logd: remove start filtration from flushTo
We have already searched for the start point, the start filter check
is paranoia that removes out-of-order entries that we are undoubtably
interested in. Out-of-order entries occur under reader pressure, as
the writer gets pushed back from in-place sorted order and lands it
at the end for the reader to pick it up. If this occurred during a
batch run or a logger thread wakeup, the entry could be filtered out
and never output to the reader.
Found one case where logcat.tail_time* tests failed which was fixed
with this adjustment.
Test: gTest logd-unit-tests, liblog-unit-tests and logcat-unit-tests
Bug: 38046067
Bug: 37791296
Change-Id: Icbde6b33dca7ab98348c3a872793aeef3997d460
diff --git a/logd/LogBuffer.cpp b/logd/LogBuffer.cpp
index ded6c8c..fbed83b 100644
--- a/logd/LogBuffer.cpp
+++ b/logd/LogBuffer.cpp
@@ -1142,10 +1142,6 @@
continue;
}
- if (element->getRealTime() <= start) {
- continue;
- }
-
// NB: calling out to another object with wrlock() held (safe)
if (filter) {
int ret = (*filter)(element, arg);
@@ -1172,11 +1168,10 @@
unlock();
// range locking in LastLogTimes looks after us
- max = element->flushTo(reader, this, privileged, sameTid);
+ log_time next = element->flushTo(reader, this, privileged, sameTid);
- if (max == element->FLUSH_ERROR) {
- return max;
- }
+ if (next == element->FLUSH_ERROR) return next;
+ if (next > max) max = next;
skip = maxSkip;
rdlock();