RE: why doesn't arm virt machine boot linux when uart address size is increased?

[ckim]

Hi Peter Maydell,
Thanks again. It's still not clear why increasing the UART address size to 
0x2000 makes it stall, 
But at least with the 'earlycon' command, I think I can later figure it out. 
(this will be helpful for analysis)
Thank you!
Chan Kim

> -----Original Message-----
> From: Peter Maydell <peter.maydell@linaro.org>
> Sent: Monday, April 12, 2021 6:39 PM
> To: Chan Kim <ckim@etri.re.kr>
> Cc: qemu-discuss <qemu-discuss@nongnu.org>
> Subject: Re: why doesn't arm virt machine boot linux when uart address
> size is increased?
> 
> On Mon, 12 Apr 2021 at 10:02, <ckim@etri.re.kr> wrote:
> > Earlier, I have copied the arm/virt machine and named it ab21q.
> >
> > I changed the memory map of ab21q to match our system (under
> development).
> >
> > Among others the original UART address/size was changed from
> > {0x09000000, 0x1000} to {0x10210000, 0xf000}. (the numbers were given
> > to me)
> >
> > I checked a couple of baremetal programs run ok on the new machine.
> >
> > But when I tried running linux on ab21q this morning, the linux kernel
> didn’t to the shell prompt.
> 
> I have no idea what other changes you've made on this frankenstein board,
> but at least for the upstream 'virt' board this address is already in use
> (it's in the middle of the PCI controller).
> 
> > I am curious about what mechanism is preventing it from booting when the
> UART address space size is bigger than necessary and also what the kernel
> is doing during the 3~4 seconds after the end of start_kernel function and
> printing “Booting from ..” message.
> 
> If the kernel crashes or panics in early bootup then it doesn't always get
> the chance to print the logging it does. In a normally configured kernel,
> log messages are stored to a buffer, and then when the console device is
> initialised (which happens fairly late on) the contents of the buffer are
> all printed out at once. This is probably what your delay is: the kernel
> is booting, but hasn't yet got to a point of being able to output to the
> console.
> 
> You can tell the kernel to make special efforts to print early log
> messages, which will allow you to read what it's doing before it crashes.
> To do this you add "earlycon" to the kernel command line (the kernel needs
> SERIAL_AMBA_PL011_CONSOLE enabled for a PL011 UART which the virt board
> has; other UART types have different config option names).
> 
> Of course, if the reason the kernel is failing to boot is because the UART
> itself is not where the kernel expects then you won't get any console
> output and you'll have to debug the kernel boot by hand the same way you
> would for hardware kernel bring-up.
> 
> thanks
> -- PMM