- 05 Apr, 2019 8 commits
-
-
Christoph Hellwig authored
Return the currently active bvec segment, potentially spanning multiple pages. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
-
Keith Busch authored
Signed-off-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de>
-
Keith Busch authored
We don't need to save the dma device as it's not used in the hot path and hasn't in a long time. Shrink the struct nvme_queue removing this unnecessary member. Signed-off-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de>
-
Keith Busch authored
A negative value for the cq_vector used to mean the queue is either disabled or a polled queue. However, we have a queue enabled flag, so the cq_vector had been serving double duty. Don't overload the meaning of cq_vector. Use a flag specific to the polled queues instead. Signed-off-by: Keith Busch <keith.busch@intel.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de>
-
Sagi Grimberg authored
TP 8000 says that the use of the SUCCESS flag depends on weather the controller support disabling sq_head pointer updates. Given that we support it by default, makes sense that we go the extra mile to actually use the SUCCESS flag. When we create the C2HData PDU header, we check if sqhd_disabled is set on our queue, if so, we set the SUCCESS flag in the PDU header and skip sending a completion response capsule. Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Oliver Smith-Denny <osmithde@cisco.com> Tested-by: Oliver Smith-Denny <osmithde@cisco.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
-
Gustavo A. R. Silva authored
Update the code to use a zero-sized array instead of a pointer in structure nvmet_fc_tgt_queue and use struct_size() in kzalloc(). Notice that one of the more common cases of allocation size calculations is finding the size of a structure that has a zero-sized array at the end, along with memory for some number of elements for that array. For example: struct foo { int stuff; struct boo entry[]; }; instance = kzalloc(sizeof(struct foo) + sizeof(struct boo) * count, GFP_KERNEL); Instead of leaving these open-coded and prone to type mistakes, we can now use the new struct_size() helper: instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL); This code was detected with the help of Coccinelle. Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com> Reviewed-by: James Smart <james.smart@broadcom.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
-
Christoph Hellwig authored
Use errno_to_nvme_status to convert from a negative errno to a nvme status field instead of going through a blk_status_t. Also remove the pointless status variable in nvmet_bdev_execute_write_zeroes. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
-
Max Gurtovoy authored
Use le16_to_cpu instead of le16_to_cpup and le64_to_cpu instead of le64_to_cpup. This will also align the code to nvme-core driver convention. Signed-off-by: Max Gurtovoy <maxg@mellanox.com> Signed-off-by: Christoph Hellwig <hch@lst.de>
-
- 04 Apr, 2019 1 commit
-
-
Johannes Thumshirn authored
With the introduction of BIO_NO_PAGE_REF we've used up all available bits in bio::bi_flags. Convert the defines of the flags to an enum and add a BUILD_BUG_ON() call to make sure no-one adds a new one and thus overrides the BVEC_POOL_IDX causing crashes. Reviewed-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
- 01 Apr, 2019 25 commits
-
-
NeilBrown authored
Currently if many flush requests are submitted to an md device is quick succession, they are serialized and can take a long to process them all. We don't really need to call flush all those times - a single flush call can satisfy all requests submitted before it started. So keep track of when the current flush started and when it finished, allow any pending flush that was requested before the flush started to complete without waiting any more. Test results from Xiao: Test is done on a raid10 device which is created by 4 SSDs. The tool is dbench. 