On Fri, 2016-11-25 at 11:06 +0100, Arend van Spriel wrote:

On 11/25/2016 9:25 AM, Johannes Berg wrote:

quoted

Sorry, forgot to reply to this until Luca's email bumped it up...

On Tue, 2016-11-22 at 21:06 +0100, Arend Van Spriel wrote:

quoted

Are we? Currently, the minimum is not checked in nl80211, but that
does not say anything about the driver which might validate the
interval as well and return an error.

Well, since drivers currently don't return an error (even if they
ignore the value!) that *does* change the API.

quoted

What made me start looking at this is that in brcmfmac the interval
in the request was ignored and a fixed interval was provisioned in
the device. I wanted to fix that but was not sure if I needed to
check it against our firmware min..max range and what the appropriate
error handling should be. If silently changing what user-space is
requesting is fine for this, I am happy to make it so. Preferably in
nl80211.

I think (agreeing with Luca) bumping it up is fine.

Fine by me although the "drift over time" reason seems only more reason 
to have minimum validation mainly because nowhere is nl80211.h it is 
stated that the interval is a "soft" requirement. Now Luca proposes 
bumping to minimum should be done in the driver. What is your opinion?

I will update the kernel doc to clarify what can be expected from the 
interval value.

Yeah, I was almost sure there was a statement somewhere that the
interval is "soft", but there isn't.  I was confusing with the match
logic, which is clearly documented as not-guaranteed: "...there is no
guarantee that the reported BSSs are fully complying with the match
sets and userspace needs to be able to ignore them by itself."

A clarification in the documentation would be great.

--
Luca.