From a56ed663047fc9927ec8b35750d23ece54f85dc7 Mon Sep 17 00:00:00 2001 From: Peter W Morreale Date: Mon, 6 Apr 2009 19:00:28 -0700 Subject: mm: fix pdflush thread creation upper bound Fix a race on creating pdflush threads. Without the patch, it is possible to create more than MAX_PDFLUSH_THREADS threads, and this has been observed in practice on IO loaded SMP machines. The fix involves moving the lock around to protect the check against the thread count and correctly dealing with thread creation failure. This fix also _mostly_ repairs a race condition on how quickly the threads are created. The original intent was to create a pdflush thread (up to the max allowed) every second. Without this patch is is possible to create NCPUS pdflush threads concurrently. The 'mostly' caveat is because an assumption is made that thread creation will be successful. If we fail to create the thread, the miss is not considered fatal. (we will try again in 1 second) Signed-off-by: Peter W Morreale Reviewed-by: Rik van Riel Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/pdflush.c | 30 ++++++++++++++++++++++++------ 1 file changed, 24 insertions(+), 6 deletions(-) (limited to 'mm/pdflush.c') diff --git a/mm/pdflush.c b/mm/pdflush.c index 118905e3d78..235ac440c44 100644 --- a/mm/pdflush.c +++ b/mm/pdflush.c @@ -98,7 +98,6 @@ static int __pdflush(struct pdflush_work *my_work) INIT_LIST_HEAD(&my_work->list); spin_lock_irq(&pdflush_lock); - nr_pdflush_threads++; for ( ; ; ) { struct pdflush_work *pdf; @@ -126,20 +125,26 @@ static int __pdflush(struct pdflush_work *my_work) (*my_work->fn)(my_work->arg0); + spin_lock_irq(&pdflush_lock); + /* * Thread creation: For how long have there been zero * available threads? + * + * To throttle creation, we reset last_empty_jifs. */ if (time_after(jiffies, last_empty_jifs + 1 * HZ)) { - /* unlocked list_empty() test is OK here */ if (list_empty(&pdflush_list)) { - /* unlocked test is OK here */ - if (nr_pdflush_threads < MAX_PDFLUSH_THREADS) + if (nr_pdflush_threads < MAX_PDFLUSH_THREADS) { + last_empty_jifs = jiffies; + nr_pdflush_threads++; + spin_unlock_irq(&pdflush_lock); start_one_pdflush_thread(); + spin_lock_irq(&pdflush_lock); + } } } - spin_lock_irq(&pdflush_lock); my_work->fn = NULL; /* @@ -236,13 +241,26 @@ int pdflush_operation(void (*fn)(unsigned long), unsigned long arg0) static void start_one_pdflush_thread(void) { - kthread_run(pdflush, NULL, "pdflush"); + struct task_struct *k; + + k = kthread_run(pdflush, NULL, "pdflush"); + if (unlikely(IS_ERR(k))) { + spin_lock_irq(&pdflush_lock); + nr_pdflush_threads--; + spin_unlock_irq(&pdflush_lock); + } } static int __init pdflush_init(void) { int i; + /* + * Pre-set nr_pdflush_threads... If we fail to create, + * the count will be decremented. + */ + nr_pdflush_threads = MIN_PDFLUSH_THREADS; + for (i = 0; i < MIN_PDFLUSH_THREADS; i++) start_one_pdflush_thread(); return 0; -- cgit v1.2.3-70-g09d2 From fafd688e4c0c34da0f3de909881117d374e4c7af Mon Sep 17 00:00:00 2001 From: Peter W Morreale Date: Mon, 6 Apr 2009 19:00:29 -0700 Subject: mm: add /proc controls for pdflush threads Add /proc entries to give the admin the ability to control the minimum and maximum number of pdflush threads. This allows finer control of pdflush on both large and small machines. The rationale is simply one size does not fit all. Admins on large and/or small systems may want to