Skip to contentSkip to conversation

x86: Add split/bus lock smoke test

PATCH Applied Sean Christopherson 1 participant 4 messages Feb 21, 2025 – Feb 24, 2025 KVM

lore
Share & tools
lore

Sean Christopherson

author reviewer
#0 lore raw

cover x86: Add split/bus lock smoke test

Add a testcase to verify the guest does NOT get hit with an #AC or #DB when generating a split-lock access. Bus Lock Detect, a.k.a. BusLockTrap, enabling on AMD exposed a bug where KVM incorrectly runs the guest with the host's DEBUGCTL MSR.

Sean Christopherson (2):
x86: Include libcflat.h in atomic.h for u64 typedef x86/debug: Add a split-lock #AC / bus-lock #DB testcase

 lib/x86/atomic.h |  2 ++
 x86/debug.c      | 45 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 47 insertions(+)

base-commit: f77fb696cfd0e4a5562cdca189be557946bf522f

Signature · 1 line
2.48.1.601.g30ceb7b040-goog
author reviewer
#1 lore raw

PATCH 1/2 x86: Include libcflat.h in atomic.h for u64 typedef

Include libcflat.h in x86's atomic.h to pick up the u64 typedef, which is used to define atomic64_t. The missing include results in build errors if a test includes atomic.h without (or before) libcflat.h.

  lib/x86/atomic.h:162:1: error: unknown type name ‘u64’
  162 | u64 atomic64_cmpxchg(atomic64_t *v, u64 old, u64 new);
---
 lib/x86/atomic.h | 2 ++
 1 file changed, 2 insertions(+)
lib/x86/atomic.h +2 -0
index 13e734bb..3c7a89eb 100644
@@ -3,6 +3,8 @@
 
 #include "asm-generic/atomic.h"
 
+#include "libcflat.h"
+
 typedef struct {
 	volatile int counter;
 } atomic_t;
Signature · 1 line
2.48.1.601.g30ceb7b040-goog
author reviewer
#2 lore raw

PATCH 2/2 x86/debug: Add a split-lock #AC / bus-lock #DB testcase

In reply to Sean Christopherson

Add a testcase to the debug test to verify that a split-lock in the guest does NOT result in an #AC or #DB (and that the test isn't killed). While KVM may run the guest with split-lock #AC enabled, KVM should never inject an #AC into the guest. And Bus Lock Detect/Trap should flat out never be enabled while KVM is running the guest.

---
 x86/debug.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 45 insertions(+)
x86/debug.c +45 -0
index f493567c..09f06ef5 100644
@@ -10,6 +10,7 @@
  */
 #include <asm/debugreg.h>
 
+#include "atomic.h"
 #include "libcflat.h"
 #include "processor.h"
 #include "desc.h"
@@ -81,6 +82,12 @@ static void handle_ud(struct ex_regs *regs)
 	got_ud = 1;
 }
 
+static bool got_ac;
+static void handle_ac(struct ex_regs *regs)
+{
+	got_ac = true;
+}
+
 typedef unsigned long (*db_test_fn)(void);
 typedef void (*db_report_fn)(unsigned long, const char *);
 
@@ -409,6 +416,41 @@ static noinline unsigned long singlestep_with_sti_hlt(void)
 	return start_rip;
 }
 
+static noinline uint64_t bus_lock(uint64_t magic)
+{
+	uint8_t buffer[128] __attribute__((aligned(64))) = { };
+	atomic64_t *val = (atomic64_t *)&buffer[60];
+
+	atomic64_cmpxchg(val, 0, magic);
+	return READ_ONCE(*(uint64_t *)val);
+}
+
+static void bus_lock_test(void)
+{
+	const uint64_t magic = 0xdeadbeefdeadbeefull;
+	bool bus_lock_db = false;
+	uint64_t val;
+
+	/*
+	 * Generate a bus lock (via a locked access that splits cache lines)
+	 * in CPL0 and again in CPL3 (Bus Lock Detect only affects CPL3), and
+	 * verify that no #AC or #DB is generated (the relevant features are
+	 * not enabled).
+	 */
+	val = bus_lock(magic);
+	report(!got_ac && !n && val == magic,
+	       "CPL0 Split Lock #AC = %u (#DB = %u), val = %lx (wanted %lx)",
+	       got_ac, n, val, magic);
+
+	val = run_in_user((usermode_func)bus_lock, DB_VECTOR, magic, 0, 0, 0, &bus_lock_db);
+	report(!bus_lock_db && val == magic,
+	       "CPL3 Bus Lock #DB = %u, val = %lx (wanted %lx)",
+	       bus_lock_db, val, magic);
+
+	n = 0;
+	got_ac = false;
+}
+
 int main(int ac, char **av)
 {
 	unsigned long cr4;
@@ -416,6 +458,9 @@ int main(int ac, char **av)
 	handle_exception(DB_VECTOR, handle_db);
 	handle_exception(BP_VECTOR, handle_bp);
 	handle_exception(UD_VECTOR, handle_ud);
+	handle_exception(AC_VECTOR, handle_ac);
+
+	bus_lock_test();
 
 	/*
 	 * DR4 is an alias for DR6 (and DR5 aliases DR7) if CR4.DE is NOT set,
Signature · 1 line
2.48.1.601.g30ceb7b040-goog
author reviewer Applied
#3 lore raw
In reply to Sean Christopherson
Quoted 10 lines · Sean Christopherson

Add a testcase to verify the guest does NOT get hit with an #AC or #DB when generating a split-lock access. Bus Lock Detect, a.k.a. BusLockTrap, enabling on AMD exposed a bug where KVM incorrectly runs the guest with the host's DEBUGCTL MSR.

Sean Christopherson (2):
x86: Include libcflat.h in atomic.h for u64 typedef x86/debug: Add a split-lock #AC / bus-lock #DB testcase

[...]

Applied to kvm-x86 next (and now pulled by Paolo), thanks!

[1/2] x86: Include libcflat.h in atomic.h for u64 typedef

      https://github.com/kvm-x86/kvm-unit-tests/commit/36fb9e84a465
[2/2] x86/debug: Add a split-lock #AC / bus-lock #DB testcase
      https://github.com/kvm-x86/kvm-unit-tests/commit/8d9218bb6b7c

--
https://github.com/kvm-x86/kvm-unit-tests/tree/next