[PATCH 1/1] x86/cpu/topology: Don't limit CPUs to 1 for Xen PV guests due to disabled APIC

[PATCH 1/1] x86/cpu/topology: Don't limit CPUs to 1 for Xen PV guests due to disabled APIC

[Petr Vaněk]

Xen PV guests in DomU have APIC disabled by design, which causes
topology_apply_cmdline_limits_early() to limit the number of possible
CPUs to 1, regardless of the configured number of vCPUs.

This is a regression introduced in version 6.9 in commit 7c0edad3643f
("x86/cpu/topology: Rework possible CPU management") which added an
early check that limits CPUs if apic_is_disabled, without accounting for
the fact that Xen PV guests always disable APIC even when SMP is
supported.

This patch fixes the issue by skipping the apic_is_disabled check for
Xen PV guests, allowing them to boot with the full set of configured vCPUs.

Fixes: 7c0edad3643f ("x86/cpu/topology: Rework possible CPU management")
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86@kernel.org
Cc: xen-devel@lists.xenproject.org
Cc: stable@vger.kernel.org # 6.9+
Signed-off-by: Petr Vaněk <arkamar@atlas.cz>
---
 arch/x86/kernel/cpu/topology.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/cpu/topology.c b/arch/x86/kernel/cpu/topology.c
index 01456236a6dd..10aa7f471ec9 100644
--- a/arch/x86/kernel/cpu/topology.c
+++ b/arch/x86/kernel/cpu/topology.c
@@ -428,8 +428,13 @@ void __init topology_apply_cmdline_limits_early(void)
 {
 	unsigned int possible = nr_cpu_ids;
 
-	/* 'maxcpus=0' 'nosmp' 'nolapic' */
-	if (!setup_max_cpus || apic_is_disabled)
+	/* 'maxcpus=0' 'nosmp' 'nolapic'
+	 *
+	 * The apic_is_disabled check is ignored for Xen PV domains because Xen
+	 * disables ACPI in unprivileged PV DomU guests, which would otherwise limit
+	 * CPUs to 1, even if multiple vCPUs were configured.
+	 */
+	if (!setup_max_cpus || (!xen_pv_domain() && apic_is_disabled))
 		possible = 1;
 
 	/* 'possible_cpus=N' */
-- 
2.48.1

Re: [PATCH 1/1] x86/cpu/topology: Don't limit CPUs to 1 for Xen PV guests due to disabled APIC

[Thomas Gleixner]

On Sat, Apr 05 2025 at 20:16, Petr Vaněk wrote:

> Xen PV guests in DomU have APIC disabled by design, which causes
> topology_apply_cmdline_limits_early() to limit the number of possible
> CPUs to 1, regardless of the configured number of vCPUs.

PV guests have a APIC emulation and there is no code which actually
disables the APIC by design unconditionally. There is one way though,
which disables the APIC indirectly.

xen_arch_setup() disables ACPI, which in turn causes acpi_mps_check() to
return 1, which disables the APIC. This only happens when the kernel
configuration has:

     CONFIG_X86_MPPARSE=n
     CONFIG_ACPI=y

If you enable MPPARSE the problem goes away, no?

> +	/* 'maxcpus=0' 'nosmp' 'nolapic'
> +	 *
> +	 * The apic_is_disabled check is ignored for Xen PV domains because Xen
> +	 * disables ACPI in unprivileged PV DomU guests, which would otherwise limit
> +	 * CPUs to 1, even if multiple vCPUs were configured.

This is the wrong place as it invalidates the effect of 'nolapic' on the
kernel command line for XEN PV.

You actually explain in the comment that XEN disables ACPI, so why are
you slapping this xen check into this code instead of doing the obvious
and prevent acpi_mps_check() to cause havoc?

