Re: [PATCH V6 4/3] brcmfmac: use wiphy_read_of_freq_limits to respect extra limits

On 5-1-2017 11:02, Rafał Miłecki wrote:

On 5 January 2017 at 10:31, Arend Van Spriel
[off-list ref] wrote:

quoted

On 4-1-2017 22:19, Rafał Miłecki wrote:

quoted

On 4 January 2017 at 21:07, Arend Van Spriel
[off-list ref] wrote:

quoted

On 4-1-2017 18:58, Rafał Miłecki wrote:

quoted

From: Rafał Miłecki <rafal@milecki.pl>

There are some devices (e.g. Netgear R8000 home router) with one chipset
model used for different radios, some of them limited to subbands. NVRAM
entries don't contain any extra info on such limitations and firmware
reports full list of channels to us. We need to store extra limitation
info in DT to support such devices properly.

Now there is a cfg80211 helper for reading such info use it in brcmfmac.
This patch adds check for channel being disabled with orig_flags which
is how this wiphy helper and wiphy_register work.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
---
This patch should probably go through wireless-driver-next which is why
it got weird number 4/3. I'm sending it just as a proof of concept.
It was succesfully tested on SmartRG SR400ac with BCM43602.

V4: Respect IEEE80211_CHAN_DISABLED in orig_flags
V5: Update commit message
V6: Call wiphy_read_of_freq_limits after brcmf_setup_wiphybands to make it work
    with helper setting "flags" instead of "orig_flags".
---
 drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
index ccae3bb..a008ba5 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c

@@ -5886,6 +5886,9 @@ static int brcmf_construct_chaninfo(struct brcmf_cfg80211_info *cfg,
                                                     band->band);
              channel[index].hw_value = ch.control_ch_num;

+             if (channel->orig_flags & IEEE80211_CHAN_DISABLED)
+                     continue;
+

So to be clear this is still needed for subsequent calls to
brcmf_setup_wiphybands(). The subsequent calls are done from the
regulatory notifier. So I think we have an issue with this approach. Say
the device comes up with US. That would set DISABLED flags for channels
12 to 14. With a country update to PL we would want to enable channels
12 and 13, right? The orig_flags are copied from the initial flags
during wiphy_register() so it seems we will skip enabling 12 and 13. I
think we should remove the check above and call
wiphy_read_of_freq_limits() as a last step within
brcmf_setup_wiphybands(). It means it is called every time, but it
safeguards that the limits in DT are always applied.

I'm not exactly happy with channels management in brcmfmac. Before
calling wiphy_register it already disables channels unavailable for
current country. This results in setting IEEE80211_CHAN_DISABLED in

What do you mean by current country. There is none that we are aware off
in the driver. So we obtain the channels for the current
country/revision in the firmware and enable those before
wiphy_register(). This all is within the probe/init sequence so I do not
really see an issue. As the wiphy object is not yet registered there is
no user-space awareness

It seems I'm terrible as describing my patches/problems/ideas :( Here
I used 1 inaccurate word and you couldn't understand my point.

Well. Because of our track record of miscommunication and other
annoyances I want to be sure this time :-p

By "current country" I meant current region (and so a set of
regulatory rules) used by the firmware. I believe firmware describes
it using "ccode" and "regrev".

Now, about the issue I see:

AFAIU if you set IEEE80211_CHAN_DISABLED in "orig_flags" it's meant to
be there for good. Some reference code that makes me believe so

Indeed. Admittedly, it is the first time I became aware of it when
Johannes brought it up.

(reg.c):
pr_debug("Disabling freq %d MHz for good\n", chan->center_freq);
chan->orig_flags |= IEEE80211_CHAN_DISABLED;

This is what happens with brcmfmac right now. If firmware doesn't
report some channels, you set "flags" to IEEE80211_CHAN_DISABLED for
them. Then you call wiphy_register which copies that "flags" to the
"orig_flags". I read it as: we are never going to use these channels.

Now, consider you support regdom change (I do with my local patches).
You translate alpha2 to a proper firmware request (board specific!),
you execute it and then firmware may allow you to use channels that
you marked as disabled for good. You would need to mess with
orig_flags to recover from this issue.

Does my explanation make more sense of this issue now?

Sure. It seems we should leave all channels enabled and disable them
after wiphy_register() based on firmware information. Or did you have
something else in mind. DFS channels may need to pass a feature check in
firmware and always be disabled otherwise.

quoted

quoted

orig_flags of channels that may become available later, after country
change. Please note it happens even right now, without this patch.

Nope. As stated earlier the country setting in firmware is not updated
unless you provide a *proper* mapping of user-space country code to
firmware country code/revision. That is the reason, ie. firmware simply
returns the same list of channels as nothing changed from its
perspective. We may actually drop 11d support.

I implemented mapping support locally, this is the feature I'm talking about.

Ok. So this is not an upstream feature. Or are you getting the mapping
from DT?

quoted

quoted

Maybe you can workaround this by ignoring orig_flags in custom
regulatory code, but I'd just prefer to have it nicely handled in the
first place.

Please care to explain your ideas before putting any effort in this
"feature". As the author of the code that makes you unhappy and as
driver maintainer I would like to get a clearer picture of your point of
view. What exactly is the issue that makes you unhappy.

I meant that problem with "orig_flags" I described in the first
paragraph. I wasn't trying to hide whatever issue I'm seeing, I swear
;)

quoted

quoted

This is the next feature I'm going to work on. AFAIU this patch won't
be applied for now (it's for wireless-drivers-next and we first need
to get wiphy_read_of_freq_limits in that tree). By the time that
happens I may have another patchset cleaning brcmfmac ready. And FWIW
this patch wouldn't make things worse *at this point* as we don't
really support country switching for any device yet.

Now who is *we*? We as Broadcom can, because we know how to map the ISO
3166-1 country code to firmware country code/revision for a specific
firmware release. Firmware uses its own regulatory rules which may
differ from what regdb has. Now I know Intel submitted a mechanism to
export firmware rules to regdb so maybe we should consider switching to
that api if that has been upstreamed. Need to check.

We as a driver developers. Please read
"we don't really support country switching for any device yet"
as
"brcmfmac doesn't really support country switching for any device yet"

Does it help to get the context?

I indeed prefer to talk about the driver instead of we. Indeed it is
true due to the orig_flags behavior although that only seems to involve
regulatory code. Could it be that brcmfmac undo that through the notifier?

Regards,
Arend