- 01 12月, 2014 1 次提交
-
-
由 Shaohua Li 提交于
We call blk_mq_alloc_tag_set() first then blk_mq_init_queue(). The requests are allocated in the former function. So the kdump check should be moved to there to really save memory. Signed-off-by: NShaohua Li <shli@fb.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 24 11月, 2014 1 次提交
-
-
由 Christoph Hellwig 提交于
Don't duplicate the code to handle the not cpu bounce case in the caller, do it inside blk_mq_hctx_next_cpu instead. Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 18 11月, 2014 2 次提交
-
-
由 Jens Axboe 提交于
It's silly to use blk_mq_free_request() which in turn maps the request to the hardware queue, for places where we already know what the hardware queue is. This saves us an extra mapping of a hardware queue on request completion, if the caller knows this information already. Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Jens Axboe 提交于
Drivers that know they are blk-mq should just use this function instead of calling through blk_put_request(). Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 12 11月, 2014 2 次提交
-
-
由 Paolo Bonzini 提交于
blk-mq is using preempt_disable/enable in order to ensure that the queue runners are placed on the right CPU. This does not work with the RT patches, because __blk_mq_run_hw_queue takes a non-raw spinlock with the preemption-disabled region. If there is contention on the lock, this violates the rules for preemption-disabled regions. While this should be easily fixable within the RT patches just by doing migrate_disable/enable, we can do better and document _why_ this particular region runs with disabled preemption. After the previous patch, it is trivial to switch it to get/put_cpu; the RT patches then can change it to get_cpu_light, which lets virtio-blk run under RT kernels. Cc: Jens Axboe <axboe@kernel.dk> Cc: Thomas Gleixner <tglx@linutronix.de> Reported-by: NClark Williams <williams@redhat.com> Tested-by: NClark Williams <williams@redhat.com> Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Paolo Bonzini 提交于
preempt_disable/enable surrounds every call to blk_mq_run_hw_queue, except the one in blk-flush.c. In fact that one is always asynchronous, and it does not need smp_processor_id(). We can do the same for all other calls, avoiding preempt_disable when async is true. This avoids peppering blk-mq.c with preemption-disabled regions. Cc: Jens Axboe <axboe@kernel.dk> Cc: Thomas Gleixner <tglx@linutronix.de> Reported-by: NClark Williams <williams@redhat.com> Tested-by: NClark Williams <williams@redhat.com> Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 30 10月, 2014 2 次提交
-
-
由 Jens Axboe 提交于
Drivers can now tell blk-mq if they take advantage of the deferred issue through 'last' or not. If they do, don't do queue-direct for sync IO. This is a preparation patch for the nvme conversion. Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Jens Axboe 提交于
Since we have the notion of a 'last' request in a chain, we can use this to have the hardware optimize the issuing of requests. Add a list_head parameter to queue_rq that the driver can use to temporarily store hw commands for issue when 'last' is true. If we are doing a chain of requests, pass in a NULL list for the first request to force issue of that immediately, then batch the remainder for deferred issue until the last request has been sent. Instead of adding yet another argument to the hot ->queue_rq path, encapsulate the passed arguments in a blk_mq_queue_data structure. This is passed as a constant, and has been tested as faster than passing 4 (or even 3) args through ->queue_rq. Update drivers for the new ->queue_rq() prototype. There are no functional changes in this patch for drivers - if they don't use the passed in list, then they will just queue requests individually like before. Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 14 10月, 2014 1 次提交
-
-
由 Jens Axboe 提交于
All other allocs are done on the specific node, somehow the cpumask for hw queue runs was missed. Fix that by using zalloc_cpumask_var_node() in blk_mq_init_queue(). Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 26 9月, 2014 7 次提交
-
-
由 Ming Lei 提交于
