Workload is build all arm defconfigs and all powerpc defconfigs and
something else in parallel from one source tree using O= and with
empty .ccache on ext2/noatime partition.

PREEMPT_RCU was again in use to do final check of Jens's patch (see URL)

Kernel is afa26be86b65a7183ceac29bdf1f51d6fc6932f0 + CFQ fixlet
http://marc.info/?l=linux-kernel&m=120946003120291&w=2 .

Oops, disassembly and .config (DEBUG_INFO was turned off) below.

Should we add PREEMPT_RCU to oops reports, BTW?

[12617.678557] general protection fault: 0000 [1] PREEMPT SMP DEBUG_PAGEALLOC
[12617.678646] CPU 1 
[12617.678695] Modules linked in: ext2 nf_conntrack_irc xt_state iptable_filter ipt_MASQUERADE iptable_nat nf_nat nf_conntrack_ipv4 nf_conntrack ip_tables x_tables usblp uhci_hcd ehci_hcd usbcore sr_mod cdrom
[12617.678953] Pid: 15988, comm: cc1 Not tainted 2.6.26-rc1-afa26be86b65a7183ceac29bdf1f51d6fc6932f0 #3
[12617.679064] RIP: 0010:[<ffffffff802a3cdc>]  [<ffffffff802a3cdc>] __d_lookup+0x8c/0x160
[12617.679171] RSP: 0018:ffff810072f3fbc8  EFLAGS: 00010202
[12617.679372] RAX: 6b6b6b6b6b6b6b6b RBX: 6b6b6b6b6b6b6b6b RCX: 0000000000000013
[12617.679441] RDX: ffff81000100c128 RSI: 0000000000000000 RDI: ffff8101233c1140
[12617.679511] RBP: ffff810072f3fc18 R08: 0000000000000000 R09: ffff8100695cc200
[12617.679581] R10: 0000000000000000 R11: 0000000000000000 R12: 6b6b6b6b6b6b6b2b
[12617.679650] R13: 0000000071bf92cd R14: ffff810129d4a138 R15: ffff810072f3fcc8
[12617.679720] FS:  00002ab45130a6f0(0000) GS:ffff81017fc44320(0000) knlGS:0000000000000000
[12617.679827] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[12617.679890] CR2: 00002b3818aba000 CR3: 000000005d00d000 CR4: 0000000000000660
[12617.679961] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[12617.680030] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[12617.680100] Process cc1 (pid: 15988, threadinfo ffff810072f3e000, task ffff8101233c1140)
[12617.680206] Stack:  ffff81000100d218 0000000000000000 ...
[ message continues ]

Given that you see it with PREEMPT_RCU, but not classic RCU, that would
make sense...

I presume that this version of the kernel has the fixes from this thread
applied?

	http://lkml.org/lkml/2008/4/19/85

--

[ message continues ]

Those are the two I am aware of.

--

[ message continues ]

Triggered second time with 2.6.24-rc4.

FWIW, workload was LTP in infinite loop + parallel compiling on 170 configs
with make O=../ on ext2

--

[ message continues ]

Thank you for the info!

So you have 170 concurrent "make O=../" builds running, with them all
putting their .o files in a common directory?  Are you using 170
specific configurations, or are the all running randconfig?

--

[ message continues ]

--

[ message continues ]

No, of course, not. :-) It's like this:

	src
	build/config-1/
	build/config-2/
		...
	.ccache/
		[.ccache stuff]
all in one ext2/noatime disk.


Very specific but this shouldn't matter. Some are fat (sparc64 allmodconfig),
some are nimble (arm defconfigs). loadavg is around 210 at the start.

--

[ message continues ]