From 44b99462d9d776522e174d6c531ce5ccef309e26 Mon Sep 17 00:00:00 2001 From: Vaibhav Nagarnaik Date: Fri, 22 Jun 2012 11:50:05 -0700 Subject: [PATCH 1/9] ring-buffer: Fix crash due to uninitialized new_pages list head The new_pages list head in the cpu_buffer is not initialized. When adding pages to the ring buffer, if the memory allocation fails in ring_buffer_resize, the clean up handler tries to free up the allocated pages from all the cpu buffers. The panic is caused by referencing the uninitialized new_pages list head. Initializing the new_pages list head in rb_allocate_cpu_buffer fixes this. Link: http://lkml.kernel.org/r/1340391005-10880-1-git-send-email-vnagarnaik@google.com Cc: Justin Teravest Cc: David Sharp Signed-off-by: Vaibhav Nagarnaik Signed-off-by: Steven Rostedt --- kernel/trace/ring_buffer.c | 1 + 1 file changed, 1 insertion(+) diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 1d0f6a8a0e5e..ba39cbabdc9f 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1075,6 +1075,7 @@ rb_allocate_cpu_buffer(struct ring_buffer *buffer, int nr_pages, int cpu) rb_init_page(bpage->page); INIT_LIST_HEAD(&cpu_buffer->reader_page->list); + INIT_LIST_HEAD(&cpu_buffer->new_pages); ret = rb_allocate_pages(cpu_buffer, nr_pages); if (ret < 0) From 48fdc72f23ad9a9956e524a47843135d0bbc3317 Mon Sep 17 00:00:00 2001 From: Vaibhav Nagarnaik Date: Fri, 29 Jun 2012 12:31:41 -0700 Subject: [PATCH 2/9] ring-buffer: Fix accounting of entries when removing pages When removing pages from the ring buffer, its state is not reset. This means that the counters need to be correctly updated to account for the pages removed. Update the overrun counter to reflect the removed events from the pages. Link: http://lkml.kernel.org/r/1340998301-1715-1-git-send-email-vnagarnaik@google.com Cc: Justin Teravest Cc: David Sharp Signed-off-by: Vaibhav Nagarnaik Signed-off-by: Steven Rostedt --- kernel/trace/ring_buffer.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index ba39cbabdc9f..f765465bffe4 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -1347,10 +1347,9 @@ rb_remove_pages(struct ring_buffer_per_cpu *cpu_buffer, unsigned int nr_pages) * If something was added to this page, it was full * since it is not the tail page. So we deduct the * bytes consumed in ring buffer from here. - * No need to update overruns, since this page is - * deleted from ring buffer and its entries are - * already accounted for. + * Increment overrun to account for the lost events. */ + local_add(page_entries, &cpu_buffer->overrun); local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes); } From 76a8349dfdb775d387e9767db3092e410403138a Mon Sep 17 00:00:00 2001 From: David Ahern Date: Thu, 14 Jun 2012 12:36:17 -0600 Subject: [PATCH 3/9] perf script: Fix format regression due to libtraceevent merge Consider the commands: perf record -e sched:sched_switch -fo /tmp/perf.data -a -- sleep 1 perf script -i /tmp/perf.data In v3.4 the output has the form (lines wrapped here) perf 29214 [005] 821043.582596: sched_switch: prev_comm=perf prev_pid=29214 prev_prio=120 prev_state=S ==> next_comm=swapper/5 next_pid=0 next_prio=120 In 3.5 that same line has become: perf 29214 [005] 821043.582596: sched_switch: <...>-29214 [005] 0.000000000: sched_switch: prev_comm=perf prev_pid=29214 prev_prio=120 prev_state=S ==> next_comm=swapper/5 next_pid=0 next_prio=120 Note the duplicates in the output -- pid, cpu, event name. With this patch the v3.4 output is restored: perf 29214 [005] 821043.582596: sched_switch: prev_comm=perf prev_pid=29214 prev_prio=120 prev_state=S ==> next_comm=swapper/5 next_pid=0 next_prio=120 v3: Remove that pesky newline too. Output now matches v3.4 (pre-libtracevent). v2: Change print_trace_event function local to perf per Steve's comments. Signed-off-by: David Ahern Acked-by: Steven Rostedt Cc: Frederic Weisbecker Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Stephane Eranian Cc: Steven Rostedt Link: http://lkml.kernel.org/r/1339698977-68962-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/trace-event-parse.