From e2d4c7a106453e8d64fc6c4a46eedf912441f927 Mon Sep 17 00:00:00 2001 From: Dan Ziemba Date: Thu, 2 Jul 2015 23:40:34 -0400 Subject: [PATCH] 4.0.7-2 --- ...loop-convert-to-per-device-workqueue.patch | 133 ++++++++++++++++++ ...oiding-too-many-pending-per-work-I-O.patch | 52 +++++++ PKGBUILD | 11 +- 3 files changed, 195 insertions(+), 1 deletion(-) create mode 100644 0001-block-loop-convert-to-per-device-workqueue.patch create mode 100644 0002-block-loop-avoiding-too-many-pending-per-work-I-O.patch diff --git a/0001-block-loop-convert-to-per-device-workqueue.patch b/0001-block-loop-convert-to-per-device-workqueue.patch new file mode 100644 index 0000000..3b426c7 --- /dev/null +++ b/0001-block-loop-convert-to-per-device-workqueue.patch @@ -0,0 +1,133 @@ +From ebcacf05168d9791d39c5434164c64b9ef937171 Mon Sep 17 00:00:00 2001 +From: Ming Lei +Date: Tue, 5 May 2015 19:49:54 +0800 +Subject: [PATCH 1/2] block: loop: convert to per-device workqueue + +Documentation/workqueue.txt: + If there is dependency among multiple work items used + during memory reclaim, they should be queued to separate + wq each with WQ_MEM_RECLAIM. + +Loop devices can be stacked, so we have to convert to per-device +workqueue. One example is Fedora live CD. + +Fixes: b5dd2f6047ca108001328aac0e8588edd15f1778 +Cc: stable@vger.kernel.org (v4.0) +Cc: Justin M. Forbes +Signed-off-by: Ming Lei +Acked-by: Tejun Heo +Signed-off-by: Jens Axboe +--- + drivers/block/loop.c | 30 ++++++++++++++---------------- + drivers/block/loop.h | 1 + + 2 files changed, 15 insertions(+), 16 deletions(-) + +diff --git a/drivers/block/loop.c b/drivers/block/loop.c +index d1f168b..fec06d5 100644 +--- a/drivers/block/loop.c ++++ b/drivers/block/loop.c +@@ -85,8 +85,6 @@ static DEFINE_MUTEX(loop_index_mutex); + static int max_part; + static int part_shift; + +-static struct workqueue_struct *loop_wq; +- + /* + * Transfer functions + */ +@@ -777,6 +775,12 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode, + size = get_loop_size(lo, file); + if ((loff_t)(sector_t)size != size) + goto out_putf; ++ error = -ENOMEM; ++ lo->wq = alloc_workqueue("kloopd%d", ++ WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 0, ++ lo->lo_number); ++ if (!lo->wq) ++ goto out_putf; + + error = 0; + +@@ -924,6 +928,8 @@ static int loop_clr_fd(struct loop_device *lo) + lo->lo_flags = 0; + if (!part_shift) + lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; ++ destroy_workqueue(lo->wq); ++ lo->wq = NULL; + mutex_unlock(&lo->lo_ctl_mutex); + /* + * Need not hold lo_ctl_mutex to fput backing file. +@@ -1477,9 +1483,13 @@ static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, + const struct blk_mq_queue_data *bd) + { + struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); ++ struct loop_device *lo = cmd->rq->q->queuedata; + + blk_mq_start_request(bd->rq); + ++ if (lo->lo_state != Lo_bound) ++ return -EIO; ++ + if (cmd->rq->cmd_flags & REQ_WRITE) { + struct loop_device *lo = cmd->rq->q->queuedata; + bool need_sched = true; +@@ -1493,9 +1503,9 @@ static int loop_queue_rq(struct blk_mq_hw_ctx *hctx, + spin_unlock_irq(&lo->lo_lock); + + if (need_sched) +- queue_work(loop_wq, &lo->write_work); ++ queue_work(lo->wq, &lo->write_work); + } else { +- queue_work(loop_wq, &cmd->read_work); ++ queue_work(lo->wq, &cmd->read_work); + } + + return BLK_MQ_RQ_QUEUE_OK; +@@ -1507,9 +1517,6 @@ static void loop_handle_cmd(struct loop_cmd *cmd) + struct loop_device *lo = cmd->rq->q->queuedata; + int ret = -EIO; + +- if (lo->lo_state != Lo_bound) +- goto failed; +- + if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) + goto failed; + +@@ -1858,13 +1865,6 @@ static int __init loop_init(void) + goto misc_out; + } + +- loop_wq = alloc_workqueue("kloopd", +- WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 0); +- if (!loop_wq) { +- err = -ENOMEM; +- goto misc_out; +- } +- + blk_register_region(MKDEV(LOOP_MAJOR, 0), range, + THIS_MODULE, loop_probe, NULL, NULL); + +@@ -1902,8 +1902,6 @@ static void __exit loop_exit(void) + blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); + unregister_blkdev(LOOP_MAJOR, "loop"); + +- destroy_workqueue(loop_wq); +- + misc_deregister(&loop_misc); + } + +diff --git a/drivers/block/loop.h b/drivers/block/loop.h +index 301c27f..49564ed 100644 +--- a/drivers/block/loop.h ++++ b/drivers/block/loop.h +@@ -54,6 +54,7 @@ struct loop_device { + gfp_t old_gfp_mask; + + spinlock_t lo_lock; ++ struct workqueue_struct *wq; + struct list_head write_cmd_head; + struct work_struct write_work; + bool write_started; +-- +2.4.5 + diff --git a/0002-block-loop-avoiding-too-many-pending-per-work-I-O.patch b/0002-block-loop-avoiding-too-many-pending-per-work-I-O.patch new file mode 100644 index 0000000..c02a0b3 --- /dev/null +++ b/0002-block-loop-avoiding-too-many-pending-per-work-I-O.patch @@ -0,0 +1,52 @@ +From 02a6b08b2706501fb0fe6ff2df2f7d8377f2f69b Mon Sep 17 00:00:00 2001 +From: Ming Lei +Date: Tue, 5 May 2015 19:49:55 +0800 +Subject: [PATCH 2/2] block: loop: avoiding too many pending per work I/O + +If there are too many pending per work I/O, too many +high priority work thread can be generated so that +system performance can be effected. + +This patch limits the max_active parameter of workqueue as 16. + +This patch fixes Fedora 22 live booting performance +regression when it is booted from squashfs over dm +based on loop, and looks the following reasons are +related with the problem: + +- not like other filesyststems(such as ext4), squashfs +is a bit special, and I observed that increasing I/O jobs +to access file in squashfs only improve I/O performance a +little, but it can make big difference for ext4 + +- nested loop: both squashfs.img and ext3fs.img are mounted +as loop block, and ext3fs.img is inside the squashfs + +- during booting, lots of tasks may run concurrently + +Fixes: b5dd2f6047ca108001328aac0e8588edd15f1778 +Cc: stable@vger.kernel.org (v4.0) +Cc: Justin M. Forbes +Signed-off-by: Ming Lei +Acked-by: Tejun Heo +Signed-off-by: Jens Axboe +--- + drivers/block/loop.c | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/drivers/block/loop.c b/drivers/block/loop.c +index fec06d5..daaaf70 100644 +--- a/drivers/block/loop.c ++++ b/drivers/block/loop.c +@@ -777,7 +777,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode, + goto out_putf; + error = -ENOMEM; + lo->wq = alloc_workqueue("kloopd%d", +- WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 0, ++ WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 16, + lo->lo_number); + if (!lo->wq) + goto out_putf; +-- +2.4.5 + diff --git a/PKGBUILD b/PKGBUILD index a75fd2d..250711a 100644 --- a/PKGBUILD +++ b/PKGBUILD @@ -3,7 +3,7 @@ pkgbase=linux-vfio _srcname=linux-4.0 pkgver=4.0.7 -pkgrel=1 +pkgrel=2 arch=('i686' 'x86_64') url="http://www.kernel.org/" license=('GPL2') @@ -17,6 +17,8 @@ source=("https://www.kernel.org/pub/linux/kernel/v4.x/${_srcname}.tar.xz" 'config' 'config.x86_64' # standard config files for mkinitcpio ramdisk 'linux.preset' + '0001-block-loop-convert-to-per-device-workqueue.patch' + '0002-block-loop-avoiding-too-many-pending-per-work-I-O.patch' 'change-default-console-loglevel.patch' 'override_for_missing_acs_capabilities.patch' 'i915_317.patch' @@ -28,6 +30,8 @@ sha256sums=('0f2f7d44979bc8f71c4fc5d3308c03499c26a824dd311fdf6eef4dee0d7d5991' 'e8d639582697f22333a96aa1614bcf5d9bcf2e6683a3d5296f9cfc64843606f1' '5dadd75693e512b77f87f5620e470405b943373613eaf4df561037e9296453be' 'f0d90e756f14533ee67afda280500511a62465b4f76adcc5effa95a40045179c' + '0682df710e8d23f0d420b3b01fbfe409b3911940b1a379b78d9f4a5ac8590386' + 'af42b1456caee0b0db8f3cc770c78083b40159260b99db4930e503ac7824eacc' '1256b241cd477b265a3c2d64bdc19ffe3c9bbcee82ea3994c590c2c76e767d99' '975f79348119bfba8dd972a9fbfe6b38484c45bfd228f2f6d48a0c02426ba149' 'f86ce528b63f198b84c4d8d92d35329aa4000d462217dc2db03bac5eb693cf19') @@ -49,6 +53,11 @@ prepare() { # add latest fixes from stable queue, if needed # http://git.kernel.org/?p=linux/kernel/git/stable/stable-queue.git + # Fix deadlock with stacked loop devices (FS#45129) + # http://marc.info/?l=linux-kernel&m=143280649731902&w=2 + patch -Np1 -i ../0001-block-loop-convert-to-per-device-workqueue.patch + patch -Np1 -i ../0002-block-loop-avoiding-too-many-pending-per-work-I-O.patch + # set DEFAULT_CONSOLE_LOGLEVEL to 4 (same value as the 'quiet' kernel param) # remove this when a Kconfig knob is made available by upstream # (relevant patch sent upstream: https://lkml.org/lkml/2011/7/26/227)