The SMT control mechanism got added as speculation attack vector
mitigation. The implemented logic relies on the primary thread mask to
be set up properly.
This turns out to be an issue with XEN/PV guests because their CPU hotplug
mechanics do not enumerate APICs and therefore the mask is never correctly
populated.
This went unnoticed so far because by chance XEN/PV ends up with
smp_num_siblings == 2. So cpu_smt_control stays at its default value
CPU_SMT_ENABLED and the primary thread mask is never evaluated in the
context of CPU hotplug.
This stopped "working" with the upcoming overhaul of the topology
evaluation which legitimately provides a fake topology for XEN/PV. That
sets smp_num_siblings to 1, which causes the core CPU hot-plug core to
refuse to bring up the APs.
This happens because cpu_smt_control is set to CPU_SMT_NOT_SUPPORTED which
causes cpu_bootable() to evaluate the unpopulated primary thread mask with
the conclusion that all non-boot CPUs are not valid to be plugged.
The core code has already been made more robust against this kind of fail,
but the primary thread mask really wants to be populated to avoid other
issues all over the place.
Just fake the mask by pretending that all XEN/PV vCPUs are primary threads,
which is consistent because all of XEN/PVs topology is fake or non-existent.
Fixes: 6a4d2657e048 ("x86/smp: Provide topology_is_primary_thread()")
Fixes: f54d4434c281 ("x86/apic: Provide cpu_primary_thread mask")
Reported-by: Juergen Gross <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Tested-by: Juergen Gross <[email protected]>
Tested-by: Sohil Mehta <[email protected]>
Tested-by: Michael Kelley <[email protected]>
---
V2: Fixup changelog - Borislav
---
arch/x86/kernel/apic/apic.c | 11 +++++++++++
1 file changed, 11 insertions(+)
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -36,6 +36,8 @@
#include <linux/smp.h>
#include <linux/mm.h>
+#include <xen/xen.h>
+
#include <asm/trace/irq_vectors.h>
#include <asm/irq_remapping.h>
#include <asm/pc-conf-reg.h>
@@ -2344,6 +2346,15 @@ static int __init smp_init_primary_threa
{
unsigned int cpu;
+ /*
+ * XEN/PV provides either none or useless topology information.
+ * Pretend that all vCPUs are primary threads.
+ */
+ if (xen_pv_domain()) {
+ cpumask_copy(&__cpu_primary_thread_mask, cpu_possible_mask);
+ return 0;
+ }
+
for (cpu = 0; cpu < nr_logical_cpuids; cpu++)
cpu_mark_primary_thread(cpu, cpuid_to_apicid[cpu]);
return 0;
The following commit has been merged into the x86/core branch of tip:
Commit-ID: 965e05ff8af98c44f9937366715c512000373164
Gitweb: https://git.kernel.org/tip/965e05ff8af98c44f9937366715c512000373164
Author: Thomas Gleixner <[email protected]>
AuthorDate: Mon, 14 Aug 2023 10:18:28 +02:00
Committer: Thomas Gleixner <[email protected]>
CommitterDate: Tue, 10 Oct 2023 14:38:17 +02:00
x86/apic: Fake primary thread mask for XEN/PV
The SMT control mechanism got added as speculation attack vector
mitigation. The implemented logic relies on the primary thread mask to
be set up properly.
This turns out to be an issue with XEN/PV guests because their CPU hotplug
mechanics do not enumerate APICs and therefore the mask is never correctly
populated.
This went unnoticed so far because by chance XEN/PV ends up with
smp_num_siblings == 2. So cpu_smt_control stays at its default value
CPU_SMT_ENABLED and the primary thread mask is never evaluated in the
context of CPU hotplug.
This stopped "working" with the upcoming overhaul of the topology
evaluation which legitimately provides a fake topology for XEN/PV. That
sets smp_num_siblings to 1, which causes the core CPU hot-plug core to
refuse to bring up the APs.
This happens because cpu_smt_control is set to CPU_SMT_NOT_SUPPORTED which
causes cpu_bootable() to evaluate the unpopulated primary thread mask with
the conclusion that all non-boot CPUs are not valid to be plugged.
The core code has already been made more robust against this kind of fail,
but the primary thread mask really wants to be populated to avoid other
issues all over the place.
Just fake the mask by pretending that all XEN/PV vCPUs are primary threads,
which is consistent because all of XEN/PVs topology is fake or non-existent.
Fixes: 6a4d2657e048 ("x86/smp: Provide topology_is_primary_thread()")
Fixes: f54d4434c281 ("x86/apic: Provide cpu_primary_thread mask")
Reported-by: Juergen Gross <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Tested-by: Juergen Gross <[email protected]>
Tested-by: Sohil Mehta <[email protected]>
Tested-by: Michael Kelley <[email protected]>
Tested-by: Peter Zijlstra (Intel) <[email protected]>
Tested-by: Zhang Rui <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
---
arch/x86/kernel/apic/apic.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
index 760adac..3cdf484 100644
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -36,6 +36,8 @@
#include <linux/smp.h>
#include <linux/mm.h>
+#include <xen/xen.h>
+
#include <asm/trace/irq_vectors.h>
#include <asm/irq_remapping.h>
#include <asm/pc-conf-reg.h>
@@ -2344,6 +2346,15 @@ static int __init smp_init_primary_thread_mask(void)
{
unsigned int cpu;
+ /*
+ * XEN/PV provides either none or useless topology information.
+ * Pretend that all vCPUs are primary threads.
+ */
+ if (xen_pv_domain()) {
+ cpumask_copy(&__cpu_primary_thread_mask, cpu_possible_mask);
+ return 0;
+ }
+
for (cpu = 0; cpu < nr_logical_cpuids; cpu++)
cpu_mark_primary_thread(cpu, cpuid_to_apicid[cpu]);
return 0;