung S5P UART"
> > > -       depends on ARCH_EXYNOS || ARCH_S5PC1XX
> > > +       depends on ARCH_APPLE || ARCH_EXYNOS || ARCH_S5PC1XX
> > >         default y
> > >         help
> > >           Select this to enable Samsung S5P UART support.
> > > diff --git a/drivers/serial/serial_s5p.c b/drivers/serial/serial_s5p.c
> > > index 6d09952a5d..eb770d9b62 100644
> > > --- a/drivers/serial/serial_s5p.c
> > > +++ b/drivers/serial/serial_s5p.c
> > > @@ -21,12 +21,21 @@
> > >
> > >  DECLARE_GLOBAL_DATA_PTR;
> > >
> > > +#ifdef CONFIG_ARCH_APPLE
> >
> > This should use the compatible string to decide which variant it is,
> > then the checks happen at runtime. We should never have arch-specific
> > #ifdefs in drivers.
> >
> > > +#define RX_FIFO_COUNT_SHIFT    0
> > > +#define RX_FIFO_COUNT_MASK     (0xf << RX_FIFO_COUNT_SHIFT)
> > > +#define RX_FIFO_FULL           (1 << 8)
> > > +#define TX_FIFO_COUNT_SHIFT    4
> > > +#define TX_FIFO_COUNT_MASK     (0xf << TX_FIFO_COUNT_SHIFT)
> > > +#define TX_FIFO_FULL           (1 << 9)
> > > +#else
> > >  #define RX_FIFO_COUNT_SHIFT    0
> > >  #define RX_FIFO_COUNT_MASK     (0xff << RX_FIFO_COUNT_SHIFT)
> > >  #define RX_FIFO_FULL           (1 << 8)
> > >  #define TX_FIFO_COUNT_SHIFT    16
> > >  #define TX_FIFO_COUNT_MASK     (0xff << TX_FIFO_COUNT_SHIFT)
> > >  #define TX_FIFO_FULL           (1 << 24)
> > > +#endif
> > >
> > >  /* Information about a serial port */
> > >  struct s5p_serial_plat {
> > > @@ -83,7 +92,11 @@ static void __maybe_unused s5p_serial_baud(struct s5p_uart *uart, uint uclk,
> > >         if (s5p_uart_divslot())
> > >                 writew(udivslot[val % 16], &uart->rest.slot);
> > >         else
> > > +#ifdef CONFIG_ARCH_APPLE
> > > +               writel(val % 16, &uart->rest.value);
> > > +#else
> > >                 writeb(val % 16, &uart->rest.value);
> > > +#endif
> > >  }
> > >
> > >  #ifndef CONFIG_SPL_BUILD
> > > @@ -148,7 +161,11 @@ static int s5p_serial_getc(struct udevice *dev)
> > >                 return -EAGAIN;
> > >
> > >         serial_err_check(uart, 0);
> > > +#ifdef CONFIG_ARCH_APPLE
> > > +       return (int)(readl(&uart->urxh) & 0xff);
> > > +#else
> > >         return (int)(readb(&uart->urxh) & 0xff);
> > > +#endif
> > >  }
> > >
> > >  static int s5p_serial_putc(struct udevice *dev, const char ch)
> > > @@ -159,7 +176,11 @@ static int s5p_serial_putc(struct udevice *dev, const char ch)
> > >         if (readl(&uart->ufstat) & TX_FIFO_FULL)
> > >                 return -EAGAIN;
> > >
> > > +#ifdef CONFIG_ARCH_APPLE
> > > +       writel(ch, &uart->utxh);
> > > +#else
> > >         writeb(ch, &uart->utxh);
> > > +#endif
> > >         serial_err_check(uart, 1);
> > >
> > >         return 0;
> > > @@ -201,6 +222,7 @@ static const struct dm_serial_ops s5p_serial_ops = {
> > >
> > >  static const struct udevice_id s5p_serial_ids[] = {
> > >         { .compatible = "samsung,exynos4210-uart" },
> > > +       { .compatible = "apple,s5l-uart" },
> > >         { }
> > >  };
> > >
> > > --
> > > 2.33.0
> > >
> >
> > Can you add debug UART support? Or does this already work?
>
> I don't think it does work.  But yes, I want to add it since it will
> help a lot with debugging.  That makes fixing some of the issues
> raised by you and bing more problematic since the debug UART support
> code:
>
> - Needs to use 32-bit access on Apple M1, and using the "reg-io-width"
>   property there isn't possible.
>
> - Relies on the FIFO control bits which are in a different position on
>   the Apple M1.
>
> So I am starting to wonder whether it makes sense to simply copy
> serial_s5p.c to serial_s5l.c and modify that one to do the right thing
> on the apple SoC.
>
> Alternatively I could introduce Kconfig stuff to switch between S5L
> and S5P support.

Well, firstly you can use a compatible string as mentioned above which
solves the normal case. You just need a flag in plat which indicates
that it should use word access for some registers. The ns16550 shows
how to call dev_get_driver_data(). You will need to add members to
s5p_serial_plat to select the tx/tx FIFO values since these need tobe
variables now.

BTW s5p_serial_baud() should have a param to indicate which access
method to use.

For the debug UART I suggest just putting an
IS_ENABLED(CONFIG_ARCH_APPLE) in there. You could add a Kconfig for
the Apple UART (even though it uses the same file) which might be a
little nicer, but it is only the debug UART, after all...

Finally, this driver is not perfect as it has CLK_EXYNOS only for some
boards. Ideally all boards would move to use the clock driver.

Regards,
Simon