Re: [XEN][PATCH] console/consoleio: account for xen serial input focus during write

[Jason Andryuk <jason.andryuk@xxxxxxx>]

On 2025-12-04 18:32, Grygorii Strashko wrote:
> From: Grygorii Strashko <grygorii_strashko@xxxxxxxx>
>
> When 2 or more domains are created and:
> - one is hwdom with "hvc0" (console_io) console
> - other DomUs with vpl011 or "hvc0" (console_io) console
> console output from hwdom may mix with output from other domains.
>
> Example:
> [    2.288816] Key type id_legacy registered
> [    2.291894] n(XEN) DOM1: [    1.016950] DMA: preallocated 128 KiB 
> GFP_KERNEL|GFP_DMA32 pool for atomic allocations
> fs4filelayout_init: NFSv4 File Layout Driver Registering...
> (XEN) DOM1: [    1.018846] audit: initializing netlink subsys (disabled)
>
> This happens because for hwdom the console output is produced by domain and
> handled by Xen as stream of chars, which can be interrupted when hwdom is
> scheduled out and so, cause console output mix.
> The Xen consoleio code trying to mimic serial HW behavior for hwdom
> unconditionally by sending available data to serial HW on arrival.
> Xen consoleio code does not account for Xen console input focus, comparing
> to emulated serial hw, like vpl011, which does the same for domain with
> active Xen console input focus only.
>
> Switching console input focus to Xen improves situation, but not enough.
>
> This patch changes consoleio code to account for domain with active Xen
> console input focus - console output will be sent directly to serial HW
> only if domain has active Xen console input focus. For other domains -
> console output will be buffered and sync on per-line basis.
>
> Example output:
> (d2) [    4.263417] Key type id_legacy registered
> (XEN) DOM1: [    4.658080] Advanced Linux Sound Architecture Driver Initialized.
> (d2) [    4.277824] nfs4filelayout_init: NFSv4 File Layout Driver Registering...
>
> Signed-off-by: Grygorii Strashko <grygorii_strashko@xxxxxxxx>
> ---
> This causes random multi-domain tests failures due to inter-domain console
> mixing which breaks console parsing checks.
Part of the motivation here is that in a downstream, I've enabled domUs 
to use the consoleio hypercalls with Hyperlunch to remove dependency on 
xenconsoled.  Grygorii can confirm if it also affects ARM sometimes.
>   xen/drivers/char/console.c | 8 +++++++-
>   1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/xen/drivers/char/console.c b/xen/drivers/char/console.c
> index a99605103552..391cefc1a7c6 100644
> --- a/xen/drivers/char/console.c
> +++ b/xen/drivers/char/console.c
> @@ -733,6 +733,8 @@ static long 
> guest_console_write(XEN_GUEST_HANDLE_PARAM(char) buffer,
>   
>       while ( count > 0 )
>       {
> +        struct domain *input;
> +
>           if ( kcount && hypercall_preempt_check() )
>               return hypercall_create_continuation(
>                   __HYPERVISOR_console_io, "iih",
> @@ -742,7 +744,9 @@ static long 
> guest_console_write(XEN_GUEST_HANDLE_PARAM(char) buffer,
>           if ( copy_from_guest(kbuf, buffer, kcount) )
>               return -EFAULT;
>   
> -        if ( is_hardware_domain(cd) )
> +        input = console_get_domain();
> +
Maybe remove this blank line?

> +        if ( cd == input )
>           {
>               /* Use direct console output as it could be interactive */
>               nrspin_lock_irq(&console_lock);
> @@ -783,6 +787,8 @@ static long 
> guest_console_write(XEN_GUEST_HANDLE_PARAM(char) buffer,
In between the hunks is:
    guest_printk(cd, XENLOG_G_DEBUG "%s%s\n", cons->buf, kbuf);

So a background dom0 would get a d0 prefix and print at debug level.

For ARM, vpl011_write_data_xen() does:

if ( d == input )
    printk("%s", intf->out);
else
    printk("DOM%u: %s", d->domain_id, intf->out);

I think d0: is okay for background printing, but we'd need to up the log 
level for at least hardware domain.  Maybe we want a properly to control 
the level for each domain?
I think the changes made in this patch make sense.

Thanks,
Jason

>               spin_unlock(&cons->lock);
>           }
>   
> +        console_put_domain(input);
> +
>           guest_handle_add_offset(buffer, kcount);
>           count -= kcount;
>       }