Issues attempting to use Raspberry Pi 4 serial console on mainline

[Peter Robinson]

On Wed, Jul 22, 2020 at 2:27 PM Nicolas Saenz Julienne
<nsaenzjulienne at suse.de> wrote:
>
> Hi Nathan,
>
> On Tue, 2020-07-21 at 15:57 -0700, Nathan Chancellor wrote:
> > Hi all,
> >
> > Thank you for all of the hard work that has been done on supporting
> > the
> > Raspberry Pi 4 upstream. It has been a great platform so far for
> > testing
> > various clang technologies on actual hardware.
>
> I'm glad it's useful :)
>
> > I am investigating a panic that occurs when running a guest under KVM
> > when clang's Shadow Call Stack is enabled and it would be handy to
> > grab
> > the kernel panic via serial console as the device panics and I lose
> > my
> > ssh connection. I picked up a USB to TTL cable and I am able to get
> > connected with the following config.txt options and cmdline.txt
> > options:
> >
> > $ head -n4 /boot/config.txt
> > enable_uart=1
> > kernel=Image
> > os_prefix=custom-mainline-arm64/
> >
> > upstream_kernel=1
>
> I'd say for rpi4 this property isn't necessary as you're creating a custom
> os_prefix anyway. But it should be harmless.
>
> That aside, the only thing I'm missing is:
>
> arm_64bit=1
>
> > $ cat /boot/custom-mainline-arm64/cmdline.txt
> > console=ttyS1,115200 console=tty1 root=PARTUUID=45a8dd8a-02
>
> Which device tree are you using? This is the right tty device if using the
> upstream one, if using the one provided by the RPi foundation it should be
> ttyS0.
>
> > rootfstype=ext4 elevator=deadline fsck.repair=yes rootwait
> > plymouth.ignore-serial-consoles
>
> For reference I just booted linux-next with this setup:
>
> boot partition:
>         ...Latest firmware files taken from the RPi firmware repo [1]...
>         Image                   #Copied from linux build
>         bcm2711-rpi-4-b.dtb     #Copied from linux build
>         config.txt
>         cmdline.txt
>
> config.txt:
>         kernel=Image
>         enable_uart=1
>         arm_64bit=1
>
> cmdline.txt:
>          console=tty console=ttyS1,115200 text root=/dev/nfs
>          nfsroot=10.42.0.1:/home/nico/netboot/root,vers=3 rw ip=dhcp
> rootwait
>          elevator=deadline
>
> > However, when I connect to the serial console via PuTTY, I get
> > nothing
> > but garbage output: https://imgur.com/a/ekFlLYq
> >
> > As I understand it, that is due to the mini UART not being as good as
> > a
> > regular PL011. On the downstream kernel, one would apply
> > 'dtoverlay=disable-bt' which would free up the first PL011 to be used
> > as
> > the primary UART but the device tree overlays do not work on a
> > mainline
> > kernel. Is there an easy way to disable Bluetooth via the device
> > tree?
> > If not, is there any recommended or documented way to use the mini
> > UART
> > successfully? I have seen information around using 'core_freq=...'
> > but
> > the documentation says that does not work for the Raspberry Pi 4.
>
> The issue with the mini UART is its clock, which is derived from VPU's, which
> is itself controlled by RPi's firmware. Changes might happen behind the
> kernel's back, and the mini UART divisors will not be updated accordingly.
> This is an area the we could do better, but no one found a good solution yet.
> That said, for now, when using the upstream kernel, VPU's clock should be
> stable as we forbid the firmware from performing frequency scaling on that
> clock.

There has actually been a regression in the firmware here, prior to
mid April if the enable_uart=1 I always had clean output on the serial
console, since that date on the rpi3/3b+/4 I get a whole bunch of
junk, revert to an older firmware (I've been using April 1st ) and it
all goes back to being fine. I've not had time to actually report it
yet, and it's still a problem with the latest firmware but a quick
look it likes similar to this issue reported in late April:

https://github.com/raspberrypi/firmware/issues/1376

Peter

> In the future, once Maxime Rippards vc4/HDMI code is merged we will most
> probably hit this issue as the core clock has to be upscaled when feeding big
> screen resolutions. As you mention, a solution to this is fixing the core
> frequency in config.txt, for example I tested this successfully:
>
> core_freq=500
> core_freq_min=500
>
> Regards,
> Nicolas
>
> [1] https://github.com/raspberrypi/firmware
>
>
> _______________________________________________
> linux-rpi-kernel mailing list
> linux-rpi-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-rpi-kernel