On Wed, Jun 06, 2012 at 08:00:00AM -0700, Chris J Arges wrote: > > > Dave Chinner wrote: > > > > It seems unlikely, but if you turn on kmemleak it might find a > > memory leak or overwrite that is causing this. > > > > Running kmemleak on a 3.0.0 kernel results in the following: > > dmesg: > [ 3855.751393] XFS (sda2): xlog_verify_grant_tail: space > > BBTOB(tail_blocks) > [22987.932317] kmemleak: 1 new suspected memory leaks (see > /sys/kernel/debug/kmemleak) > > kmemleak: > unreferenced object 0xffff88015612a340 (size 208): > comm "copy-files", pid 1483, jiffies 4310560285 (age 16571.656s) > hex dump (first 32 bytes): > 00 f3 1b 57 01 88 ff ff 00 08 00 58 01 88 ff ff ...W.......X.... > a0 a7 32 a0 ff ff ff ff 04 00 00 00 00 00 00 00 ..2............. > backtrace: > [<ffffffff815c6196>] kmemleak_alloc+0x26/0x50 > [<ffffffff811542e3>] kmem_cache_alloc+0x123/0x190 > [<ffffffffa0313967>] kmem_zone_alloc+0x67/0xe0 [xfs] > [<ffffffffa03139fd>] kmem_zone_zalloc+0x1d/0x50 [xfs] > [<ffffffffa02b033f>] xfs_allocbt_init_cursor+0xdf/0x130 [xfs] > [<ffffffffa02adb4c>] xfs_alloc_ag_vextent_near+0x6c/0xd80 [xfs] > [<ffffffffa02aea88>] xfs_alloc_ag_vextent+0x228/0x290 [xfs] > [<ffffffffa02af7d9>] xfs_alloc_vextent+0x649/0x8c0 [xfs] > [<ffffffffa02bcfc6>] xfs_bmap_btalloc+0x286/0x7c0 [xfs] > [<ffffffffa02bd521>] xfs_bmap_alloc+0x21/0x40 [xfs] > [<ffffffffa02c6ba3>] xfs_bmapi+0xdc3/0x1950 [xfs] > [<ffffffffa02f5059>] xfs_iomap_write_allocate+0x179/0x340 [xfs] > [<ffffffffa03147d5>] xfs_map_blocks+0x215/0x380 [xfs] > [<ffffffffa0315792>] xfs_vm_writepage+0x1b2/0x510 [xfs] > [<ffffffff811142e7>] __writepage+0x17/0x40 > [<ffffffff8111485d>] write_cache_pages+0x20d/0x460 > > analysis: > I turned on kmemleak, function tracing, and xfs debugging in the build > in which I ran this. So far I’ve been able to run the copy-files script for > about 24 hrs without failure. I’m not sure if this is because all these > features are turned on and it has slowed something down (so it takes longer > to reproduce), or if the debugging code is changing the behavior. I’m not > sure if this backtrace is valid, so I’m attaching an annotated objdump of my > xfs module. OK, this is not what we are looking for. Yes, there's a cursor leak that I can find just by looking at the code, but that's definitely not related the AIL issue.... I'll send a patch to fix this leak soon. Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs