slub: Enable backtrace for create/delete points
authorBen Greear <greearb@candelatech.com>
Thu, 7 Jul 2011 18:36:36 +0000 (11:36 -0700)
committerPekka Enberg <penberg@kernel.org>
Thu, 7 Jul 2011 19:17:03 +0000 (22:17 +0300)
This patch attempts to grab a backtrace for the creation
and deletion points of the slub object.  When a fault is
detected, we can then get a better idea of where the item
was deleted.

Example output from debugging some funky nfs/rpc behaviour:

=============================================================================
BUG kmalloc-64: Object is on free-list
-----------------------------------------------------------------------------

INFO: Allocated in rpcb_getport_async+0x39c/0x5a5 [sunrpc] age=381 cpu=3 pid=3750
       __slab_alloc+0x348/0x3ba
       kmem_cache_alloc_trace+0x67/0xe7
       rpcb_getport_async+0x39c/0x5a5 [sunrpc]
       call_bind+0x70/0x75 [sunrpc]
       __rpc_execute+0x78/0x24b [sunrpc]
       rpc_execute+0x3d/0x42 [sunrpc]
       rpc_run_task+0x79/0x81 [sunrpc]
       rpc_call_sync+0x3f/0x60 [sunrpc]
       rpc_ping+0x42/0x58 [sunrpc]
       rpc_create+0x4aa/0x527 [sunrpc]
       nfs_create_rpc_client+0xb1/0xf6 [nfs]
       nfs_init_client+0x3b/0x7d [nfs]
       nfs_get_client+0x453/0x5ab [nfs]
       nfs_create_server+0x10b/0x437 [nfs]
       nfs_fs_mount+0x4ca/0x708 [nfs]
       mount_fs+0x6b/0x152
INFO: Freed in rpcb_map_release+0x3f/0x44 [sunrpc] age=30 cpu=2 pid=29049
       __slab_free+0x57/0x150
       kfree+0x107/0x13a
       rpcb_map_release+0x3f/0x44 [sunrpc]
       rpc_release_calldata+0x12/0x14 [sunrpc]
       rpc_free_task+0x59/0x61 [sunrpc]
       rpc_final_put_task+0x82/0x8a [sunrpc]
       __rpc_execute+0x23c/0x24b [sunrpc]
       rpc_async_schedule+0x10/0x12 [sunrpc]
       process_one_work+0x230/0x41d
       worker_thread+0x133/0x217
       kthread+0x7d/0x85
       kernel_thread_helper+0x4/0x10
INFO: Slab 0xffffea00029aa470 objects=20 used=9 fp=0xffff8800be7830d8 flags=0x20000000004081
INFO: Object 0xffff8800be7830d8 @offset=4312 fp=0xffff8800be7827a8

Bytes b4 0xffff8800be7830c8:  87 a8 96 00 01 00 00 00 5a 5a 5a 5a 5a 5a 5a 5a .�......ZZZZZZZZ
 Object 0xffff8800be7830d8:  6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b kkkkkkkkkkkkkkkk
 Object 0xffff8800be7830e8:  6b 6b 6b 6b 01 08 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b kkkk..kkkkkkkkkk
 Object 0xffff8800be7830f8:  6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b kkkkkkkkkkkkkkkk
 Object 0xffff8800be783108:  6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b 6b a5 kkkkkkkkkkkkkkk�
 Redzone 0xffff8800be783118:  bb bb bb bb bb bb bb bb                         �������������
 Padding 0xffff8800be783258:  5a 5a 5a 5a 5a 5a 5a 5a                         ZZZZZZZZ
Pid: 29049, comm: kworker/2:2 Not tainted 3.0.0-rc4+ #8
Call Trace:
 [<ffffffff811055c3>] print_trailer+0x131/0x13a
 [<ffffffff81105601>] object_err+0x35/0x3e
 [<ffffffff8110746f>] verify_mem_not_deleted+0x7a/0xb7
 [<ffffffffa02851b5>] rpcb_getport_done+0x23/0x126 [sunrpc]
 [<ffffffffa027d0ba>] rpc_exit_task+0x3f/0x6d [sunrpc]
 [<ffffffffa027d4ab>] __rpc_execute+0x78/0x24b [sunrpc]
 [<ffffffffa027d6c0>] ? rpc_execute+0x42/0x42 [sunrpc]
 [<ffffffffa027d6d0>] rpc_async_schedule+0x10/0x12 [sunrpc]
 [<ffffffff810611b7>] process_one_work+0x230/0x41d
 [<ffffffff81061102>] ? process_one_work+0x17b/0x41d
 [<ffffffff81063613>] worker_thread+0x133/0x217
 [<ffffffff810634e0>] ? manage_workers+0x191/0x191
 [<ffffffff81066e10>] kthread+0x7d/0x85
 [<ffffffff81485924>] kernel_thread_helper+0x4/0x10
 [<ffffffff8147eb18>] ? retint_restore_args+0x13/0x13
 [<ffffffff81066d93>] ? __init_kthread_worker+0x56/0x56
 [<ffffffff81485920>] ? gs_change+0x13/0x13

Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Ben Greear <greearb@candelatech.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
mm/slub.c

index 7be0223531b090dbd9f716104714bc4d9ca9c7b5..c9050995bc87ffc136a83f75af8a7a8a5c1bb086 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -191,8 +191,12 @@ static LIST_HEAD(slab_caches);
 /*
  * Tracking user of a slab.
  */
+#define TRACK_ADDRS_COUNT 16
 struct track {
        unsigned long addr;     /* Called from address */
+#ifdef CONFIG_STACKTRACE
+       unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
+#endif
        int cpu;                /* Was running on cpu */
        int pid;                /* Pid context */
        unsigned long when;     /* When did the operation occur */
@@ -420,6 +424,24 @@ static void set_track(struct kmem_cache *s, void *object,
        struct track *p = get_track(s, object, alloc);
 
        if (addr) {
+#ifdef CONFIG_STACKTRACE
+               struct stack_trace trace;
+               int i;
+
+               trace.nr_entries = 0;
+               trace.max_entries = TRACK_ADDRS_COUNT;
+               trace.entries = p->addrs;
+               trace.skip = 3;
+               save_stack_trace(&trace);
+
+               /* See rant in lockdep.c */
+               if (trace.nr_entries != 0 &&
+                   trace.entries[trace.nr_entries - 1] == ULONG_MAX)
+                       trace.nr_entries--;
+
+               for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
+                       p->addrs[i] = 0;
+#endif
                p->addr = addr;
                p->cpu = smp_processor_id();
                p->pid = current->pid;
@@ -444,6 +466,16 @@ static void print_track(const char *s, struct track *t)
 
        printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
                s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
+#ifdef CONFIG_STACKTRACE
+       {
+               int i;
+               for (i = 0; i < TRACK_ADDRS_COUNT; i++)
+                       if (t->addrs[i])
+                               printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]);
+                       else
+                               break;
+       }
+#endif
 }
 
 static void print_tracking(struct kmem_cache *s, void *object)