diff --git a/staticlibs/native/bpf_headers/BpfRingbufTest.cpp b/staticlibs/native/bpf_headers/BpfRingbufTest.cpp
index e4de812..e81fb92 100644
--- a/staticlibs/native/bpf_headers/BpfRingbufTest.cpp
+++ b/staticlibs/native/bpf_headers/BpfRingbufTest.cpp
@@ -74,11 +74,27 @@
     ASSERT_RESULT_OK(result);
     EXPECT_TRUE(result.value()->isEmpty());
 
+    struct timespec t1, t2;
+    EXPECT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &t1));
+    EXPECT_FALSE(result.value()->wait(1000 /*ms*/));  // false because wait should timeout
+    EXPECT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &t2));
+    long long time1 = t1.tv_sec * 1000000000LL + t1.tv_nsec;
+    long long time2 = t2.tv_sec * 1000000000LL + t2.tv_nsec;
+    EXPECT_GE(time2 - time1, 1000000000 /*ns*/);  // 1000 ms as ns
+
     for (int i = 0; i < n; i++) {
       RunProgram();
     }
 
     EXPECT_FALSE(result.value()->isEmpty());
+
+    EXPECT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &t1));
+    EXPECT_TRUE(result.value()->wait());
+    EXPECT_EQ(0, clock_gettime(CLOCK_MONOTONIC, &t2));
+    time1 = t1.tv_sec * 1000000000LL + t1.tv_nsec;
+    time2 = t2.tv_sec * 1000000000LL + t2.tv_nsec;
+    EXPECT_LE(time2 - time1, 1000000 /*ns*/);  // in x86 CF testing < 5000 ns
+
     EXPECT_THAT(result.value()->ConsumeAll(callback), HasValue(n));
     EXPECT_TRUE(result.value()->isEmpty());
     EXPECT_EQ(output, TEST_RINGBUF_MAGIC_NUM);
diff --git a/staticlibs/native/bpf_headers/include/bpf/BpfRingbuf.h b/staticlibs/native/bpf_headers/include/bpf/BpfRingbuf.h
index 9aff790..d716358 100644
--- a/staticlibs/native/bpf_headers/include/bpf/BpfRingbuf.h
+++ b/staticlibs/native/bpf_headers/include/bpf/BpfRingbuf.h
@@ -19,6 +19,7 @@
 #include <android-base/result.h>
 #include <android-base/unique_fd.h>
 #include <linux/bpf.h>
+#include <poll.h>
 #include <sys/mman.h>
 #include <utils/Log.h>
 
@@ -41,6 +42,9 @@
 
   bool isEmpty(void);
 
+  // returns !isEmpty() for convenience
+  bool wait(int timeout_ms = -1);
+
  protected:
   // Non-initializing constructor, used by Create.
   BpfRingbufBase(size_t value_size) : mValueSize(value_size) {}
@@ -200,12 +204,21 @@
 }
 
 inline bool BpfRingbufBase::isEmpty(void) {
-  uint32_t prod_pos = mProducerPos->load(std::memory_order_acquire);
-  // Only userspace writes to mConsumerPos, so no need to use std::memory_order_acquire
+  uint32_t prod_pos = mProducerPos->load(std::memory_order_relaxed);
   uint64_t cons_pos = mConsumerPos->load(std::memory_order_relaxed);
   return (cons_pos & 0xFFFFFFFF) == prod_pos;
 }
 
+inline bool BpfRingbufBase::wait(int timeout_ms) {
+  // possible optimization: if (!isEmpty()) return true;
+  struct pollfd pfd = {  // 1-element array
+    .fd = mRingFd.get(),
+    .events = POLLIN,
+  };
+  (void)poll(&pfd, 1, timeout_ms);  // 'best effort' poll
+  return !isEmpty();
+}
+
 inline base::Result<int> BpfRingbufBase::ConsumeAll(
     const std::function<void(const void*)>& callback) {
   int64_t count = 0;