diff options
author | Nicholas Piggin <npiggin@gmail.com> | 2020-06-25 15:34:03 +1000 |
---|---|---|
committer | Steven Rostedt (VMware) <rostedt@goodmis.org> | 2020-06-30 17:18:56 -0400 |
commit | b23d7a5f4a07af02343cdd28fe1f7488bac3afda (patch) | |
tree | c775987b3e84915ec4c22a410f65c8c145ee70ce /kernel/trace/ring_buffer.c | |
parent | 10464b4aa605ef93c937452f442e74cc0a4a6ceb (diff) | |
download | linux-b23d7a5f4a07af02343cdd28fe1f7488bac3afda.tar.bz2 |
ring-buffer: speed up buffer resets by avoiding synchronize_rcu for each CPU
On a 144 thread system, `perf ftrace` takes about 20 seconds to start
up, due to calling synchronize_rcu() for each CPU.
cat /proc/108560/stack
0xc0003e7eb336f470
__switch_to+0x2e0/0x480
__wait_rcu_gp+0x20c/0x220
synchronize_rcu+0x9c/0xc0
ring_buffer_reset_cpu+0x88/0x2e0
tracing_reset_online_cpus+0x84/0xe0
tracing_open+0x1d4/0x1f0
On a system with 10x more threads, it starts to become an annoyance.
Batch these up so we disable all the per-cpu buffers first, then
synchronize_rcu() once, then reset each of the buffers. This brings
the time down to about 0.5s.
Link: https://lkml.kernel.org/r/20200625053403.2386972-1-npiggin@gmail.com
Tested-by: Anton Blanchard <anton@ozlabs.org>
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Diffstat (limited to 'kernel/trace/ring_buffer.c')
-rw-r--r-- | kernel/trace/ring_buffer.c | 85 |
1 files changed, 71 insertions, 14 deletions
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c index 802bb38d9c81..ed1941304f69 100644 --- a/kernel/trace/ring_buffer.c +++ b/kernel/trace/ring_buffer.c @@ -270,6 +270,9 @@ EXPORT_SYMBOL_GPL(ring_buffer_event_data); #define for_each_buffer_cpu(buffer, cpu) \ for_each_cpu(cpu, buffer->cpumask) +#define for_each_online_buffer_cpu(buffer, cpu) \ + for_each_cpu_and(cpu, buffer->cpumask, cpu_online_mask) + #define TS_SHIFT 27 #define TS_MASK ((1ULL << TS_SHIFT) - 1) #define TS_DELTA_TEST (~TS_MASK) @@ -4790,6 +4793,26 @@ rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) rb_head_page_activate(cpu_buffer); } +/* Must have disabled the cpu buffer then done a synchronize_rcu */ +static void reset_disabled_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer) +{ + unsigned long flags; + + raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); + + if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) + goto out; + + arch_spin_lock(&cpu_buffer->lock); + + rb_reset_cpu(cpu_buffer); + + arch_spin_unlock(&cpu_buffer->lock); + + out: + raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); +} + /** * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer * @buffer: The ring buffer to reset a per cpu buffer of @@ -4798,7 +4821,6 @@ rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer) void ring_buffer_reset_cpu(struct trace_buffer *buffer, int cpu) { struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; - unsigned long flags; if (!cpumask_test_cpu(cpu, buffer->cpumask)) return; @@ -4809,24 +4831,42 @@ void ring_buffer_reset_cpu(struct trace_buffer *buffer, int cpu) /* Make sure all commits have finished */ synchronize_rcu(); - raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags); + reset_disabled_cpu_buffer(cpu_buffer); - if (RB_WARN_ON(cpu_buffer, local_read(&cpu_buffer->committing))) - goto out; + atomic_dec(&cpu_buffer->record_disabled); + atomic_dec(&cpu_buffer->resize_disabled); +} +EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); - arch_spin_lock(&cpu_buffer->lock); +/** + * ring_buffer_reset_cpu - reset a ring buffer per CPU buffer + * @buffer: The ring buffer to reset a per cpu buffer of + * @cpu: The CPU buffer to be reset + */ +void ring_buffer_reset_online_cpus(struct trace_buffer *buffer) +{ + struct ring_buffer_per_cpu *cpu_buffer; + int cpu; - rb_reset_cpu(cpu_buffer); + for_each_online_buffer_cpu(buffer, cpu) { + cpu_buffer = buffer->buffers[cpu]; - arch_spin_unlock(&cpu_buffer->lock); + atomic_inc(&cpu_buffer->resize_disabled); + atomic_inc(&cpu_buffer->record_disabled); + } - out: - raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); + /* Make sure all commits have finished */ + synchronize_rcu(); - atomic_dec(&cpu_buffer->record_disabled); - atomic_dec(&cpu_buffer->resize_disabled); + for_each_online_buffer_cpu(buffer, cpu) { + cpu_buffer = buffer->buffers[cpu]; + + reset_disabled_cpu_buffer(cpu_buffer); + + atomic_dec(&cpu_buffer->record_disabled); + atomic_dec(&cpu_buffer->resize_disabled); + } } -EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); /** * ring_buffer_reset - reset a ring buffer @@ -4834,10 +4874,27 @@ EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu); */ void ring_buffer_reset(struct trace_buffer *buffer) { + struct ring_buffer_per_cpu *cpu_buffer; int cpu; - for_each_buffer_cpu(buffer, cpu) - ring_buffer_reset_cpu(buffer, cpu); + for_each_buffer_cpu(buffer, cpu) { + cpu_buffer = buffer->buffers[cpu]; + + atomic_inc(&cpu_buffer->resize_disabled); + atomic_inc(&cpu_buffer->record_disabled); + } + + /* Make sure all commits have finished */ + synchronize_rcu(); + + for_each_buffer_cpu(buffer, cpu) { + cpu_buffer = buffer->buffers[cpu]; + + reset_disabled_cpu_buffer(cpu_buffer); + + atomic_dec(&cpu_buffer->record_disabled); + atomic_dec(&cpu_buffer->resize_disabled); + } } EXPORT_SYMBOL_GPL(ring_buffer_reset); |