This patch supports to run one single flush machinery for each blk-mq dispatch queue, so that: - current init_request and exit_request callbacks can cover flush request too, then the buggy copying way of initializing flush request's pdu can be fixed - flushing performance gets improved in case of multi hw-queue In fio sync write test over virtio-blk(4 hw queues, ioengine=sync, iodepth=64, numjobs=4, bs=4K), it is observed that througput gets increased a lot over my test environment: - throughput: +70% in case of virtio-blk over null_blk - throughput: +30% in case of virtio-blk over SSD image The multi virtqueue feature isn't merged to QEMU yet, and patches for the feature can be found in below tree: git://kernel.ubuntu.com/ming/qemu.git v2.1.0-mq.4 And simply passing 'num_queues=4 vectors=5' should be enough to enable multi queue(quad queue) feature for QEMU virtio-blk. Suggested-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Ming Lei 提交于
This patch adds 'blk_mq_ctx' parameter to blk_get_flush_queue(), so that this function can find the corresponding blk_flush_queue bound with current mq context since the flush queue will become per hw-queue. For legacy queue, the parameter can be simply 'NULL'. For multiqueue case, the parameter should be set as the context from which the related request is originated. With this context info, the hw queue and related flush queue can be found easily. Signed-off-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Ming Lei 提交于
Now mission of the two helpers is over, and just call blk_alloc_flush_queue() and blk_free_flush_queue() directly. Signed-off-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Ming Lei 提交于
This patch introduces 'struct blk_flush_queue' and puts all flush machinery related fields into this structure, so that - flush implementation details aren't exposed to driver - it is easy to convert to per dispatch-queue flush machinery This patch is basically a mechanical replacement. Reviewed-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Ming Lei 提交于
These two temporary functions are introduced for holding flush initialization and de-initialization, so that we can introduce 'flush queue' easier in the following patch. And once 'flush queue' and its allocation/free functions are ready, they will be removed for sake of code readability. Reviewed-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Ming Lei 提交于
It is reasonable to allocate flush req in blk_mq_init_flush(). Reviewed-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Ming Lei 提交于
Failure of initializing one hctx isn't handled, so this patch introduces blk_mq_init_hctx() and its pair to handle it explicitly. Also this patch makes code cleaner. Reviewed-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 25 9月, 2014 3 次提交
-
-
由 Tejun Heo 提交于
blk-mq uses percpu_ref for its usage counter which tracks the number of in-flight commands and used to synchronously drain the queue on freeze. percpu_ref shutdown takes measureable wallclock time as it involves a sched RCU grace period. This means that draining a blk-mq takes measureable wallclock time. One would think that this shouldn't matter as queue shutdown should be a rare event which takes place asynchronously w.r.t. userland. Unfortunately, SCSI probing involves synchronously setting up and then tearing down a lot of request_queues back-to-back for non-existent LUNs. This means that SCSI probing may take above ten seconds when scsi-mq is used. [ 0.949892] scsi host0: Virtio SCSI HBA [ 1.007864] scsi 0:0:0:0: Direct-Access QEMU QEMU HARDDISK 1.1. PQ: 0 ANSI: 5 [ 1.021299] scsi 0:0:1:0: Direct-Access QEMU QEMU HARDDISK 1.1. PQ: 0 ANSI: 5 [ 1.520356] tsc: Refined TSC clocksource calibration: 2491.910 MHz <stall> [ 16.186549] sd 0:0:0:0: Attached scsi generic sg0 type 0 [ 16.190478] sd 0:0:1:0: Attached scsi generic sg1 type 0 [ 16.194099] osd: LOADED open-osd 0.2.1 [ 16.203202] sd 0:0:0:0: [sda] 31457280 512-byte logical blocks: (16.1 GB/15.0 GiB) [ 16.208478] sd 0:0:0:0: [sda] Write Protect is off [ 16.211439] sd 0:0:0:0: [sda] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA [ 16.218771] sd 0:0:1:0: [sdb] 31457280 512-byte logical blocks: (16.1 GB/15.0 GiB) [ 16.223264] sd 0:0:1:0: [sdb] Write Protect is off [ 16.225682] sd 0:0:1:0: [sdb] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA This is also the reason why request_queues start in bypass mode which is ended on blk_register_queue() as shutting down a fully functional queue also involves a RCU grace period and the queues for non-existent SCSI devices never reach registration. blk-mq basically needs to do the same thing - start the mq in a degraded mode which is faster to shut down and then make it fully functional only after the queue reaches registration. percpu_ref recently grew facilities to force atomic operation until explicitly switched to percpu mode, which can be used for this purpose. This patch makes blk-mq initialize q->mq_usage_counter in atomic