diff --git a/src/rt/arch/i386/context.h b/src/rt/arch/i386/context.h
index 71cd7aeada2..ae65cdb5813 100644
--- a/src/rt/arch/i386/context.h
+++ b/src/rt/arch/i386/context.h
@@ -29,8 +29,6 @@ struct registers_t {
   uint32_t eip;
 } __attribute__((aligned(16)));
 
-extern "C" void __morestack(void *args, void *fn_ptr, uintptr_t stack_ptr);
-
 class context {
 public:
   registers_t regs;
@@ -41,10 +39,6 @@ public:
 
   void swap(context &out);
   void call(void *f, void *arg, void *sp);
-
-  void call_and_change_stacks(void *args, void *fn_ptr) {
-      __morestack(args, fn_ptr, regs.esp);
-  }
 };
 
 #endif
diff --git a/src/rt/arch/x86_64/context.h b/src/rt/arch/x86_64/context.h
index 2b9b5d4c6c7..412d968599f 100644
--- a/src/rt/arch/x86_64/context.h
+++ b/src/rt/arch/x86_64/context.h
@@ -28,8 +28,6 @@ struct registers_t {
     uint64_t data[RUSTRT_MAX];
 } __attribute__((aligned(16)));
 
-extern "C" void __morestack(void *args, void *fn_ptr, uintptr_t stack_ptr);
-
 class context {
 public:
     registers_t regs;
@@ -40,10 +38,6 @@ public:
     
     void swap(context &out);
     void call(void *f, void *arg, void *sp);
-
-    void call_and_change_stacks(void *args, void *fn_ptr) {
-        __morestack(args, fn_ptr, regs.data[RUSTRT_RSP]);
-    }
 };
 
 #endif
diff --git a/src/rt/rust_task.cpp b/src/rt/rust_task.cpp
index 912225fba3f..a025800e4ae 100644
--- a/src/rt/rust_task.cpp
+++ b/src/rt/rust_task.cpp
@@ -88,7 +88,9 @@ rust_task::rust_task(rust_task_thread *thread, rust_task_list *state,
     propagate_failure(true),
     dynastack(this),
     cc_counter(0),
-    total_stack_sz(0)
+    total_stack_sz(0),
+    c_stack(NULL),
+    next_c_sp(0)
 {
     LOGPTR(thread, "new task", (uintptr_t)this);
     DLOG(thread, task, "sizeof(task) = %d (0x%x)", sizeof *this, sizeof *this);
@@ -166,7 +168,6 @@ cleanup_task(cleanup_args *args) {
 }
 
 // This runs on the Rust stack
-extern "C" CDECL
 void task_start_wrapper(spawn_args *a)
 {
     rust_task *task = a->task;
@@ -180,8 +181,15 @@ void task_start_wrapper(spawn_args *a)
         A(task->thread, ex == task,
           "Expected this task to be thrown for unwinding");
         threw_exception = true;
+
+        if (task->c_stack) {
+            task->return_c_stack();
+        }
     }
 
+    // We should have returned any C stack by now
+    I(task->thread, task->c_stack == NULL);
+
     rust_opaque_box* env = a->envptr;
     if(env) {
         // free the environment (which should be a unique closure).
@@ -722,10 +730,35 @@ rust_task::config_notify(chan_handle chan) {
     notify_chan = chan;
 }
 
+extern "C" void __morestack(void *args, void *fn_ptr, uintptr_t stack_ptr);
+
 void
 rust_task::call_on_c_stack(void *args, void *fn_ptr) {
     I(thread, on_rust_stack());
-    thread->c_context.call_and_change_stacks(args, fn_ptr);
+
+    bool borrowed_a_c_stack = false;
+    if (c_stack == NULL) {
+        c_stack = thread->borrow_c_stack();
+        next_c_sp = align_down(c_stack->end);
+        borrowed_a_c_stack = true;
+    }
+
+    __morestack(args, fn_ptr, next_c_sp);
+
+    // Note that we may not actually get here if we threw an exception,
+    // in which case we will return the c stack when the exception is caught.
+    if (borrowed_a_c_stack) {
+        return_c_stack();
+    }
+}
+
+void
+rust_task::return_c_stack() {
+    I(thread, on_rust_stack());
+    I(thread, c_stack != NULL);
+    thread->return_c_stack(c_stack);
+    c_stack = NULL;
+    next_c_sp = 0;
 }
 
 //
diff --git a/src/rt/rust_task.h b/src/rt/rust_task.h
index c4939d208c1..5ac2d7a212e 100644
--- a/src/rt/rust_task.h
+++ b/src/rt/rust_task.h
@@ -36,6 +36,8 @@ typedef unsigned long task_result;
 #define tr_success 0
 #define tr_failure 1
 