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/tools/perf/util/trace-event-parse.c b/tools/perf/util/trace-event-parse.c index df2fddbf0cd2..5dd3b5ec8411 100644 --- a/tools/perf/util/trace-event-parse.c +++ b/tools/perf/util/trace-event-parse.c @@ -198,9 +198,8 @@ void print_trace_event(int cpu, void *data, int size) record.data = data; trace_seq_init(&s); - pevent_print_event(pevent, &s, &record); + pevent_event_info(&s, event, &record); trace_seq_do_printf(&s); - printf("\n"); } void print_event(int cpu, void *data, int size, unsigned long long nsecs, From 207b5792696206663a38e525b9793644895bad3b Mon Sep 17 00:00:00 2001 From: David Ahern Date: Sun, 1 Jul 2012 16:11:37 -0600 Subject: [PATCH 4/9] perf kvm: Fix regression with guest machine creation Commit 743eb868657bdb1b26c7b24077ca21c67c82c777 reworked when the machines were created. Prior to this commit guest machines could be created in perf_event__process_kernel_mmap() while processing kernel MMAP events. This commit assumes that the machines exist by the time perf_session_deliver_event is called (e.g., during processing of build id events) - which is not always correct. One example is the use of default guest args (--guestkallsyms and --guestmodules) for short times where no samples hit within a guest module. For this case no build id is added to the file header. No build id == no machine created. That leads to the next example -- the use of no-buildid (-B) on the record for all perf-kvm invocations. In both cases perf report dies with a SEGFAULT of the form: (gdb) bt 0 0x000000000046dd7b in machine__mmap_name (self=0x0, bf=0x7fffffffbd20 "q\021", size=4096) at util/map.c:715 1 0x0000000000444161 in perf_event__process_kernel_mmap (tool=0x7fffffffdd80, event=0x7ffff7fb4120, machine=0x0) at util/event.c:562 2 0x0000000000444642 in perf_event__process_mmap (tool=0x7fffffffdd80, event=0x7ffff7fb4120, sample=0x7fffffffd210, machine=0x0) at util/event.c:668 3 0x0000000000470e0b in perf_session_deliver_event (session=0x915ca0, event=0x7ffff7fb4120, sample=0x7fffffffd210, tool=0x7fffffffdd80, file_offset=8480) at util/session.c:979 4 0x000000000047032e in flush_sample_queue (s=0x915ca0, tool=0x7fffffffdd80) at util/session.c:679 5 0x0000000000471c8d in __perf_session__process_events (session=0x915ca0, data_offset=400, data_size=150448, file_size=150848, tool= 0x7fffffffdd80) at util/session.c:1363 6 0x0000000000471d42 in perf_session__process_events (self=0x915ca0, tool=0x7fffffffdd80) at util/session.c:1379 7 0x000000000042484a in __cmd_report (rep=0x7fffffffdd80) at builtin-report.c:368 8 0x0000000000425bf1 in cmd_report (argc=0, argv=0x915b00, prefix=0x0) at builtin-report.c:756 9 0x0000000000438505 in __cmd_report (argc=4, argv=0x7fffffffe260) at builtin-kvm.c:84 10 0x000000000043882a in cmd_kvm (argc=4, argv=0x7fffffffe260, prefix=0x0) at builtin-kvm.c:131 11 0x00000000004152cd in run_builtin (p=0x7a54e8, argc=9, argv=0x7fffffffe260) at perf.c:273 12 0x00000000004154c7 in handle_internal_command (argc=9, argv=0x7fffffffe260) at perf.c:345 13 0x0000000000415613 in run_argv (argcp=0x7fffffffe14c, argv=0x7fffffffe140) at perf.c:389 14 0x0000000000415899 in main (argc=9, argv=0x7fffffffe260) at perf.c:487 Fix by allowing the machine to be created in perf_session_deliver_event. Tested with --guestmount option and default guest args, with and without -B arg on record for both and for short (10 seconds) and long (10 minutes) windows. Reported-by: Pradeep Kumar Surisetty Signed-off-by: David Ahern Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Pradeep Kumar Surisetty Link: http://lkml.kernel.org/r/1341180697-64515-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/session.