From: Sean Christopherson <seanjc@google.com>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: kvm@vger.kernel.org, Sean Christopherson <seanjc@google.com>,
Mathias Krause <minipli@grsecurity.net>,
Andrew Jones <andrew.jones@linux.dev>
Subject: [kvm-unit-tests PATCH v3 18/20] x86/vmx: Write to KVM's WALL_CLOCK MSR via VM-Entry load list sync in SIPI test
Date: Thu, 14 May 2026 14:04:58 -0700 [thread overview]
Message-ID: <20260514210500.1626871-19-seanjc@google.com> (raw)
In-Reply-To: <20260514210500.1626871-1-seanjc@google.com>
In the VMX Wait-for-SIPI => SIPI VM-Exit test, signal that the AP has
entered the guest by writing to MSR_KVM_WALL_CLOCK_NEW (when supported)
via the VM-Entry MSR load list instead of writing to memory from the AP
_before_ actually doing VM-Enter. Abusing the MSR load list ensures that
the AP's "ready" signal to the BSP happens atomically with respect to
VM-Enter, and thus fixes a race where the BSP can see "ready" and send the
SIPI before the AP has executed VM-Enter. E.g. with a delay inserted on
the AP, and no delay on the BSP, the test will hang 100% of the time.
Use MSR_KVM_WALL_CLOCK_NEW as it is pretty much the only MSR that KVM
emulates as a per-VM MSR, and that has a high likelihood of being
available.
Keep the BSP's delay before send the SIPI so that the test continues to
work if MSR_KVM_WALL_CLOCK_NEW isn't available, e.g. in bare metal (and
most KVM) setups, hitting the race is practically impossible.
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
x86/vmx_tests.c | 26 +++++++++++++++++++++++++-
1 file changed, 25 insertions(+), 1 deletion(-)
diff --git a/x86/vmx_tests.c b/x86/vmx_tests.c
index 31c7672c..ac0250b7 100644
--- a/x86/vmx_tests.c
+++ b/x86/vmx_tests.c
@@ -6,6 +6,7 @@
#include <asm/debugreg.h>
+#include "kvmclock.h"
#include "vmx.h"
#include "msr.h"
#include "processor.h"
@@ -10155,6 +10156,8 @@ static void vmx_init_signal_test(void)
*/
}
+static bool use_kvm_wall_clock;
+
#define SIPI_SIGNAL_TEST_DELAY 100000000ULL
static void vmx_sipi_test_guest(void)
@@ -10199,6 +10202,11 @@ static void vmx_sipi_test_guest(void)
static void sipi_test_ap_thread(void *data)
{
+ const struct vmx_msr_entry msr_load_wall_clock = {
+ .index = MSR_KVM_WALL_CLOCK_NEW,
+ .reserved = 0,
+ .value = 1,
+ };
struct guest_regs *regs = this_cpu_guest_regs();
struct vmcs *ap_vmcs;
u64 *ap_vmxon_region;
@@ -10231,7 +10239,13 @@ static void sipi_test_ap_thread(void *data)
/* Set guest activity state to wait-for-SIPI state */
vmcs_write(GUEST_ACTV_STATE, ACTV_WAIT_SIPI);
- vmx_set_test_stage(1);
+ if (use_kvm_wall_clock) {
+ wrmsr(MSR_KVM_WALL_CLOCK_NEW, 0);
+ vmcs_write(ENT_MSR_LD_CNT, 1);
+ vmcs_write(ENTER_MSR_LD_ADDR, virt_to_phys(&msr_load_wall_clock));
+ } else {
+ vmx_set_test_stage(1);
+ }
/* AP enter guest */
enter_guest();
@@ -10274,6 +10288,9 @@ static void vmx_sipi_signal_test(void)
u64 cpu_ctrl_0 = CPU_SECONDARY;
u64 cpu_ctrl_1 = 0;
+ use_kvm_wall_clock = this_cpu_has_kvm() &&
+ this_cpu_has(KVM_FEATURE_CLOCKSOURCE2);
+
/* passthrough lapic to L2 */
disable_intercept_for_x2apic_msrs();
vmcs_write(PIN_CONTROLS, vmcs_read(PIN_CONTROLS) & ~PIN_EXTINT);
@@ -10290,6 +10307,13 @@ static void vmx_sipi_signal_test(void)
/* start AP */
on_cpu_async(1, sipi_test_ap_thread, NULL);
+ if (use_kvm_wall_clock) {
+ while (rdmsr(MSR_KVM_WALL_CLOCK_NEW) != 1)
+ cpu_relax();
+
+ vmx_set_test_stage(1);
+ }
+
/* BSP enter guest */
enter_guest();
}
--
2.54.0.563.g4f69b47b94-goog
next prev parent reply other threads:[~2026-05-14 21:05 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-05-14 21:04 [kvm-unit-tests PATCH v3 00/20] x86: Better backtraces for leaf functions Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 01/20] x86/vmx: Drop unused SYSENTER "support" in nested VMX infrastructure Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 02/20] x86/vmx: Drop unused guest_regs " Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 03/20] x86/svm: Sort (and swap) GPRs by their index, not alphabetically Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 04/20] x86: Dedup guest/host context switch of registers across SVM and VMX Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 05/20] x86/virt: Use macro shenanigans to get reg offsets when swapping guest/host regs Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 06/20] x86/virt: Track "guest regs" using per-CPU variable Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 07/20] x86/svm: Don't VMLOAD/VMSAVE "guest" state around VMRUN Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 08/20] x86/vmx: Use separate VMCSes for BSP vs. AP in INIT test Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 09/20] x86/vmx: Swap GPRs after checking "launched" status Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 10/20] x86/vmx: Track VMCS "launched" state per-CPU Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 11/20] x86/vmx: Track "is this CPU in guest mode" per-CPU Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 12/20] x86/vmx: Communicate hypercalls via RAX, not a global field Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 13/20] x86/vmx: Initialize test stage in SIPI test *before* launching AP thread Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 14/20] x86/kvmclock: Replace spaces with tabs Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 15/20] x86/kvmclock: Skip kvmclock test when not running on KVM with CLOCKSOURCE2 Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 16/20] x86/vmx: Tag "struct vmx_msr_entry" as needing to be 16-byte aligned Sean Christopherson
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 17/20] x86/smp: Align the stack to a 16-byte boundary when invoking SMP function calls Sean Christopherson
2026-05-14 21:04 ` Sean Christopherson [this message]
2026-05-14 21:04 ` [kvm-unit-tests PATCH v3 19/20] x86: Better backtraces for leaf functions Sean Christopherson
2026-05-14 21:05 ` [kvm-unit-tests PATCH v3 20/20] x86: Prevent realmode test code instrumentation with nop-mcount Sean Christopherson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260514210500.1626871-19-seanjc@google.com \
--to=seanjc@google.com \
--cc=andrew.jones@linux.dev \
--cc=kvm@vger.kernel.org \
--cc=minipli@grsecurity.net \
--cc=pbonzini@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.