Raspberrypi3: Mini UART driver

Christian Mauderer list at c-mauderer.de
Mon Dec 23 21:50:00 UTC 2019


Hello Niteesh,

quite a lot of questions. I'll try to answer them. Note that it has been
some time since I had a detailed look at that code so if something I
tell seems odd please don't hesitate to question it.

Please note that in RTEMS their are more or less two "levels" of support
for a serial console:

1. A very basic polled system console (also known as "debug-console" in
some BSPs). This one is used for printk and should work in basically
every case. It is used for critical system messages like printing the
exception frame. For that a BSP has to provide a "BSP_output_char" function.

2. A full featured UART driver integrated into Termios. That one will be
used for all normal I/O on the UARTs.

As far as I know the "console_tbl Console_Configuration_Ports" belongs
to a table based legacy interface. It is handled in the file
bsps/shared/dev/serial/legacy-console.c. I'm not sure whether it is
documented in the BSP guide because it shouldn't be used for new BSPs.
Same is true for the "major" and "minor" stuff: It's not really used for
new drivers.

Newer drivers use the initialization that is described in the manual
that you have already found. Basically they use
"rtems_termios_device_install" to register a new UART as
"/dev/ttySomething". Some recent (ARM) BSPs that do that are the imx or
the atsam.

The console that is used for stdin, stdout and stderr (printf, scanf,
...) is the one called "/dev/console" (defined in CONSOLE_DEVICE_NAME).
For the legacy table based interface it's the one with the index of
"Console_Port_Minor".


If you want to access any UART other than the one for stdin and stdout
you do that the same way like on Linux: Just use the "open" function on
the "/dev/ttySomething" and use "read", "write" and simmilar or use
"fopen" together with "fread", "fwrite", "fprintf", ...


"printf" (and family) is a function belonging to the C library. In our
case that's newlib. It will format your message and after some other
preprocessing will call the "write" function of the file that is opened
as stdout (which is "/dev/console" in the default case).


I hope that I helped you with that explanation. Please feel free to ask
anything if it isn't clear.

