Re: Server regeneration no longer works
On Thu, Oct 30, 2003 at 08:51:48AM +0100, Frank Gießler wrote: David Dawes wrote: snip cvs rdiff -D 15 October 2003 -D 16 October 2003 Xserver/os gives: snip Thanks, it is indeed the same bug. I've marked the one in Bugzilla as fixed. Is there an easy way to figure out the dates for the diffs? I chose them based on the date of the message of mine you quoted that mentioned the fix :-) Also, cvsweb.xfree86.org allows easy viewing of changes. That's how I found out. But one has to know which files to investigate, and if more than one file is involved it can be a real PITA. Or am I missing something? The cvs-commit list (and its archive) are useful for this. The commit messages will show you exactly what the revisions are for each changed file. I'd recommend that anyone following XFree86 development subscribe to cvs-commit (see http://www.xfree86.org/lists.html). David -- David Dawes X-Oz Technologies www.XFree86.org/~dawes www.x-oz.com ___ Devel mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/devel
Re: Server regeneration no longer works
On Thu, 30 Oct 2003, [ISO-8859-1] Frank Gießler wrote: snip cvs rdiff -D 15 October 2003 -D 16 October 2003 Xserver/os gives: snip Thanks, it is indeed the same bug. I've marked the one in Bugzilla as fixed. Is there an easy way to figure out the dates for the diffs? Also, cvsweb.xfree86.org allows easy viewing of changes. That's how I found out. But one has to know which files to investigate, and if more than one file is involved it can be a real PITA. Or am I missing something? Search cvs-commits list archives, and/or use the cvsps tool to generate patchsets. You can find the latter with google. The documentation blows goats but the tool works awesome once you mess with it for hours and hours wasting time until you figure out the right magic. ;o) It's a godsend. -- Mike A. Harris ___ Devel mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/devel
Re: Server regeneration no longer works
On Tue, Oct 28, 2003 at 12:48:36PM +0100, Frank Gießler wrote:
David Dawes wrote:
On Wed, Oct 15, 2003 at 04:17:35PM -0400, David Dawes wrote:
On Wed, Oct 15, 2003 at 12:53:32PM -0400, Mark Vojkovich wrote:
Start the server with no clients and access control disabled
XFree86 -ac. Run a client and kill it, or run a client that
terminates itself (like xset) and the server segfaults on the
server regeneration. I'm having a hard time getting a back trace.
It looks like the stack is trashed pretty badly. This appears
to be a recent regression.
It looks like a malloc/free bug -- maybe a double free somewhere. I
see it crashing in free(), called from CMapUnwrapScreen(). If I add
some debugging ErrorF's the caller of the crashing free() moves. I'll
try a build on FreeBSD and enable some of its malloc debugging flags.
That located the problem nicely, aborting at the actual double-free.
The screen saver timers get freed in TimerInit() for each generation.
However, the timer pointers kept by the screen saver code weren't being
cleared for each generation, so the old values kept being used and added
back to the timers list. Then they'd get freed again in TimerInit() at
the second regeneration. Not to mention that freed memory was still
being used. I'm going to commit a fix that clears the Screensaver and
DPMS timers at each regeneration.
David
Is this fix available as a single patch? I would like to check whether this also
fixes
Bugzilla's #776.
