On Thu, 29 Jan 2026, Jason Andryuk wrote:
> On 2026-01-29 17:08, Stefano Stabellini wrote:
> > Enable dom0less guests on ARM to use console_io hypercalls:
> > - set input_allow = true for dom0less domains
> > - update the in-code comment in console.c
> > - prioritize the VUART check to retain the same behavior as today
> > 
> > Signed-off-by: Stefano Stabellini <[email protected]>
> > ---
> > Changes in v8:
> > - move in-code comment update to previous patch
> > - add in-code comment about is_focus_domain() check
> > ---
> >   xen/common/device-tree/dom0less-build.c |  2 ++
> >   xen/drivers/char/console.c              | 16 ++++++++++------
> >   2 files changed, 12 insertions(+), 6 deletions(-)
> > 
> > diff --git a/xen/common/device-tree/dom0less-build.c
> > b/xen/common/device-tree/dom0less-build.c
> > index 840d14419d..cb7026fa7e 100644
> > --- a/xen/common/device-tree/dom0less-build.c
> > +++ b/xen/common/device-tree/dom0less-build.c
> > @@ -829,6 +829,8 @@ static int __init construct_domU(struct kernel_info
> > *kinfo,
> >         rangeset_destroy(kinfo->xen_reg_assigned);
> >   +    d->console->input_allowed = true;
> > +
> >       return rc;
> >   }
> >   diff --git a/xen/drivers/char/console.c b/xen/drivers/char/console.c
> > index ed8f1ad8f2..418d194cef 100644
> > --- a/xen/drivers/char/console.c
> > +++ b/xen/drivers/char/console.c
> > @@ -613,11 +613,20 @@ static void __serial_rx(char c)
> >       if ( console_rx == 0 )
> >           return handle_keypress(c, false);
> >   +    /* Includes an is_focus_domain() check. */
> >       d = console_get_domain();
> >       if ( !d )
> >           return;
> >   -    if ( is_hardware_domain(d) )
> > +#ifdef CONFIG_SBSA_VUART_CONSOLE
> > +    /* Prioritize vpl011 if enabled for this domain */
> > +    if ( d->arch.vpl011.base_addr )
> > +    {
> > +        /* Deliver input to the emulated UART. */
> > +        rc = vpl011_rx_char_xen(d, c);
> > +    }
> > +    else
> > +#endif
> >       {
> >           unsigned long flags;
> >   @@ -636,11 +645,6 @@ static void __serial_rx(char c)
> >            */
> >           send_global_virq(VIRQ_CONSOLE);
> 
> I think we need an additional patch, or included in one of these two, to
> change VIRQ_CONSOLE to a VIRQ_DOMAIN.  Otherwise only hwdom could bind to the
> virq, I think?  It would be the two changes below:

Thank you Jason. I didn't notice this problem because Linux is able to
silently fallback to polling which works surprisingly well. I didn't
notice the difference.

I confirm that the error you highlighted is real and that with the patch
below the error goes away.

My preference is to submit it as a separate patch, it can still be part
of this series. Especially as the other two patches have already been
reviewed and tested independently a few times. But I am happy either
way.

Jan, what do you think?

 
> diff --git a/xen/common/event_channel.c b/xen/common/event_channel.c
> index 67700b050a..dab123f20d 100644
> --- a/xen/common/event_channel.c
> +++ b/xen/common/event_channel.c
> @@ -138,6 +138,7 @@ static enum virq_type get_virq_type(unsigned int virq)
>          return VIRQ_VCPU;
> 
>      case VIRQ_ARGO:
> +    case VIRQ_CONSOLE:
>          return VIRQ_DOMAIN;
> 
>      case VIRQ_ARCH_0 ... VIRQ_ARCH_7:
> diff --git a/xen/drivers/char/console.c b/xen/drivers/char/console.c
> index 903ad912cc..138eeaa14d 100644
> --- a/xen/drivers/char/console.c
> +++ b/xen/drivers/char/console.c
> @@ -611,7 +611,7 @@ static void __serial_rx(char c)
>           * Always notify the hardware domain: prevents receive path from
>           * getting stuck.
>           */
> -        send_global_virq(VIRQ_CONSOLE);
> +        send_guest_domain_virq(d, VIRQ_CONSOLE);
>      }
>  #ifdef CONFIG_SBSA_VUART_CONSOLE
>      else
> 

Reply via email to