John Fastabend <john.fastabend@xxxxxxxxx> writes: > Björn Töpel wrote: >> From: Björn Töpel <bjorn.topel@xxxxxxxxx> >> >> Now that all XDP maps that can be used with bpf_redirect_map() tracks >> entries to be flushed in a global fashion, there is not need to track >> that the map has changed and flush from xdp_do_generic_map() >> anymore. All entries will be flushed in xdp_do_flush_map(). >> >> This means that the map_to_flush can be removed, and the corresponding >> checks. Moving the flush logic to one place, xdp_do_flush_map(), give >> a bulking behavior and performance boost. >> >> Acked-by: Toke Høiland-Jørgensen <toke@xxxxxxxxxx> >> Signed-off-by: Björn Töpel <bjorn.topel@xxxxxxxxx> >> --- > > __dev_map_flush() still has rcu_read_lock/unlock() around flush_list by > this point, assuming I've followed along correctly. Can we drop those > now seeing its per CPU and all list ops are per-cpu inside napi context? Hmm, I guess so? :) > Two reasons to consider, with this patch dev_map_flush() is always > called even if the list is empty so even in TX case without redirect. > But probably more important it makes the locking requirements more clear. > Could probably be done in a follow up patch but wanted to bring it up. This series was already merged, but I'll follow up with the non-map redirect change. This requires a bit of refactoring anyway, so I can incorporate the lock removal into that... -Toke