tune the min/max pdflush thread count to best suit their needs. Right now the min/max is hardcoded to 2/8. While probably a fair estimate for smaller machines, large machines with large numbers of CPUs and large numbers of filesystems/block devices may benefit from larger numbers of threads working on different block devices. Even if the background flushing algorithm is radically changed, it is still likely that multiple threads will be involved and admins would still desire finer control on the min/max other than to have to recompile the kernel. The patch adds '/proc/sys/vm/nr_pdflush_threads_min' and '/proc/sys/vm/nr_pdflush_threads_max' with r/w permissions. The minimum value for nr_pdflush_threads_min is 1 and the maximum value is the current value of nr_pdflush_threads_max. This minimum is required since additional thread creation is performed in a pdflush thread itself. The minimum value for nr_pdflush_threads_max is the current value of nr_pdflush_threads_min and the maximum value can be 1000. Documentation/sysctl/vm.txt is also updated. [akpm@linux-foundation.org: fix comment, fix whitespace, use __read_mostly] Signed-off-by: Peter W Morreale Reviewed-by: Rik van Riel Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- Documentation/sysctl/vm.txt | 28 ++++++++++++++++++++++++++++ include/linux/writeback.h | 2 ++ kernel/sysctl.c | 23 +++++++++++++++++++++++ mm/pdflush.c | 31 +++++++++++++++++++------------ 4 files changed, 72 insertions(+), 12 deletions(-) (limited to 'mm/pdflush.c') diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt index 3197fc83bc5..97c4b328432 100644 --- a/Documentation/sysctl/vm.txt +++ b/Documentation/sysctl/vm.txt @@ -39,6 +39,8 @@ Currently, these files are in /proc/sys/vm: - nr_hugepages - nr_overcommit_hugepages - nr_pdflush_threads +- nr_pdflush_threads_min +- nr_pdflush_threads_max - nr_trim_pages (only if CONFIG_MMU=n) - numa_zonelist_order - oom_dump_tasks @@ -463,6 +465,32 @@ The default value is 0. ============================================================== +nr_pdflush_threads_min + +This value controls the minimum number of pdflush threads. + +At boot time, the kernel will create and maintain 'nr_pdflush_threads_min' +threads for the kernel's lifetime. + +The default value is 2. The minimum value you can specify is 1, and +the maximum value is the current setting of 'nr_pdflush_threads_max'. + +See 'nr_pdflush_threads_max' below for more information. + +============================================================== + +nr_pdflush_threads_max + +This value controls the maximum number of pdflush threads that can be +created. The pdflush algorithm will create a new pdflush thread (up to +this maximum) if no pdflush threads have been available for >= 1 second. + +The default value is 8. The minimum value you can specify is the +current value of 'nr_pdflush_threads_min' and the +maximum is 1000. + +============================================================== + overcommit_memory: This value contains a flag that enables memory overcommitment. diff --git a/include/linux/writeback.h b/include/linux/writeback.h index 93445477f86..9c1ed1fb6dd 100644 --- a/include/linux/writeback.h +++ b/include/linux/writeback.h @@ -168,6 +168,8 @@ void writeback_set_ratelimit(void); /* pdflush.c */ extern int nr_pdflush_threads; /* Global so it can be exported to sysctl read-only. */ +extern int