cvs rdiff -D 15 October 2003 -D 16 October 2003 Xserver/os
gives:
Index: xc/programs/Xserver/os/WaitFor.c
diff -u xc/programs/Xserver/os/WaitFor.c:3.41 xc/programs/Xserver/os/WaitFor.c:3.42
--- xc/programs/Xserver/os/WaitFor.c:3.41 Thu Sep 25 09:26:27 2003
+++ xc/programs/Xserver/os/WaitFor.cWed Oct 15 21:33:35 2003
@@ -1,4 +1,4 @@
-/* $XFree86: xc/programs/Xserver/os/WaitFor.c,v 3.41 2003/09/25 13:26:27 pascal Exp $
*/
+/* $XFree86: xc/programs/Xserver/os/WaitFor.c,v 3.42 2003/10/16 01:33:35 dawes Exp $
*/
/***
Copyright 1987, 1998 The Open Group
@@ -112,7 +112,7 @@
};
static void DoTimer(OsTimerPtr timer, CARD32 now, OsTimerPtr *prev);
-static OsTimerPtr timers;
+static OsTimerPtr timers = NULL;
/*
* WaitForSomething:
@@ -568,10 +568,15 @@
}
static OsTimerPtr ScreenSaverTimer = NULL;
+static int ScreenSaverGeneration = -1;
void
SetScreenSaverTimer(void)
{
+if (ScreenSaverGeneration != serverGeneration) {
+ ScreenSaverTimer = NULL;
+ ScreenSaverGeneration = serverGeneration;
+}
if (ScreenSaverTime 0) {
ScreenSaverTimer = TimerSet(ScreenSaverTimer, 0, ScreenSaverTime,
ScreenSaverTimeoutExpire, NULL);
@@ -586,6 +591,7 @@
static OsTimerPtr DPMSStandbyTimer = NULL;
static OsTimerPtr DPMSSuspendTimer = NULL;
static OsTimerPtr DPMSOffTimer = NULL;
+static int DPMSGeneration = -1;
static CARD32
DPMSStandbyTimerExpire(OsTimerPtr timer,CARD32 now,pointer arg)
@@ -635,12 +641,19 @@
if (!DPMSEnabled)
return;
+if (DPMSGeneration != serverGeneration) {
+ DPMSStandbyTimer = NULL;
+ DPMSSuspendTimer = NULL;
+ DPMSOffTimer = NULL;
+ DPMSGeneration = serverGeneration;
+}
+
if (DPMSStandbyTime 0) {
DPMSStandbyTimer = TimerSet(DPMSStandbyTimer, 0, DPMSStandbyTime,
DPMSStandbyTimerExpire, NULL);
}
if (DPMSSuspendTime 0) {
-DPMSStandbyTimer = TimerSet(DPMSSuspendTimer, 0, DPMSSuspendTime,
+DPMSSuspendTimer = TimerSet(DPMSSuspendTimer, 0, DPMSSuspendTime,
DPMSSuspendTimerExpire, NULL);
}
if (DPMSOffTime 0) {
Also, cvsweb.xfree86.org allows easy viewing of changes.
A better approach is to explicitly free the timers at the end of each
generation. I committed that last night.
David
--
David Dawes X-Oz Technologies
www.XFree86.org/~dawes www.x-oz.com
___
Devel mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/devel
Re: Server regeneration no longer works
David Dawes wrote: On Wed, Oct 15, 2003 at 04:17:35PM -0400, David Dawes wrote: On Wed, Oct 15, 2003 at 12:53:32PM -0400, Mark Vojkovich wrote: Start the server with no clients and access control disabled XFree86 -ac. Run a client and kill it, or run a client that terminates itself (like xset) and the server segfaults on the server regeneration. I'm having a hard time getting a back trace. It looks like the stack is trashed pretty badly. This appears to be a recent regression. It looks like a malloc/free bug -- maybe a double free somewhere. I see it crashing in free(), called from CMapUnwrapScreen(). If I add some debugging ErrorF's the caller of the crashing free() moves. I'll try a build on FreeBSD and enable some of its malloc debugging flags. That located the problem nicely, aborting at the actual double-free. The screen saver timers get freed in TimerInit() for each generation. However, the timer pointers kept by the screen saver code weren't being cleared for each generation, so the old values kept being used and added back to the timers list. Then they'd get freed again in TimerInit() at the second regeneration. Not to mention that freed memory was still being used. I'm going to commit a fix that clears the Screensaver and DPMS timers at each regeneration. David Is this fix available as a single patch? I would like to check whether this also fixes Bugzilla's #776. Thanks, Frank. -- Frank Giessler Klinikum der Universitaet Jena Tel.: +49-3641-9 32 57 80 Biomagnetisches Zentrum Fax : +49-3641-9 32 57 72 ___ Devel mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/devel
