Pass down "stack start" variables from closer to the top of the stack

The implementation of `native_thread_init_stack` for the various
threading models can use the address of a local variable as part of the
calculation of the machine stack extents:

* pthreads uses it as a lower-bound on the start of the stack, because
  glibc (and maybe other libcs) can store its own data on the stack
  before calling into user code on thread creation.
* win32 uses it as an argument to VirtualQuery, which gets the extent of
  the memory mapping which contains the variable

However, the local being used for this is actually allocated _inside_
the `native_thread_init_stack` frame; that means the caller might
allocate a VALUE on the stack that actually lies outside the bounds
stored in machine.stack_{start,end}.

A local variable from one level above the topmost frame that stores
VALUEs on the stack must be drilled down into the call to
`native_thread_init_stack` to be used in the calculation. This probably
doesn't _really_ matter for the win32 case (they'll be in the same
memory mapping so VirtualQuery should return the same thing), but
definitely could matter for the pthreads case.

[Bug #20001]

1 files changed, 16 insertions, 5 deletions

diff --git a/thread_pthread.c b/thread_pthread.c
index 6d2f55a957..a6a6c9d127 100644
--- a/thread_pthread.c
+++ b/thread_pthread.c
@@ -1964,7 +1964,7 @@ reserve_stack(volatile char *limit, size_t size)
 
 #undef ruby_init_stack
 void
-ruby_init_stack(volatile VALUE *addr)
+ruby_init_stack(volatile void *addr)
 {
     native_main_thread.id = pthread_self();
 
@@ -2049,7 +2049,7 @@ ruby_init_stack(volatile VALUE *addr)
     {int err = (expr); if (err) {rb_bug_errno(#expr, err);}}
 
 static int
-native_thread_init_stack(rb_thread_t *th)
+native_thread_init_stack(rb_thread_t *th, void *local_in_parent_frame)
 {
     rb_nativethread_id_t curr = pthread_self();
 
@@ -2064,8 +2064,8 @@ native_thread_init_stack(rb_thread_t *th)
             size_t size;
 
             if (get_stack(&start, &size) == 0) {
-                uintptr_t diff = (uintptr_t)start - (uintptr_t)&curr;
-                th->ec->machine.stack_start = (VALUE *)&curr;
+                uintptr_t diff = (uintptr_t)start - (uintptr_t)local_in_parent_frame;
+                th->ec->machine.stack_start = (uintptr_t)local_in_parent_frame;
                 th->ec->machine.stack_maxsize = size - diff;
             }
         }
@@ -2185,8 +2185,19 @@ native_thread_create_dedicated(rb_thread_t *th)
 static void
 call_thread_start_func_2(rb_thread_t *th)
 {
-    native_thread_init_stack(th);
+    /* Capture the address of a local in this stack frame to mark the beginning of the
+       machine stack for this thread. This is required even if we can tell the real
+       stack beginning from the pthread API in native_thread_init_stack, because
+       glibc stores some of its own data on the stack before calling into user code
+       on a new thread, and replacing that data on fiber-switch would break it (see
+       bug #13887) */
+    VALUE stack_start = 0;
+    VALUE *stack_start_addr = &stack_start;
+    native_thread_init_stack(th, stack_start_addr);
     thread_start_func_2(th, th->ec->machine.stack_start);
+
+    /* Ensure that stack_start really was spilled to the stack */
+    RB_GC_GUARD(stack_start)
 }
 
 static void *