am 2be511d4: Merge "Improve stack overflow diagnostics (take 2)."

* commit '2be511d405d47eccc61a6e3c338d1877bf33b4fa':
  Improve stack overflow diagnostics (take 2).

libc/bionic/pthread.c

@@ -31,6 +31,7 @@
 #include <errno.h>
 #include <limits.h>
 #include <sys/atomics.h>
+#include <sys/mman.h>
 #include <unistd.h>
 
 #include "bionic_atomic_inline.h"
@@ -102,6 +103,18 @@ void pthread_exit(void * retval)
     // space (see pthread_key_delete)
     pthread_key_clean_all();
 
+    if (thread->alternate_signal_stack != NULL) {
+      // Tell the kernel to stop using the alternate signal stack.
+      stack_t ss;
+      ss.ss_sp = NULL;
+      ss.ss_flags = SS_DISABLE;
+      sigaltstack(&ss, NULL);
+
+      // Free it.
+      munmap(thread->alternate_signal_stack, SIGSTKSZ);
+      thread->alternate_signal_stack = NULL;
+    }
+
     // if the thread is detached, destroy the pthread_internal_t
     // otherwise, keep it in memory and signal any joiners.
     pthread_mutex_lock(&gThreadListLock);

libc/bionic/pthread_attr.cpp

@@ -30,12 +30,16 @@
 
 #include "pthread_internal.h"
 
-#define DEFAULT_STACK_SIZE (1024 * 1024)
+// Traditionally we give threads a 1MiB stack. When we started allocating per-thread
+// alternate signal stacks to ease debugging of stack overflows, we subtracted the
+// same amount we were using there from the default thread stack size. This should
+// keep memory usage roughly constant.
+#define DEFAULT_THREAD_STACK_SIZE ((1 * 1024 * 1024) - SIGSTKSZ)
 
 int pthread_attr_init(pthread_attr_t* attr) {
   attr->flags = 0;
   attr->stack_base = NULL;
-  attr->stack_size = DEFAULT_STACK_SIZE;
+  attr->stack_size = DEFAULT_THREAD_STACK_SIZE;
   attr->guard_size = PAGE_SIZE;
   attr->sched_policy = SCHED_NORMAL;
   attr->sched_priority = 0;

libc/bionic/pthread_create.cpp

@@ -69,9 +69,22 @@ void  __init_tls(pthread_internal_t* thread) {
   thread->tls[TLS_SLOT_STACK_GUARD] = (void*) __stack_chk_guard;
 
   __set_tls(thread->tls);
+
+  // Create and set an alternate signal stack.
+  // This must happen after __set_tls, in case a system call fails and tries to set errno.
+  stack_t ss;
+  ss.ss_sp = mmap(NULL, SIGSTKSZ, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
+  if (ss.ss_sp != MAP_FAILED) {
+    ss.ss_size = SIGSTKSZ;
+    ss.ss_flags = 0;
+    sigaltstack(&ss, NULL);
+    thread->alternate_signal_stack = ss.ss_sp;
+  }
 }
 
-// This trampoline is called from the assembly _pthread_clone() function.
+// This trampoline is called from the assembly _pthread_clone function.
+// Our 'tls' and __pthread_clone's 'child_stack' are one and the same, just growing in
+// opposite directions.
 extern "C" void __thread_entry(void* (*func)(void*), void* arg, void** tls) {
   // Wait for our creating thread to release us. This lets it have time to
   // notify gdb about this thread before we start doing anything.
@@ -187,8 +200,12 @@ int pthread_create(pthread_t* thread_out, pthread_attr_t const* attr,
     thread->attr.flags |= PTHREAD_ATTR_FLAG_USER_STACK;
   }
 
-  // Make room for TLS.
+  // Make room for the TLS area.
+  // The child stack is the same address, just growing in the opposite direction.
+  // At offsets >= 0, we have the TLS slots.
+  // At offsets < 0, we have the child stack.
   void** tls = (void**)((uint8_t*)(thread->attr.stack_base) + thread->attr.stack_size - BIONIC_TLS_SLOTS * sizeof(void*));
+  void* child_stack = tls;
 
   // Create a mutex for the thread in TLS_SLOT_SELF to wait on once it starts so we can keep
   // it from doing anything until after we notify the debugger about it
@@ -204,7 +221,7 @@ int pthread_create(pthread_t* thread_out, pthread_attr_t const* attr,
 
   int flags = CLONE_FILES | CLONE_FS | CLONE_VM | CLONE_SIGHAND | CLONE_THREAD | CLONE_SYSVSEM;
 
-  int tid = __pthread_clone(start_routine, tls, flags, arg);
+  int tid = __pthread_clone(start_routine, child_stack, flags, arg);
   if (tid < 0) {
     int clone_errno = errno;
     if ((thread->attr.flags & PTHREAD_ATTR_FLAG_USER_STACK) == 0) {

libc/bionic/pthread_internal.h

@@ -47,6 +47,8 @@ typedef struct pthread_internal_t
     __pthread_cleanup_t*        cleanup_stack;
     void**                      tls;         /* thread-local storage area */
 
+    void* alternate_signal_stack;
+
     /*
      * The dynamic linker implements dlerror(3), which makes it hard for us to implement this
      * per-thread buffer by simply using malloc(3) and free(3).