Skip to contentSkip to conversation

KVM: arm64: selftests: arch_timer_edge_cases fixes

PATCH Applied Sebastian Ott 3 participants 8 messages Jun 5, 2025 – Jun 10, 2025

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

v3 cover KVM: arm64: selftests: arch_timer_edge_cases fixes

Some small fixes for arch_timer_edge_cases that I stumbled upon while debugging failures for this selftest on ampere-one.

Changes since v1:
* determine effective counter width based on suggestions from Marc
Changes since v2:
* new patch to fix xval initialization

I've done tests with this on various machines - no issues during several hundreds of test runs.

v1: https://lore.kernel.org/kvmarm/[email protected]/
v2: https://lore.kernel.org/kvmarm/[email protected]/

Sebastian Ott (4):
KVM: arm64: selftests: fix help text for arch_timer_edge_cases
KVM: arm64: selftests: fix thread migration in arch_timer_edge_cases
KVM: arm64: selftests: arch_timer_edge_cases - fix xval init
KVM: arm64: selftests: arch_timer_edge_cases - determine effective counter width

 .../kvm/arm64/arch_timer_edge_cases.c         | 39 ++++++++++++-------
 1 file changed, 25 insertions(+), 14 deletions(-)

base-commit: 0ff41df1cb268fc69e703a08a57ee14ae967d0ca

Signature · 1 line
2.49.0
#1 lore raw

PATCH v3 1/4 KVM: arm64: selftests: fix help text for arch_timer_edge_cases

Fix the help text for arch_timer_edge_cases to show the correct option for setting the wait time.

---
 tools/testing/selftests/kvm/arm64/arch_timer_edge_cases.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/kvm/arm64/arch_timer_edge_cases.c +1 -1
index a36a7e2db434..c4716e0c1438 100644
@@ -986,7 +986,7 @@ static void test_print_help(char *name)
 	pr_info("\t-b: Test both physical and virtual timers (default: true)\n");
 	pr_info("\t-l: Delta (in ms) used for long wait time test (default: %u)\n",
 	     LONG_WAIT_TEST_MS);
-	pr_info("\t-l: Delta (in ms) used for wait times (default: %u)\n",
+	pr_info("\t-w: Delta (in ms) used for wait times (default: %u)\n",
 		WAIT_TEST_MS);
 	pr_info("\t-p: Test physical timer (default: true)\n");
 	pr_info("\t-v: Test virtual timer (default: true)\n");
Signature · 1 line
2.49.0
#2 lore raw

PATCH v3 2/4 KVM: arm64: selftests: fix thread migration in arch_timer_edge_cases

In reply to Sebastian Ott

arch_timer_edge_cases tries to migrate itself across host cpus. Before the first test it migrates to cpu 0 by setting up an affinity mask with only bit 0 set. After that it looks for the next possible cpu in the current affinity mask which still has only bit 0 set. So there is no migration at all.

Fix this by reading the default mask at start and use this to find the next cpu in each iteration.

---
 tools/testing/selftests/kvm/arm64/arch_timer_edge_cases.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
tools/testing/selftests/kvm/arm64/arch_timer_edge_cases.c +5 -3
index c4716e0c1438..a813b4c6c817 100644
@@ -849,17 +849,17 @@ static void guest_code(enum arch_timer timer)
 	GUEST_DONE();
 }
 
+static cpu_set_t default_cpuset;
+
 static uint32_t next_pcpu(void)
 {
 	uint32_t max = get_nprocs();
 	uint32_t cur = sched_getcpu();
 	uint32_t next = cur;
-	cpu_set_t cpuset;
+	cpu_set_t cpuset = default_cpuset;
 
 	TEST_ASSERT(max > 1, "Need at least two physical cpus");
 
-	sched_getaffinity(0, sizeof(cpuset), &cpuset);
-
 	do {
 		next = (next + 1) % CPU_SETSIZE;
 	} while (!CPU_ISSET(next, &cpuset));
@@ -1046,6 +1046,8 @@ int main(int argc, char *argv[])
 	if (!parse_args(argc, argv))
 		exit(KSFT_SKIP);
 
+	sched_getaffinity(0, sizeof(default_cpuset), &default_cpuset);
+
 	if (test_args.test_virtual) {
 		test_vm_create(&vm, &vcpu, VIRTUAL);
 		test_run(vm, vcpu);
Signature · 1 line
2.49.0
#3 lore raw

PATCH v3 3/4 KVM: arm64: selftests: arch_timer_edge_cases - fix xval init

In reply to Sebastian Ott

arch_timer_edge_cases hits the following assertion in < 10% of the test runs:

==== Test Assertion Failure ====
arm64/arch_timer_edge_cases.c:490: timer_get_cntct(timer) >= DEF_CNT + (timer_get_cntfrq() * (uint64_t)(delta_2_ms) / 1000) pid=17110 tid=17110 errno=4 - Interrupted system call

     1  0x0000000000404ec7: test_run at arch_timer_edge_cases.c:945
     2  0x0000000000401fa3: main at arch_timer_edge_cases.c:1074
     3  0x0000ffffa774b587: ?? ??:0
     4  0x0000ffffa774b65f: ?? ??:0
     5  0x000000000040206f: _start at ??:?

timer_get_cntct(timer) >= DEF_CNT + msec_to_cycles(delta_2_ms)

Enabling the timer without proper xval initialization in set_tval_irq() resulted in an early interrupt during timer reprogramming. Make sure to set the xval before setting the enable bit.

