From: Nicholas Nethercote <njn@valgrind.org>
Date: Thu, 26 Jun 2003 14:57:01 +0000 (+0000)
Subject: comment only
X-Git-Tag: svn/VALGRIND_2_0_0~76
X-Git-Url: http://git.ipfire.org/gitweb.cgi?a=commitdiff_plain;h=c10bae9afe3475912b5fa65fbb96ecb2c6beb69d;p=thirdparty%2Fvalgrind.git

comment only


git-svn-id: svn://svn.valgrind.org/valgrind/trunk@1708
---

diff --git a/coregrind/vg_syscalls.c b/coregrind/vg_syscalls.c
index 59092a74b0..b79ada6d24 100644
--- a/coregrind/vg_syscalls.c
+++ b/coregrind/vg_syscalls.c
@@ -2704,7 +2704,20 @@ void VG_(perform_assumed_nonblocking_syscall) ( ThreadId tid )
          }
          /* For some unknown reason, %ebx sometimes gets changed by poll...
             let the skin know (using the `post_reg_write_syscall_return'
-            event isn't ideal... */
+            event isn't ideal... 
+          
+            Update from Lennert Buytenhek <buytenh@gnu.org>: 
+            This came up a while ago on linux-kernel, search for a posting
+            by dvorak <dvorak@xs4all.nl>.
+
+            Basically, the linux syscall handler pushes the arguments to the
+            stack, and inside sys_poll (or another routine), one of the
+            parameters to the function is changed.  gcc uses the stack space
+            the parameter came in as the storage location for that variable,
+            and will sometimes spill that variable back to stack.  The linux
+            syscall handler pops all registers on return, and that is how
+            %ebx can get changed sometimes.
+          */
          if (arg1 != tst->m_ebx) {
             VG_TRACK( post_reg_write_syscall_return, tid, R_EBX );
          }