Christian Lamparter wrote:
> The issue with r92su is: The driver is ready, but it's fundamentally
^^^^^^^^
> incompatible with the rtlwifi framework and mac80211.
_anything_ is better than a staging driver.
There are distributions that don't build staging drivers.
So, please, replace and delete staging/rtl8712 with r92su.
thank you.
On 11/04/2013 08:17 AM, Xose Vazquez Perez wrote:
> Christian Lamparter wrote:
>
>> The issue with r92su is: The driver is ready, but it's fundamentally
> ^^^^^^^^
>> incompatible with the rtlwifi framework and mac80211.
>
> _anything_ is better than a staging driver.
>
> There are distributions that don't build staging drivers.
As a result, they force their users to build out-of-tree drivers!
> So, please, replace and delete staging/rtl8712 with r92su.
You seem not to understand. Any new wireless driver that does not use mac80211
*must* go into staging. AFAIK, that rule is absolute, and r92su cannot go into
the regular wireless tree.
If you do not like having drivers for RTL8192SU in staging, then write one that
uses mac80211. You have access to as much information about the chips as anyone
outside Realtek. They have no interest in doing the mac80211 version, and I have
many other tasks with higher priority.
Larry
On Monday, November 04, 2013 05:04:20 PM you wrote:
> On Mon, 2013-11-04 at 16:47 +0100, Christian Lamparter wrote:
>
> > Be careful ;-). mac80211 is great for softmac/umac/halfmac/semimac devices.
> > However the rtl8712 firmware interface is fullmac-like, so it needs cfg80211.
> > ... and r92su uses cfg80211 exclusively. There's no more WEXT or realtek ioctl
> > legacy API support.
>
> still seems there's duplicate stuff in there like the michael MIC, at
> least lib80211 has that?
Yes, there is even more duplicated code if you factor in the (de-)fragmentation
and deaggregation code from mac80211. Furthermore r92su could be sharing a bit
of infrastructure with rtlwifi.
I didn't include lib80211 since it reports mic failures via WEXT:
wireless_send_event(dev, IWEVMICHAELMICFAILURE, &wrqu, (char *)&ev);
(And I don't think this can be changed easily without touching ipw2200
and hostap-driver).
Regards,
Christian
On 11/04/2013 11:04 AM, Johannes Berg wrote:
> On Mon, 2013-11-04 at 16:47 +0100, Christian Lamparter wrote:
>
>> Be careful ;-). mac80211 is great for softmac/umac/halfmac/semimac devices.
>> However the rtl8712 firmware interface is fullmac-like, so it needs cfg80211.
>> ... and r92su uses cfg80211 exclusively. There's no more WEXT or realtek ioctl
>> legacy API support.
>
> still seems there's duplicate stuff in there like the michael MIC, at
> least lib80211 has that? OTOH, that's not all that much code.
>
> johannes
>
FWIW I separately wrote a "rtlfmac" cfg80211 driver that, after all was
said and done, looked a lot like a barebones r92su. I have been using
my driver for a few weeks now and it looks like multiple people have
reported success using r92su. I believe a fullmac/cfg80211 approach is
completely sufficient and far superior to the mess that is rtl8712. I
would be in favor of r92su replacing staging/rtl8712/.
Josh
On Monday, November 04, 2013 09:34:28 AM you wrote:
> On 11/04/2013 08:17 AM, Xose Vazquez Perez wrote:
> > Christian Lamparter wrote:
> >
> >> The issue with r92su is: The driver is ready, but it's fundamentally
> > ^^^^^^^^
> >> incompatible with the rtlwifi framework and mac80211.
> >
> > _anything_ is better than a staging driver.
> >
> > There are distributions that don't build staging drivers.
>
> As a result, they force their users to build out-of-tree drivers!
>
> > So, please, replace and delete staging/rtl8712 with r92su.
>
> You seem not to understand. Any new wireless driver that does not use mac80211
> *must* go into staging. AFAIK, that rule is absolute, and r92su cannot go into
> the regular wireless tree.
Be careful ;-). mac80211 is great for softmac/umac/halfmac/semimac devices.
However the rtl8712 firmware interface is fullmac-like, so it needs cfg80211.
... and r92su uses cfg80211 exclusively. There's no more WEXT or realtek ioctl
legacy API support.
Regards,
Christian
On Mon, 2013-11-04 at 17:19 +0100, Christian Lamparter wrote:
> Yes, there is even more duplicated code if you factor in the (de-)fragmentation
> and deaggregation code from mac80211. Furthermore r92su could be sharing a bit
> of infrastructure with rtlwifi.
Yeah, that some of that seems to be duplicated in mwifiex too. I think
there'd be some benefit in trying to share more. Maybe moving some to
cfg80211? Or some new module?
> I didn't include lib80211 since it reports mic failures via WEXT:
> wireless_send_event(dev, IWEVMICHAELMICFAILURE, &wrqu, (char *)&ev);
> (And I don't think this can be changed easily without touching ipw2200
> and hostap-driver).
Yeah that'd be tricky, though I guess you could pass a wdev pointer and
allow it to be NULL or so.
johannes
On 11/04/2013 04:34 PM, Larry Finger wrote:
>> _anything_ is better than a staging driver.
>>
>> There are distributions that don't build staging drivers.