1. The latest linux stable kernel Operation Count AvgLat MaxLat -------------------------------------------------- Deltree 768 10.509 78.305 Flush 2078376 0.013 10.094 Close 21787697 0.019 18.821 LockX 96580 0.007 3.184 Mkdir 384 0.008 0.062 Rename 1255883 0.191 23.534 ReadX 46495589 0.020 14.230 WriteX 14790591 7.123 60.706 Unlink 5989118 0.440 54.551 UnlockX 96580 0.005 2.736 FIND_FIRST 10393845 0.042 12.079 SET_FILE_INFORMATION 2415558 0.129 10.088 QUERY_FILE_INFORMATION 4711725 0.005 8.462 QUERY_PATH_INFORMATION 26883327 0.032 21.715 QUERY_FS_INFORMATION 4929409b 0.010 8.238 NTCreateX 29660080 0.100 53.268 Throughput 1034.88 MB/sec (sync open) 128 clients 128 procs max_latency=60.712 ms 2. With patch1 "Revert "MD: fix lock contention for flush bios"" Operation Count AvgLat MaxLat -------------------------------------------------- Deltree 256 8.326 36.761 Flush 693291 3.974 180.269 Close 7266404 0.009 36.929 LockX 32160 0.006 0.840 Mkdir 128 0.008 0.021 Rename 418755 0.063 29.945 ReadX 15498708 0.007 7.216 WriteX 4932310 22.482 267.928 Unlink 1997557 0.109 47.553 UnlockX 32160 0.004 1.110 FIND_FIRST 3465791 0.036 7.320 SET_FILE_INFORMATION 805825 0.015 1.561 QUERY_FILE_INFORMATION 1570950 0.005 2.403 QUERY_PATH_INFORMATION 8965483 0.013 14.277 QUERY_FS_INFORMATION 1643626 0.009 3.314 NTCreateX 9892174 0.061 41.278 Throughput 345.009 MB/sec (sync open) 128 clients 128 procs max_latency=267.939 m 3. With patch1 and patch2 Operation Count AvgLat MaxLat -------------------------------------------------- Deltree 768 9.570 54.588 Flush 2061354 0.666 15.102 Close 21604811 0.012 25.697 LockX 95770 0.007 1.424 Mkdir 384 0.008 0.053 Rename 1245411 0.096 12.263 ReadX 46103198 0.011 12.116 WriteX 14667988 7.375 60.069 Unlink 5938936 0.173 30.905 UnlockX 95770 0.005 4.147 FIND_FIRST 10306407 0.041 11.715 SET_FILE_INFORMATION 2395987 0.048 7.640 QUERY_FILE_INFORMATION 4672371 0.005 9.291 QUERY_PATH_INFORMATION 26656735 0.018 19.719 QUERY_FS_INFORMATION 4887940 0.010 7.654 NTCreateX 29410811 0.059 28.551 Throughput 1026.21 MB/sec (sync open) 128 clients 128 procs max_latency=60.075 ms Cc: <stable@vger.kernel.org> # v4.19+ Tested-by: Xiao Ni <xni@redhat.com> Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: Song Liu <songliubraving@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
NeilBrown authored
This reverts commit 5a409b4f. This patch has two problems. 1/ it make multiple calls to submit_bio() from inside a make_request_fn. The bios thus submitted will be queued on current->bio_list and not submitted immediately. As the bios are allocated from a mempool, this can theoretically result in a deadlock - all the pool of requests could be in various ->bio_list queues and a subsequent mempool_alloc could block waiting for one of them to be released. 2/ It aims to handle a case when there are many concurrent flush requests. It handles this by submitting many requests in parallel - all of which are identical and so most of which do nothing useful. It would be more efficient to just send one lower-level request, but allow that to satisfy multiple upper-level requests. Fixes: 5a409b4f ("MD: fix lock contention for flush bios") Cc: <stable@vger.kernel.org> # v4.19+ Tested-by: Xiao Ni <xni@redhat.com> Signed-off-by: NeilBrown <neilb@suse.com> Signed-off-by: Song Liu <songliubraving@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Nigel Croxon authored
Changing state from check_state_check_result to check_state_compute_result not only is unsafe but also doesn't appear to serve a valid purpose. A raid6 check should only be pushing out extra writes if doing repair and a mis-match occurs. The stripe dev management will already try and do repair writes for failing sectors. This patch makes the raid6 check_state_check_result handling work more like raid5's. If somehow too many failures for a check, just quit the check operation for the stripe. When any checks pass, don't try and use check_state_compute_result for a purpose it isn't needed for and is unsafe for. Just mark the stripe as in sync for passing its parity checks and let the stripe dev read/write code and the bad blocks list do their job handling I/O errors. Repro steps from Xiao: These are the steps to reproduce this problem: 1. redefined OPT_MEDIUM_ERR_ADDR to 12000 in scsi_debug.c 2. insmod scsi_debug.ko dev_size_mb=11000 max_luns=1 num_tgts=1 3. mdadm --create /dev/md127 --level=6 --raid-devices=5 /dev/sde1 /dev/sde2 /dev/sde3 /dev/sde5 /dev/sde6 sde is the disk created by scsi_debug 4. echo "2" >/sys/module/scsi_debug/parameters/opts 5. raid-check It panic: [ 4854.730899] md: data-check of RAID array md127 [ 4854.857455] sd 5:0:0:0: [sdr] tag#80 FAILED Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE [ 4854.859246] sd 5:0:0:0: [sdr] tag#80 Sense Key : Medium Error [current] [ 4854.860694] sd 5:0:0:0: [sdr] tag#80 Add. Sense: Unrecovered read error [ 4854.862207] sd 5:0:0:0: [sdr] tag#80 CDB: Read(10) 28 00 00 00 2d 88 00 04 00 