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c index c3e399bcf18d..56142d0fb8d7 100644 --- a/tools/perf/util/session.c +++ b/tools/perf/util/session.c @@ -926,7 +926,7 @@ static struct machine * else pid = event->ip.pid; - return perf_session__find_machine(session, pid); + return perf_session__findnew_machine(session, pid); } return perf_session__find_host_machine(session); From 7ed97ad41ffa94040dfd593948962a7e9e7b0db9 Mon Sep 17 00:00:00 2001 From: David Ahern Date: Mon, 2 Jul 2012 09:12:57 -0600 Subject: [PATCH 5/9] perf kvm: Fix segfault with report and mixed guestmount use Using the guestmount option on record: $ perf kvm --guest --host --guestmount=/tmp/guest-mount record -ag But not the subsequent report: $ perf kvm report causes a SEGFAULT in the usual place: (gdb) bt 0 0x0000000000470356 in machine__mmap_name (self=0x0, bf=0x7fffffffbdb0 " z\370\367\377\177", size= 4096) at util/map.c:712 1 0x00000000004453e8 in perf_event__process_kernel_mmap (tool=0x7fffffffde10, event=0x7ffff7f87e38, machine=0x0) at util/event.c:550 2 0x00000000004458c9 in perf_event__process_mmap (tool=0x7fffffffde10, event=0x7ffff7f87e38, sample= 0x7fffffffd2a0, machine=0x0) at util/event.c:656 3 0x00000000004733e0 in perf_session_deliver_event (session=0x91aca0, event=0x7ffff7f87e38, sample= 0x7fffffffd2a0, tool=0x7fffffffde10, file_offset=7736) at util/session.c:979 ... The MMAP events in this case already contain the full path to the module. No need to require it for the report path to. Signed-off-by: David Ahern Cc: Frederic Weisbecker Cc: Ingo Molnar Cc: Jiri Olsa Cc: Namhyung Kim Cc: Peter Zijlstra Link: http://lkml.kernel.org/r/1341241977-71535-1-git-send-email-dsahern@gmail.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/util/map.c | 29 +++++++++++++++-------------- 1 file changed, 15 insertions(+), 14 deletions(-) diff --git a/tools/perf/util/map.c b/tools/perf/util/map.c index 35ae56864e4f..a1f4e3669142 100644 --- a/tools/perf/util/map.c +++ b/tools/perf/util/map.c @@ -669,25 +669,26 @@ struct machine *machines__find(struct rb_root *self, pid_t pid) struct machine *machines__findnew(struct rb_root *self, pid_t pid) { char path[PATH_MAX]; - const char *root_dir; + const char *root_dir = ""; struct machine *machine = machines__find(self, pid); - if (!machine || machine->pid != pid) { - if (pid == HOST_KERNEL_ID || pid == DEFAULT_GUEST_KERNEL_ID) - root_dir = ""; - else { - if (!symbol_conf.guestmount) - goto out; - sprintf(path, "%s/%d", symbol_conf.guestmount, pid); - if (access(path, R_OK)) { - pr_err("Can't access file %s\n", path); - goto out; - } - root_dir = path; + if (machine && (machine->pid == pid)) + goto out; + + if ((pid != HOST_KERNEL_ID) && + (pid != DEFAULT_GUEST_KERNEL_ID) && + (symbol_conf.guestmount)) { + sprintf(path, "%s/%d", symbol_conf.guestmount, pid); + if (access(path, R_OK)) { + pr_err("Can't access file %s\n", path); + machine = NULL; + goto out; } - machine = machines__add(self, pid, root_dir); + root_dir = path; } + machine = machines__add(self, pid, root_dir); + out: return machine; } From cba6d0d64ee53772b285d0c0c288deefbeaf7775 Mon Sep 17 00:00:00 2001 From: "Paul E. McKenney" Date: Mon, 2 Jul 2012 07:08:42 -0700 Subject: [PATCH 6/9] Revert "rcu: Move PREEMPT_RCU preemption to switch_to() invocation" This reverts commit 616c310e83b872024271c915c1b9ab505b9efad9. (Move PREEMPT_RCU preemption to switch_to() invocation). Testing by Sasha Levin showed that this can result in deadlock due to invoking the scheduler when one of the runqueue locks is held. Because this commit was simply a performance optimization, revert it. Reported-by: Sasha Levin Signed-off-by: Paul E. McKenney Tested-by: Sasha Levin --- arch/um/drivers/mconsole_kern.c | 1 - include/linux/rcupdate.h | 1 - include/linux/rcutiny.h | 6 ++++++ include/linux/sched.h | 10 ---------- kernel/rcutree.c | 1 + kernel/rcutree.h | 1 + kernel/rcutree_plugin.h | 14 +++++++++++--- kernel/sched/core.c | 1 - 8 files changed, 19 insertions(+), 16 deletions(-) diff --git a/arch/um/drivers/mconsole_kern.c b/arch/um/drivers/mconsole_kern.c index 88e466b159dc..43b39d61b538 100644 --- a/arch/um/drivers/mconsole_kern.c +++ b/arch/um/drivers/mconsole_kern.c @@ -705,7 +705,6 @@ static void stack_proc(void *arg) struct task_struct *from = current, *to = arg; to->thread.saved_task = from; - rcu_switch_from(from); switch_to(from, to, from); } diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h index 26d1a47591f1..9cac722b169c 100644 --- a/include/linux/rcupdate.h +++ b/include/linux/rcupdate.h @@ -184,7 +184,6 @@ static inline int rcu_preempt_depth(void) /* Internal to kernel */ extern void rcu_sched_qs(int cpu); extern void rcu_bh_qs(int cpu); -extern