>
> As a result, they force their users to build out-of-tree drivers!
reformulating: As a result, they force their users to _have a stable kernel_
Staging is only a working area. And the rtl8712-TODO is unreal:
- merge Realtek's bugfixes and new features into the driver
- switch to use LIB80211
- switch to use MAC80211
- checkpatch.pl fixes - only a few remain
>> So, please, replace and delete staging/rtl8712 with r92su.
>
> You seem not to understand. Any new wireless driver that does not
> use mac80211 *must* go into staging. AFAIK, that rule is absolute,
^^^^^^^^^^^^^^^^^^^^^
> and r92su cannot go into the regular wireless tree.
Who said that ? Then replace staging/rtl8712 with staging/r92su
Anyway, r92su is a _linux_ driver with a cfg80211 interface.
staging/rtl8712 is crap?
1152k drivers/staging/rtl8712/
340k r92su/
> If you do not like having drivers for RTL8192SU in staging, then
> write one that uses mac80211. You have access to as much information
> about the chips as anyone outside Realtek. They have no interest
> in doing the mac80211 version, and I have many other tasks with
> higher priority.
Nobody said that you have to write a rtl8712-mac80211.
But right now there is a better alternative, and it should be used.
As I understand it, there is a will to replace r8712u with r92su. I
was thinking about it this weekend and if it is saner and nearer to
where the community wants to bring the driver, why not go with it?
Could we think about merging it in the staging drivers for kernel 3.13
and tag r8712u as deprecated? One could always build it or try r92su
instead, which would give a better overlook about its shape and
performance, don't you think?
Alexandre
On Mon, Nov 4, 2013 at 1:32 PM, Johannes Berg <[email protected]> wrote:
> On Mon, 2013-11-04 at 17:19 +0100, Christian Lamparter wrote:
>
>> Yes, there is even more duplicated code if you factor in the (de-)fragmentation
>> and deaggregation code from mac80211. Furthermore r92su could be sharing a bit
>> of infrastructure with rtlwifi.
>
> Yeah, that some of that seems to be duplicated in mwifiex too. I think
> there'd be some benefit in trying to share more. Maybe moving some to
> cfg80211? Or some new module?
>
>> I didn't include lib80211 since it reports mic failures via WEXT:
>> wireless_send_event(dev, IWEVMICHAELMICFAILURE, &wrqu, (char *)&ev);
>> (And I don't think this can be changed easily without touching ipw2200
>> and hostap-driver).
>
> Yeah that'd be tricky, though I guess you could pass a wdev pointer and
> allow it to be NULL or so.
>
> johannes
>
On Mon, 2013-11-04 at 15:17 +0100, Xose Vazquez Perez wrote:
> Christian Lamparter wrote:
>
> > The issue with r92su is: The driver is ready, but it's fundamentally
> ^^^^^^^^
> > incompatible with the rtlwifi framework and mac80211.
>
> _anything_ is better than a staging driver.
That's still only staging quality though.
johannes
On 11/04/2013 10:06 AM, Xose Vazquez Perez wrote:
> On 11/04/2013 04:34 PM, Larry Finger wrote:
>
>>> _anything_ is better than a staging driver.
>>>
>>> There are distributions that don't build staging drivers.
>>
>> As a result, they force their users to build out-of-tree drivers!
>
> reformulating: As a result, they force their users to _have a stable kernel_
Not all staging drivers are unstable, just as not all regular drivers are stable.
> Staging is only a working area. And the rtl8712-TODO is unreal:
> - merge Realtek's bugfixes and new features into the driver
> - switch to use LIB80211
> - switch to use MAC80211
> - checkpatch.pl fixes - only a few remain
>
>>> So, please, replace and delete staging/rtl8712 with r92su.
>>
>> You seem not to understand. Any new wireless driver that does not
>> use mac80211 *must* go into staging. AFAIK, that rule is absolute,
> ^^^^^^^^^^^^^^^^^^^^^
>> and r92su cannot go into the regular wireless tree.
>
> Who said that ? Then replace staging/rtl8712 with staging/r92su
>
> Anyway, r92su is a _linux_ driver with a cfg80211 interface.
> staging/rtl8712 is crap?
>
> 1152k drivers/staging/rtl8712/
> 340k r92su/
>
>> If you do not like having drivers for RTL8192SU in staging, then
>> write one that uses mac80211. You have access to as much information
>> about the chips as anyone outside Realtek. They have no interest
>> in doing the mac80211 version, and I have many other tasks with
>> higher priority.
>
> Nobody said that you have to write a rtl8712-mac80211.
> But right now there is a better alternative, and it should be used.
Go ahead and submit it and remove r8712u. I would be more than happy to be able
to just bitch about bugs the way other people do! :)
Larry
On Mon, 2013-11-04 at 16:47 +0100, Christian Lamparter wrote:
> Be careful ;-). mac80211 is great for softmac/umac/halfmac/semimac devices.
> However the rtl8712 firmware interface is fullmac-like, so it needs cfg80211.
> ... and r92su uses cfg80211 exclusively. There's no more WEXT or realtek ioctl
> legacy API support.
still seems there's duplicate stuff in there like the michael MIC, at
least lib80211 has that? OTOH, that's not all that much code.
johannes