Re: [Intel-wired-lan] [PATCH iwl-next] ice: allow hot-swapping XDP programs

["Toke Høiland-Jørgensen" <toke@kernel.org>]

Maciej Fijalkowski <maciej.fijalkowski@intel.com> writes:

> On Wed, Jun 14, 2023 at 02:40:07PM +0200, Alexander Lobakin wrote:
>> From: Toke Høiland-Jørgensen <toke@kernel.org>
>> Date: Tue, 13 Jun 2023 19:59:37 +0200
>> 
>> > Maciej Fijalkowski <maciej.fijalkowski@intel.com> writes:
>> > 
>> >> On Tue, Jun 13, 2023 at 05:15:15PM +0200, Alexander Lobakin wrote:
>> >>> From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
>> >>> Date: Tue, 13 Jun 2023 17:10:05 +0200
>> 
>> [...]
>> 
>> >> Since we removed rcu sections from driver sides and given an assumption
>> >> that local_bh_{dis,en}able() pair serves this purpose now i believe this
>> >> is safe. Are you aware of:
>> >>
>> >> https://lore.kernel.org/bpf/20210624160609.292325-1-toke@redhat.com/
>> 
>> Why [0] then? Added in [1] precisely for the sake of safe XDP prog
>> access and wasn't removed :s I was relying on that one in my suggestions
>> and code :D
>> 
>> > 
>> > As the author of that series, I agree that it's not necessary to add
>> > additional RCU protection. ice_vsi_assign_bpf_prog() already uses xchg()
>> > and WRITE_ONCE() which should protect against tearing, and the xdp_prog
>> > pointer being passed to ice_run_xdp() is a copy residing on the stack,
>> > so it will only be read once per NAPI cycle anyway (which is in line
>> > with how most other drivers do it).
>> 
>> What if a NAPI polling cycle is being run on one core while at the very
>> same moment I'm replacing the XDP prog on another core? Not in terms of
>> pointer tearing, I see now that this is handled correctly, but in terms
>> of refcounts? Can't bpf_prog_put() free it while the polling is still
>> active?
>
> Hmm you mean we should do bpf_prog_put() *after* we update bpf_prog on
> ice_rx_ring? I think this is a fair point as we don't bump the refcount
> per each Rx ring that holds the ptr to bpf_prog, we just rely on the main
> one from VSI.

Yes, that's true, the duplication of the pointer in all the ring
structures can lead to problems there (why is that done in the first
place?). I agree that swapping the order of the pointer assignments
should be enough to fix this.

>> > It *would* be nice to add an __rcu annotation to ice_vsi->xdp_prog and
>> > ice_rx_ring->xdp_prog (and move to using rcu_dereference(),
>> > rcu_assign_pointer() etc), but this is more a documentation/static
>> > checker thing than it's a "correctness of the generated code" thing :)
>
> Agree but I would rather address the rest of Intel drivers in the
> series.

That's fair :)

-Toke