void rcu_preempt_note_context_switch(void); extern void rcu_check_callbacks(int cpu, int user); struct notifier_block; extern void rcu_idle_enter(void); diff --git a/include/linux/rcutiny.h b/include/linux/rcutiny.h index 854dc4c5c271..4e56a9c69a35 100644 --- a/include/linux/rcutiny.h +++ b/include/linux/rcutiny.h @@ -87,6 +87,10 @@ static inline void kfree_call_rcu(struct rcu_head *head, #ifdef CONFIG_TINY_RCU +static inline void rcu_preempt_note_context_switch(void) +{ +} + static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) { *delta_jiffies = ULONG_MAX; @@ -95,6 +99,7 @@ static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) #else /* #ifdef CONFIG_TINY_RCU */ +void rcu_preempt_note_context_switch(void); int rcu_preempt_needs_cpu(void); static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) @@ -108,6 +113,7 @@ static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies) static inline void rcu_note_context_switch(int cpu) { rcu_sched_qs(cpu); + rcu_preempt_note_context_switch(); } /* diff --git a/include/linux/sched.h b/include/linux/sched.h index 4059c0f33f07..06a4c5f4f55c 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1871,22 +1871,12 @@ static inline void rcu_copy_process(struct task_struct *p) INIT_LIST_HEAD(&p->rcu_node_entry); } -static inline void rcu_switch_from(struct task_struct *prev) -{ - if (prev->rcu_read_lock_nesting != 0) - rcu_preempt_note_context_switch(); -} - #else static inline void rcu_copy_process(struct task_struct *p) { } -static inline void rcu_switch_from(struct task_struct *prev) -{ -} - #endif #ifdef CONFIG_SMP diff --git a/kernel/rcutree.c b/kernel/rcutree.c index 38ecdda3f55f..4b97bba7396e 100644 --- a/kernel/rcutree.c +++ b/kernel/rcutree.c @@ -201,6 +201,7 @@ void rcu_note_context_switch(int cpu) { trace_rcu_utilization("Start context switch"); rcu_sched_qs(cpu); + rcu_preempt_note_context_switch(cpu); trace_rcu_utilization("End context switch"); } EXPORT_SYMBOL_GPL(rcu_note_context_switch); diff --git a/kernel/rcutree.h b/kernel/rcutree.h index ea056495783e..19b61ac1079f 100644 --- a/kernel/rcutree.h +++ b/kernel/rcutree.h @@ -444,6 +444,7 @@ DECLARE_PER_CPU(char, rcu_cpu_has_work); /* Forward declarations for rcutree_plugin.h */ static void rcu_bootup_announce(void); long rcu_batches_completed(void); +static void rcu_preempt_note_context_switch(int cpu); static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp); #ifdef CONFIG_HOTPLUG_CPU static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h index 5271a020887e..3e4899459f3d 100644 --- a/kernel/rcutree_plugin.h +++ b/kernel/rcutree_plugin.h @@ -153,7 +153,7 @@ static void rcu_preempt_qs(int cpu) * * Caller must disable preemption. */ -void rcu_preempt_note_context_switch(void) +static void rcu_preempt_note_context_switch(int cpu) { struct task_struct *t = current; unsigned long flags; @@ -164,7 +164,7 @@ void rcu_preempt_note_context_switch(void) (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) { /* Possibly blocking in an RCU read-side critical section. */ - rdp = __this_cpu_ptr(rcu_preempt_state.rda); + rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu); rnp = rdp->mynode; raw_spin_lock_irqsave(&rnp->lock, flags); t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED; @@ -228,7 +228,7 @@ void rcu_preempt_note_context_switch(void) * means that we continue to block the current grace period. */ local_irq_save(flags); - rcu_preempt_qs(smp_processor_id()); + rcu_preempt_qs(cpu); local_irq_restore(flags); } @@ -1001,6 +1001,14 @@ void rcu_force_quiescent_state(void) } EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); +/* + * Because preemptible RCU does not exist, we never have to check for + * CPUs being in quiescent states. + */ +static void rcu_preempt_note_context_switch(int cpu) +{ +} + /* * Because preemptible RCU does not exist, there are never any preempted * RCU readers. diff --git a/kernel/sched/core.c b/kernel/sched/core.c index d5594a4268d4..eaead2df6aa8 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -2081,7 +2081,6 @@ context_switch(struct rq *rq, struct task_struct *prev, #endif /* Here we just switch the register