On 10/31/20 12:45 AM, Marek Behún wrote:

On Fri, 30 Oct 2020 23:37:52 +0100
Jacek Anaszewski [off-list ref] wrote:

quoted

Hi Marek,

Bitops are guaranteed to be atomic and this was done for a reason.

Hmm okay...
Sooo, netdev_trig_work cannot be executed at the same time as the
link/linkup/rx/tx changing stuff from netdev_trig_notify,
interval_store or netdev_led_attr_store, because all these functions
ensure cancelation of netdev_trig_work by calling
cancel_delayed_work_sync. Doesn't this somehow prevent the need for
memory barriers provided by atomic bitops?

That's true. But unless there is proven decline in performance related
to use of bitops, I don't see any gain in removing those from this
trigger. They improve code readability.

BTW Jacek what do you think about the other patches?

I like the idea, but I'd need to spend more time reviewing it.
One thing coming to mind at first glance - it would be good to get rid
of blink_set op at this occasion since this is just another case of
trigger offloading. It would however need touching many drivers, so
that could possibly be done as a follow-up.

-- 
Best regards,
Jacek Anaszewski