mode and switch it to percpu mode only once blk_register_queue() is reached. Note that this issue was previously worked around by 0a30288d ("blk-mq, percpu_ref: implement a kludge for SCSI blk-mq stall during probe") for v3.17. The temp fix was reverted in preparation of adding persistent atomic mode to percpu_ref by 9eca8046 ("Revert "blk-mq, percpu_ref: implement a kludge for SCSI blk-mq stall during probe""). This patch and the prerequisite percpu_ref changes will be merged during v3.18 devel cycle. Signed-off-by: NTejun Heo <tj@kernel.org> Reported-by: NChristoph Hellwig <hch@infradead.org> Link: http://lkml.kernel.org/g/20140919113815.GA10791@lst.de Fixes: add703fd ("blk-mq: use percpu_ref for mq usage count") Reviewed-by: NKent Overstreet <kmo@daterainc.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Johannes Weiner <hannes@cmpxchg.org>
-
由 Tejun Heo 提交于
With the recent addition of percpu_ref_reinit(), percpu_ref now can be used as a persistent switch which can be turned on and off repeatedly where turning off maps to killing the ref and waiting for it to drain; however, there currently isn't a way to initialize a percpu_ref in its off (killed and drained) state, which can be inconvenient for certain persistent switch use cases. Similarly, percpu_ref_switch_to_atomic/percpu() allow dynamic selection of operation mode; however, currently a newly initialized percpu_ref is always in percpu mode making it impossible to avoid the latency overhead of switching to atomic mode. This patch adds @flags to percpu_ref_init() and implements the following flags. * PERCPU_REF_INIT_ATOMIC : start ref in atomic mode * PERCPU_REF_INIT_DEAD : start ref killed and drained These flags should be able to serve the above two use cases. v2: target_core_tpg.c conversion was missing. Fixed. Signed-off-by: NTejun Heo <tj@kernel.org> Reviewed-by: NKent Overstreet <kmo@daterainc.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Christoph Hellwig <hch@infradead.org> Cc: Johannes Weiner <hannes@cmpxchg.org>
-
由 Tejun Heo 提交于
This reverts commit 0a30288d, which was a temporary fix for SCSI blk-mq stall issue. The following patches will fix the issue properly by introducing atomic mode to percpu_ref. Signed-off-by: NTejun Heo <tj@kernel.org> Cc: Kent Overstreet <kmo@daterainc.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Christoph Hellwig <hch@lst.de>
-
- 24 9月, 2014 1 次提交
-
-
由 Tejun Heo 提交于
blk-mq uses percpu_ref for its usage counter which tracks the number of in-flight commands and used to synchronously drain the queue on freeze. percpu_ref shutdown takes measureable wallclock time as it involves a sched RCU grace period. This means that draining a blk-mq takes measureable wallclock time. One would think that this shouldn't matter as queue shutdown should be a rare event which takes place asynchronously w.r.t. userland. Unfortunately, SCSI probing involves synchronously setting up and then tearing down a lot of request_queues back-to-back for non-existent LUNs. This means that SCSI probing may take more than ten seconds when scsi-mq is used. This will be properly fixed by implementing a mechanism to keep q->mq_usage_counter in atomic mode till genhd registration; however, that involves rather big updates to percpu_ref which is difficult to apply late in the devel cycle (v3.17-rc6 at the moment). As a stop-gap measure till the proper fix can be implemented in the next cycle, this patch introduces __percpu_ref_kill_expedited() and makes blk_mq_freeze_queue() use it. This is heavy-handed but should work for testing the experimental SCSI blk-mq implementation. Signed-off-by: NTejun Heo <tj@kernel.org> Reported-by: NChristoph Hellwig <hch@infradead.org> Link: http://lkml.kernel.org/g/20140919113815.GA10791@lst.de Fixes: add703fd ("blk-mq: use percpu_ref for mq usage count") Cc: Kent Overstreet <kmo@daterainc.com> Cc: Jens Axboe <axboe@kernel.dk> Tested-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 23 9月, 2014 14 次提交
-
-
由 Christoph Hellwig 提交于
Signed-off-by: NChristoph Hellwig <hch@lst.de> Moved blk_mq_rq_timed_out() definition to the private blk-mq.h header. Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Jens Axboe 提交于
It's not uncommon for crash dump kernels to be limited to 128MB or something low in that area. This is normally not a problem for devices as we don't use that much memory, but for some shared SCSI setups with huge queue depths, it can potentially fill most of memory with tons of request allocations. blk-mq does scale back when it fails to allocate memory, but it scales back just enough so that blk-mq succeeds. This could still leave the system with not enough memory to make any real progress. Check if we are in a kdump environment and limit the hardware queues and tag depth. Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Ming Lei 提交于