+struct spawn_args;
+
 // std::lib::task::task_notification
 //
 // since it's currently a unary tag, we only add the fields.
@@ -104,6 +106,11 @@ rust_task : public kernel_owned<rust_task>, rust_cond
     size_t total_stack_sz;
 
 private:
+
+    // The stack used for running C code, borrowed from the scheduler thread
+    stk_seg *c_stack;
+    uintptr_t next_c_sp;
+
     // Called when the atomic refcount reaches zero
     void delete_this();
 
@@ -112,6 +119,10 @@ private:
     void free_stack(stk_seg *stk);
     size_t get_next_stack_size(size_t min, size_t current, size_t requested);
 
+    void return_c_stack();
+
+    friend void task_start_wrapper(spawn_args *a);
+
 public:
 
     // Only a pointer to 'name' is kept, so it must live as long as this task.
diff --git a/src/rt/rust_task_thread.cpp b/src/rt/rust_task_thread.cpp
index 4129969057c..48bd7f16bdf 100644
--- a/src/rt/rust_task_thread.cpp
+++ b/src/rt/rust_task_thread.cpp
@@ -61,10 +61,6 @@ rust_task_thread::~rust_task_thread() {
 #ifndef __WIN32__
     pthread_attr_destroy(&attr);
 #endif
-
-    if (cached_c_stack) {
-        destroy_stack(kernel, cached_c_stack);
-    }
 }
 
 void
@@ -72,7 +68,9 @@ rust_task_thread::activate(rust_task *task) {
     task->ctx.next = &c_context;
     DLOG(this, task, "descheduling...");
     lock.unlock();
+    prepare_c_stack();
     task->ctx.swap(c_context);
+    unprepare_c_stack();
     lock.lock();
     DLOG(this, task, "task has returned");
 }
@@ -287,6 +285,13 @@ rust_task_thread::start_main_loop() {
     DLOG(this, dom, "finished main-loop %d", id);
 
     lock.unlock();
+
+    I(this, !extra_c_stack);
+    if (cached_c_stack) {
+        unconfig_valgrind_stack(cached_c_stack);
+        destroy_stack(kernel, cached_c_stack);
+        cached_c_stack = NULL;
+    }
 }
 
 rust_crate_cache *
@@ -374,24 +379,51 @@ rust_task_thread::exit() {
     lock.signal();
 }
 
-stk_seg *
-rust_task_thread::borrow_c_stack() {
-
-    if (cached_c_stack) {
-        stk_seg *your_stack = cached_c_stack;
-        cached_c_stack = NULL;
-        return your_stack;
-    } else {
-        return create_stack(kernel, C_STACK_SIZE);
+// Before activating each task, make sure we have a C stack available.
+// It needs to be allocated ahead of time (while we're on our own
+// stack), because once we're on the Rust stack we won't have enough
+// room to do the allocation
+void
+rust_task_thread::prepare_c_stack() {
+    I(this, !extra_c_stack);
+    if (!cached_c_stack) {
+        cached_c_stack = create_stack(kernel, C_STACK_SIZE);
     }
 }
 
 void
-rust_task_thread::return_c_stack(stk_seg *stack) {
-    if (cached_c_stack) {
-        destroy_stack(kernel, stack);
+rust_task_thread::unprepare_c_stack() {
+    if (extra_c_stack) {
+        destroy_stack(kernel, extra_c_stack);
+        extra_c_stack = NULL;
+    }
+}
+
+// NB: Runs on the Rust stack
+stk_seg *
+rust_task_thread::borrow_c_stack() {
+    I(this, cached_c_stack);
+    stk_seg *your_stack;
+    if (extra_c_stack) {
+        your_stack = extra_c_stack;
+        extra_c_stack = NULL;
     } else {
+        your_stack = cached_c_stack;
+        cached_c_stack = NULL;
+    }
+    config_valgrind_stack(your_stack);
+    return your_stack;
+}
+
+// NB: Runs on the Rust stack
+void
+rust_task_thread::return_c_stack(stk_seg *stack) {
+    I(this, !extra_c_stack);
+    unconfig_valgrind_stack(stack);
+    if (!cached_c_stack) {
         cached_c_stack = stack;
+    } else {
+        extra_c_stack = stack;
     }
 }
 
diff --git a/src/rt/rust_task_thread.h b/src/rt/rust_task_thread.h
index 360e4e44e0b..82609bc887a 100644
--- a/src/rt/rust_task_thread.h
+++ b/src/rt/rust_task_thread.h
@@ -95,6 +95,10 @@ struct rust_task_thread : public kernel_owned<rust_task_thread>,
 private:
 
     stk_seg *cached_c_stack;
+    stk_seg *extra_c_stack;
+
+    void prepare_c_stack();
+    void unprepare_c_stack();
 
 public: