misc/133264: [build] make buildworld on RELENG_6_4 is broken on SMP machines

[Victor Sudakov]

The following reply was made to PR misc/133264; it has been noted by GNATS.

From: Victor Sudakov <sudakov at sibptus.tomsk.ru>
To: John Baldwin <jhb at freebsd.org>
Cc: bug-followup at freebsd.org
Subject: Re: misc/133264: [build] make buildworld on RELENG_6_4 is broken on SMP machines
Date: Tue, 14 Apr 2009 08:47:09 +0700

 John Baldwin wrote:
 > > > > > > > tree.  If that is the case, a build without -j4 should also fail 
 > and 
 > > > > > should 
 > > > > > > > show you where you have object files lying around.
 > > > > > > 
 > > > > > > Building on VMWare ESXi 3.5.0 Update 3 (FreeBSD as a guest OS) still
 > > > > > > fails occasionally if more than 1 CPU is allocated to the virtual
 > > > > > > machine. The failures are due to processes like sh, sed or cc1 
 > dupming
 > > > > > > core on signal 11 during the build.
 > > > > > > 
 > > > > > > I am using a pristine source tree now.
 > > > > > > 
 > > > > > > The problem seems to be SMP related because enabling only 1 virtual
 > > > > > > CPU removes the problem.
 > > > > > > 
 > > > > > > Should I open a new PR?
 > > > > > 
 > > > > > Yes, this seems to be a different problem. :(  Do you see this 
 > building other 
 > > > > > source trees as well?  (e.g. RELENG_6 or RELENG_7).
 > > > > 
 > > > > RELENG_7_1 is not affected.
 > > > > RELENG_6 is not affected.
 > > > 
 > > > This is all while running the same kernel?
 > > 
 > > Each its own kernel.
 > > 
 > > On the 6.4-RELEASE SMP kernel, any large compilation will crash
 > > eventually with signal 11. Using "-j4" or higher seems to aggravate
 > > the problem.
 > 
 > Hmm, ok.  I'm not really sure why you are seeing this.  I don't know of 
 > anything in RELENG_6 that isn't in 6.4 that would explain that.  Are you sure 
 > your hardware is ok?
 
 As I stated above, my hardware is virtualized, it is a VMWare ESXi
 3.5.0 Update 3. Of course I am not sure it is ok (i.e. compatible with
 FreeBSD). However RELENG_7_1 compiles fine on it.
 
 I have had various problems with FreeBSD on different virtual
 platforms. For example, FreeBSD on Microsoft VPC has timing issues
 (the "microuptime went backwards" message). On Sun's VirtualBox,
 FreeBSD processes hang mysteriously with the obscure 
 "sigreturn: eflags = 0x80286" kernel message. Maybe FreeBSD does not
 feel like being a guest OS.
 
 
 -- 
 Victor Sudakov,  VAS4-RIPE, VAS47-RIPN
 sip:sudakov at sibptus.tomsk.ru