This patch removes two unnecessary blk_clear_rq_complete(), the REQ_ATOM_COMPLETE flag is cleared inside blk_mq_start_request(), so: - The blk_clear_rq_complete() in blk_flush_restore_request() needn't because the request will be freed later, and clearing it here may open a small race window with timeout. - The blk_clear_rq_complete() in blk_mq_requeue_request() isn't necessary too, even though REQ_ATOM_STARTED is cleared in __blk_mq_requeue_request(), in theory it still may cause a small race window with timeout since the two clear_bit() may be reordered. Signed-off-by: NMing Lei <ming.lei@canoical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Christoph Hellwig 提交于
Allow blk-mq to pass an argument to the timeout handler to indicate if we're timing out a reserved or regular command. For many drivers those need to be handled different. Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Christoph Hellwig 提交于
Duplicate the (small) timeout handler in blk-mq so that we can pass arguments more easily to the driver timeout handler. This enables the next patch. Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Christoph Hellwig 提交于
Don't do a kmalloc from timer to handle timeouts, chances are we could be under heavy load or similar and thus just miss out on the timeouts. Fortunately it is very easy to just iterate over all in use tags, and doing this properly actually cleans up the blk_mq_busy_iter API as well, and prepares us for the next patch by passing a reserved argument to the iterator. Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Christoph Hellwig 提交于
Now that we've changed the driver API on the submission side use the opportunity to fix up the name on the completion side to fit into the general scheme. Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Christoph Hellwig 提交于
When we call blk_mq_start_request from the core blk-mq code before calling into ->queue_rq there is a racy window where the timeout handler can hit before we've fully set up the driver specific part of the command. Move the call to blk_mq_start_request into the driver so the driver can start the request only once it is fully set up. Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Christoph Hellwig 提交于
Pass an explicit parameter for the last request in a batch to ->queue_rq instead of using a request flag. Besides being a cleaner and non-stateful interface this is also required for the next patch, which fixes the blk-mq I/O submission code to not start a time too early. Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Jens Axboe 提交于
When requests are retried due to hw or sw resource shortages, we often stop the associated hardware queue. So ensure that we restart the queues when running the requeue work, otherwise the queue run will be a no-op. Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Jens Axboe 提交于
__blk_mq_alloc_rq_maps() can be invoked multiple times, if we scale back the queue depth if we are low on memory. So don't clear set->tags when we fail, this is handled directly in the parent function, blk_mq_alloc_tag_set(). Reported-by: NRobert Elliott <Elliott@hp.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Christoph Hellwig 提交于
We should not insert requests into the flush state machine from blk_mq_insert_request. All incoming flush requests come through blk_{m,s}q_make_request and are handled there, while blk_execute_rq_nowait should only be called for BLOCK_PC requests. All other callers deal with requests that already went through the flush statemchine and shouldn't be reinserted into it. Reported-by: NRobert Elliott <Elliott@hp.com> Debugged-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NChristoph Hellwig <hch@lst.de> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 David Hildenbrand 提交于
This patch should fix the bug reported in https://lkml.org/lkml/2014/9/11/249. We have to initialize at least the atomic_flags and the cmd_flags when allocating storage for the requests. Otherwise blk_mq_timeout_check() might dereference uninitialized pointers when racing with the creation of a request. Also move the reset of cmd_flags for the initializing code to the point where a request is freed. So we will never end up with pending flush request indicators that might trigger dereferences of invalid pointers in blk_mq_timeout_check(). Cc: stable@vger.kernel.org Signed-off-by: NDavid Hildenbrand <dahi@linux.vnet.ibm.com> Reported-by: NPaulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com> Tested-by: NPaulo De Rezende Pinatti <ppinatti@linux.vnet.ibm.com> Acked-by: NChristian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
