From 020b0f0a31920e5b7e7e120d4560453b67b70733 Mon Sep 17 00:00:00 2001 From: Ming Lei Date: Fri, 22 Jan 2021 10:33:17 +0800 Subject: scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval Cc: Kashyap Desai Cc: Sumanesh Samanta Cc: Ewan D. Milne Cc: Hannes Reinecke Tested-by: Sumanesh Samanta Reviewed-by: Hannes Reinecke Signed-off-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_scan.c | 23 +++++++++++++++++++++-- 1 file changed, 21 insertions(+), 2 deletions(-) (limited to 'drivers/scsi/scsi_scan.c') diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c index 9af50e6f94c4..9f1b7f3c650a 100644 --- a/drivers/scsi/scsi_scan.c +++ b/drivers/scsi/scsi_scan.c @@ -215,6 +215,7 @@ static void scsi_unlock_floptical(struct scsi_device *sdev, static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget, u64 lun, void *hostdata) { + unsigned int depth; struct scsi_device *sdev; int display_failure_msg = 1, ret; struct Scsi_Host *shost = dev_to_shost(starget->dev.parent); @@ -276,8 +277,25 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget, WARN_ON_ONCE(!blk_get_queue(sdev->request_queue)); sdev->request_queue->queuedata = sdev; - scsi_change_queue_depth(sdev, sdev->host->cmd_per_lun ? - sdev->host->cmd_per_lun : 1); + depth = sdev->host->cmd_per_lun ?: 1; + + /* + * Use .can_queue as budget map's depth because we have to + * support adjusting queue depth from sysfs. Meantime use + * default device queue depth to figure out sbitmap shift + * since we use this queue depth most of times. + */ + if (sbitmap_init_node(&sdev->budget_map, + scsi_device_max_queue_depth(sdev), + sbitmap_calculate_shift(depth), + GFP_KERNEL, sdev->request_queue->node, + false, true)) { + put_device(&starget->dev); + kfree(sdev); + goto out; + } + + scsi_change_queue_depth(sdev, depth); scsi_sysfs_device_initialize(sdev); @@ -979,6 +997,7 @@ static int scsi_add_lun(struct scsi_device *sdev, unsigned char *inq_result, scsi_attach_vpd(sdev); sdev->max_queue_depth = sdev->queue_depth; + WARN_ON_ONCE(sdev->max_queue_depth > sdev->budget_map.depth); sdev->sdev_bflags = *bflags; /* -- cgit v1.2.3