00 [ 4854.864196] print_req_error: critical medium error, dev sdr, sector 11656 flags 0 [ 4854.867409] sd 5:0:0:0: [sdr] tag#100 FAILED Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE [ 4854.869469] sd 5:0:0:0: [sdr] tag#100 Sense Key : Medium Error [current] [ 4854.871206] sd 5:0:0:0: [sdr] tag#100 Add. Sense: Unrecovered read error [ 4854.872858] sd 5:0:0:0: [sdr] tag#100 CDB: Read(10) 28 00 00 00 2e e0 00 00 08 00 [ 4854.874587] print_req_error: critical medium error, dev sdr, sector 12000 flags 4000 [ 4854.876456] sd 5:0:0:0: [sdr] tag#101 FAILED Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE [ 4854.878552] sd 5:0:0:0: [sdr] tag#101 Sense Key : Medium Error [current] [ 4854.880278] sd 5:0:0:0: [sdr] tag#101 Add. Sense: Unrecovered read error [ 4854.881846] sd 5:0:0:0: [sdr] tag#101 CDB: Read(10) 28 00 00 00 2e e8 00 00 08 00 [ 4854.883691] print_req_error: critical medium error, dev sdr, sector 12008 flags 4000 [ 4854.893927] sd 5:0:0:0: [sdr] tag#166 FAILED Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE [ 4854.896002] sd 5:0:0:0: [sdr] tag#166 Sense Key : Medium Error [current] [ 4854.897561] sd 5:0:0:0: [sdr] tag#166 Add. Sense: Unrecovered read error [ 4854.899110] sd 5:0:0:0: [sdr] tag#166 CDB: Read(10) 28 00 00 00 2e e0 00 00 10 00 [ 4854.900989] print_req_error: critical medium error, dev sdr, sector 12000 flags 0 [ 4854.902757] md/raid:md127: read error NOT corrected!! (sector 9952 on sdr1). [ 4854.904375] md/raid:md127: read error NOT corrected!! (sector 9960 on sdr1). [ 4854.906201] ------------[ cut here ]------------ [ 4854.907341] kernel BUG at drivers/md/raid5.c:4190! raid5.c:4190 above is this BUG_ON: handle_parity_checks6() ... BUG_ON(s->uptodate < disks - 1); /* We don't need Q to recover */ Cc: <stable@vger.kernel.org> # v3.16+ OriginalAuthor: David Jeffery <djeffery@redhat.com> Cc: Xiao Ni <xni@redhat.com> Tested-by: David Jeffery <djeffery@redhat.com> Signed-off-by: David Jeffy <djeffery@redhat.com> Signed-off-by: Nigel Croxon <ncroxon@redhat.com> Signed-off-by: Song Liu <songliubraving@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
loop is one block device, for any bio submitted to this device, the upper layer does guarantee that pages added to loop's bio won't go away when the bio is in-flight. So mark loop's bvec as ITER_BVEC_FLAG_NO_REF then get_page/put_page can be saved for serving loop's IO. Cc: linux-fsdevel@vger.kernel.org Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
Now both passthrough and FS IO have supported multi-page bvec, and bvec merging has been handled actually when adding page to bio, then adjacent bvecs won't be mergeable any more if they belong to same bio. So only try to merge bvecs if they are from different bios. Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
Inside __blk_segment_map_sg(), page sized bvec mapping is optimized a bit with one standalone branch. So reuse __blk_bvec_map_sg() to do that. Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
The argument of 'request_queue' isn't used by __blk_bvec_map_sg(), so remove it. Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
Now block IO stack is basically ready for supporting multi-page bvec, however it isn't enabled on passthrough IO. One reason is that passthrough IO is dispatched to LLD directly and bio split is bypassed, so the bio has to be built correctly for dispatch to LLD from the beginning. Implement multi-page support for passthrough IO by limitting each bvec as block device's segment and applying all kinds of queue limit in blk_add_pc_page(). Then we don't need to calculate segments any more for passthrough IO any more, turns out code is simplified much. Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
When the added page is merged to last same page in bio_add_pc_page(), the user may need to put this page for avoiding page leak. bio_map_user_iov() needs this kind of handling, and now it deals with it by itself in hack style. Moves the handling of put page into __bio_add_pc_page(), so bio_map_user_iov() may be simplified a bit, and maybe more users can benefit from this change. Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
Now the check for deciding if one page is mergeable to current bvec becomes a bit complicated, and we need to reuse the code before adding pc page. So move the check in one dedicated helper. No function change. Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