由 Jens Axboe 提交于
When we start the request, we set the deadline and flip the bits marking the request as started and non-complete. However, it's important that the deadline store is ordered before flipping the bits, otherwise we could have a small window where the request is marked started but with an invalid deadline. This can confuse the timeout handling. Suggested-by: NMing Lei <tom.leiming@gmail.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 10 9月, 2014 1 次提交
-
-
由 Jens Axboe 提交于
If we are running in a kdump environment, resources are scarce. For some SCSI setups with a huge set of shared tags, we run out of memory allocating what the drivers is asking for. So implement a scale back logic to reduce the tag depth for those cases, allowing the driver to successfully load. We should extend this to detect low memory situations, and implement a sane fallback for those (1 queue, 64 tags, or something like that). Tested-by: NRobert Elliott <elliott@hp.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 08 9月, 2014 1 次提交
-
-
由 Tejun Heo 提交于
Percpu allocator now supports allocation mask. Add @gfp to percpu_ref_init() so that !GFP_KERNEL allocation masks can be used with percpu_refs too. This patch doesn't make any functional difference. v2: blk-mq conversion was missing. Updated. Signed-off-by: NTejun Heo <tj@kernel.org> Cc: Kent Overstreet <koverstreet@google.com> Cc: Benjamin LaHaise <bcrl@kvack.org> Cc: Li Zefan <lizefan@huawei.com> Cc: Nicholas A. Bellinger <nab@linux-iscsi.org> Cc: Jens Axboe <axboe@kernel.dk>
-
- 04 9月, 2014 1 次提交
-
-
由 Robert Elliott 提交于
In blk-mq.c blk_mq_alloc_tag_set, if: set->tags = kmalloc_node() succeeds, but one of the blk_mq_init_rq_map() calls fails, goto out_unwind; needs to free set->tags so the caller is not obligated to do so. None of the current callers (null_blk, virtio_blk, virtio_blk, or the forthcoming scsi-mq) do so. set->tags needs to be set to NULL after doing so, so other tag cleanup logic doesn't try to free a stale pointer later. Also set it to NULL in blk_mq_free_tag_set. Tested with error injection on the forthcoming scsi-mq + hpsa combination. Signed-off-by: NRobert Elliott <elliott@hp.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 29 8月, 2014 1 次提交
-
-
由 Joe Lawrence 提交于
The blk_get_request function may fail in low-memory conditions or during device removal (even if __GFP_WAIT is set). To distinguish between these errors, modify the blk_get_request call stack to return the appropriate ERR_PTR. Verify that all callers check the return status and consider IS_ERR instead of a simple NULL pointer check. For consistency, make a similar change to the blk_mq_alloc_request leg of blk_get_request. It may fail if the queue is dead, or the caller was unwilling to wait. Signed-off-by: NJoe Lawrence <joe.lawrence@stratus.com> Acked-by: Jiri Kosina <jkosina@suse.cz> [for pktdvd] Acked-by: Boaz Harrosh <bharrosh@panasas.com> [for osd] Reviewed-by: NJeff Moyer <jmoyer@redhat.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 23 8月, 2014 1 次提交
-
-
由 Tony Battersby 提交于
This patch fixes code such as the following with scsi-mq enabled: rq = blk_get_request(...); blk_rq_set_block_pc(rq); rq->cmd = my_cmd_buffer; /* separate CDB buffer */ blk_execute_rq_nowait(...); Code like this appears in e.g. sg_start_req() in drivers/scsi/sg.c (for large CDBs only). Without this patch, scsi_mq_prep_fn() will set rq->cmd back to rq->__cmd, causing the wrong CDB to be sent to the device. Signed-off-by: NTony Battersby <tonyb@cybernetics.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-
- 22 8月, 2014 1 次提交
-
-
由 Tejun Heo 提交于
While converting to percpu_ref for freezing, add703fd ("blk-mq: use percpu_ref for mq usage count") incorrectly made blk_mq_freeze_queue() misbehave when freezing is nested due to percpu_ref_kill() being invoked on an already killed ref. Fix it by making blk_mq_freeze_queue() kill and kick the queue only for the outermost freeze attempt. All the nested ones can simply wait for the ref to reach zero. While at it, remove unnecessary @wake initialization from blk_mq_unfreeze_queue(). Signed-off-by: NTejun Heo <tj@kernel.org> Reported-by: NMing Lei <ming.lei@canonical.com> Signed-off-by: NJens Axboe <axboe@fb.com>
-