x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

[PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Roger Pau Monne 2 days, 10 hours ago

Reading from the E9 port if the emergency console is active should return
0xe9 according to the documentation from Bochs:

https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html

See `port_e9_hack` section description.

Fix Xen so it also returns the port address.  OSes can use it to detect
whether the emergency console is available or not.

Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
---
 xen/arch/x86/hvm/hvm.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index 23bd7f078a1d..0c60faa39d7b 100644
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -567,9 +567,15 @@ static int cf_check hvm_print_line(
 
     ASSERT(bytes == 1 && port == XEN_HVM_DEBUGCONS_IOPORT);
 
-    /* Deny any input requests. */
-    if ( dir != IOREQ_WRITE )
-        return X86EMUL_UNHANDLEABLE;
+    if ( dir == IOREQ_READ )
+    {
+        /*
+         * According to Bochs documentation, reading from the E9 port should
+         * return 0xE9 if the "port E9 hack" is implemented.
+         */
+        *val = XEN_HVM_DEBUGCONS_IOPORT;
+        return X86EMUL_OKAY;
+    }
 
     if ( !is_console_printable(c) )
         return X86EMUL_OKAY;
-- 
2.51.0

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Andrew Cooper 2 days, 9 hours ago

On 02/10/2025 11:22 am, Roger Pau Monne wrote:
> Reading from the E9 port if the emergency console is active should return
> 0xe9 according to the documentation from Bochs:
>
> https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
>
> See `port_e9_hack` section description.
>
> Fix Xen so it also returns the port address.  OSes can use it to detect
> whether the emergency console is available or not.
>
> Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
> Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>

Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>

That's been wrong for rather a long time.  How did you find it?

CC-ing Oleksii as you've tagged this for 4.21.

> ---
>  xen/arch/x86/hvm/hvm.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
> index 23bd7f078a1d..0c60faa39d7b 100644
> --- a/xen/arch/x86/hvm/hvm.c
> +++ b/xen/arch/x86/hvm/hvm.c
> @@ -567,9 +567,15 @@ static int cf_check hvm_print_line(
>  
>      ASSERT(bytes == 1 && port == XEN_HVM_DEBUGCONS_IOPORT);
>  
> -    /* Deny any input requests. */
> -    if ( dir != IOREQ_WRITE )
> -        return X86EMUL_UNHANDLEABLE;
> +    if ( dir == IOREQ_READ )
> +    {
> +        /*
> +         * According to Bochs documentation, reading from the E9 port should
> +         * return 0xE9 if the "port E9 hack" is implemented.
> +         */
> +        *val = XEN_HVM_DEBUGCONS_IOPORT;
> +        return X86EMUL_OKAY;
> +    }
>  
>      if ( !is_console_printable(c) )
>          return X86EMUL_OKAY;

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Roger Pau Monné 2 days, 6 hours ago

On Thu, Oct 02, 2025 at 11:37:36AM +0100, Andrew Cooper wrote:
> On 02/10/2025 11:22 am, Roger Pau Monne wrote:
> > Reading from the E9 port if the emergency console is active should return
> > 0xe9 according to the documentation from Bochs:
> >
> > https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
> >
> > See `port_e9_hack` section description.
> >
> > Fix Xen so it also returns the port address.  OSes can use it to detect
> > whether the emergency console is available or not.
> >
> > Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
> > Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
> 
> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
> 
> That's been wrong for rather a long time.  How did you find it?

I came across the documentation above and I didn't remember Xen
returning any value for reads, which sadly was indeed true.

This was because I had the intention to suggest Alejandro to (also?) use
the port 0xe9 hack for printing from XTF, which should work for both
Xen and QEMU.

> CC-ing Oleksii as you've tagged this for 4.21.

I was told that bugfixes didn't need a release-ack until hard code
freeze, which is the 31st of October?

Thanks, Roger.

