Re: Console vs serial

[dmkhn@xxxxxxxxx]

On Fri, Aug 01, 2025 at 12:44:57PM +0100, Frediano Ziglio wrote:
> On Fri, Aug 1, 2025 at 12:25 PM Andrew Cooper <andrew.cooper3@xxxxxxxxxx> 
> wrote:
> >
> > Hello,
> >
> > There's a long-standing irritation in Xen, whereby printk()'s which
> > occur prior to console_init_preirq() get half-lost.
> >
> > They end up in the console ring, and hence are available via `xl dmesg`,
> > but they're missing from serial console.
> >
> > e.g. these 4 lines don't escape
> >
> >     (XEN) Enabling lockdown mode because Secure Boot is enabled
> >     (XEN) Lockdown mode is enabled
> >     (XEN) Ignoring cmdline option 'hap_2mb' in lockdown mode
> >     (XEN) Ignoring cmdline option 'hap_1gb' in lockdown mode
> >      __  __            _  _    _  ___   ____
> >      \ \/ /___ _ __   | || |  / |/ _ \ |___ \
> >
> >
> > This is from a testcase failure I've just debugged, but the issue is not
> > specific to secureboot.  More generally, any diagnostic from cmdline
> > parsing ends up in this state.
> >
> > For PVShim, we did a gross hack to assume the PV console is usable right
> > from the outset, but all other console configurations don't start
> > working until the cmdline has been parsed.
> >
> > Is anything stopping us from feeding the current contents of the console
> > ring into serial in console_init_preirq()?
> >
> > ~Andrew
> >
> 
> I don't think there's a problem replaying the console logs into the
> serial once the serial is set up.
> It will slow down the boot maybe but beside that it should be fine.

Replaying console logs into the serial is already a case on staging branch:
console_init_preirq() calls conring_flush(..), which will replay everything
from conring into all available console devices.

--
Denis

> 
> Frediano
>