state and the stack. */ - rcu_switch_from(prev); switch_to(prev, next, prev); barrier(); From 164c33c6adee609b8b9062cce4c10f764d0dce13 Mon Sep 17 00:00:00 2001 From: Salman Qazi Date: Mon, 25 Jun 2012 18:18:15 -0700 Subject: [PATCH 7/9] sched: Fix fork() error path to not crash In dup_task_struct(), if arch_dup_task_struct() fails, the clean up code fails to clean up correctly. That's because the clean up code depends on unininitalized ti->task pointer. We fix this by making sure that the task and thread_info know about each other before we attempt to take the error path. Signed-off-by: Salman Qazi Signed-off-by: Peter Zijlstra Link: http://lkml.kernel.org/r/20120626011815.11323.5533.stgit@dungbeetle.mtv.corp.google.com Signed-off-by: Ingo Molnar --- kernel/fork.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/kernel/fork.c b/kernel/fork.c index ab5211b9e622..f00e319d8376 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -304,12 +304,17 @@ static struct task_struct *dup_task_struct(struct task_struct *orig) } err = arch_dup_task_struct(tsk, orig); + + /* + * We defer looking at err, because we will need this setup + * for the clean up path to work correctly. + */ + tsk->stack = ti; + setup_thread_stack(tsk, orig); + if (err) goto out; - tsk->stack = ti; - - setup_thread_stack(tsk, orig); clear_user_return_notifier(tsk); clear_tsk_need_resched(tsk); stackend = end_of_stack(tsk); From 5167e8d5417bf5c322a703d2927daec727ea40dd Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Fri, 22 Jun 2012 15:52:09 +0200 Subject: [PATCH 8/9] sched/nohz: Rewrite and fix load-avg computation -- again Thanks to Charles Wang for spotting the defects in the current code: - If we go idle during the sample window -- after sampling, we get a negative bias because we can negate our own sample. - If we wake up during the sample window we get a positive bias because we push the sample to a known active period. So rewrite the entire nohz load-avg muck once again, now adding copious documentation to the code. Reported-and-tested-by: Doug Smythies Reported-and-tested-by: Charles Wang Signed-off-by: Peter Zijlstra Cc: Linus Torvalds Cc: Andrew Morton Cc: stable@kernel.org Link: http://lkml.kernel.org/r/1340373782.18025.74.camel@twins [ minor edits ] Signed-off-by: Ingo Molnar --- include/linux/sched.h | 8 ++ kernel/sched/core.c | 277 ++++++++++++++++++++++++++++----------- kernel/sched/idle_task.c | 1 - kernel/sched/sched.h | 2 - kernel/time/tick-sched.c | 2 + 5 files changed, 214 insertions(+), 76 deletions(-) diff --git a/include/linux/sched.h b/include/linux/sched.h index 4059c0f33f07..20cb7497c59c 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1909,6 +1909,14 @@ static inline int set_cpus_allowed_ptr(struct task_struct *p, } #endif +#ifdef CONFIG_NO_HZ +void calc_load_enter_idle(void); +void calc_load_exit_idle(void); +#else +static inline void calc_load_enter_idle(void) { } +static inline void calc_load_exit_idle(void) { } +#endif /* CONFIG_NO_HZ */ + #ifndef CONFIG_CPUMASK_OFFSTACK static inline int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) { diff --git a/kernel/sched/core.c b/kernel/sched/core.c index d5594a4268d4..bb840405335d 100644 --- a/kernel/sched/core.c +++ b/kernel/sched/core.c @@ -2161,11 +2161,73 @@ unsigned long this_cpu_load(void) } +/* + * Global load-average calculations + * + * We take a distributed and async approach to calculating the global load-avg + * in order to minimize overhead. + * + * The global load average is an exponentially decaying average of nr_running + + * nr_uninterruptible. + * + * Once every LOAD_FREQ: + * + * nr_active = 0; + * for_each_possible_cpu(cpu) + * nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible; + * + * avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n) + * + * Due to a number of reasons the above turns in the mess below: + * + * - for_each_possible_cpu() is prohibitively expensive on machines with + * serious number of cpus, therefore we need to take a distributed approach + * to calculating nr_active. + * + * \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0 + * = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) } + * + * So assuming nr_active := 0 when we start out -- true per definition, we + * can simply take per-cpu deltas and fold those into a global accumulate + * to obtain the same result. See calc_load_fold_active(). + * + * Furthermore, in order to avoid synchronizing all per-cpu delta folding + * across the machine, we assume 10 ticks is sufficient time for every + * cpu to have completed this task. + * + * This places an upper-bound on the IRQ-off latency of the machine. Then + * again, being late doesn't loose the delta, just wrecks the sample. + * + * - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because + * this would add another cross-cpu cacheline miss and atomic operation + * to the wakeup path. Instead we increment on whatever cpu the task ran + * when it went into uninterruptible state and decrement on whatever cpu + * did the wakeup. This means that only the sum of nr_uninterruptible over + * all cpus yields the correct result. + * + * This covers the NO_HZ=n code, for extra head-aches, see the comment below. + */ + /* Variables and functions for calc_load */ static atomic_long_t calc_load_tasks; static unsigned long calc_load_update; unsigned long avenrun[3]; -EXPORT_SYMBOL(avenrun); +EXPORT_SYMBOL(avenrun); /* should be removed */ + +/** + * get_avenrun - get the load average array + * @loads: pointer to dest load array + * @offset: offset to add + * @shift: shift count to shift the result left + * + * These values are estimates at best, so no need for locking. + */ +void get_avenrun(unsigned long *loads, unsigned long offset, int shift) +{ + loads[0] = (avenrun[0] + offset) << shift; + loads[1] = (avenrun[1] + offset) << shift; + loads[2] = (avenrun[2] + offset) << shift; +} static long calc_load_fold_active(struct rq *this_rq) { @@ -2182,6 +2244,9 @@ static long calc_load_fold_active(struct rq *this_rq) return delta; } +/* + * a1 = a0 * e + a * (1 - e) + */ static unsigned long calc_load(unsigned long load, unsigned long exp, unsigned long active) { @@ -2193,30 +2258,118 @@ calc_load(unsigned long load, unsigned long exp, unsigned long active) #ifdef CONFIG_NO_HZ /* - * For NO_HZ we delay the active fold to the next LOAD_FREQ update. + * Handle NO_HZ for the global load-average. + * + * Since the above described distributed algorithm to compute the global + * load-average relies on per-cpu sampling from the tick, it is affected by + * NO_HZ. + * + * The basic idea is to fold the nr_active delta into a global idle-delta upon + * entering NO_HZ state such that we can include this as an 'extra' cpu delta + * when we read the global state. + * + * Obviously reality has to ruin such a delightfully simple scheme: + * + * - When we go NO_HZ idle during the window, we can negate our sample + * contribution, causing under-accounting. + * + * We avoid this by keeping two idle-delta counters and flipping them + * when the window starts, thus separating old and new NO_HZ load. + * + * The only trick is the slight shift in index flip for read vs write. + * + * 0s 5s 10s 15s + * +10 +10 +10 +10 + * |-|-----------|-|-----------|-|-----------|-| + * r:0 0 1 1 0 0 1 1 0 + * w:0 1 1 0 0 1 1 0 0 + * + * This ensures we'll fold the old idle contribution in this window while + * accumlating the new one. + * + * - When we wake up from NO_HZ idle during the window, we push up our + * contribution, since we effectively move our sample point to a known + * busy state. + * + * This is solved by pushing the window forward, and thus skipping the + * sample, for this cpu (effectively using the idle-delta for this cpu which + * was in effect at the time the window opened). This also solves the issue + * of having to deal with a cpu having been in NOHZ idle for multiple + * LOAD_FREQ intervals. * * When making the ILB scale, we should try to pull this in as well. */ -static atomic_long_t calc_load_tasks_idle; +static atomic_long_t calc_load_idle[2]; +static int calc_load_idx; -void calc_load_account_idle(struct rq *this_rq) +static inline int calc_load_write_idx(void) { + int idx = calc_load_idx; + + /* + * See calc_global_nohz(), if we observe the new index, we also + * need