diff --git a/src/eval.rs b/src/eval.rs
index ab82c39836b2..c5a04d75858c 100644
--- a/src/eval.rs
+++ b/src/eval.rs
@@ -222,7 +222,8 @@ pub fn eval_main<'tcx>(tcx: TyCtxt<'tcx>, main_id: DefId, config: MiriConfig) ->
         // Read the return code pointer *before* we run TLS destructors, to assert
         // that it was written to by the time that `start` lang item returned.
         let return_code = ecx.read_scalar(ret_place.into())?.not_undef()?.to_machine_isize(&ecx)?;
-        // Global destructors.
+        // Run Windows destructors. (We do not support concurrency on Windows
+        // yet, so we run the destructor of the main thread separately.)
         ecx.run_windows_tls_dtors()?;
         Ok(return_code)
     })();
diff --git a/src/shims/tls.rs b/src/shims/tls.rs
index c08ec78c136d..31a9ee3c9425 100644
--- a/src/shims/tls.rs
+++ b/src/shims/tls.rs
@@ -6,7 +6,6 @@ use std::collections::HashSet;
 
 use log::trace;
 
-use rustc_index::vec::Idx;
 use rustc_middle::ty;
 use rustc_target::abi::{Size, HasDataLayout};
 
@@ -201,7 +200,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
             return Ok(());
         }
         let active_thread = this.get_active_thread()?;
-        assert_eq!(active_thread.index(), 0, "concurrency on Windows not supported");
+        assert_eq!(this.get_total_thread_count()?, 1, "concurrency on Windows not supported");
         assert!(!this.machine.tls.dtors_running.contains(&active_thread), "running TLS dtors twice");
         this.machine.tls.dtors_running.insert(active_thread);
         // Windows has a special magic linker section that is run on certain events.
diff --git a/src/thread.rs b/src/thread.rs
index 657792bd2c67..8c353d6a8853 100644
--- a/src/thread.rs
+++ b/src/thread.rs
@@ -227,6 +227,11 @@ impl<'mir, 'tcx: 'mir> ThreadManager<'mir, 'tcx> {
         self.active_thread
     }
 
+    /// Get the total number of threads that were ever spawn by this program.
+    fn get_total_thread_count(&self) -> usize {
+        self.threads.len()
+    }
+
     /// Has the given thread terminated?
     fn has_terminated(&self, thread_id: ThreadId) -> bool {
         self.threads[thread_id].state == ThreadState::Terminated
@@ -492,6 +497,12 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         Ok(this.machine.threads.get_active_thread_id())
     }
 
+    #[inline]
+    fn get_total_thread_count(&self) -> InterpResult<'tcx, usize> {
+        let this = self.eval_context_ref();
+        Ok(this.machine.threads.get_total_thread_count())
+    }
+
     #[inline]
     fn has_terminated(&self, thread_id: ThreadId) -> InterpResult<'tcx, bool> {
         let this = self.eval_context_ref();
diff --git a/tests/compile-fail/concurrency/dangling_tls_lib.rs b/tests/compile-fail/concurrency/dangling_tls_lib.rs
index 684dd0e86f60..6be5538bb444 100644
--- a/tests/compile-fail/concurrency/dangling_tls_lib.rs
+++ b/tests/compile-fail/concurrency/dangling_tls_lib.rs
@@ -1,5 +1,8 @@
 // ignore-windows: Concurrency on Windows is not supported yet.
 
+//! Check that we catch if a thread local is accessed after the thread has
+//! terminated.
+
 #![feature(thread_local_internals)]
 
 use std::cell::RefCell;
diff --git a/tests/run-pass/concurrency/tls_lib_drop.rs b/tests/run-pass/concurrency/tls_lib_drop.rs
new file mode 100644
index 000000000000..c9b04a728258
--- /dev/null
+++ b/tests/run-pass/concurrency/tls_lib_drop.rs
@@ -0,0 +1,46 @@
+// ignore-windows: Concurrency on Windows is not supported yet.
+
+//! Check that destructors of the library thread locals are executed immediately
+//! after a thread terminates.
+
+#![feature(thread_local_internals)]
+
+use std::cell::RefCell;
+use std::thread;
+
+struct TestCell {
+    value: RefCell<u8>,
+}
+
+impl Drop for TestCell {
+    fn drop(&mut self) {
+        println!("Dropping: {}", self.value.borrow())
+    }
+}
+
+static A: std::thread::LocalKey<TestCell> = {
+    #[inline]
+    fn __init() -> TestCell {
+        TestCell { value: RefCell::new(0) }
+    }
+
+    unsafe fn __getit() -> Option<&'static TestCell> {
+        static __KEY: std::thread::__OsLocalKeyInner<TestCell> =
+            std::thread::__OsLocalKeyInner::new();
+        __KEY.get(__init)
+    }
+
+    unsafe { std::thread::LocalKey::new(__getit) }
+};
+
+fn main() {
+    thread::spawn(|| {
+        A.with(|f| {
+            assert_eq!(*f.value.borrow(), 0);
+            *f.value.borrow_mut() = 5;
+        });
+    })
+    .join()
+    .unwrap();
+    println!("Continue main.")
+}
diff --git a/tests/run-pass/concurrency/tls_lib_drop.stderr b/tests/run-pass/concurrency/tls_lib_drop.stderr
new file mode 100644
index 000000000000..2dbfb7721d36
--- /dev/null
+++ b/tests/run-pass/concurrency/tls_lib_drop.stderr
@@ -0,0 +1,2 @@
+warning: thread support is experimental. For example, Miri does not detect data races yet.
+
diff --git a/tests/run-pass/concurrency/tls_lib_drop.stdout b/tests/run-pass/concurrency/tls_lib_drop.stdout
new file mode 100644
index 000000000000..d2bbb866b77e
--- /dev/null
+++ b/tests/run-pass/concurrency/tls_lib_drop.stdout
@@ -0,0 +1,2 @@
+Dropping: 5
+Continue main.