REQ_PC is out of date, so replace it with passthrough IO. Also remove the local variable of 'prev' since we can reuse the top local variable of 'bvec'. No function change. Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
For normal filesystem IO, each page is added via blk_add_page(), in which bvec(page) merge has been handled already, and basically not possible to merge two adjacent bvecs in one bio. So not try to merge two adjacent bvecs in blk_queue_split(). Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
XEN has special page merge requirement, see xen_biovec_phys_mergeable(). We can't merge pages into one bvec simply for XEN. So move XEN's specific check on page merge into __bio_try_merge_page(), then abvoid to break XEN by multi-page bvec. Cc: ris Ostrovsky <boris.ostrovsky@oracle.com> Cc: xen-devel@lists.xenproject.org Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Juergen Gross <jgross@suse.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Ming Lei authored
xen_biovec_phys_mergeable() only needs .bv_page of the 2nd bio bvec for checking if the two bvecs can be merged, so pass page to xen_biovec_phys_mergeable() directly. No function change. Cc: ris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Juergen Gross <jgross@suse.com> Cc: xen-devel@lists.xenproject.org Cc: Omar Sandoval <osandov@fb.com> Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Holger Hoffstätte authored
The loop driver always declares the rotational flag of its device as rotational, even when the device of the mapped file is nonrotational, as is the case with SSDs or on tmpfs. This can confuse filesystem tools which are SSD-aware; in my case I frequently forget to tell mkfs.btrfs that my loop device on tmpfs is nonrotational, and that I really don't need any automatic metadata redundancy. The attached patch fixes this by introspecting the rotational flag of the mapped file's underlying block device, if it exists. If the mapped file's filesystem has no associated block device - as is the case on e.g. tmpfs - we assume nonrotational storage. If there is a better way to identify such non-devices I'd love to hear them. Cc: Jens Axboe <axboe@kernel.dk> Cc: linux-block@vger.kernel.org Cc: holger@applied-asynchrony.com Signed-off-by: Holger Hoffstätte <holger.hoffstaette@googlemail.com> Signed-off-by: Gwendal Grignou <gwendal@chromium.org> Signed-off-by: Benjamin Gordon <bmgordon@chromium.org> Reviewed-by: Guenter Roeck <groeck@chromium.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Paolo Valente authored
The execution time of BFQ has been slightly lowered. Report the new execution time in BFQ documentation. Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Francesco Pollicino authored
bfq saves the state of a queue each time a merge occurs, to be able to resume such a state when the queue is associated again with its original process, on a split. Unfortunately bfq does not save & restore also the weight of the queue. If the weight is not correctly resumed when the queue is recycled, then the weight of the recycled queue could differ from the weight of the original queue. This commit adds the missing save & resume of the weight. Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Signed-off-by: Francesco Pollicino <fra.fra.800@gmail.com> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Francesco Pollicino authored
The function "bfq_log_bfqq" prints the pid of the process associated with the queue passed as input. Unfortunately, if the queue is shared, then more than one process is associated with the queue. The pid that gets printed in this case is the pid of one of the associated processes. Which process gets printed depends on the exact sequence of merge events the queue underwent. So printing such a pid is rather useless and above all is often rather confusing because it reports a random pid between those of the associated processes. This commit addresses this issue by printing SHARED instead of a pid if the queue is shared. Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Signed-off-by: Francesco Pollicino <fra.fra.800@gmail.com> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Paolo Valente authored
