On Fri, 7 Jun 2024 10:29:03 +0200 Petr Pavlu <petr.pavlu@xxxxxxxx> wrote: > Another option could be to try traversing the whole list in smaller > parts and give up the reader_lock in between them. This would need some > care to make sure that the operation completes, e.g. the code would need > to bail out if it detects a change on cpu_buffer->pages_read. I think I like this approach the most. Perhaps even have a counter that gets incremented everything a new reader page is taken. And if it detects that, it restarts the check? To prevent a DOS, we restart 3 times at most, and then just say "the list is OK" and exit. So basically, we release the lock within the loop per each sub-buffer, and then check if the reader touch it when reacquiring the lock. -- Steve