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Oleksii Kurochko 1 day, 11 hours ago

On 10/2/25 3:38 PM, Roger Pau Monné wrote:
> On Thu, Oct 02, 2025 at 11:37:36AM +0100, Andrew Cooper wrote:
>> On 02/10/2025 11:22 am, Roger Pau Monne wrote:
>>> Reading from the E9 port if the emergency console is active should return
>>> 0xe9 according to the documentation from Bochs:
>>>
>>> https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
>>>
>>> See `port_e9_hack` section description.
>>>
>>> Fix Xen so it also returns the port address.  OSes can use it to detect
>>> whether the emergency console is available or not.
>>>
>>> Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
>>> Signed-off-by: Roger Pau Monné<roger.pau@citrix.com>
>> Reviewed-by: Andrew Cooper<andrew.cooper3@citrix.com>
>>
>> That's been wrong for rather a long time.  How did you find it?
> I came across the documentation above and I didn't remember Xen
> returning any value for reads, which sadly was indeed true.
>
> This was because I had the intention to suggest Alejandro to (also?) use
> the port 0xe9 hack for printing from XTF, which should work for both
> Xen and QEMU.
>
>> CC-ing Oleksii as you've tagged this for 4.21.
> I was told that bugfixes didn't need a release-ack until hard code
> freeze, which is the 31st of October?

I meant until the start of the hard code freeze, which is on October 4.
So there is no need for a release-ack for this patch.

Thanks.

~ Oleksii

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Alejandro Vallejo 2 days, 6 hours ago

On Thu Oct 2, 2025 at 3:38 PM CEST, Roger Pau Monné wrote:
> On Thu, Oct 02, 2025 at 11:37:36AM +0100, Andrew Cooper wrote:
>> On 02/10/2025 11:22 am, Roger Pau Monne wrote:
>> > Reading from the E9 port if the emergency console is active should return
>> > 0xe9 according to the documentation from Bochs:
>> >
>> > https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
>> >
>> > See `port_e9_hack` section description.
>> >
>> > Fix Xen so it also returns the port address.  OSes can use it to detect
>> > whether the emergency console is available or not.
>> >
>> > Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
>> > Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
>> 
>> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
>> 
>> That's been wrong for rather a long time.  How did you find it?
>
> I came across the documentation above and I didn't remember Xen
> returning any value for reads, which sadly was indeed true.
>
> This was because I had the intention to suggest Alejandro to (also?) use
> the port 0xe9 hack for printing from XTF, which should work for both
> Xen and QEMU.

QEMU doesn't support 0xE9 though?

Cheers,
Alejandro

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Roger Pau Monné 2 days, 6 hours ago

On Thu, Oct 02, 2025 at 04:02:00PM +0200, Alejandro Vallejo wrote:
> On Thu Oct 2, 2025 at 3:38 PM CEST, Roger Pau Monné wrote:
> > On Thu, Oct 02, 2025 at 11:37:36AM +0100, Andrew Cooper wrote:
> >> On 02/10/2025 11:22 am, Roger Pau Monne wrote:
> >> > Reading from the E9 port if the emergency console is active should return
> >> > 0xe9 according to the documentation from Bochs:
> >> >
> >> > https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
> >> >
> >> > See `port_e9_hack` section description.
> >> >
> >> > Fix Xen so it also returns the port address.  OSes can use it to detect
> >> > whether the emergency console is available or not.
> >> >
> >> > Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
> >> > Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
> >> 
> >> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
> >> 
> >> That's been wrong for rather a long time.  How did you find it?
> >
> > I came across the documentation above and I didn't remember Xen
> > returning any value for reads, which sadly was indeed true.
> >
> > This was because I had the intention to suggest Alejandro to (also?) use
> > the port 0xe9 hack for printing from XTF, which should work for both
> > Xen and QEMU.
> 
> QEMU doesn't support 0xE9 though?

AFAICT it does:

https://wiki.osdev.org/QEMU#The_debug_console

However when running QEMU on Xen as a device model writes to 0xe9 are
handled in the hypervisor, and never forwarded to any device model.

Regards, Roger.

