On Fri, 11 Jul 2025 13:26:28 -0700 Namhyung Kim <namhyung@xxxxxxxxxx> wrote: > > - Removed use of timestamp. The deferred unwind has gone back to using > > cookies, and perf doesn't use the cookie. This means the > > struct perf_callchain_deferred_event is not modified. > > What about adding the cookies in the records to handle lost data? Even > if it's not necessary to match callchains to samples, it still needs to > reject invalid callchains across the losts. Maybe it can just flush > pending samples when it sees LOST records and not try to match them but > having the cookies will handle it more accurately as some callchains may > be valid after the LOST. Sure, I can add that back, with the added comments needed that Jens suggested. Thanks, -- Steve