If many bfq_queues belonging to the same group happen to be created shortly after each other, then the processes associated with these queues have typically a common goal. In particular, bursts of queue creations are usually caused by services or applications that spawn many parallel threads/processes. Examples are systemd during boot, or git grep. If there are no other active queues, then, to help these processes get their job done as soon as possible, the best thing to do is to reach a high throughput. To this goal, it is usually better to not grant either weight-raising or device idling to the queues associated with these processes. And this is exactly what BFQ currently does. There is however a drawback: if, in contrast, some other queues are already active, then the newly created queues must be protected from the I/O flowing through the already existing queues. In this case, the best thing to do is the opposite as in the other case: it is much better to grant weight-raising and device idling to the newly-created queues, if they deserve it. This commit addresses this issue by doing so if there are already other active queues. This change also helps eliminating false positives, which occur when the newly-created queues do not belong to an actual large burst of creations, but some background task (e.g., a service) happens to trigger the creation of new queues in the middle, i.e., very close to when the victim queues are created. These false positive may cause total loss of control on process latencies. Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Paolo Valente authored
Sync random I/O is likely to be confused with soft real-time I/O, because it is characterized by limited throughput and apparently isochronous arrival pattern. To avoid false positives, this commits prevents bfq_queues containing only random (seeky) I/O from being tagged as soft real-time. Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Paolo Valente authored
To boost throughput with a set of processes doing interleaved I/O (i.e., a set of processes whose individual I/O is random, but whose merged cumulative I/O is sequential), BFQ merges the queues associated with these processes, i.e., redirects the I/O of these processes into a common, shared queue. In the shared queue, I/O requests are ordered by their position on the medium, thus sequential I/O gets dispatched to the device when the shared queue is served. Queue merging costs execution time, because, to detect which queues to merge, BFQ must maintain a list of the head I/O requests of active queues, ordered by request positions. Measurements showed that this costs about 10% of BFQ's total per-request processing time. Request processing time becomes more and more critical as the speed of the underlying storage device grows. Yet, fortunately, queue merging is basically useless on the very devices that are so fast to make request processing time critical. To reach a high throughput, these devices must have many requests queued at the same time. But, in this configuration, the internal scheduling algorithms of these devices do also the job of queue merging: they reorder requests so as to obtain as much as possible a sequential I/O pattern. As a consequence, with processes doing interleaved I/O, the throughput reached by one such device is likely to be the same, with and without queue merging. In view of this fact, this commit disables queue merging, and all related housekeeping, for non-rotational devices with internal queueing. The total, single-lock-protected, per-request processing time of BFQ drops to, e.g., 1.9 us on an Intel Core i7-2760QM@2.40GHz (time measured with simple code instrumentation, and using the throughput-sync.sh script of the S suite [1], in performance-profiling mode). To put this result into context, the total, single-lock-protected, per-request execution time of the lightest I/O scheduler available in blk-mq, mq-deadline, is 0.7 us (mq-deadline is ~800 LOC, against ~10500 LOC for BFQ). Disabling merging provides a further, remarkable benefit in terms of throughput. Merging tends to make many workloads artificially more uneven, mainly because of shared queues remaining non empty for incomparably more time than normal queues. So, if, e.g., one of the queues in a set of merged queues has a higher weight than a normal queue, then the shared queue may inherit such a high weight and, by staying almost always active, may force BFQ to perform I/O plugging most of the time. This evidently makes it harder for BFQ to let the device reach a high throughput. As a practical example of this problem, and of the benefits of this commit, we measured again the throughput in the nasty scenario considered in previous commit messages: dbench test (in the Phoronix suite), with 6 