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Alejandro Vallejo 2 days, 5 hours ago

On Thu Oct 2, 2025 at 4:17 PM CEST, Roger Pau Monné wrote:
> On Thu, Oct 02, 2025 at 04:02:00PM +0200, Alejandro Vallejo wrote:
>> On Thu Oct 2, 2025 at 3:38 PM CEST, Roger Pau Monné wrote:
>> > On Thu, Oct 02, 2025 at 11:37:36AM +0100, Andrew Cooper wrote:
>> >> On 02/10/2025 11:22 am, Roger Pau Monne wrote:
>> >> > Reading from the E9 port if the emergency console is active should return
>> >> > 0xe9 according to the documentation from Bochs:
>> >> >
>> >> > https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
>> >> >
>> >> > See `port_e9_hack` section description.
>> >> >
>> >> > Fix Xen so it also returns the port address.  OSes can use it to detect
>> >> > whether the emergency console is available or not.
>> >> >
>> >> > Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
>> >> > Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
>> >> 
>> >> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
>> >> 
>> >> That's been wrong for rather a long time.  How did you find it?
>> >
>> > I came across the documentation above and I didn't remember Xen
>> > returning any value for reads, which sadly was indeed true.
>> >
>> > This was because I had the intention to suggest Alejandro to (also?) use
>> > the port 0xe9 hack for printing from XTF, which should work for both
>> > Xen and QEMU.
>> 
>> QEMU doesn't support 0xE9 though?
>
> AFAICT it does:
>
> https://wiki.osdev.org/QEMU#The_debug_console
>
> However when running QEMU on Xen as a device model writes to 0xe9 are
> handled in the hypervisor, and never forwarded to any device model.
>
> Regards, Roger.

The more you know. I searched for it once upon a time and couldn't find it.
Thanks for the pointer.

Regardless, "-debugcon stdio" is functionally equivalent to "-serial stdio"
and the latter can be adapted to work on real hardware too.

Cheers,
Alejandro

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Roger Pau Monné 1 day, 12 hours ago

On Thu, Oct 02, 2025 at 04:56:48PM +0200, Alejandro Vallejo wrote:
> On Thu Oct 2, 2025 at 4:17 PM CEST, Roger Pau Monné wrote:
> > On Thu, Oct 02, 2025 at 04:02:00PM +0200, Alejandro Vallejo wrote:
> >> On Thu Oct 2, 2025 at 3:38 PM CEST, Roger Pau Monné wrote:
> >> > On Thu, Oct 02, 2025 at 11:37:36AM +0100, Andrew Cooper wrote:
> >> >> On 02/10/2025 11:22 am, Roger Pau Monne wrote:
> >> >> > Reading from the E9 port if the emergency console is active should return
> >> >> > 0xe9 according to the documentation from Bochs:
> >> >> >
> >> >> > https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
> >> >> >
> >> >> > See `port_e9_hack` section description.
> >> >> >
> >> >> > Fix Xen so it also returns the port address.  OSes can use it to detect
> >> >> > whether the emergency console is available or not.
> >> >> >
> >> >> > Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
> >> >> > Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
> >> >> 
> >> >> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
> >> >> 
> >> >> That's been wrong for rather a long time.  How did you find it?
> >> >
> >> > I came across the documentation above and I didn't remember Xen
> >> > returning any value for reads, which sadly was indeed true.
> >> >
> >> > This was because I had the intention to suggest Alejandro to (also?) use
> >> > the port 0xe9 hack for printing from XTF, which should work for both
> >> > Xen and QEMU.
> >> 
> >> QEMU doesn't support 0xE9 though?
> >
> > AFAICT it does:
> >
> > https://wiki.osdev.org/QEMU#The_debug_console
> >
> > However when running QEMU on Xen as a device model writes to 0xe9 are
> > handled in the hypervisor, and never forwarded to any device model.
> >
> > Regards, Roger.
> 
> The more you know. I searched for it once upon a time and couldn't find it.
> Thanks for the pointer.
> 
> Regardless, "-debugcon stdio" is functionally equivalent to "-serial stdio"
> and the latter can be adapted to work on real hardware too.

