Received-SPF: pass (google.com: best guess record for domain of linux-wireless-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200324075216.22553-1-yhchuang@realtek.com> <20200324075216.22553-2-yhchuang@realtek.com>
 <20200324165105.GA99185@google.com> <c17c6e7243e642679a38854e6e6cda96@realtek.com>
In-Reply-To: <c17c6e7243e642679a38854e6e6cda96@realtek.com>
From:   Brian Norris <briannorris@chromium.org>
Date:   Wed, 25 Mar 2020 18:45:19 -0700
Message-ID: <CA+ASDXM49NW1atMoGgEgU0R7siCnhLf6eRawReGiD=PNe-hMrA@mail.gmail.com>
Subject: Re: [PATCH v6 1/2] rtw88: add regulatory process strategy for
 different chipset
To:     Andy Huang <tehuang@realtek.com>
Cc:     Tony Chuang <yhchuang@realtek.com>,
        "kvalo@codeaurora.org" <kvalo@codeaurora.org>,
        "linux-wireless@vger.kernel.org" <linux-wireless@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-wireless-owner@vger.kernel.org
Precedence: bulk

[ I'll preface this by saying that the more I look at the regulatory
core, the more I realize I'm confused or wrong at times. So forgive me
if I've made errors along the way, and please do correct me. ]

On Tue, Mar 24, 2020 at 8:11 PM Andy Huang <tehuang@realtek.com> wrote:
> > On Tue, Mar 24, 2020 at 03:52:15PM +0800, yhchuang@realtek.com wrote:
> > > --- a/drivers/net/wireless/realtek/rtw88/Kconfig
> > > +++ b/drivers/net/wireless/realtek/rtw88/Kconfig

> > I'm still not sure why rtw88 needs this, and nobody else does. I read
>
> I think in Atheros driver, ATH_REG_DYNAMIC_USER_REG_HINTS config serves
> the same purpose.

Ah, I forgot about that one, sorry.

> > your commit message, but that doesn't sound like something that belongs
> > in a single driver still.
> >
>
> As our previous commit message claims, it is due to FCC [...]

Yes, I saw that: my point was that effectively all drivers are subject
to this FCC rule, and so this could be a common CONFIG_*. But if we
already have the ATH_* one (I missed that, above), I guess we can have
an rtw88 one too. It might be less confusing (and more
straightforwardly-implemented) if we moved this stuff to the core
someday, though.

> > > +   ret = regulatory_hint(hw->wiphy, rtwdev->efuse.country_code);
> > > +   if (ret)
> > > +           rtw_warn(rtwdev, "failed to hint regulatory: %d\n", ret);
> >
> > I don't think this is what you want; you had it right in previous
> > revisions:
> >
> >       if (!rtwdev->efuse.country_worldwide) {
> >               if (regulatory_hint(hw->wiphy, rtwdev->efuse.country_code))
> >                       rtw_err( ... );
> >       }
> >
> > Without the 'country_worlwide' check, you start "hinting" (even on
> > worldwide chips) that you really wanted "country" 00 only, and so we
> > *never* adapt to more strict country settings. That's not how world-wide
> > settings are supposed to work.
>
> It doesn't mean that we want country 00 only, we will get country notifies
> from stack, and we will apply it if we accept it. We don't want stack to change
> the channel plan for us.

I noted this to you privately, but I don't believe it's expected to
call regulatory_hint() with "00". See the kerneldoc:

 * @alpha2: the ISO/IEC 3166 alpha2 the driver claims its regulatory domain
 *      should be in. If @rd is set this should be NULL. Note that if you
 *      set this to NULL you should still set rd->alpha2 to some accepted
 *      alpha2.

Note that "00" is *not* actually an ISO 3166 alpha2 code.

The key problem I'm seeing: once you do this, you establish a
wiphy-specific regd, and this regd never updates its country code or
DFS region according to IE updates. So attributes like
NL80211_ATTR_DFS_REGION and NL80211_ATTR_REG_ALPHA2 remain unset.

Your previous revision -- which for WW settings used
wiphy_apply_custom_regulatory() and *not* regulatory_hint() -- did not
have that problem.

> > Why are you ignoring SET_BY_DRIVER?
>
> Since the notification with NL80211_REGDOM_SET_BY_DRIVER flag might
> comes from an another chipset's regulatory_hint().

Ack.

Brian