On Tue, Aug 16 2016 at 1:32pm -0400, Bart Van Assche <bart.vanassche@xxxxxxxxxxx> wrote: > Hello Mike, > > If I trigger failover and failback with kernel v4.8-rc2 and ib_srp then I > see the following: > > BUG: unable to handle kernel NULL pointer dereference at 0000000000000080 > IP: [<ffffffff8130d03b>] blk_mq_insert_request+0x3b/0xc0 > CPU: 4 PID: 12606 Comm: kdmwork-254:1 Not tainted 4.8.0-rc2-dbg+ #1 > Hardware name: Dell Inc. PowerEdge R430/03XKDV, BIOS 1.0.2 11/17/2014 > task: ffff880363d3e240 task.stack: ffff8803618ac000 > RIP: 0010:[<ffffffff8130d03b>] [<ffffffff8130d03b>] blk_mq_insert_request+0x3b/0xc0 > Call Trace: > [<ffffffff81300da9>] blk_insert_cloned_request+0xa9/0x1e0 > [<ffffffffa04302f0>] map_request+0x190/0x2d0 [dm_mod] > [<ffffffffa043044d>] map_tio_request+0x1d/0x40 [dm_mod] > [<ffffffff81087101>] kthread_worker_fn+0xd1/0x1b0 > [<ffffffff81086fba>] kthread+0xea/0x100 > [<ffffffff8162d53f>] ret_from_fork+0x1f/0x40 > > (gdb) list *(blk_mq_insert_request+0x3b) > 0xffffffff8130d03b is in blk_mq_insert_request (block/blk-mq.c:1078). > 1073 struct request_queue *q = rq->q; > 1074 struct blk_mq_hw_ctx *hctx; > 1075 struct blk_mq_ctx *ctx = rq->mq_ctx, *current_ctx; > 1076 > 1077 current_ctx = blk_mq_get_ctx(q); > 1078 if (!cpu_online(ctx->cpu)) > 1079 rq->mq_ctx = ctx = current_ctx; > 1080 > 1081 hctx = q->mq_ops->map_queue(q, ctx->cpu); > 1082 > (gdb) print &((struct blk_mq_ctx*)0)->cpu > $1 = (unsigned int *) 0x80 > > I think this means that ctx->cpu == NULL was hit. > > This was observed with the same test software and configuration I > used for my kernel v4.7 tests (CONFIG_SCSI_MQ_DEFAULT=y and > CONFIG_DM_MQ_DEFAULT=n). > > The above callstack was not observed while testing kernel v4.7. I only applied the 3 patches that I asked you to include in your v4.7 kernel(s), which I made available with this branch: https://git.kernel.org/cgit/linux/kernel/git/device-mapper/linux-dm.git/log/?h=dm-4.7-mpath-fixes Could be that something changed in v4.8 block core's blk-mq code that needs to be taken into account, we'll see. > Can you have a look at this? I'll get linux-dm.git's 'dm-4.8' branch (v4.8-rc2) loaded on one of my testbed systems to run the mptest testsuite. It should provide coverage for simple failover and failback. -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
