Re: [PATCH 1/2] mwifiex: Use non-posted PCI register writes

[Brian Norris]

On Wed, Sep 1, 2021 at 1:41 PM Andy Shevchenko
<andy.shevchenko@xxxxxxxxx> wrote:
>
> On Wed, Sep 1, 2021 at 11:25 PM Brian Norris <briannorris@xxxxxxxxxxxx> wrote:
> > On Mon, Aug 30, 2021 at 5:37 AM Jonas Dreßler <verdre@xxxxxxx> wrote:
>
> ...
>
> > This might be good for many cases, but please read this commit:
> >
> > https://git.kernel.org/linus/062e008a6e83e7c4da7df0a9c6aefdbc849e2bb3
> > mwifiex: pcie: use posted write to wake up firmware
> >
> > It's very much intentional that this is a posted write in some cases.
> >
> > Without ensuring this doesn't regress, NAK from me.
>
> Can you ensure that from Chrome / Google perspective, please?

I mean, the context link should be pretty obvious. But sure: with this patch:

(a) all my least-favorite audio issues to debug return (audio is
especially latency-sensitive on the Rockchip RK3399 systems I have)
(b) latency spikes to ~6ms:
# trace-cmd record -p function_graph -O funcgraph-abstime -l
mwifiex_pm_wakeup_card
# trace-cmd report
   kworker/u13:0-199   [003]   348.987306: funcgraph_entry:      #
6219.500 us |  mwifiex_pm_wakeup_card();
   kworker/u13:0-199   [003]   349.316312: funcgraph_entry:      #
6267.625 us |  mwifiex_pm_wakeup_card();
   kworker/u13:3-4057  [001]   352.238530: funcgraph_entry:      #
6184.250 us |  mwifiex_pm_wakeup_card();
   kworker/u13:0-199   [002]   356.626366: funcgraph_entry:      #
6553.166 us |  mwifiex_pm_wakeup_card();
   kworker/u13:3-4057  [002]   356.709389: funcgraph_entry:      #
6212.500 us |  mwifiex_pm_wakeup_card();
   kworker/u13:3-4057  [002]   356.847215: funcgraph_entry:      #
6230.292 us |  mwifiex_pm_wakeup_card();
   kworker/u13:3-4057  [000]   356.897576: funcgraph_entry:      #
6451.667 us |  mwifiex_pm_wakeup_card();
   kworker/u13:0-199   [004]   357.175025: funcgraph_entry:      #
6204.042 us |  mwifiex_pm_wakeup_card();

whereas it used to look more like:

   kworker/u13:1-173   [005]   212.230542: funcgraph_entry:
7.000 us   |  mwifiex_pm_wakeup_card();
   kworker/u13:3-1768  [005]   213.886063: funcgraph_entry:
9.334 us   |  mwifiex_pm_wakeup_card();
   kworker/u13:3-1768  [002]   214.473273: funcgraph_entry:      +
11.375 us  |  mwifiex_pm_wakeup_card();
   kworker/u13:3-1768  [005]   214.530705: funcgraph_entry:
5.542 us   |  mwifiex_pm_wakeup_card();
   kworker/u13:1-173   [002]   215.050168: funcgraph_entry:      +
13.125 us  |  mwifiex_pm_wakeup_card();
   kworker/u13:1-173   [002]   215.106492: funcgraph_entry:      +
11.959 us  |  mwifiex_pm_wakeup_card();
   kworker/u13:3-1768  [005]   215.484807: funcgraph_entry:
8.459 us   |  mwifiex_pm_wakeup_card();
   kworker/u13:1-173   [003]   215.515238: funcgraph_entry:      +
15.166 us  |  mwifiex_pm_wakeup_card();
   kworker/u13:3-1768  [001]   217.175691: funcgraph_entry:      +
11.083 us  |  mwifiex_pm_wakeup_card();

I think it would be enough to only modify mwifiex_pm_wakeup_card() to
retain the posted version, but I'm not sure how that fits in with the
rest of the author's work. I suppose I can give a later version a run
if needed.

Brian