Patchwork [committed] GDB/libiberty: Mingw32 remote debug hang on close

login
register
mail settings
Submitter Maciej W. Rozycki
Date Aug. 20, 2010, 10:40 p.m.
Message ID <alpine.DEB.1.10.1008202224000.4295@tp.orcam.me.uk>
Download mbox | patch
Permalink /patch/62324/
State New
Headers show

Comments

Maciej W. Rozycki - Aug. 20, 2010, 10:40 p.m.
Hi,

 The change I proposed here:

http://gcc.gnu.org/ml/gcc-patches/2009-12/msg01089.html
http://gcc.gnu.org/ml/gcc-patches/2010-04/msg01214.html

has been reviewed by Mark Mitchell off the list and so I have applied it 
now.  For a reference I'm including the original submission below.

  Maciej

---------- Forwarded message ----------
Message-ID: <alpine.DEB.1.10.0912181134140.4720@tp.orcam.me.uk>
Date: Wed, 23 Dec 2009 19:25:13 +0000 (GMT)
From: Maciej W. Rozycki <macro@codesourcery.com>
To: gcc-patches@gcc.gnu.org
Cc: gdb-patches@sourceware.org, Maciej W. Rozycki <macro@linux-mips.org>
Subject: [PATCH] GDB/libiberty: Mingw32 remote debug hang on close

Hello,

 Noticed while debugging a remote target using pipes to communicate with a 
debug stub both running on a Mingw32 host.

 When a debug session is terminated from the GDB side, like with:

(gdb) kill

and the stub outputs any data to its stderr file descriptor (such as 
diagnostic information) afterwards before it terminates, then both 
programs hang.

 I have tracked down the cause to be the parent pipe descriptor 
corresponding to child's stderr having been left open while a wait() call 
is made in the parent to await child's termination.  Meanwhile the child 
would keep producing diagnostic output to its stderr up till the 
exhaustion of the pipe system buffer at which point the child would hang 
as the parent busy with the wait() would not pay attention to its 
corresponding descriptor anymore.

 Here's a change that works for me.  It makes sure the parent pipe 
descriptors corresponding to child's stderr and stdout are closed before 
the call to wait() is made, making any writes to these pipes in the child 
fail straight away.

2009-12-23  Maciej W. Rozycki  <macro@codesourcery.com>

	libiberty/
	* pex-common.c (pex_read_err): Set stderr_pipe to -1 if a 
	corresponding stream has been opened.
	(pex_free): Close pipe file descriptors corresponding to child's 
	stdout and stderr before waiting.

 Regression tested successfully with GDB testsuite the i386-mingw32 host 
and the mips-sde-elf target, using QEMU as the remote system.  OK to 
apply?

  Maciej

gdb-6.8.50-20091205-pex-free-0.patch

Patch

Index: libiberty/pex-common.c
===================================================================
--- libiberty/pex-common.c	(revision 268718)
+++ libiberty/pex-common.c	(working copy)
@@ -505,6 +505,7 @@  pex_read_err (struct pex_obj *obj, int b
   if (o < 0 || o == STDIN_FILE_NO)
     return NULL;
   obj->read_err = obj->funcs->fdopenr (obj, o, binary);
+  obj->stderr_pipe = -1;
   return obj->read_err;    
 }
 
@@ -597,8 +598,17 @@  pex_get_times (struct pex_obj *obj, int 
 void
 pex_free (struct pex_obj *obj)
 {
+  /* Close pipe file descriptors corresponding to child's stdout and
+     stderr so that the child does not hang trying to output anything
+     while we're waiting for it.  */
   if (obj->next_input >= 0 && obj->next_input != STDIN_FILE_NO)
     obj->funcs->close (obj, obj->next_input);
+  if (obj->stderr_pipe >= 0 && obj->stderr_pipe != STDIN_FILE_NO)
+    obj->funcs->close (obj, obj->stderr_pipe);
+  if (obj->read_output != NULL)
+    fclose (obj->read_output);
+  if (obj->read_err != NULL)
+    fclose (obj->read_err);
 
   /* If the caller forgot to wait for the children, we do it here, to
      avoid zombies.  */
@@ -619,10 +629,6 @@  pex_free (struct pex_obj *obj)
     free (obj->status);
   if (obj->time != NULL)
     free (obj->time);
-  if (obj->read_output != NULL)
-    fclose (obj->read_output);
-  if (obj->read_err != NULL)
-    fclose (obj->read_err);
 
   if (obj->remove_count > 0)
     {