---
 tools/testing/selftests/kvm/arm64/arch_timer_edge_cases.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/kvm/arm64/arch_timer_edge_cases.c +1 -1
index a813b4c6c817..be8bbedc933b 100644
@@ -191,8 +191,8 @@ static void set_tval_irq(enum arch_timer timer, uint64_t tval_cycles,
 {
 	atomic_set(&shared_data.handled, 0);
 	atomic_set(&shared_data.spurious, 0);
-	timer_set_ctl(timer, ctl);
 	timer_set_tval(timer, tval_cycles);
+	timer_set_ctl(timer, ctl);
 }
 
 static void set_xval_irq(enum arch_timer timer, uint64_t xval, uint32_t ctl,
Signature · 1 line
2.49.0
#4 lore raw

PATCH v3 4/4 KVM: arm64: selftests: arch_timer_edge_cases - determine effective counter width

In reply to Sebastian Ott

arch_timer_edge_cases uses ~0 as the maximum counter value, however there's no architectural guarantee that this is valid.

Figure out the effective counter width based on the effective frequency like it's done by the kernel.

This also serves as a workaround for AC03_CPU_14 that led to the following assertion failure on ampere-one machines:

==== Test Assertion Failure ====
arm64/arch_timer_edge_cases.c:169: timer_condition == istatus pid=11236 tid=11236 errno=4 - Interrupted system call

     1  0x0000000000404ce7: test_run at arch_timer_edge_cases.c:938
     2  0x0000000000401ebb: main at arch_timer_edge_cases.c:1053
     3  0x0000ffff9fa8625b: ?? ??:0
     4  0x0000ffff9fa8633b: ?? ??:0
     5  0x0000000000401fef: _start at ??:?

0x1 != 0x0 (timer_condition != istatus)

Note that the following subtest only worked since the counter initialized with CVAL_MAX would instantly overflow (which is no longer the case):

	test_set_cnt_after_cval_no_irq(timer, 0, DEF_CNT, CVAL_MAX, sm);

To fix this we could swap CVAL_MAX for 0 here but since that is already done by test_move_counters_behind_timers() let's remove that subtest.

---
 .../kvm/arm64/arch_timer_edge_cases.c         | 27 ++++++++++++-------
 1 file changed, 18 insertions(+), 9 deletions(-)
tools/testing/selftests/kvm/arm64/arch_timer_edge_cases.c +18 -9
index be8bbedc933b..b4d22b3ab7cc 100644
@@ -22,7 +22,8 @@
 #include "gic.h"
 #include "vgic.h"
 
-static const uint64_t CVAL_MAX = ~0ULL;
+/* Depends on counter width. */
+static uint64_t CVAL_MAX;
 /* tval is a signed 32-bit int. */
 static const int32_t TVAL_MAX = INT32_MAX;
 static const int32_t TVAL_MIN = INT32_MIN;
@@ -30,8 +31,8 @@ static const int32_t TVAL_MIN = INT32_MIN;
 /* After how much time we say there is no IRQ. */
 static const uint32_t TIMEOUT_NO_IRQ_US = 50000;
 
-/* A nice counter value to use as the starting one for most tests. */
-static const uint64_t DEF_CNT = (CVAL_MAX / 2);
+/* Counter value to use as the starting one for most tests. Set to CVAL_MAX/2 */
+static uint64_t DEF_CNT;
 
 /* Number of runs. */
 static const uint32_t NR_TEST_ITERS_DEF = 5;
@@ -732,12 +733,6 @@ static void test_move_counters_ahead_of_timers(enum arch_timer timer)
 		test_set_cnt_after_tval(timer, 0, tval, (uint64_t) tval + 1,
 					wm);
 	}
-
-	for (i = 0; i < ARRAY_SIZE(sleep_method); i++) {
-		sleep_method_t sm = sleep_method[i];
-
-		test_set_cnt_after_cval_no_irq(timer, 0, DEF_CNT, CVAL_MAX, sm);
-	}
 }
 
