Skip to content

hbt/bperf: replace compiler barrier with CPU fence std::atomic_thread_fence(std::memory_order_consume) #470

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
luciang wants to merge 1 commit into
base: main
Choose a base branch
from
Open

hbt/bperf: replace compiler barrier with CPU fence std::atomic_thread_fence(std::memory_order_consume) #470

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 3 additions & 4 deletions hbt/src/perf_event/BPerfPerThreadReader.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
#include "hbt/src/perf_event/BPerfPerThreadReader.h"
#include <bpf/bpf.h>
#include <time.h>
#include <atomic>
#include "hbt/src/perf_event/BPerfEventsGroup.h"

namespace facebook::hbt::perf_event {
Expand Down Expand Up @@ -186,8 +187,6 @@ BPerfPerThreadReader::~BPerfPerThreadReader() {
disable();
}

#define barrier() asm volatile("" ::: "memory")

int BPerfPerThreadReader::read(struct BPerfThreadData* data) {
struct bperf_clock_param *ptr = &data_->tsc_param, p;
struct bperf_thread_data raw_thread_data;
Expand All @@ -203,7 +202,7 @@ int BPerfPerThreadReader::read(struct BPerfThreadData* data) {

do {
lock = data_->lock;
barrier();
std::atomic_thread_fence(std::memory_order_consume);
tsc = rdtsc();
p = *ptr;
raw_thread_data = *data_;
Expand All @@ -216,7 +215,7 @@ int BPerfPerThreadReader::read(struct BPerfThreadData* data) {
pmc_val[i] = 0;
}
}
barrier();
std::atomic_thread_fence(std::memory_order_consume);
} while (lock != data_->lock);

data->monoTime = (((__uint128_t)tsc * p.multi) >> p.shift) + p.offset +
Expand Down