On Thu, 22 Dec 2011 15:20:36 -0800 Tejun Heo <tj@xxxxxxxxxx> wrote: > Hello, Andrew. > > On Thu, Dec 22, 2011 at 03:08:36PM -0800, Andrew Morton wrote: > > > [ 558.576528] SysRq : Show Blocked State > > > [ 558.576633] task PC stack pid father > > > [ 558.576738] sh D 0000000000000001 0 4701 4700 0x00000080 > > > [ 558.576882] ffff8802493f78b8 0000000000000046 000000014a1121c0 ffff8802493f6010 > > > [ 558.577109] ffff88024a1121c0 00000000001d1100 ffff8802493f7fd8 0000000000004000 > > > [ 558.577336] ffff8802493f7fd8 00000000001d1100 ffff880255db66c0 ffff88024a1121c0 > > > [ 558.577568] Call Trace: > > > [ 558.577905] [<ffffffff813d2744>] schedule+0x55/0x57 > > > [ 558.577960] [<ffffffff813d27cd>] io_schedule+0x87/0xca > > > [ 558.578017] [<ffffffff811a1e72>] get_request_wait+0xbd/0x19e > > > [ 558.578182] [<ffffffff811a20cc>] blk_queue_bio+0x179/0x271 > > > [ 558.578238] [<ffffffff811a02a9>] generic_make_request+0x9c/0xde > > > [ 558.578293] [<ffffffff811a03a4>] submit_bio+0xb9/0xc4 > > > [ 558.578348] [<ffffffff810ffcc6>] submit_bh+0xe6/0x108 > > > [ 558.578404] [<ffffffff8110273c>] __block_write_full_page+0x1ec/0x2e3 > > > [ 558.578518] [<ffffffff811028fb>] block_write_full_page_endio+0xc8/0xcc > > > [ 558.578573] [<ffffffff8110290f>] block_write_full_page+0x10/0x12 > > > [ 558.578631] [<ffffffff811417cd>] ext3_writeback_writepage+0xaa/0x11d > > > [ 558.578690] [<ffffffff810a0ed0>] __writepage+0x15/0x34 > > > [ 558.578744] [<ffffffff810a1913>] write_cache_pages+0x240/0x33e > > > [ 558.578911] [<ffffffff810a1a54>] generic_writepages+0x43/0x5a > > > [ 558.578967] [<ffffffff810a1a91>] do_writepages+0x26/0x28 > > > [ 558.579022] [<ffffffff8109a8cf>] __filemap_fdatawrite_range+0x4e/0x50 > > > [ 558.579078] [<ffffffff8109aee8>] filemap_flush+0x17/0x19 > > > [ 558.579134] [<ffffffff8113f2c2>] ext3_release_file+0x2e/0xa4 > > > [ 558.579190] [<ffffffff810dbdce>] fput+0x10f/0x1cd > > > [ 558.579244] [<ffffffff810d86e0>] filp_close+0x70/0x7b > > > [ 558.579300] [<ffffffff8102c09b>] put_files_struct+0x16c/0x2c1 > > > [ 558.579412] [<ffffffff8102c236>] exit_files+0x46/0x4e > > > [ 558.579465] [<ffffffff8102dd2a>] do_exit+0x246/0x73c > > > [ 558.579576] [<ffffffff8102e2a4>] do_group_exit+0x84/0xb2 > > > [ 558.579743] [<ffffffff8102e2e4>] sys_exit_group+0x12/0x16 > > > [ 558.579910] [<ffffffff813d9562>] system_call_fastpath+0x16/0x1b > > Hmmm... probably cic allocation failure? Dunno, it's an 8Gb 8 CPU x86_64 box. > > A large amount of block core code was merged in the Dec 15 - Dec 21 > > window. Tejun... > > Yeah, those are blk-ioc cleanup patches. I was wishing to merge them > earlier. > > > revert-f2dbd76a0a994bc1d5a3d0e7c844cc373832e86c.patch BAD > > revert-1238033c79e92e5c315af12e45396f1a78c73dec.patch > > revert-b50b636bce6293fa858cc7ff6c3ffe4920d90006.patch > > revert-b9a1920837bc53430d339380e393a6e4c372939f.patch > > revert-b2efa05265d62bc29f3a64400fad4b44340eedb8.patch > > revert-f1a4f4d35ff30a328d5ea28f6cc826b2083111d2.patch > > revert-216284c352a0061f5b20acff2c4e50fb43fea183.patch > > revert-dc86900e0a8f665122de6faadd27fb4c6d2b3e4d.patch > > revert-283287a52e3c3f7f8f9da747f4b8c5202740d776.patch > > revert-09ac46c429464c919d04bb737b27edd84d944f02.patch BAD > > revert-6e736be7f282fff705db7c34a15313281b372a76.patch GOOD > > revert-42ec57a8f68311bbbf4ff96a5d33c8a2e90b9d05.patch GOOD > > revert-a73f730d013ff2788389fd0c46ad3e5510f124e6.patch > > revert-8ba61435d73f2274e12d4d823fde06735e8f6a54.patch GOOD > > revert-481a7d64790cd7ca61a8bbcbd9d017ce58e6fe39.patch > > revert-34f6055c80285e4efb3f602a9119db75239744dc.patch > > revert-1ba64edef6051d2ec79bb2fbd3a0c8f0df00ab55.patch GOOD > > > > At the f2dbd76a0a994bc1d5a3d0e7c844cc373832e86 pivot point the kernel > > went odd, got stuck, slowly emitting "cfq: cic link failed!" messages. > > So we've added yet another bisection hole in there somewhere. > > You were likely seeing the same problem, just showing up differently. > Hmm.... we always had the problem of allocation failure in cfq could > lead to deadlock. This looks like a lost I/O completion. > It's just that those cases happened infrequently > enough that nobody really noticed (or at least tracked it down). How > can you reproduce the problem? Easily. One time it got to a login prompt and hung quickly during a make. Every other time (ten times, maybe) it hung during initscripts. -- To unsubscribe from this list: send the line "unsubscribe linux-next" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