clients, on a filesystem with journaling, and with the journaling daemon enjoying a higher weight than normal processes. With this commit, the throughput grows from ~150 MB/s to ~200 MB/s on a PLEXTOR PX-256M5 SSD. This is the same peak throughput reached by any of the other I/O schedulers. As such, this is also likely to be the maximum possible throughput reachable with this workload on this device, because I/O is mostly random, and the other schedulers basically just pass I/O requests to the drive as fast as possible. [1] https://github.com/Algodev-github/STested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Tested-by: Francesco Pollicino <fra.fra.800@gmail.com> Signed-off-by: Alessio Masola <alessio.masola@gmail.com> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Paolo Valente authored
The processes associated with a bfq_queue, say Q, may happen to generate their cumulative I/O at a lower rate than the rate at which the device could serve the same I/O. This is rather probable, e.g., if only one process is associated with Q and the device is an SSD. It results in Q becoming often empty while in service. If BFQ is not allowed to switch to another queue when Q becomes empty, then, during the service of Q, there will be frequent "service holes", i.e., time intervals during which Q gets empty and the device can only consume the I/O already queued in its hardware queues. This easily causes considerable losses of throughput. To counter this problem, BFQ implements a request injection mechanism, which tries to fill the above service holes with I/O requests taken from other bfq_queues. The hard part in this mechanism is finding the right amount of I/O to inject, so as to both boost throughput and not break Q's bandwidth and latency guarantees. To this goal, the current version of this mechanism measures the bandwidth enjoyed by Q while it is being served, and tries to inject the maximum possible amount of extra service that does not cause Q's bandwidth to decrease too much. This solution has an important shortcoming. For bandwidth measurements to be stable and reliable, Q must remain in service for a much longer time than that needed to serve a single I/O request. Unfortunately, this does not hold with many workloads. This commit addresses this issue by changing the way the amount of injection allowed is dynamically computed. It tunes injection as a function of the service times of single I/O requests of Q, instead of Q's bandwidth. Single-request service times are evidently meaningful even if Q gets very few I/O requests completed while it is in service. As a testbed for this new solution, we measured the throughput reached by BFQ for one of the nastiest workloads and configurations for this scheduler: the workload generated by the dbench test (in the Phoronix suite), with 6 clients, on a filesystem with journaling, and with the journaling daemon enjoying a higher weight than normal processes. With this commit, the throughput grows from ~100 MB/s to ~150 MB/s on a PLEXTOR PX-256M5. Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Tested-by: Francesco Pollicino <fra.fra.800@gmail.com> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Paolo Valente authored
In most cases, it is detrimental for throughput to plug I/O dispatch when the in-service bfq_queue becomes temporarily empty (plugging is performed to wait for the possible arrival, soon, of new I/O from the in-service queue). There is however a case where plugging is needed for service guarantees. If a bfq_queue, say Q, has a higher weight than some other active bfq_queue, and is sync, i.e., contains sync I/O, then, to guarantee that Q does receive a higher share of the throughput than other lower-weight queues, it is necessary to plug I/O dispatch when Q remains temporarily empty while being served. For this reason, BFQ performs I/O plugging when some active bfq_queue has a higher weight than some other active bfq_queue. But this is unnecessarily overkill. In fact, if the in-service bfq_queue actually has a weight lower than or equal to the other queues, then the queue *must not* be guaranteed a higher share of the throughput than the other queues. So, not plugging I/O cannot cause any harm to the queue. And can boost throughput. Taking advantage of this fact, this commit does not plug I/O for sync bfq_queues with a weight lower than or equal to the weights of the other queues. Here is an example of the resulting throughput boost with the dbench workload, which is particularly nasty for BFQ. With the dbench test in the Phoronix suite, BFQ reaches its lowest total throughput with 6 clients on a filesystem with journaling, in case the journaling daemon has a higher weight than normal processes. Before this commit, the total throughput was ~80 MB/sec on a PLEXTOR PX-256M5, after this commit it is ~100 MB/sec. Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Paolo Valente authored