 /*
@@ -975,6 +970,8 @@ static void test_vm_create(struct kvm_vm **vm, struct kvm_vcpu **vcpu,
 	test_init_timer_irq(*vm, *vcpu);
 	vgic_v3_setup(*vm, 1, 64);
 	sync_global_to_guest(*vm, test_args);
+	sync_global_to_guest(*vm, CVAL_MAX);
+	sync_global_to_guest(*vm, DEF_CNT);
 }
 
 static void test_print_help(char *name)
@@ -1035,6 +1032,17 @@ static bool parse_args(int argc, char *argv[])
 	return false;
 }
 
+static void set_counter_defaults(void)
+{
+	const uint64_t MIN_ROLLOVER_SECS = 40ULL * 365 * 24 * 3600;
+	uint64_t freq = read_sysreg(CNTFRQ_EL0);
+	uint64_t width = ilog2(MIN_ROLLOVER_SECS * freq);
+
+	width = clamp(width, 56, 64);
+	CVAL_MAX = GENMASK_ULL(width - 1, 0);
+	DEF_CNT = CVAL_MAX / 2;
+}
+
 int main(int argc, char *argv[])
 {
 	struct kvm_vcpu *vcpu;
@@ -1047,6 +1055,7 @@ int main(int argc, char *argv[])
 		exit(KSFT_SKIP);
 
 	sched_getaffinity(0, sizeof(default_cpuset), &default_cpuset);
+	set_counter_defaults();
 
 	if (test_args.test_virtual) {
 		test_vm_create(&vm, &vcpu, VIRTUAL);
Signature · 1 line
2.49.0
maintainer Applied
#5 lore raw
In reply to Sebastian Ott
Quoted 9 lines · Sebastian Ott

Some small fixes for arch_timer_edge_cases that I stumbled upon while debugging failures for this selftest on ampere-one.

Changes since v1:
* determine effective counter width based on suggestions from Marc
Changes since v2:
* new patch to fix xval initialization

[...]

Applied to fixes, thanks!

[1/4] KVM: arm64: selftests: fix help text for arch_timer_edge_cases

      commit: 9a9864fd09c7e52440c05e70609bf5ee8da425d0
[2/4] KVM: arm64: selftests: fix thread migration in arch_timer_edge_cases
      commit: 050632ae6571d0f148fa0d4b90b6409a12645461
[3/4] KVM: arm64: selftests: arch_timer_edge_cases - fix xval init
      commit: 05ce38d489dbc96eb1dd700b287f949cb8cc0610
[4/4] KVM: arm64: selftests: arch_timer_edge_cases - determine effective counter width
      commit: fad4cf944839da7f5c3376243aa353295c88f588

Cheers,

	M.
Signature · 1 line
Without deviation from the norm, progress is not possible.
#6 lore raw
In reply to Sebastian Ott

Hi Sebastian,

Thanks in advance.

Quoted 5 lines · Sebastian Ott

Some small fixes for arch_timer_edge_cases that I stumbled upon while debugging failures for this selftest on ampere-one.

I’ve tested this test under three circumstances:

1) Original (!HAS_EL2 && !HAS_EL2_E2H0)

2) HAS_EL2 && HAS_EL2_E2H0

3) HAS_EL2 && !HAS_EL2_E2H0

Tests 1) and 2) returned in approx. the same amount of real time (about 12s) although 3) doesn’t seem to return at all.

I’ve tested it on ampere-one.

Miguel

Quoted 26 lines

Changes since v1:
* determine effective counter width based on suggestions from Marc Changes since v2:
* new patch to fix xval initialization

I've done tests with this on various machines - no issues during several hundreds of test runs.

v1: https://lore.kernel.org/kvmarm/[email protected]/
v2: https://lore.kernel.org/kvmarm/[email protected]/

Sebastian Ott (4):
KVM: arm64: selftests: fix help text for arch_timer_edge_cases
KVM: arm64: selftests: fix thread migration in arch_timer_edge_cases
KVM: arm64: selftests: arch_timer_edge_cases - fix xval init
KVM: arm64: selftests: arch_timer_edge_cases - determine effective counter width

.../kvm/arm64/arch_timer_edge_cases.c | 39 ++++++++++++-------

1 file changed, 25 insertions(+), 14 deletions(-)

base-commit: 0ff41df1cb268fc69e703a08a57ee14ae967d0ca

Signature · 1 line
2.49.0
#7 lore raw

Hej Miguel,

Quoted 8 lines · Miguel Luis
Quoted 5 lines · Sebastian Ott

Some small fixes for arch_timer_edge_cases that I stumbled upon while debugging failures for this selftest on ampere-one.

I’ve tested this test under three circumstances:

Quoted 8 lines

1) Original (!HAS_EL2 && !HAS_EL2_E2H0)

2) HAS_EL2 && HAS_EL2_E2H0

3) HAS_EL2 && !HAS_EL2_E2H0

Tests 1) and 2) returned in approx. the same amount of real time (about 12s) although 3) doesn’t seem to return at all.

I've never really looked at nested stuff, I'll try to set smth up and see if I can reproduce that..

Thanks,
Sebastian