Thanks,

        tglx

Re: [PATCH 1/1] x86/cpu/topology: Don't limit CPUs to 1 for Xen PV guests due to disabled APIC

[Petr Vaněk]

On Sun, Apr 06, 2025 at 12:08:29PM +0200, Thomas Gleixner wrote:
> On Sat, Apr 05 2025 at 20:16, Petr Vaněk wrote:
> 
> > Xen PV guests in DomU have APIC disabled by design, which causes
> > topology_apply_cmdline_limits_early() to limit the number of possible
> > CPUs to 1, regardless of the configured number of vCPUs.
> 
> PV guests have a APIC emulation and there is no code which actually
> disables the APIC by design unconditionally. There is one way though,
> which disables the APIC indirectly.

It seems I have got a bit lost in APIC/ACPI abbreviations. Sorry.

> xen_arch_setup() disables ACPI, which in turn causes acpi_mps_check() to
> return 1, which disables the APIC. This only happens when the kernel
> configuration has:
> 
>      CONFIG_X86_MPPARSE=n
>      CONFIG_ACPI=y
> 
> If you enable MPPARSE the problem goes away, no?

Yes, it goes away.

> > +	/* 'maxcpus=0' 'nosmp' 'nolapic'
> > +	 *
> > +	 * The apic_is_disabled check is ignored for Xen PV domains because Xen
> > +	 * disables ACPI in unprivileged PV DomU guests, which would otherwise limit
> > +	 * CPUs to 1, even if multiple vCPUs were configured.
> 
> This is the wrong place as it invalidates the effect of 'nolapic' on the
> kernel command line for XEN PV.
> 
> You actually explain in the comment that XEN disables ACPI, so why are
> you slapping this xen check into this code instead of doing the obvious
> and prevent acpi_mps_check() to cause havoc?

Thank you for your explanation and suggestion. I will correct acpi_mps_check()
in following patch.

Thanks,
Petr

[PATCH v2 1/1] x86/acpi: Don't limit CPUs to 1 for Xen PV guests due to disabled ACPI

[Petr Vaněk]

Xen disables ACPI for PV guests in DomU, which causes acpi_mps_check()
to return 1 when CONFIG_X86_MPPARSE is not set. As a result, APIC is
disabled and the guest is later limited to a single vCPU, despite being
configured with more.

This regression was introduced in version 6.9 in commit 7c0edad3643f
("x86/cpu/topology: Rework possible CPU management"), which added an
early check that limits CPUs to 1 if apic_is_disabled.

This patch updates acpi_mps_check() logic to return 0 early when running
as a Xen PV guest in DomU, preventing APIC from being disabled in this
specific case and restoring correct multi-vCPU behaviour.

Fixes: 7c0edad3643f ("x86/cpu/topology: Rework possible CPU management")
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86@kernel.org
Cc: xen-devel@lists.xenproject.org
Cc: stable@vger.kernel.org # 6.9+
Signed-off-by: Petr Vaněk <arkamar@atlas.cz>
---
 arch/x86/kernel/acpi/boot.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/arch/x86/kernel/acpi/boot.c b/arch/x86/kernel/acpi/boot.c
index dae6a73be40e..9fa321a95eb3 100644
--- a/arch/x86/kernel/acpi/boot.c
+++ b/arch/x86/kernel/acpi/boot.c
@@ -23,6 +23,8 @@
 #include <linux/serial_core.h>
 #include <linux/pgtable.h>
 
+#include <xen/xen.h>
+
 #include <asm/e820/api.h>
 #include <asm/irqdomain.h>
 #include <asm/pci_x86.h>
@@ -1729,6 +1731,15 @@ int __init acpi_mps_check(void)
 {
 #if defined(CONFIG_X86_LOCAL_APIC) && !defined(CONFIG_X86_MPPARSE)
 /* mptable code is not built-in*/
+
+	/*
+	 * Xen disables ACPI in PV DomU guests but it still emulates APIC and
+	 * supports SMP. Returning early here ensures that APIC is not disabled
+	 * unnecessarily and the guest is not limited to a single vCPU.
+	 */
+	if (xen_pv_domain() && !xen_initial_domain())
+		return 0;
+
 	if (acpi_disabled || acpi_noirq) {
 		pr_warn("MPS support code is not built-in, using acpi=off or acpi=noirq or pci=noacpi may have problem\n");
 		return 1;
-- 
2.48.1