aarch64: Try to free the GCS of makecontext

Free GCS after a makecontext start func returns and at thread exit, so
assume makecontext cannot outlive the thread where it was created.

This is an attempt to bound the lifetime of the GCS allocated for
makecontext, but it is still possible to have significant GCS leaks,
new GCS aware APIs could solve that, but that would not allow using
GCS with existing code transparently.

include/set-freeres.h

@@ -78,6 +78,10 @@ extern void __nss_database_freeres (void) attribute_hidden;
 extern int _IO_cleanup (void) attribute_hidden;;
 /* From dlfcn/dlerror.c */
 extern void __libc_dlerror_result_free (void) attribute_hidden;
+/* From libc.so, arch specific.  */
+#ifdef ARCH_THREAD_FREERES
+extern void ARCH_THREAD_FREERES (void) attribute_hidden;
+#endif
 
 /* From either libc.so or libpthread.so  */
 extern void __libpthread_freeres (void) attribute_hidden;

malloc/thread-freeres.c

@@ -29,6 +29,9 @@
 void
 __libc_thread_freeres (void)
 {
+#ifdef ARCH_THREAD_FREERES
+  call_function_static_weak (ARCH_THREAD_FREERES);
+#endif
 #if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_32)
   __rpc_thread_destroy ();
 #endif

sysdeps/unix/sysv/linux/aarch64/makecontext.c

@@ -20,7 +20,9 @@
 #include <sysdep.h>
 #include <stdarg.h>
 #include <stdint.h>
+#include <stdlib.h>
 #include <ucontext.h>
+#include <sys/mman.h>
 
 #define GCS_MAGIC 0x47435300
 
@@ -29,6 +31,47 @@ static struct _aarch64_ctx *extension (void *p)
   return p;
 }
 
+struct gcs_list {
+  struct gcs_list *next;
+  void *base;
+  size_t size;
+};
+
+static __thread struct gcs_list *gcs_list_head = NULL;
+
+static void
+record_gcs (void *base, size_t size)
+{
+  struct gcs_list *p = malloc (sizeof *p);
+  if (p == NULL)
+    abort ();
+  p->base = base;
+  p->size = size;
+  p->next = gcs_list_head;
+  gcs_list_head = p;
+}
+
+static void
+free_gcs_list (void)
+{
+  for (;;)
+    {
+      struct gcs_list *p = gcs_list_head;
+      if (p == NULL)
+	break;
+      gcs_list_head = p->next;
+      __munmap (p->base, p->size);
+      free (p);
+    }
+}
+
+/* Called during thread shutdown to free resources.  */
+void
+__libc_aarch64_thread_freeres (void)
+{
+  free_gcs_list ();
+}
+
 #ifndef __NR_map_shadow_stack
 # define __NR_map_shadow_stack 453
 #endif
@@ -58,6 +101,9 @@ alloc_makecontext_gcs (size_t stack_size)
   if (base == (void *) -1)
     /* ENOSYS, bad size or OOM.  */
     abort ();
+
+  record_gcs (base, size);
+
   uint64_t *gcsp = (uint64_t *) ((char *) base + size);
   /* Skip end of GCS token.  */
   gcsp--;
@@ -69,6 +115,25 @@ alloc_makecontext_gcs (size_t stack_size)
   return gcsp + 1;
 }
 
+void
+__free_makecontext_gcs (void *gcs)
+{
+  struct gcs_list *p = gcs_list_head;
+  struct gcs_list **q = &gcs_list_head;
+  for (;;)
+    {
+      if (p == NULL)
+	abort ();
+      if (gcs == p->base + p->size - 8)
+	break;
+      q = &p->next;
+      p = p->next;
+    }
+  *q = p->next;
+  __munmap (p->base, p->size);
+  free (p);
+}
+
 /* makecontext sets up a stack and the registers for the
    user context.  The stack looks like this:
 

sysdeps/unix/sysv/linux/aarch64/setcontext.S

@@ -34,6 +34,9 @@
 	.text
 
 ENTRY (__setcontext)
+	/* If x10 is set then old GCS is freed.  */
+	mov	x10, 0
+__setcontext_internal:
 	PTR_ARG (0)
 	/* Save a copy of UCP.  */
 	mov	x9, x0
@@ -145,7 +148,8 @@ ENTRY (__setcontext)
 	ldr	x3, [x2, #oGCSPR]
 	MRS_GCSPR (x2)
 	mov	x4, x3
-	/* x2: GCSPR now.  x3, x4: target GCSPR.  x5, x6: tmp regs.  */
+	mov	x1, x2
+	/* x1, x2: GCSPR now.  x3, x4: target GCSPR.  x5, x6: tmp regs.  */
 L(gcs_scan):
 	cmp	x2, x4
 	b.eq	L(gcs_pop)
@@ -162,10 +166,18 @@ L(gcs_switch):
 	GCSSS2 (xzr)
 L(gcs_pop):
 	cmp	x2, x3
-	b.eq	L(gcs_done)
+	b.eq	L(gcs_free_old)
 	GCSPOPM (xzr)
 	add	x2, x2, 8
 	b	L(gcs_pop)
+L(gcs_free_old):
+	cbz	x10, L(gcs_done)
+	mov	x28, x0
+	mov	x0, x1
+	bl	__free_makecontext_gcs
+	mov	x0, x28
+	ldp	x28, x29, [x0, oX0 + 28 * SZREG]
+	ldr     x30,      [x0, oX0 + 30 * SZREG]
 L(gcs_done):
 
 2:
@@ -186,6 +198,7 @@ ENTRY (__startcontext)
 	cfi_undefined (x30)
 	blr	x20
 	mov	x0, x19
-	cbnz	x0, __setcontext
+	mov	x10, 1
+	cbnz	x0, __setcontext_internal
 1:	b       HIDDEN_JUMPTARGET (exit)
 END (__startcontext)

sysdeps/unix/sysv/linux/aarch64/sysdep.h

@@ -29,8 +29,12 @@
 
 #include <tls.h>
 
-/* In order to get __set_errno() definition in INLINE_SYSCALL.  */
 #ifndef __ASSEMBLER__
+/* Thread cleanup function.  */
+#define ARCH_THREAD_FREERES __libc_aarch64_thread_freeres
+void __libc_aarch64_thread_freeres (void) attribute_hidden;
+
+/* In order to get __set_errno() definition in INLINE_SYSCALL.  */
 #include <errno.h>
 #endif