Best regards

Christian

On 23/12/2019 19:50, Niteesh wrote:
> And finally, how does printf work? It is a macro? In that case, how does
> any write to
> a console work?
> 
> On Tue, Dec 24, 2019 at 12:18 AM Niteesh <gsnb.gn at gmail.com
> <mailto:gsnb.gn at gmail.com>> wrote:
> 
>     Is the correct port minor number set during the initialization? What
>     is the application want's to
>     access some other port?
> 
>     On Tue, Dec 24, 2019 at 12:16 AM Niteesh <gsnb.gn at gmail.com
>     <mailto:gsnb.gn at gmail.com>> wrote:
> 
>         I would like to clarify my doubts regarding the console driver.
>         I went through the documentation
>         for the console
>         driver https://docs.rtems.org/branches/master/bsp-howto/console.html#introduction.
>         But it is quite different from how some BSPs initialize.
>         Correct me if I am wrong
>         The console_tbl contains the various entries of serial ports.
>         The console_fns is a struct of function pointers, which point to
>         the BSP uart functions.
>         The BSP_output_char_function_type is what will be called for
>         printing a char on to the console.
>         How does RTEMS initialize the uart? It's seems not to be same
>         for all BSPs.
>         The doc says that the driver's initialization function is called
>         once during the rtems initialization process.
>         The console init function install the serial driver using
>         rtems_termios_device_install but there seems to be
>         no such function in the raspberry pi? But there is a entry in
>         console_fns for init function, but then how does it
>         gets called?
>         And for BSP's with multiple serial's, the output function
>         chooses the right serial using console_port_minor,
>         Is it during initialization?
>         What is the need for get and set register functions? 
> 
>         On Mon, Dec 23, 2019 at 1:04 AM Christian Mauderer
>         <list at c-mauderer.de <mailto:list at c-mauderer.de>> wrote:
> 
>             On 22/12/2019 19:45, Joel Sherrill wrote:
>             >
>             >
>             > On Sun, Dec 22, 2019, 12:29 PM Niteesh <gsnb.gn at gmail.com
>             <mailto:gsnb.gn at gmail.com>
>             > <mailto:gsnb.gn at gmail.com <mailto:gsnb.gn at gmail.com>>> wrote:
>             >
>             >     On Sun, Dec 22, 2019 at 8:44 PM Christian Mauderer
>             >     <list at c-mauderer.de <mailto:list at c-mauderer.de>
>             <mailto:list at c-mauderer.de <mailto:list at c-mauderer.de>>> wrote:
>             >
>             >         Hello Niteesh,
>             >
>             >         thanks for doing that work.
>             >
>             >         On 22/12/2019 12:10, Niteesh wrote:
>             >         > The rpi1 and rpi2 use the PL011 UART, whereas,
>             with RPI's
>             >         equipped with
>             >         > wireless/Bluetooth module, the PL011 is
>             connected to the Bluetooth
>             >         > module, and the mini UART is used as the primary
>             UART.
>             >
>             >         In my opinion it would be great if you could use
>             the FDT to
>             >         distinguish
>             >         between the boards. That should allow to add
>             raspberry 3 (and
>             >         maybe 4)
>             >         support without adding another BSP. More BSPs mean
>             a bigger
>             >         maintenance
>             >         effort for the RTEMS community.
>             >
>             >     Learning more about FDT is on my list for a long
>             time.  I would love
>             >     to work on that
>             >     but I have almost no exp with FDT's.
>             >     But another thing could also be done, in
>             >     raspberrypi/start/bspstart.c we get the revision and
>             >     model of the board using the mailbox. Every board has
>             a unique id,
>             >     which we could use to initialize
>             >     the BSP. But using FDT seems to be a more elegant
>             option, it is a
>             >     lot of work I think, but we could take
>             >     help from libbsd and linux I suppose. What do you think?
>             >
>             >
>             > I think there are almost always two steps to a project
>             like this: get it
>             > to work and make it nice. :)
>             >
>             > If you fix the startup code to read the board revision and
>             memory size,
>             > you can get a working BSP that dynamically adapts to the
>             models and
>             > memory variations with minimal modifications. If you want
>             to then
>             > convert the BSP to FDT, it will be a LOT easier to debug
>             with a working BSP.
>             >
>             > Plus you may be able to identify every variation point
>             based on just the
>             > model info. Then FDT is just a matter of switching the
>             source of
>             > some/all of the info.
>             >
>             > That would be my work plan anyway.
> 
>             I agree with Joel that a secure development basis (also
>             known as "hack")
>             as a first step is a good idea. You maybe even just make the
>             mini UART
>             the default driver while you are developing. Then you can be
>             sure that
>             you have the right driver.
> 
>             As soon as that works you can either change to the revision
>             method or
>             (better) to the FDT one and after that the patches can be
>             merged. Using
>             the FDT isn't that complicated. Basically you search for a
>             node based on
>             different parameters. For an example you can take a look at
>             the imx BSP.
>             In imx_uart_probe (bsps/arm/imx/console/console-config.c) a
>             fdt node is
>             searched and based on that a UART driver is used. But again:
>             Follow
>             Joels suggestion to start simple and secure.
> 
>             >
>             >         >
>             >         >
>             >       
>              https://www.raspberrypi.org/app/uploads/2012/02/BCM2835-ARM-Peripherals.pdf
>             >         > But from the above doc (PAGE 10), the mini uart
>             has 16550 like
>             >         registers
>             >         > and RTEMS already has the driver for it
>             >         > bsps/shared/dev/serial/ns16550.c. But I am not
>             sure how
>             >         compatible they
>             >         > are? Should a new driver be implemented from
>             scratch or use
>             >         ns16550 if
>             >         > possible?
>             >
>             >         In general it's better to re-use existing code.
>             That has multiple
>             >         advantages:
>             >
>             >         - It reduces the maintenance effort. Fewer code
>             means fewer work.
>             >         - If you have multiple driver for the same or
>             similar hardware
>             >         it can
>             >         happen that a bug is fixed in one but not the other.
>             >         - It's simpler to find a hardware to test changes.
>             >         - The driver becomes more universal with every new
>             supported
>             >         hardware.
>             >         That increases the chance that it fits the next
>             new hardware.
>             >
>             >         I'm sure there are some more if you ask someone else.
>             >
>             >     I do understand the issues, I just spent some time
>             reading the
>             >     driver code.
>             >     I think we could most probably use it. I will take a
>             closer look and
>             >     will update.
>             >
> 
>             Great.
> 
>             >          
>             >
>             >         >
>             >         > Also, the core clock on which the PL011 is based
>             on is changed
>             >         in rpi3.
>             >         > Rpi1 and 2 use 250Mhz as the default clock but
>             it was changed
>             >         to 400Mhz
>             >         > in Rpi3 and newer
>             >
>             >         Again: Would be great if that could be adapted
>             based on FDT or by
>             >         reading the right registers.
>             >
>             >         >
>             >         > Few differences between PL011 and Mini uart
>             >         > The mini UART has smaller FIFOs. Combined with
>             the lack of
>             >         flow control,
>             >         > this makes it more prone to losing characters at
>             higher baud
>             >         rates. It
>             >         > is also generally less capable than the PL011,
>             mainly due to
>             >         its baud
>             >         > rate link to the VPU clock speed.
>             >
>             >         That shouldn't really be a problem for the system
>             console.
>             >
>             >         >
>             >         > The particular deficiencies of the mini UART
>             compared to the
>             >         PL011 are :
>             >         >
>             >         > No break detection
>             >         > No framing errors detection
>             >         > No parity bit
>             >         > No receive timeout interrupt
>             >         > No DCD, DSR, DTR or RI signals
>             >         >
>             >
>             >     _______________________________________________
>             >     devel mailing list
>             >     devel at rtems.org <mailto:devel at rtems.org>
>             <mailto:devel at rtems.org <mailto:devel at rtems.org>>
>             >     http://lists.rtems.org/mailman/listinfo/devel
>             >
> 


More information about the devel mailing list