On Mon, Feb 15, 2021 at 12:35:31PM +0200, Mikko Perttunen wrote:
> On 2/15/21 12:25 PM, Thierry Reding wrote:
> > On Sat, Feb 13, 2021 at 01:58:24PM +0200, Mikko Perttunen wrote:
> > > Add an earlycon driver for platforms with TCU, namely Tegra194.
> > > The driver is compatible with boot parameters passed by NVIDIA
> > > boot chains.
> >
> > I'm not sure I understand the latter part of this description. What boot
> > parameters is this compatible with? Looking at the setup function there
> > doesn't seem to be anything out of the ordinary here, so I'm wondering
> > if that's just confusing. If there's anything special, it might be worth
> > specifically pointing out what that is. Perhaps both in the commit
> > message and in a code comment, so it's properly documented.
>
> It's that the name of the driver 'tegra_comb_uart' matches what the boot
> chain passes; and that OF_EARLYCON_DECLARE is not used. (OF_EARLYCON_DECLARE
> cannot anyway be used due to the mailbox indirection in device tree).
This is all not immediately obvious. Perhaps you can add more of this
into the commit message and perhaps provide an example of how this would
be used on the kernel command-line.
You say "mailbox indirection" and looking at the implementation this
does seem to use the mailbox's base address as a sort of TX FIFO, which
I think is all good. However, I'm wondering if we couldn't somehow
detect this all dynamically at runtime. Don't we have access to the
device tree node at this point? If so, couldn't we parse all the
necessary information from the DT instead of relying on the user
providing the mailbox address on the command-line?
I realize that this would all make things a bit more complicated in this
driver, but at the same time it'd make life so much easier for users, so
I think it's worth at least considering.
To elaborate on this a bit, I think it'd be much more useful if users
could specify something like this:
earlycon=tegra-tcu
rather than:
earlycon=tegra_comb_uart,0xc150000
Note that I'm not even sure if that's a correct address. It'd be even
better if all of this can just be derived from the device tree. My
recollection is that earlycon always needs to be explicitly enabled, but
I thought it was also possible to derive which console to use from the
/chose/stdout-path property in device tree.
> > > Signed-off-by: Mikko Perttunen <mperttunen@xxxxxxxxxx>
> > > ---
> > > drivers/tty/serial/Kconfig | 12 +++++
> > > drivers/tty/serial/Makefile | 1 +
> > > drivers/tty/serial/tegra-tcu-earlycon.c | 72 +++++++++++++++++++++++++
> > > 3 files changed, 85 insertions(+)
> > > create mode 100644 drivers/tty/serial/tegra-tcu-earlycon.c
> > >
> > > diff --git a/drivers/tty/serial/Kconfig b/drivers/tty/serial/Kconfig
> > > index 34a2899e69c0..d941785e3f46 100644
> > > --- a/drivers/tty/serial/Kconfig
> > > +++ b/drivers/tty/serial/Kconfig
> > > @@ -331,6 +331,18 @@ config SERIAL_TEGRA_TCU_CONSOLE
> > > If unsure, say Y.
> > > +config SERIAL_TEGRA_TCU_EARLYCON
> > > + bool "Earlycon on NVIDIA Tegra Combined UART"
> > > + depends on ARCH_TEGRA || COMPILE_TEST
> > > + select SERIAL_EARLYCON
> > > + select SERIAL_CORE_CONSOLE
> > > + default y if SERIAL_TEGRA_TCU_CONSOLE
> > > + help
> > > + If you say Y here, TCU output will be supported during the earlycon
> > > + phase of the boot.
> > > +
> > > + If unsure, say Y.
