[02/15] coroutine-ucontext: add a switch to monitor maximum stack size

diff --git a/configure b/configure
index 5929aba..82bcc25 100755
--- a/configure
+++ b/configure
@@ -296,6 +296,7 @@  libiscsi=""
 libnfs=""
 coroutine=""
 coroutine_pool=""
+coroutine_stack_size_debug="no"
 seccomp=""
 glusterfs=""
 glusterfs_xlator_opt="no"
@@ -1004,6 +1005,8 @@  for opt do
   ;;
   --enable-coroutine-pool) coroutine_pool="yes"
   ;;
+  --enable-coroutine-stack-size-debug) coroutine_stack_size_debug="yes"
+  ;;
   --disable-docs) docs="no"
   ;;
   --enable-docs) docs="yes"
@@ -1361,6 +1364,8 @@  disabled with --disable-FEATURE, default is enabled if available:
                   (for reading bzip2-compressed dmg images)
   seccomp         seccomp support
   coroutine-pool  coroutine freelist (better performance)
+  coroutine-stack-size-debug
+                  report coroutine max stack usage (only for debugging)
   glusterfs       GlusterFS backend
   archipelago     Archipelago backend
   tpm             TPM support
@@ -4298,6 +4303,15 @@  fi
 if test "$coroutine" = "gthread" -a "$coroutine_pool" = "yes"; then
   error_exit "'gthread' coroutine backend does not support pool (use --disable-coroutine-pool)"
 fi
+if test "$coroutine_stack_size_debug" = "yes"; then
+  if test "$coroutine" != "ucontext"; then
+    error_exit "coroutine stack size debugging currently only works with ucontext"
+  fi
+  if test "$coroutine_pool" = "yes"; then
+    echo "WARN: disabling coroutine pool for stack size debugging"
+    coroutine_pool=no
+  fi
+fi
 
 ##########################################
 # check if we have open_by_handle_at
@@ -4866,6 +4880,7 @@  echo "QGA MSI support   $guest_agent_msi"
 echo "seccomp support   $seccomp"
 echo "coroutine backend $coroutine"
 echo "coroutine pool    $coroutine_pool"
+echo "coroutine stack size debug $coroutine_stack_size_debug"
 echo "GlusterFS support $glusterfs"
 echo "Archipelago support $archipelago"
 echo "gcov              $gcov_tool"
@@ -5335,6 +5350,9 @@  if test "$coroutine_pool" = "yes" ; then
 else
   echo "CONFIG_COROUTINE_POOL=0" >> $config_host_mak
 fi
+if test "$coroutine_stack_size_debug" = "yes" ; then
+  echo "CONFIG_COROUTINE_STACK_SIZE_DEBUG=y" >> $config_host_mak
+fi
 
 if test "$open_by_handle_at" = "yes" ; then
   echo "CONFIG_OPEN_BY_HANDLE=y" >> $config_host_mak
diff --git a/util/coroutine-ucontext.c b/util/coroutine-ucontext.c
index 841e7db..27c61f3 100644
--- a/util/coroutine-ucontext.c
+++ b/util/coroutine-ucontext.c
@@ -31,6 +31,10 @@ 
 #include <valgrind/valgrind.h>
 #endif
 
+#ifdef CONFIG_COROUTINE_STACK_SIZE_DEBUG
+#include "qemu/error-report.h"
+#endif
+
 typedef struct {
     Coroutine base;
     void *stack;
@@ -48,6 +52,10 @@  typedef struct {
 static __thread CoroutineUContext leader;
 static __thread Coroutine *current;
 
+#ifdef CONFIG_COROUTINE_STACK_SIZE_DEBUG
+static uint32_t max_stack_usage;
+#endif
+
 /*
  * va_args to makecontext() must be type 'int', so passing
  * the pointer we need may require several int args. This
@@ -88,6 +96,9 @@  Coroutine *qemu_coroutine_new(void)
     ucontext_t old_uc, uc;
     sigjmp_buf old_env;
     union cc_arg arg = {0};
+#ifdef CONFIG_COROUTINE_STACK_SIZE_DEBUG
+    void *ptr;
+#endif
 
     /* The ucontext functions preserve signal masks which incurs a
      * system call overhead.  sigsetjmp(buf, 0)/siglongjmp() does not
@@ -118,6 +129,13 @@  Coroutine *qemu_coroutine_new(void)
     co->stack = g_malloc(stack_size);
 #endif
 
+#ifdef CONFIG_COROUTINE_STACK_SIZE_DEBUG
+    for (ptr = co->stack + getpagesize();
+         ptr < co->stack + COROUTINE_STACK_SIZE; ptr += sizeof(u_int32_t)) {
+        *(u_int32_t *)ptr = 0xdeadbeaf;
+    }
+#endif
+
     co->base.entry_arg = &old_env; /* stash away our jmp_buf */
 
     uc.uc_link = &old_uc;
@@ -161,6 +179,20 @@  void qemu_coroutine_delete(Coroutine *co_)
 {
     CoroutineUContext *co = DO_UPCAST(CoroutineUContext, base, co_);
 
+#ifdef CONFIG_COROUTINE_STACK_SIZE_DEBUG
+    void *ptr;
+    for (ptr = co->stack + getpagesize();
+         ptr < co->stack + COROUTINE_STACK_SIZE; ptr += sizeof(u_int32_t)) {
+        if (*(u_int32_t *)ptr != 0xdeadbeaf) {
+            break;
+        }
+    }
+    /* we only want to estimate the max stack usage, the OR will overestimate
+     * the stack usage, but this is ok here and avoids the usage of a mutex */
+    atomic_or(&max_stack_usage,
+              COROUTINE_STACK_SIZE - (uintptr_t) (ptr - co->stack));
+#endif
+
 #ifdef CONFIG_VALGRIND_H
     valgrind_stack_deregister(co);
 #endif
@@ -210,3 +242,11 @@  bool qemu_in_coroutine(void)
 {
     return current && current->caller;
 }
+
+#ifdef CONFIG_COROUTINE_STACK_SIZE_DEBUG
+static void __attribute__((destructor)) print_max_stack_usage(void)
+{
+   error_report("coroutine-ucontext: max stack usage was less or equal to "
+                "%"PRIu32" bytes.", max_stack_usage);
+}
+#endif