The emulated serial is possibly equivalent to the 0xe9 hack in QEMU,
as I expect the serial is mostly setup to "just work".  On native you
likely need to configure the baud rate &c, plus implement flow
control?

Is there any testing XTF should do to ensure the serial is there,
before blindly writing at 0x3f8?

Thanks, Roger.

Re: [PATCH for-4.21] x86/hvm: fix reading from 0xe9 IO port if port E9 hack is active

Posted by Alejandro Vallejo 1 day, 11 hours ago

On Fri Oct 3, 2025 at 10:18 AM CEST, Roger Pau Monné wrote:
> On Thu, Oct 02, 2025 at 04:56:48PM +0200, Alejandro Vallejo wrote:
>> On Thu Oct 2, 2025 at 4:17 PM CEST, Roger Pau Monné wrote:
>> > On Thu, Oct 02, 2025 at 04:02:00PM +0200, Alejandro Vallejo wrote:
>> >> On Thu Oct 2, 2025 at 3:38 PM CEST, Roger Pau Monné wrote:
>> >> > On Thu, Oct 02, 2025 at 11:37:36AM +0100, Andrew Cooper wrote:
>> >> >> On 02/10/2025 11:22 am, Roger Pau Monne wrote:
>> >> >> > Reading from the E9 port if the emergency console is active should return
>> >> >> > 0xe9 according to the documentation from Bochs:
>> >> >> >
>> >> >> > https://bochs.sourceforge.io/doc/docbook/user/bochsrc.html
>> >> >> >
>> >> >> > See `port_e9_hack` section description.
>> >> >> >
>> >> >> > Fix Xen so it also returns the port address.  OSes can use it to detect
>> >> >> > whether the emergency console is available or not.
>> >> >> >
>> >> >> > Fixes: d1bd157fbc9b ("Big merge the HVM full-virtualisation abstractions.")
>> >> >> > Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
>> >> >> 
>> >> >> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
>> >> >> 
>> >> >> That's been wrong for rather a long time.  How did you find it?
>> >> >
>> >> > I came across the documentation above and I didn't remember Xen
>> >> > returning any value for reads, which sadly was indeed true.
>> >> >
>> >> > This was because I had the intention to suggest Alejandro to (also?) use
>> >> > the port 0xe9 hack for printing from XTF, which should work for both
>> >> > Xen and QEMU.
>> >> 
>> >> QEMU doesn't support 0xE9 though?
>> >
>> > AFAICT it does:
>> >
>> > https://wiki.osdev.org/QEMU#The_debug_console
>> >
>> > However when running QEMU on Xen as a device model writes to 0xe9 are
>> > handled in the hypervisor, and never forwarded to any device model.
>> >
>> > Regards, Roger.
>> 
>> The more you know. I searched for it once upon a time and couldn't find it.
>> Thanks for the pointer.
>> 
>> Regardless, "-debugcon stdio" is functionally equivalent to "-serial stdio"
>> and the latter can be adapted to work on real hardware too.
>
> The emulated serial is possibly equivalent to the 0xe9 hack in QEMU,
> as I expect the serial is mostly setup to "just work".  On native you
> likely need to configure the baud rate &c, plus implement flow
> control?

You'd need to boot XTF via GRUB (or similar), which would've pre-configured
the serial port for you.

And baud rate, flow control and the like are ignored when emulated, so that's
all fine.

>
> Is there any testing XTF should do to ensure the serial is there,
> before blindly writing at 0x3f8?

Closest would be parsing the FADT for LEGACY_DEVICES being set, but if that was
clear we'd be out of options because there's nothing else to write to. I don't
intend to make XTF rock solid on real hardware, just "workable" on typical
everyday devices/servers with GRUB's help.

FWIW Xen just assumes it's there if com1= is given in the cmdline AFAICS. The
most likely case for COM1 not being there is on PV(H), which XTF already uses
the PV console for.

Cheers,
Alejandro