Re: [PATCH] mt76: mt7603: add debugfs knob to enable/disable edcca

[Lorenzo Bianconi <lorenzo@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 3644 bytes --]

> On Wed, May 15, 2019 at 01:13:49PM +0200, Lorenzo Bianconi wrote:
> > > On Wed, May 15, 2019 at 12:03:44PM +0200, Lorenzo Bianconi wrote:
> > > > > On Wed, May 15, 2019 at 11:43:55AM +0200, Lorenzo Bianconi wrote:
> > > > > > > On Mon, May 13, 2019 at 11:48:37AM +0200, Stanislaw Gruszka wrote:
> > > > > > > > On Mon, May 13, 2019 at 10:41:28AM +0200, Lorenzo Bianconi wrote:
> > > > > > > > > > Lorenzo Bianconi <lorenzo@kernel.org> writes:
> > > > > > > > > > 
> > > > > > > > > > > Introduce a knob in mt7603 debugfs in order to enable/disable
> > > > > > > > > > > edcca processing
> > > > > > > > > > >
> > > > > > > > > > > Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> > > > > > > > > > 
> > > > > > > > > > It's good to explain what edcca does and how the file is used supposed
> > > > > > > > > > to be used. In other words, have a small introduction for the user.
> > > > > > > > > 
> > > > > > > > > Hi Kalle,
> > > > > > > > > 
> > > > > > > > > edcca is used for adjusting energy detect based on CCA thresholds.
> > > > > > > > > The code was already there so I just reported the acronym.
> > > > > > > > 
> > > > > > > > What for it is needed ?
> > > > > > > 
> > > > > > > Care to comment why EDCCA is needed at all ?
> > > > > > > 
> > > > > > > Taking that debugfs file that enable it is read-only, it looks like
> > > > > > > feature that nobody needs nor tests.
> > > > > > 
> > > > > > already fixed in v2
> > > > > > https://patchwork.kernel.org/patch/10940645/
> > > > > 
> > > > > I'm aware of this patch and other one for mt76x02. But so far in the
> > > > > sources EDCCA is disabled for mt76x02 without possibility to enable it
> > > > > (and this permission file issue was pointed by Kalle during review, not
> > > > > by someone who want to test EDCCA). So again, what for EDCCA is needed ?
> > > > 
> > > > As I have already written in a previous email, ED/CCA is used to control tx power
> > > > according to the CCA MIB counters (e.g do not transmit if the channel busy time
> > > > is higher than 90% for given amount of time in a row). I guess it is required
> > > > by ETSI regulatory.
> > > But what is user case for that, i.e. who need this (it wasn't implemented in
> > > mt76x2 since you added it on Dec 2018). What will happen if it will be removed?
> > > 
> > > > Regarding file permission for mt76x02 debugfs edcca node is a typo.
> > > Typo or not, effectively disable the feature and show nobody is
> > > testing it.
> > > 
> > > The reason I'm asking is that seems EDCCA is the main reason to
> > > implement watchod for mt76x2, it wasn't necessary to have a watchdog
> > > as seems devices did not hung before EDCCA was added.
> > 
> > IIRC I added the first watchdog implementation to fix tx hangs that occur
> > under heavy load even using FCC regulatory (so when EDCCA processing is
> > disabled)
> 
> There was changes in various registers programming introduced by EDCCA
> support, even with EDCCA disabled. It's rally not convenient that
> watchdog and EDCCA are not related, since you added tx hung watchdog
> 2 weeks after adding EDCCA.
> 
> You can look at this report:
> https://github.com/openwrt/mt76/issues/246
> Before mt76x2e worked without hungs & watchodg. Now, even with EDCCA
> disabled watchdog and HW restarts are required to fix hungs on runtime.

Tx hangs occur in very particular conditions (e.g 200Mbps bidirectional
traffic) and moreover they do not always occur so I am not convinced they
are always EDCCA related and so I am not confident to remove the watchdog

Lorenzo

> 
> Stanislaw

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]