From 815c2c543d3aeb914a361f981440ece552778724 Mon Sep 17 00:00:00 2001 From: Shaohua Li Date: Wed, 11 Sep 2013 14:20:30 -0700 Subject: swap: make swap discard async swap can do cluster discard for SSD, which is good, but there are some problems here: 1. swap do the discard just before page reclaim gets a swap entry and writes the disk sectors. This is useless for high end SSD, because an overwrite to a sector implies a discard to original sector too. A discard + overwrite == overwrite. 2. the purpose of doing discard is to improve SSD firmware garbage collection. Idealy we should send discard as early as possible, so firmware can do something smart. Sending discard just after swap entry is freed is considered early compared to sending discard before write. Of course, if workload is already bound to gc speed, sending discard earlier or later doesn't make 3. block discard is a sync API, which will delay scan_swap_map() significantly. 4. Write and discard command can be executed parallel in PCIe SSD. Making swap discard async can make execution more efficiently. This patch makes swap discard async and moves discard to where swap entry is freed. Discard and write have no dependence now, so above issues can be avoided. Idealy we should do discard for any freed sectors, but some SSD discard is very slow. This patch still does discard for a whole cluster. My test does a several round of 'mmap, write, unmap', which will trigger a lot of swap discard. In a fusionio card, with this patch, the test runtime is reduced to 18% of the time without it, so around 5.5x faster. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Shaohua Li Cc: Rik van Riel Cc: Minchan Kim Cc: Kyungmin Park Cc: Hugh Dickins Cc: Rafael Aquini Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/swapfile.c | 192 ++++++++++++++++++++++++++++++++++------------------------ 1 file changed, 114 insertions(+), 78 deletions(-) (limited to 'mm') diff --git a/mm/swapfile.c b/mm/swapfile.c index d1fbeb486de5..dac47c66055c 100644 --- a/mm/swapfile.c +++ b/mm/swapfile.c @@ -175,12 +175,6 @@ static void discard_swap_cluster(struct swap_info_struct *si, } } -static int wait_for_discard(void *word) -{ - schedule(); - return 0; -} - #define SWAPFILE_CLUSTER 256 #define LATENCY_LIMIT 256 @@ -242,6 +236,90 @@ static inline void cluster_set_null(struct swap_cluster_info *info) info->data = 0; } +/* Add a cluster to discard list and schedule it to do discard */ +static void swap_cluster_schedule_discard(struct swap_info_struct *si, + unsigned int idx) +{ + /* + * If scan_swap_map() can't find a free cluster, it will check + * si->swap_map directly. To make sure the discarding cluster isn't + * taken by scan_swap_map(), mark the swap entries bad (occupied). It + * will be cleared after discard + */ + memset(si->swap_map + idx * SWAPFILE_CLUSTER, + SWAP_MAP_BAD, SWAPFILE_CLUSTER); + + if (cluster_is_null(&si->discard_cluster_head)) { + cluster_set_next_flag(&si->discard_cluster_head, + idx, 0); + cluster_set_next_flag(&si->discard_cluster_tail, + idx, 0); + } else { + unsigned int tail = cluster_next(&si->discard_cluster_tail); + cluster_set_next(&si->cluster_info[tail], idx); + cluster_set_next_flag(&si->discard_cluster_tail, + idx, 0); + } + + schedule_work(&si->discard_work); +} + +/* + * Doing discard actually. After a cluster discard is finished, the cluster + * will be added to free cluster list. caller should hold si->lock. +*/ +static void swap_do_scheduled_discard(struct swap_info_struct *si) +{ + struct swap_cluster_info *info; + unsigned int idx; + + info = si->cluster_info; + + while (!cluster_is_null(&si->discard_cluster_head)) { + idx = cluster_next(&si->discard_cluster_head); + + cluster_set_next_flag(&si->discard_cluster_head, + cluster_next(&info[idx]), 0); + if (cluster_next(&si->discard_cluster_tail) == idx) { + cluster_set_null(&si->discard_cluster_head); + cluster_set_null(&si->discard_cluster_tail); + } + spin_unlock(&si->lock); + + discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, + SWAPFILE_CLUSTER); + + spin_lock(&si->lock); + cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE); + if (cluster_is_null(&si->free_cluster_head)) { + cluster_set_next_flag(&si->free_cluster_head, + idx, 0); + cluster_set_next_flag(&si->free_cluster_tail, + idx, 0); + } else { + unsigned int tail; + + tail = cluster_next(&si->free_cluster_tail); + cluster_set_next(&info[tail], idx); + cluster_set_next_flag(&si->free_cluster_tail, + idx, 0); + } + memset(si->swap_map + idx * SWAPFILE_CLUSTER, + 0, SWAPFILE_CLUSTER); + } +} + +static void swap_discard_work(struct work_struct *work) +{ + struct swap_info_struct *si; + + si = container_of(work, struct swap_info_struct, discard_work); + + spin_lock(&si->lock); + swap_do_scheduled_discard(si); + spin_unlock(&si->lock); +} + /* * The cluster corresponding to page_nr will be used. The cluster will be * removed from free cluster list and its usage counter will be increased. @@ -287,6 +365,16 @@ static void dec_cluster_info_page(struct swap_info_struct *p, cluster_count(&cluster_info[idx]) - 1); if (cluster_count(&cluster_info[idx]) == 