KVM: selftests: Fix target thread to be migrated in rseq_test
authorGavin Shan <gshan@redhat.com>
Tue, 19 Jul 2022 02:08:30 +0000 (10:08 +0800)
committerPaolo Bonzini <pbonzini@redhat.com>
Tue, 19 Jul 2022 13:03:49 +0000 (09:03 -0400)
In rseq_test, there are two threads, which are vCPU thread and migration
worker separately. Unfortunately, the test has the wrong PID passed to
sched_setaffinity() in the migration worker. It forces migration on the
migration worker because zeroed PID represents the calling thread, which
is the migration worker itself. It means the vCPU thread is never enforced
to migration and it can migrate at any time, which eventually leads to
failure as the following logs show.

  host# uname -r
  5.19.0-rc6-gavin+
  host# # cat /proc/cpuinfo | grep processor | tail -n 1
  processor    : 223
  host# pwd
  /home/gavin/sandbox/linux.main/tools/testing/selftests/kvm
  host# for i in `seq 1 100`; do \
        echo "--------> $i"; ./rseq_test; done
  --------> 1
  --------> 2
  --------> 3
  --------> 4
  --------> 5
  --------> 6
  ==== Test Assertion Failure ====
    rseq_test.c:265: rseq_cpu == cpu
    pid=3925 tid=3925 errno=4 - Interrupted system call
       1  0x0000000000401963: main at rseq_test.c:265 (discriminator 2)
       2  0x0000ffffb044affb: ?? ??:0
       3  0x0000ffffb044b0c7: ?? ??:0
       4  0x0000000000401a6f: _start at ??:?
    rseq CPU = 4, sched CPU = 27

Fix the issue by passing correct parameter, TID of the vCPU thread, to
sched_setaffinity() in the migration worker.

Fixes: 61e52f1630f5 ("KVM: selftests: Add a test for KVM_RUN+rseq to detect task migration bugs")
Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Gavin Shan <gshan@redhat.com>
Reviewed-by: Oliver Upton <oliver.upton@linux.dev>
Message-Id: <20220719020830.3479482-1-gshan@redhat.com>
Reviewed-by: Andrew Jones <andrew.jones@linux.dev>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
tools/testing/selftests/kvm/rseq_test.c

index 4158da0..2237d1a 100644 (file)
@@ -82,8 +82,9 @@ static int next_cpu(int cpu)
        return cpu;
 }
 
-static void *migration_worker(void *ign)
+static void *migration_worker(void *__rseq_tid)
 {
+       pid_t rseq_tid = (pid_t)(unsigned long)__rseq_tid;
        cpu_set_t allowed_mask;
        int r, i, cpu;
 
@@ -106,7 +107,7 @@ static void *migration_worker(void *ign)
                 * stable, i.e. while changing affinity is in-progress.
                 */
                smp_wmb();
-               r = sched_setaffinity(0, sizeof(allowed_mask), &allowed_mask);
+               r = sched_setaffinity(rseq_tid, sizeof(allowed_mask), &allowed_mask);
                TEST_ASSERT(!r, "sched_setaffinity failed, errno = %d (%s)",
                            errno, strerror(errno));
                smp_wmb();
@@ -231,7 +232,8 @@ int main(int argc, char *argv[])
        vm = vm_create_default(VCPU_ID, 0, guest_code);
        ucall_init(vm, NULL);
 
-       pthread_create(&migration_thread, NULL, migration_worker, 0);
+       pthread_create(&migration_thread, NULL, migration_worker,
+                      (void *)(unsigned long)gettid());
 
        for (i = 0; !done; i++) {
                vcpu_run(vm, VCPU_ID);