If a sync bfq_queue has a higher weight than some other queue, and remains temporarily empty while in service, then, to preserve the bandwidth share of the queue, it is necessary to plug I/O dispatching until a new request arrives for the queue. In addition, a timeout needs to be set, to avoid waiting for ever if the process associated with the queue has actually finished its I/O. Even with the above timeout, the device is however not fed with new I/O for a while, if the process has finished its I/O. If this happens often, then throughput drops and latencies grow. For this reason, the timeout is kept rather low: 8 ms is the current default. Unfortunately, such a low value may cause, on the opposite end, a violation of bandwidth guarantees for a process that happens to issue new I/O too late. The higher the system load, the higher the probability that this happens to some process. This is a problem in scenarios where service guarantees matter more than throughput. One important case are weight-raised queues, which need to be granted a very high fraction of the bandwidth. To address this issue, this commit lower-bounds the plugging timeout for weight-raised queues to 20 ms. This simple change provides relevant benefits. For example, on a PLEXTOR PX-256M5S, with which gnome-terminal starts in 0.6 seconds if there is no other I/O in progress, the same applications starts in - 0.8 seconds, instead of 1.2 seconds, if ten files are being read sequentially in parallel - 1 second, instead of 2 seconds, if, in parallel, five files are being read sequentially, and five more files are being written sequentially Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Signed-off-by: Paolo Valente <paolo.valente@linaro.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
Konstantin Khlebnikov authored
Replace BFQ_GROUP_IOSCHED_ENABLED with CONFIG_BFQ_GROUP_IOSCHED. Code under these ifdefs never worked, something might be broken. Fixes: 0471559c ("block, bfq: add/remove entity weights correctly") Fixes: 73d58118 ("block, bfq: consider also ioprio classes in symmetry detection") Reviewed-by: Holger Hoffstätte <holger@applied-asynchrony.com> Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Signed-off-by: Jens Axboe <axboe@kernel.dk>
-
- 31 Mar, 2019 6 commits
-
-
Linus Torvalds authored
-
git://git.kernel.org/pub/scm/virt/kvm/kvmLinus Torvalds authored
Pull KVM fixes from Paolo Bonzini: "A collection of x86 and ARM bugfixes, and some improvements to documentation. On top of this, a cleanup of kvm_para.h headers, which were exported by some architectures even though they not support KVM at all. This is responsible for all the Kbuild changes in the diffstat" * tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (28 commits) Documentation: kvm: clarify KVM_SET_USER_MEMORY_REGION KVM: doc: Document the life cycle of a VM and its resources KVM: selftests: complete IO before migrating guest state KVM: selftests: disable stack protector for all KVM tests KVM: selftests: explicitly disable PIE for tests KVM: selftests: assert on exit reason in CR4/cpuid sync test KVM: x86: update %rip after emulating IO x86/kvm/hyper-v: avoid spurious pending stimer on vCPU init kvm/x86: Move MSR_IA32_ARCH_CAPABILITIES to array emulated_msrs KVM: x86: Emulate MSR_IA32_ARCH_CAPABILITIES on AMD hosts kvm: don't redefine flags as something else kvm: mmu: Used range based flushing in slot_handle_level_range KVM: export <linux/kvm_para.h> and <asm/kvm_para.h> iif KVM is supported KVM: x86: remove check on nr_mmu_pages in kvm_arch_commit_memory_region() kvm: nVMX: Add a vmentry check for HOST_SYSENTER_ESP and HOST_SYSENTER_EIP fields KVM: SVM: Workaround errata#1096 (insn_len maybe zero on SMAP violation) KVM: Reject device ioctls from processes other than the VM's creator KVM: doc: Fix incorrect word ordering regarding supported use of APIs KVM: x86: fix handling of role.cr4_pae and rename it to 'gpte_size' KVM: nVMX: Do not inherit quadrant and invalid for the root shadow EPT ...
