On 04/21/2017 10:48 AM, Jens Axboe wrote: > On 04/21/2017 10:40 AM, Bart Van Assche wrote: >> On Fri, 2017-04-21 at 10:33 -0600, Jens Axboe wrote: >>> On 04/21/2017 10:31 AM, Bart Van Assche wrote: >>>> On Fri, 2017-04-21 at 10:25 -0600, Jens Axboe wrote: >>>>> On 04/21/2017 09:32 AM, Bart Van Assche wrote: >>>>>> Hello Jens, >>>>>> >>>>>> Since yesterday the following complaint is reported frequently after having >>>>>> installed the for-4.12/block branch on my test setup. Unless someone has a >>>>>> better proposal, I will run a bisect. >>>>>> >>>>>> BUG: sleeping function called from invalid context at ./include/linux/buffer_head.h:349 >>>>>> in_atomic(): 1, irqs_disabled(): 0, pid: 8019, name: find >>>>>> CPU: 10 PID: 8019 Comm: find Tainted: G W I 4.11.0-rc4-dbg+ #2 >>>>>> Call Trace: >>>>>> dump_stack+0x68/0x93 >>>>>> ___might_sleep+0x16e/0x230 >>>>>> __might_sleep+0x4a/0x80 >>>>>> __ext4_get_inode_loc+0x1e0/0x4e0 >>>>>> ext4_iget+0x70/0xbc0 >>>>>> ext4_iget_normal+0x2f/0x40 >>>>>> ext4_lookup+0xb6/0x1f0 >>>>>> lookup_slow+0x104/0x1e0 >>>>>> walk_component+0x19a/0x330 >>>>>> path_lookupat+0x4b/0x100 >>>>>> filename_lookup+0x9a/0x110 >>>>>> user_path_at_empty+0x36/0x40 >>>>>> vfs_statx+0x67/0xc0 >>>>>> SYSC_newfstatat+0x20/0x40 >>>>>> SyS_newfstatat+0xe/0x10 >>>>>> entry_SYSCALL_64_fastpath+0x18/0xad >>>>> >>>>> How are you reproducing this? I've been running testing on the test box >>>>> and I run it on my laptop as well, but I haven't seen anything odd. >>>> >>>> Hello Jens, >>>> >>>> All I have to do to reproduce this is to build, install and boot the kernel. >>>> Maybe we are using a different kernel config? >>> >>> I'd say odds are good we are not using an identical kernel config :-) >>> What is your root device? Is it using mq and scheduling, or what's >>> the config? >> >> Hello Jens, >> >> The boot device is a SATA disk: >> # lsscsi >> [0:0:0:0] disk ATA ST1000NM0033-9ZM GA67 /dev/sda >> >> SCSI-mq is enabled and the default I/O scheduler is the deadline scheduler. >> From the kernel .config: >> CONFIG_DEFAULT_IOSCHED="deadline" >> CONFIG_SCSI_MQ_DEFAULT=y > > I wonder if it's an imbalance in the preempt count. Looking at it, it > looks like we're not clearing the alloc data. But I would think that > would potentially cause much worse problems, but maybe we got lucky? > > Let me generate a cleanup patch for that. Something like the below. diff --git a/block/blk-mq.c b/block/blk-mq.c index 0b2f60407748..519e5a08c0c6 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -279,7 +279,7 @@ EXPORT_SYMBOL_GPL(__blk_mq_alloc_request); struct request *blk_mq_alloc_request(struct request_queue *q, int rw, unsigned int flags) { - struct blk_mq_alloc_data alloc_data = { .flags = flags }; + struct blk_mq_alloc_data alloc_data; struct request *rq; int ret; @@ -287,6 +287,7 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw, if (ret) return ERR_PTR(ret); + blk_mq_init_alloc_data(&alloc_data, flags); rq = blk_mq_sched_get_request(q, NULL, rw, &alloc_data); blk_mq_put_ctx(alloc_data.ctx); @@ -305,7 +306,7 @@ EXPORT_SYMBOL(blk_mq_alloc_request); struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw, unsigned int flags, unsigned int hctx_idx) { - struct blk_mq_alloc_data alloc_data = { .flags = flags }; + struct blk_mq_alloc_data alloc_data; struct request *rq; unsigned int cpu; int ret; @@ -326,6 +327,8 @@ struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw, if (ret) return ERR_PTR(ret); + blk_mq_init_alloc_data(&alloc_data, flags); + /* * Check if the hardware context is actually mapped to anything. * If not tell the caller that it should skip this queue. @@ -1531,7 +1534,7 @@ static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio) { const int is_sync = op_is_sync(bio->bi_opf); const int is_flush_fua = op_is_flush(bio->bi_opf); - struct blk_mq_alloc_data data = { .flags = 0 }; + struct blk_mq_alloc_data data; struct request *rq; unsigned int request_count = 0; struct blk_plug *plug; @@ -1559,6 +1562,7 @@ static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio) trace_block_getrq(q, bio, bio->bi_opf); + blk_mq_init_alloc_data(&data, 0); rq = blk_mq_sched_get_request(q, bio, bio->bi_opf, &data); if (unlikely(!rq)) { __wbt_done(q->rq_wb, wb_acct); diff --git a/block/blk-mq.h b/block/blk-mq.h index 524f44742816..36cfc21c7cc6 100644 --- a/block/blk-mq.h +++ b/block/blk-mq.h @@ -148,6 +148,16 @@ struct blk_mq_alloc_data { struct blk_mq_hw_ctx *hctx; }; +static inline void blk_mq_init_alloc_data(struct blk_mq_alloc_data *data, + unsigned int flags) +{ + data->q = NULL; + data->flags = flags; + data->shallow_depth = 0; + data->ctx = NULL; + data->hctx = NULL; +} + static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data) { if (data->flags & BLK_MQ_REQ_INTERNAL) -- Jens Axboe