0) { + /* + * If the swap is discardable, prepare discard the cluster + * instead of free it immediately. The cluster will be freed + * after discard. + */ + if (p->flags & SWP_PAGE_DISCARD) { + swap_cluster_schedule_discard(p, idx); + return; + } + cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); if (cluster_is_null(&p->free_cluster_head)) { cluster_set_next_flag(&p->free_cluster_head, idx, 0); @@ -319,7 +407,6 @@ static unsigned long scan_swap_map(struct swap_info_struct *si, unsigned long scan_base; unsigned long last_in_cluster = 0; int latency_ration = LATENCY_LIMIT; - int found_free_cluster = 0; /* * We try to cluster swap pages by allocating them sequentially @@ -340,19 +427,6 @@ static unsigned long scan_swap_map(struct swap_info_struct *si, si->cluster_nr = SWAPFILE_CLUSTER - 1; goto checks; } - if (si->flags & SWP_PAGE_DISCARD) { - /* - * Start range check on racing allocations, in case - * they overlap the cluster we eventually decide on - * (we scan without swap_lock to allow preemption). - * It's hardly conceivable that cluster_nr could be - * wrapped during our scan, but don't depend on it. - */ - if (si->lowest_alloc) - goto checks; - si->lowest_alloc = si->max; - si->highest_alloc = 0; - } check_cluster: if (!cluster_is_null(&si->free_cluster_head)) { offset = cluster_next(&si->free_cluster_head) * @@ -360,15 +434,27 @@ check_cluster: last_in_cluster = offset + SWAPFILE_CLUSTER - 1; si->cluster_next = offset; si->cluster_nr = SWAPFILE_CLUSTER - 1; - found_free_cluster = 1; goto checks; } else if (si->cluster_info) { + /* + * we don't have free cluster but have some clusters in + * discarding, do discard now and reclaim them + */ + if (!cluster_is_null(&si->discard_cluster_head)) { + si->cluster_nr = 0; + swap_do_scheduled_discard(si); + scan_base = offset = si->cluster_next; + if (!si->cluster_nr) + goto check_cluster; + si->cluster_nr--; + goto checks; + } + /* * Checking free cluster is fast enough, we can do the * check every time */ si->cluster_nr = 0; - si->lowest_alloc = 0; goto checks; } @@ -395,7 +481,6 @@ check_cluster: offset -= SWAPFILE_CLUSTER - 1; si->cluster_next = offset; si->cluster_nr = SWAPFILE_CLUSTER - 1; - found_free_cluster = 1; goto checks; } if (unlikely(--latency_ration < 0)) { @@ -416,7 +501,6 @@ check_cluster: offset -= SWAPFILE_CLUSTER - 1; si->cluster_next = offset; si->cluster_nr = SWAPFILE_CLUSTER - 1; - found_free_cluster = 1; goto checks; } if (unlikely(--latency_ration < 0)) { @@ -428,7 +512,6 @@ check_cluster: offset = scan_base; spin_lock(&si->lock); si->cluster_nr = SWAPFILE_CLUSTER - 1; - si->lowest_alloc = 0; } checks: @@ -470,59 +553,6 @@ checks: si->cluster_next = offset + 1; si->flags -= SWP_SCANNING; - if (si->lowest_alloc) { - /* - * Only set when SWP_PAGE_DISCARD, and there's a scan - * for a free cluster in progress or just completed. - */ - if (found_free_cluster) { - /* - * To optimize wear-levelling, discard the - * old data of the cluster, taking care not to - * discard any of its pages that have already - * been allocated by racing tasks (offset has - * already stepped over any at the beginning). - */ - if (offset < si->highest_alloc && - si->lowest_alloc <= last_in_cluster) - last_in_cluster = si->lowest_alloc - 1; - si->flags |= SWP_DISCARDING; - spin_unlock(&si->lock); - - if (offset < last_in_cluster) - discard_swap_cluster(si, offset, - last_in_cluster - offset + 1); - - spin_lock(&si->lock); - si->lowest_alloc = 0; - si->flags &= ~SWP_DISCARDING; - - smp_mb(); /* wake_up_bit advises this */ - wake_up_bit(&si->flags, ilog2(SWP_DISCARDING)); - - } else if (si->flags & SWP_DISCARDING) { - /* - * Delay using pages allocated by racing tasks - * until the whole discard has been issued. We - * could defer that delay until swap_writepage, - * but it's easier to keep this self-contained. - */ - spin_unlock(&si->lock); - wait_on_bit(&si->flags, ilog2(SWP_DISCARDING), - wait_for_discard, TASK_UNINTERRUPTIBLE); - spin_lock(&si->lock); - } else { - /* - * Note pages allocated by racing tasks while - * scan for a free cluster is in progress, so - * that its final discard can exclude them. - */ - if (offset < si->lowest_alloc) - si->lowest_alloc = offset; - if (offset > si->highest_alloc) - si->highest_alloc = offset; - } - } return offset; scan: @@ -1806,6 +1836,8 @@ SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) goto out_dput; } + flush_work(&p->discard_work); + destroy_swap_extents(p); if (p->flags & SWP_CONTINUED) free_swap_count_continuations(p); @@ -2172,6 +2204,8 @@ static int setup_swap_map_and_extents(struct swap_info_struct *p, cluster_set_null(&p->free_cluster_head); cluster_set_null(&p->free_cluster_tail); + cluster_set_null(&p->discard_cluster_head); + cluster_set_null(&p->discard_cluster_tail); for (i = 0; i < swap_header->info.nr_badpages; i++) { unsigned int page_nr = swap_header->info.badpages[i]; @@ -2281,6 +2315,8 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) if (IS_ERR(p)) return PTR_ERR(p); + INIT_WORK(&p->discard_work, swap_discard_work); + name = getname(specialfile); if (IS_ERR(name)) { error = PTR_ERR(name); -- cgit v1.2.3