> > > +
> > > config SERIAL_MAX3100
> > > tristate "MAX3100 support"
> > > depends on SPI
> > > diff --git a/drivers/tty/serial/Makefile b/drivers/tty/serial/Makefile
> > > index b85d53f9e9ff..408144326fed 100644
> > > --- a/drivers/tty/serial/Makefile
> > > +++ b/drivers/tty/serial/Makefile
> > > @@ -72,6 +72,7 @@ obj-$(CONFIG_SERIAL_XILINX_PS_UART) += xilinx_uartps.o
> > > obj-$(CONFIG_SERIAL_SIRFSOC) += sirfsoc_uart.o
> > > obj-$(CONFIG_SERIAL_TEGRA) += serial-tegra.o
> > > obj-$(CONFIG_SERIAL_TEGRA_TCU) += tegra-tcu.o
> > > +obj-$(CONFIG_SERIAL_TEGRA_TCU_EARLYCON) += tegra-tcu-earlycon.o
> > > obj-$(CONFIG_SERIAL_AR933X) += ar933x_uart.o
> > > obj-$(CONFIG_SERIAL_EFM32_UART) += efm32-uart.o
> > > obj-$(CONFIG_SERIAL_ARC) += arc_uart.o
> > > diff --git a/drivers/tty/serial/tegra-tcu-earlycon.c b/drivers/tty/serial/tegra-tcu-earlycon.c
> > > new file mode 100644
> > > index 000000000000..9decfbced0a7
> > > --- /dev/null
> > > +++ b/drivers/tty/serial/tegra-tcu-earlycon.c
> > > @@ -0,0 +1,72 @@
> > > +// SPDX-License-Identifier: GPL-2.0
> > > +/*
> > > + * Copyright (c) 2017-2021, NVIDIA CORPORATION. All rights reserved.
> > > + */
> > > +
> > > +#include <linux/console.h>
> > > +#include <linux/io.h>
> > > +#include <linux/serial_core.h>
> > > +
> > > +#define NUM_BYTES_FIELD_BIT 24
> > > +#define FLUSH_BIT 26
> >
> > This one seems to be unused.
>
> True, I'll remove it.
>
> >
> > > +#define INTR_TRIGGER_BIT 31
> >
> > I wonder if this could somehow be integrated with the existing TCU
> > driver since we have these bits defined there already. And really this
> > is basically a skeleton version of the same driver.
> >
> > > +/*
> > > + * This function splits the string to be printed (const char *s) into multiple
> > > + * packets. Each packet contains a max of 3 characters. Packets are sent to the
> > > + * SPE-based combined UART server for printing. Communication with SPE is done
> > > + * through mailbox registers which can generate interrupts for SPE.
> > > + */
> > > +static void early_tcu_write(struct console *console, const char *s, unsigned int count)
> > > +{
> > > + struct earlycon_device *device = console->data;
> > > + u8 __iomem *addr = device->port.membase;
> > > + u32 mbox_val = BIT(INTR_TRIGGER_BIT);
> > > + unsigned int i;
> > > +
> > > + /* Loop for processing each 3 char packet */
> > > + for (i = 0; i < count; i++) {
> > > + if (s[i] == '\n')
> > > + mbox_val = update_and_send_mbox(addr, mbox_val, '\r');
> > > + mbox_val = update_and_send_mbox(addr, mbox_val, s[i]);
> > > + }
> > > +
> > > + if ((mbox_val >> NUM_BYTES_FIELD_BIT) & 0x3) {
> > > + while (readl(addr) & BIT(INTR_TRIGGER_BIT))
> > > + cpu_relax();
> > > + writel(mbox_val, addr);
> > > + }
> > > +}
> >
> > For example this function already exists in the Tegra TCU driver and
> > perhaps some of that could be refactored to work for both cases.
>
> This is very similar to the main tegra_tcu driver, but considering how
> simple this driver is, and the main driver using the mailbox framework
> making the actual implementation incompatible, I was thinking that it's
> easier to just have this be independent.
I don't have a strong objection to keeping these functions separate,
especially since they are fairly small and not likely to ever change, so
the maintenance burden is going to be small in any case.
But even so it might be nice to stash this all into the same file. After
all, people aren't going to enable this configuration option if they
have the Tegra TCU driver disabled. Once these are integrated, it's also
likely not worth even having a separate Kconfig option because the added
code is so little.
Thierry
Attachment:
signature.asc
Description: PGP signature