nr_pdflush_threads_max; /* Global so it can be exported to sysctl */ +extern int nr_pdflush_threads_min; /* Global so it can be exported to sysctl */ #endif /* WRITEBACK_H */ diff --git a/kernel/sysctl.c b/kernel/sysctl.c index b125e338756..72eb1a41dca 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -101,6 +101,7 @@ static int __maybe_unused one = 1; static int __maybe_unused two = 2; static unsigned long one_ul = 1; static int one_hundred = 100; +static int one_thousand = 1000; /* this is needed for the proc_dointvec_minmax for [fs_]overflow UID and GID */ static int maxolduid = 65535; @@ -1026,6 +1027,28 @@ static struct ctl_table vm_table[] = { .mode = 0444 /* read-only*/, .proc_handler = &proc_dointvec, }, + { + .ctl_name = CTL_UNNUMBERED, + .procname = "nr_pdflush_threads_min", + .data = &nr_pdflush_threads_min, + .maxlen = sizeof nr_pdflush_threads_min, + .mode = 0644 /* read-write */, + .proc_handler = &proc_dointvec_minmax, + .strategy = &sysctl_intvec, + .extra1 = &one, + .extra2 = &nr_pdflush_threads_max, + }, + { + .ctl_name = CTL_UNNUMBERED, + .procname = "nr_pdflush_threads_max", + .data = &nr_pdflush_threads_max, + .maxlen = sizeof nr_pdflush_threads_max, + .mode = 0644 /* read-write */, + .proc_handler = &proc_dointvec_minmax, + .strategy = &sysctl_intvec, + .extra1 = &nr_pdflush_threads_min, + .extra2 = &one_thousand, + }, { .ctl_name = VM_SWAPPINESS, .procname = "swappiness", diff --git a/mm/pdflush.c b/mm/pdflush.c index 235ac440c44..f2caf96993f 100644 --- a/mm/pdflush.c +++ b/mm/pdflush.c @@ -57,6 +57,14 @@ static DEFINE_SPINLOCK(pdflush_lock); */ int nr_pdflush_threads = 0; +/* + * The max/min number of pdflush threads. R/W by sysctl at + * /proc/sys/vm/nr_pdflush_threads_max/min + */ +int nr_pdflush_threads_max __read_mostly = MAX_PDFLUSH_THREADS; +int nr_pdflush_threads_min __read_mostly = MIN_PDFLUSH_THREADS; + + /* * The time at which the pdflush thread pool last went empty */ @@ -68,7 +76,7 @@ static unsigned long last_empty_jifs; * Thread pool management algorithm: * * - The minimum and maximum number of pdflush instances are bound - * by MIN_PDFLUSH_THREADS and MAX_PDFLUSH_THREADS. + * by nr_pdflush_threads_min and nr_pdflush_threads_max. * * - If there have been no idle pdflush instances for 1 second, create * a new one. @@ -134,14 +142,13 @@ static int __pdflush(struct pdflush_work *my_work) * To throttle creation, we reset last_empty_jifs. */ if (time_after(jiffies, last_empty_jifs + 1 * HZ)) { - if (list_empty(&pdflush_list)) { - if (nr_pdflush_threads < MAX_PDFLUSH_THREADS) { - last_empty_jifs = jiffies; - nr_pdflush_threads++; - spin_unlock_irq(&pdflush_lock); - start_one_pdflush_thread(); - spin_lock_irq(&pdflush_lock); - } + if (list_empty(&pdflush_list) && + nr_pdflush_threads < nr_pdflush_threads_max) { + last_empty_jifs = jiffies; + nr_pdflush_threads++; + spin_unlock_irq(&pdflush_lock); + start_one_pdflush_thread(); + spin_lock_irq(&pdflush_lock); } } @@ -153,7 +160,7 @@ static int __pdflush(struct pdflush_work *my_work) */ if (list_empty(&pdflush_list)) continue; - if (nr_pdflush_threads <= MIN_PDFLUSH_THREADS) + if (nr_pdflush_threads <= nr_pdflush_threads_min) continue; pdf = list_entry(pdflush_list.prev, struct pdflush_work, list); if (time_after(jiffies, pdf->when_i_went_to_sleep + 1 * HZ)) { @@ -259,9 +266,9 @@ static int __init pdflush_init(void) * Pre-set nr_pdflush_threads... If we fail