Issue 19 - compcache - crash with repeated cycles of memhog (x86 and x64)

Status:	Fixed
Owner:	nitingupta910
Closed:	Jan 2009

Sign in to add a comment

Reported by nitingupta910, Dec 31, 2008

What steps will reproduce the problem?
1. run:

a=1
while ((1)); do
    echo "Loop: $a"
    a=$((a = a + 1))"
    cat /proc/swaps
    cat /proc/compcache
    memhog 512m
done

2. Both on x86 and x86_64, system freezes within 5 iterations (1st
iteration is always successful).

3. System: Fedora Linux VMs:
    arch: x86, x64
    mem: 512m
    vcpus: 1

What is the expected output? What do you see instead?

System freezes with nothing useful in syslog. Issued flushes after crash
using sysrq but still nothing useful in logs.

Please use labels and text to provide additional information.

Screenshots attached.

	Fedora10Freeze-x86.jpg 56.3 KB View Download

	Fedora10Freeze-x64.jpg 48.8 KB View Download

Comment 1 by nitingupta910, Jan 04, 2009

This system freeze is happening since we are calling kmap() inside spin lock in
xvmalloc. kmap() can cause a reschedule which is a big NO inside spin lock.

Here is the serial o/p for crash:

BUG: spinlock cpu recursion on CPU#0, kswapd0/262  <--- cause: sched under spin lock
 lock: debc8044, .magic: dead4ead, .owner: memhog/2690, .owner_cpu: 0
Pid: 262, comm: kswapd0 Not tainted 2.6.28 #2
Call Trace:
 [<c075f8c6>] ? printk+0x1d/0x1f
 [<c0599bb3>] spin_bug+0xa3/0xf0
 [<c0599d14>] _raw_spin_lock+0x74/0x140
 [<e15853b5>] ? xvFree+0x35/0x1c0 [xvmalloc]
 [<e1585599>] ? xvMalloc+0x59/0x200 [xvmalloc]
 [<c0762de7>] _spin_lock+0x57/0x70
 [<e15853b5>] ? xvFree+0x35/0x1c0 [xvmalloc]
 [<e15853b5>] xvFree+0x35/0x1c0 [xvmalloc]
 [<e15b53d4>] compcache_make_request+0x2c4/0x4c0 [compcache]
 [<c057c6f4>] generic_make_request+0x3c4/0x4b0
 [<c0463457>] ? __lock_acquire+0x297/0x1160
 [<c04c6355>] ? kmem_cache_alloc+0xc5/0x120
 [<c049e423>] ? mempool_alloc_slab+0x13/0x20
 [<c0409531>] ? sched_clock+0x11/0x20
 [<c046195d>] ? lock_release_holdtime+0x2d/0x210
 [<c04a9faa>] ? inc_zone_page_state+0x5a/0x90
 [<c057c847>] submit_bio+0x67/0xf0
 [<c0462f0b>] ? trace_hardirqs_on+0xb/0x10
 [<c04a3b95>] ? test_set_page_writeback+0x75/0x120
 [<c04b8e26>] swap_writepage+0x76/0xb0
 [<c04b8e60>] ? end_swap_bio_write+0x0/0x90
 [<c04a7245>] shrink_page_list+0x615/0x830
 [<c0409531>] ? sched_clock+0x11/0x20
 [<c046195d>] ? lock_release_holdtime+0x2d/0x210
 [<c04a7f8c>] shrink_zone+0x4dc/0x8f0
 [<c0762c02>] ? _spin_unlock+0x22/0x30
 [<c04a8c3b>] kswapd+0x52b/0x550
 [<c04a7530>] ? isolate_pages_global+0x0/0x200
 [<c04509f0>] ? autoremove_wake_function+0x0/0x50
 [<c04a8710>] ? kswapd+0x0/0x550
 [<c04506a1>] kthread+0x41/0x80
 [<c0450660>] ? kthread+0x0/0x80
 [<c0404c7b>] kernel_thread_helper+0x7/0x10

Comment 2 by nitingupta910, Jan 04, 2009

This is fixed in SVN rev #151

Comment 3 by nitingupta910, Jan 04, 2009

(No comment was entered for this change.)

Status: Fixed

Comment 4 by nitingupta910, Jan 04, 2009

Tested on both x86 and x64 with 50+ memhog cycles.

Status: Verified

Comment 5 by nitingupta910, Jan 27, 2009

Fixed in compcache-0.5.1

Status: Fixed
Labels: -Type-Defect -Priority-Medium

Issue 19:	crash with repeated cycles of memhog (x86 and x64)
		Back to list