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull x86 fixes from Thomas Gleixner: "A pile of x86 updates: - Prevent exceeding he valid physical address space in the /dev/mem limit checks. - Move all header content inside the header guard to prevent compile failures. - Fix the bogus __percpu annotation in this_cpu_has() which makes sparse very noisy. - Disable switch jump tables completely when retpolines are enabled. - Prevent leaking the trampoline address" * 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: x86/realmode: Make set_real_mode_mem() static inline x86/cpufeature: Fix __percpu annotation in this_cpu_has() x86/mm: Don't exceed the valid physical address space x86/retpolines: Disable switch jump tables when retpolines are enabled x86/realmode: Don't leak the trampoline kernel address x86/boot: Fix incorrect ifdeffery scope x86/resctrl: Remove unused variable
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull perf tooling fixes from Thomas Gleixner: "Core libraries: - Fix max perf_event_attr.precise_ip detection. - Fix parser error for uncore event alias - Fixup ordering of kernel maps after obtaining the main kernel map address. Intel PT: - Fix TSC slip where A TSC packet can slip past MTC packets so that the timestamp appears to go backwards. - Fixes for exported-sql-viewer GUI conversion to python3. ARM coresight: - Fix the build by adding a missing case value for enumeration value introduced in newer library, that now is the required one. tool headers: - Syncronize kernel headers with the kernel, getting new io_uring and pidfd_send_signal syscalls so that 'perf trace' can handle them" * 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: perf pmu: Fix parser error for uncore event alias perf scripts python: exported-sql-viewer.py: Fix python3 support perf scripts python: exported-sql-viewer.py: Fix never-ending loop perf machine: Update kernel map address and re-order properly tools headers uapi: Sync powerpc's asm/kvm.h copy with the kernel sources tools headers: Update x86's syscall_64.tbl and uapi/asm-generic/unistd tools headers uapi: Update drm/i915_drm.h tools arch x86: Sync asm/cpufeatures.h with the kernel sources tools headers uapi: Sync linux/fcntl.h to get the F_SEAL_FUTURE_WRITE addition tools headers uapi: Sync asm-generic/mman-common.h and linux/mman.h perf evsel: Fix max perf_event_attr.precise_ip detection perf intel-pt: Fix TSC slip perf cs-etm: Add missing case value
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull CPU hotplug fixes from Thomas Gleixner: "Two SMT/hotplug related fixes: - Prevent crash when HOTPLUG_CPU is disabled and the CPU bringup aborts. This is triggered with the 'nosmt' command line option, but can happen by any abort condition. As the real unplug code is not compiled in, prevent the fail by keeping the CPU in zombie state. - Enforce HOTPLUG_CPU for SMP on x86 to avoid the above situation completely. With 'nosmt' being a popular option it's required to unplug the half brought up sibling CPUs (due to the MCE wreckage) completely" * 'smp-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: x86/smp: Enforce CONFIG_HOTPLUG_CPU when SMP=y cpu/hotplug: Prevent crash when CPU bringup fails on CONFIG_HOTPLUG_CPU=n
-
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tipLinus Torvalds authored
Pull locking fixlet from Thomas Gleixner: "Trivial update to the maintainers file" * 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: MAINTAINERS: Remove deleted file from futex file pattern
-