projects / pve-kernel-jessie.git / blame
| Commit | Line | Data |
|---|---|---|
| d354e29e FG |
1 | From 3168fc7faf603da9d523c9dffbec6fee5b1a8a04 Mon Sep 17 00:00:00 2001 |
| 2 | From: =?UTF-8?q?Fabian=20Gr=C3=BCnbichler?= <f.gruenbichler@proxmox.com> | |
| 3 | Date: Wed, 4 Jan 2017 11:29:00 +0100 | |
| 4 | Subject: [PATCH 1/2] Revert "mm: throttle on IO only when there are too many | |
| 5 | dirty and writeback pages" | |
| 6 | MIME-Version: 1.0 | |
| 7 | Content-Type: text/plain; charset=UTF-8 | |
| 8 | Content-Transfer-Encoding: 8bit | |
| 9 | ||
| 10 | This reverts commit 57e9ef475661f46769cad6c0ed9a13f0cec1dbd8. | |
| 11 | ||
| 12 | Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com> | |
| 13 | --- | |
| 14 | mm/backing-dev.c | 20 +++++++++++++++++--- | |
| 15 | mm/page_alloc.c | 41 ++++------------------------------------- | |
| 16 | 2 files changed, 21 insertions(+), 40 deletions(-) | |
| 17 | ||
| 18 | diff --git a/mm/backing-dev.c b/mm/backing-dev.c | |
| 19 | index a1aef87..9ef80bf 100644 | |
| 20 | --- a/mm/backing-dev.c | |
| 21 | +++ b/mm/backing-dev.c | |
| 22 | @@ -976,8 +976,9 @@ EXPORT_SYMBOL(congestion_wait); | |
| 23 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
| 24 | * or a write to complete. | |
| 25 | * | |
| 26 | - * In the absence of zone congestion, cond_resched() is called to yield | |
| 27 | - * the processor if necessary but otherwise does not sleep. | |
| 28 | + * In the absence of zone congestion, a short sleep or a cond_resched is | |
| 29 | + * performed to yield the processor and to allow other subsystems to make | |
| 30 | + * a forward progress. | |
| 31 | * | |
| 32 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
| 33 | * it is the number of jiffies that were still remaining when the function | |
| 34 | @@ -997,7 +998,20 @@ long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
| 35 | */ | |
| 36 | if (atomic_read(&nr_wb_congested[sync]) == 0 || | |
| 37 | !test_bit(ZONE_CONGESTED, &zone->flags)) { | |
| 38 | - cond_resched(); | |
| 39 | + | |
| 40 | + /* | |
| 41 | + * Memory allocation/reclaim might be called from a WQ | |
| 42 | + * context and the current implementation of the WQ | |
| 43 | + * concurrency control doesn't recognize that a particular | |
| 44 | + * WQ is congested if the worker thread is looping without | |
| 45 | + * ever sleeping. Therefore we have to do a short sleep | |
| 46 | + * here rather than calling cond_resched(). | |
| 47 | + */ | |
| 48 | + if (current->flags & PF_WQ_WORKER) | |
| 49 | + schedule_timeout_uninterruptible(1); | |
| 50 | + else | |
| 51 | + cond_resched(); | |
| 52 | + | |
| 53 | /* In case we scheduled, work out time remaining */ | |
| 54 | ret = timeout - (jiffies - start); | |
| 55 | if (ret < 0) | |
| 56 | diff --git a/mm/page_alloc.c b/mm/page_alloc.c | |
| 57 | index aadbd7e..f13b503 100644 | |
| 58 | --- a/mm/page_alloc.c | |
| 59 | +++ b/mm/page_alloc.c | |
| 60 | @@ -3038,9 +3038,8 @@ should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
| 61 | for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx, | |
| 62 | ac->nodemask) { | |
| 63 | unsigned long available; | |
| 64 | - unsigned long reclaimable; | |
| 65 | ||
| 66 | - available = reclaimable = zone_reclaimable_pages(zone); | |
| 67 | + available = zone_reclaimable_pages(zone); | |
| 68 | available -= DIV_ROUND_UP(no_progress_loops * available, | |
| 69 | MAX_RECLAIM_RETRIES); | |
| 70 | available += zone_page_state_snapshot(zone, NR_FREE_PAGES); | |
| 71 | @@ -3050,41 +3049,9 @@ should_reclaim_retry(gfp_t gfp_mask, unsigned order, | |
| 72 | * available? | |
| 73 | */ | |
| 74 | if (__zone_watermark_ok(zone, order, min_wmark_pages(zone), | |
| 75 | - ac->classzone_idx, alloc_flags, available)) { | |
| 76 | - /* | |
| 77 | - * If we didn't make any progress and have a lot of | |
| 78 | - * dirty + writeback pages then we should wait for | |
| 79 | - * an IO to complete to slow down the reclaim and | |
| 80 | - * prevent from pre mature OOM | |
| 81 | - */ | |
| 82 | - if (!did_some_progress) { | |
| 83 | - unsigned long writeback; | |
| 84 | - unsigned long dirty; | |
| 85 | - | |
| 86 | - writeback = zone_page_state_snapshot(zone, | |
| 87 | - NR_WRITEBACK); | |
| 88 | - dirty = zone_page_state_snapshot(zone, NR_FILE_DIRTY); | |
| 89 | - | |
| 90 | - if (2*(writeback + dirty) > reclaimable) { | |
| 91 | - congestion_wait(BLK_RW_ASYNC, HZ/10); | |
| 92 | - return true; | |
| 93 | - } | |
| 94 | - } | |
| 95 | - | |
| 96 | - /* | |
| 97 | - * Memory allocation/reclaim might be called from a WQ | |
| 98 | - * context and the current implementation of the WQ | |
| 99 | - * concurrency control doesn't recognize that | |
| 100 | - * a particular WQ is congested if the worker thread is | |
| 101 | - * looping without ever sleeping. Therefore we have to | |
| 102 | - * do a short sleep here rather than calling | |
| 103 | - * cond_resched(). | |
| 104 | - */ | |
| 105 | - if (current->flags & PF_WQ_WORKER) | |
| 106 | - schedule_timeout_uninterruptible(1); | |
| 107 | - else | |
| 108 | - cond_resched(); | |
| 109 | - | |
| 110 | + ac->high_zoneidx, alloc_flags, available)) { | |
| 111 | + /* Wait for some write requests to complete then retry */ | |
| 112 | + wait_iff_congested(zone, BLK_RW_ASYNC, HZ/50); | |
| 113 | return true; | |
| 114 | } | |
| 115 | } | |
| 116 | -- | |
| 117 | 2.1.4 | |
| 118 |