to observe the new update time. + */ + smp_rmb(); + + /* + * If the folding window started, make sure we start writing in the + * next idle-delta. + */ + if (!time_before(jiffies, calc_load_update)) + idx++; + + return idx & 1; +} + +static inline int calc_load_read_idx(void) +{ + return calc_load_idx & 1; +} + +void calc_load_enter_idle(void) +{ + struct rq *this_rq = this_rq(); long delta; + /* + * We're going into NOHZ mode, if there's any pending delta, fold it + * into the pending idle delta. + */ delta = calc_load_fold_active(this_rq); - if (delta) - atomic_long_add(delta, &calc_load_tasks_idle); + if (delta) { + int idx = calc_load_write_idx(); + atomic_long_add(delta, &calc_load_idle[idx]); + } +} + +void calc_load_exit_idle(void) +{ + struct rq *this_rq = this_rq(); + + /* + * If we're still before the sample window, we're done. + */ + if (time_before(jiffies, this_rq->calc_load_update)) + return; + + /* + * We woke inside or after the sample window, this means we're already + * accounted through the nohz accounting, so skip the entire deal and + * sync up for the next window. + */ + this_rq->calc_load_update = calc_load_update; + if (time_before(jiffies, this_rq->calc_load_update + 10)) + this_rq->calc_load_update += LOAD_FREQ; } static long calc_load_fold_idle(void) { + int idx = calc_load_read_idx(); long delta = 0; - /* - * Its got a race, we don't care... - */ - if (atomic_long_read(&calc_load_tasks_idle)) - delta = atomic_long_xchg(&calc_load_tasks_idle, 0); + if (atomic_long_read(&calc_load_idle[idx])) + delta = atomic_long_xchg(&calc_load_idle[idx], 0); return delta; } @@ -2302,66 +2455,39 @@ static void calc_global_nohz(void) { long delta, active, n; - /* - * If we crossed a calc_load_update boundary, make sure to fold - * any pending idle changes, the respective CPUs might have - * missed the tick driven calc_load_account_active() update - * due to NO_HZ. - */ - delta = calc_load_fold_idle(); - if (delta) - atomic_long_add(delta, &calc_load_tasks); + if (!time_before(jiffies, calc_load_update + 10)) { + /* + * Catch-up, fold however many we are behind still + */ + delta = jiffies - calc_load_update - 10; + n = 1 + (delta / LOAD_FREQ); + + active = atomic_long_read(&calc_load_tasks); + active = active > 0 ? active * FIXED_1 : 0; + + avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); + avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); + avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); + + calc_load_update += n * LOAD_FREQ; + } /* - * It could be the one fold was all it took, we done! + * Flip the idle index... + * + * Make sure we first write the new time then flip the index, so that + * calc_load_write_idx() will see the new time when it reads the new + * index, this avoids a double flip messing things up. */ - if (time_before(jiffies, calc_load_update + 10)) - return; - - /* - * Catch-up, fold however many we are behind still - */ - delta = jiffies - calc_load_update - 10; - n = 1 + (delta / LOAD_FREQ); - - active = atomic_long_read(&calc_load_tasks); - active = active > 0 ? active * FIXED_1 : 0; - - avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); - avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); - avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); - - calc_load_update += n * LOAD_FREQ; -} -#else -void calc_load_account_idle(struct rq *this_rq) -{ + smp_wmb(); + calc_load_idx++; } +#else /* !CONFIG_NO_HZ */ -static inline long calc_load_fold_idle(void) -{ - return 0; -} +static inline long calc_load_fold_idle(void) { return 0; } +static inline void calc_global_nohz(void) { } -static void calc_global_nohz(void) -{ -} -#endif - -/** - * get_avenrun - get the load average array - * @loads: pointer to dest load array - * @offset: offset to add - * @shift: shift count to shift the result left - * - * These values are estimates at best, so no need for locking. - */ -void get_avenrun(unsigned long *loads, unsigned long offset, int shift) -{ - loads[0] = (avenrun[0] + offset) << shift; - loads[1] = (avenrun[1] + offset) << shift; - loads[2] = (avenrun[2] + offset) << shift; -} +#endif /* CONFIG_NO_HZ */ /* * calc_load - update the avenrun load estimates 10 ticks after the @@ -2369,11 +2495,18 @@ void get_avenrun(unsigned long *loads, unsigned long offset, int shift) */ void calc_global_load(unsigned long ticks) { - long active; + long active, delta; if (time_before(jiffies, calc_load_update + 10)) return; + /* + * Fold the 'old' idle-delta to include all NO_HZ cpus. + */ + delta = calc_load_fold_idle(); + if (delta) + atomic_long_add(delta, &calc_load_tasks); + active = atomic_long_read(&calc_load_tasks); active = active > 0 ? active * FIXED_1 : 0; @@ -2384,12 +2517,7 @@ void calc_global_load(unsigned long ticks) calc_load_update += LOAD_FREQ; /* - * Account one period with whatever state we found before - * folding in the nohz state and ageing the entire idle period. - * - * This avoids loosing a sample when we go idle between - * calc_load_account_active() (10 ticks ago) and now and thus - * under-accounting. + * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk. */ calc_global_nohz(); } @@ -2406,13 +2534,16 @@ static void calc_load_account_active(struct rq *this_rq) return; delta = calc_load_fold_active(this_rq); - delta += calc_load_fold_idle(); if (delta) atomic_long_add(delta, &calc_load_tasks); this_rq->calc_load_update += LOAD_FREQ; } +/* + * End of global load-average stuff + */ + /* * The exact cpuload at various idx values, calculated at every tick would be * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load diff --git a/kernel/sched/idle_task.c b/kernel/sched/idle_task.c index b44d604b35d1..b6baf370cae9 100644 --- a/kernel/sched/idle_task.c +++ b/kernel/sched/idle_task.c @@ -25,7 +25,6 @@ static void check_preempt_curr_idle(struct rq *rq, struct task_struct *p, int fl static struct task_struct *pick_next_task_idle(struct rq *rq) { schedstat_inc(rq, sched_goidle); - calc_load_account_idle(rq); return rq->idle; } diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h index 6d52cea7f33d..55844f24435a 100644 --- a/kernel/sched/sched.h +++ b/kernel/sched/sched.h @@ -942,8 +942,6 @@ static inline u64 sched_avg_period(void) return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; } -void calc_load_account_idle(struct rq *this_rq); - #ifdef CONFIG_SCHED_HRTICK /* diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c index 869997833928..4a08472c3ca7 100644 --- a/kernel/time/tick-sched.c +++ b/kernel/time/tick-sched.c @@ -406,6 +406,7 @@ static void tick_nohz_stop_sched_tick(struct tick_sched *ts) */ if (!ts->tick_stopped) { select_nohz_load_balancer(1); + calc_load_enter_idle(); ts->idle_tick = hrtimer_get_expires(&ts->sched_timer); ts->tick_stopped = 1; @@ -597,6 +598,7 @@ void tick_nohz_idle_exit(void) account_idle_ticks(ticks); #endif + calc_load_exit_idle(); touch_softlockup_watchdog(); /* * Cancel the scheduled timer and restore the tick From 95c0d71dcb853c2eca5f2231ebbd4c1d3af775b7 Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Tue, 3 Jul 2012 23:37:31 +0900 Subject: [PATCH 9/9] MAINTAINERS/sched: Update scheduler file pattern The commit 391e43da797a ("sched: Move all scheduler bits into kernel/sched/") moved all scheduler codes to the kernel/sched/ directory, but missed the MAINTAINERS. Since it still expects files from kernel/ directory, get_maintainer script has to rely on the git (log) fallback mechanism. $ scripts/get_maintainer.pl -f kernel/sched/core.c --nogit-fallback linux-kernel@vger.kernel.org (open list) With this patch: $ scripts/get_maintainer.pl -f kernel/sched/core.c --nogit-fallback Ingo Molnar (maintainer:SCHEDULER) Peter Zijlstra (maintainer:SCHEDULER) linux-kernel@vger.kernel.org (open list) Signed-off-by: Namhyung Kim Acked-by: Peter Zijlstra Link: http://lkml.kernel.org/r/1341326251-4140-1-git-send-email-namhyung@kernel.org Signed-off-by: Ingo Molnar --- MAINTAINERS | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/MAINTAINERS b/MAINTAINERS index 03df1d15ebf3..214d75489640 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -5909,7 +5909,7 @@ M: Ingo Molnar M: Peter Zijlstra T: git git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/core S: Maintained -F: kernel/sched* +F: kernel/sched/ F: include/linux/sched.h SCORE ARCHITECTURE