to create, * the count will be decremented. */ - nr_pdflush_threads = MIN_PDFLUSH_THREADS; + nr_pdflush_threads = nr_pdflush_threads_min; - for (i = 0; i < MIN_PDFLUSH_THREADS; i++) + for (i = 0; i < nr_pdflush_threads_min; i++) start_one_pdflush_thread(); return 0; } -- cgit v1.2.3-70-g09d2 From cd17cbfda004fe5f406c01b318c6378d9895896f Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Fri, 15 May 2009 11:32:24 +0200 Subject: Revert "mm: add /proc controls for pdflush threads" This reverts commit fafd688e4c0c34da0f3de909881117d374e4c7af. Work is progressing to switch away from pdflush as the process backing for flushing out dirty data. So it seems pointless to add more knobs to control pdflush threads. The original author of the patch did not have any specific use cases for adding the knobs, so we can easily revert this before 2.6.30 to avoid having to maintain this API forever. Signed-off-by: Jens Axboe --- Documentation/sysctl/vm.txt | 28 ---------------------------- include/linux/writeback.h | 2 -- kernel/sysctl.c | 23 ----------------------- mm/pdflush.c | 31 ++++++++++++------------------- 4 files changed, 12 insertions(+), 72 deletions(-) (limited to 'mm/pdflush.c') diff --git a/Documentation/sysctl/vm.txt b/Documentation/sysctl/vm.txt index b716d33912d..c302ddf629a 100644 --- a/Documentation/sysctl/vm.txt +++ b/Documentation/sysctl/vm.txt @@ -39,8 +39,6 @@ Currently, these files are in /proc/sys/vm: - nr_hugepages - nr_overcommit_hugepages - nr_pdflush_threads -- nr_pdflush_threads_min -- nr_pdflush_threads_max - nr_trim_pages (only if CONFIG_MMU=n) - numa_zonelist_order - oom_dump_tasks @@ -469,32 +467,6 @@ The default value is 0. ============================================================== -nr_pdflush_threads_min - -This value controls the minimum number of pdflush threads. - -At boot time, the kernel will create and maintain 'nr_pdflush_threads_min' -threads for the kernel's lifetime. - -The default value is 2. The minimum value you can specify is 1, and -the maximum value is the current setting of 'nr_pdflush_threads_max'. - -See 'nr_pdflush_threads_max' below for more information. - -============================================================== - -nr_pdflush_threads_max - -This value controls the maximum number of pdflush threads that can be -created. The pdflush algorithm will create a new pdflush thread (up to -this maximum) if no pdflush threads have been available for >= 1 second. - -The default value is 8. The minimum value you can specify is the -current value of 'nr_pdflush_threads_min' and the -maximum is 1000. - -============================================================== - overcommit_memory: This value contains a flag that enables memory overcommitment. diff --git a/include/linux/writeback.h b/include/linux/writeback.h index 9c1ed1fb6dd..93445477f86 100644 --- a/include/linux/writeback.h +++ b/include/linux/writeback.h @@ -168,8 +168,6 @@ void writeback_set_ratelimit(void); /* pdflush.c */ extern int nr_pdflush_threads; /* Global so it can be exported to sysctl read-only. */ -extern int nr_pdflush_threads_max; /* Global so it can be exported to sysctl */ -extern int nr_pdflush_threads_min; /* Global so it can be exported to sysctl */ #endif /* WRITEBACK_H */ diff --git a/kernel/sysctl.c b/kernel/sysctl.c index ea78fa101ad..b2970d56fb7 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -101,7 +101,6 @@ static int __maybe_unused one = 1; static int __maybe_unused two = 2; static unsigned long one_ul = 1; static int one_hundred = 100; -static int one_thousand = 1000; /* this is needed for the proc_doulongvec_minmax of vm_dirty_bytes */ static unsigned long dirty_bytes_min = 2 * PAGE_SIZE; @@ -1033,28 +1032,6 @@ static struct ctl_table vm_table[] = { .mode = 0444 /* read-only*/, .proc_handler = &proc_dointvec, }, - { - .ctl_name = CTL_UNNUMBERED, - .procname = "nr_pdflush_threads_min", - .data = &nr_pdflush_threads_min, - .maxlen = sizeof nr_pdflush_threads_min, - .mode = 0644 /* read-write */, - .proc_handler = &proc_dointvec_minmax, - .strategy = &sysctl_intvec, - .extra1 = &one, - .extra2 = &nr_pdflush_threads_max, - }, - { - .ctl_name = CTL_UNNUMBERED, - .procname = "nr_pdflush_threads_max", - .data = &nr_pdflush_threads_max, - .maxlen = sizeof nr_pdflush_threads_max, - .mode = 0644 /* read-write */, - .proc_handler = &proc_dointvec_minmax, - .strategy = &sysctl_intvec, - .extra1 = &nr_pdflush_threads_min, - .extra2 = &one_thousand, - }, { .ctl_name = VM_SWAPPINESS, .procname = "swappiness", diff --git a/mm/pdflush.c b/mm/pdflush.c index f2caf96993f..235ac440c44 100644 --- a/mm/pdflush.c +++ b/mm/pdflush.c @@ -57,14 +57,6 @@ static DEFINE_SPINLOCK(pdflush_lock); */ int nr_pdflush_threads = 0; -/* - * The max/min number of pdflush threads. R/W by sysctl at - * /proc/sys/vm/nr_pdflush_threads_max/min - */ -int nr_pdflush_threads_max __read_mostly = MAX_PDFLUSH_THREADS; -int nr_pdflush_threads_min __read_mostly = MIN_PDFLUSH_THREADS; - - /* * The time at which the pdflush thread pool last went empty */ @@ -76,7 +68,7 @@ static unsigned long last_empty_jifs; * Thread pool management algorithm: * * - The minimum and maximum number of pdflush instances are bound - * by nr_pdflush_threads_min and nr_pdflush_threads_max. + * by MIN_PDFLUSH_THREADS and MAX_PDFLUSH_THREADS. * * - If there have been no idle pdflush instances for 1 second, create * a new one. @@ -142,13 +134,14 @@ static int __pdflush(struct pdflush_work *my_work) * To throttle creation, we reset last_empty_jifs. */ if (time_after(jiffies, last_empty_jifs + 1 * HZ)) { - if (list_empty(&pdflush_list) && - nr_pdflush_threads < nr_pdflush_threads_max) { - last_empty_jifs = jiffies; - nr_pdflush_threads++; - spin_unlock_irq(&pdflush_lock); - start_one_pdflush_thread(); - spin_lock_irq(&pdflush_lock); + if (list_empty(&pdflush_list)) { + if (nr_pdflush_threads < MAX_PDFLUSH_THREADS) { + last_empty_jifs = jiffies; + nr_pdflush_threads++; + spin_unlock_irq(&pdflush_lock); + start_one_pdflush_thread(); + spin_lock_irq(&pdflush_lock); + } } } @@ -160,7 +153,7 @@ static int __pdflush(struct pdflush_work *my_work) */ if (list_empty(&pdflush_list)) continue; - if (nr_pdflush_threads <= nr_pdflush_threads_min) + if (nr_pdflush_threads <= MIN_PDFLUSH_THREADS) continue; pdf = list_entry(pdflush_list.prev, struct pdflush_work, list); if (time_after(jiffies, pdf->when_i_went_to_sleep + 1 * HZ)) { @@ -266,9 +259,9 @@ static int __init pdflush_init(void) * Pre-set nr_pdflush_threads... If we fail to create, * the count will be decremented. */ - nr_pdflush_threads = nr_pdflush_threads_min; + nr_pdflush_threads = MIN_PDFLUSH_THREADS; - for (i = 0; i < nr_pdflush_threads_min; i++) + for (i = 0; i < MIN_PDFLUSH_THREADS; i++) start_one_pdflush_thread(); return 0; } -- cgit v1.2.3-70-g09d2