Skip to contentSkip to conversation

cfg80211: Add nl80211 antenna configuration

PATCH Superseded Bruno Randolf 4 participants 21 messages Jul 29, 2010 – Aug 2, 2010 802.11 (including cfg80211/NL80211)

v5 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

PATCH v5 1/1 cfg80211: Add nl80211 antenna configuration

Allow setting of TX and RX antenna configuration via nl80211.

The antenna configuration is defined as a bitmap of allowed antennas to use. This API can be used to mask out antennas which are not attached or should not be used for other reasons like regulatory concerns or special setups.

Separate bitmaps are used for RX and TX to allow configuring different antennas for receiving and transmitting. The bitmap is 8 bit long, each bit representing one antenna, starting with antenna 1 at the first bit. If an antenna bit is set, this means the driver is allowed to use this antenna for RX or TX respectively; if the bit is not set the hardware is not allowed to use this antenna.

Using bitmaps has the benefit of allowing for a flexible configuration interface which can support many different configurations and which can be used for 802.11n as well as non-802.11n devices. Instead of relying on some hardware specific assumptions, drivers can use this information to know which antennas are actually attached to the system and derive their capabilities based on that.

802.11n devices should enable or disable chains, based on which antennas are present (If all antennas belonging to a particular chain are disabled, the entire chain should be disabled). HT capabilities (like STBC, TX Beamforming, Antenna selection) should be calculated based on the available chains after applying the antenna masks. Should a 802.11n device have diversity antennas attached to one of their chains, diversity can be enabled or disabled based on the antenna information.

Non-802.11n drivers can use the antenna masks to enable or disable antenna diversity.

While covering chainmasks for 802.11n and the standard "legacy" modes "fixed antenna 1", "fixed antenna 2" and "diversity" this API also allows more rare, but useful configurations as follows:

1) Send on antenna 1, receive on antenna 2 (or vice versa). This can be used to have a low gain antenna for TX in order to keep within the regulatory constraints and a high gain antenna for RX in order to receive weaker signals ("speak softly, but listen harder"). This can be useful for building long-shot outdoor links. Another usage of this setup is having a low-noise pre-amplifier on antenna 1 and a power amplifier on the other antenna. This way transmit noise is mostly kept out of the low noise receive channel.
(This would be bitmaps: tx 1 rx 2).

2) Another similar setup is: Use RX diversity on both antennas, but always send on antenna 1. Again that would allow us to benefit from a higher gain RX antenna, while staying within the legal limits.
(This would be: tx 0 rx 3).

3) And finally there can be special experimental setups in research and development even with pre 802.11n hardware where more than 2 antennas are available. It's good to keep the API simple, yet flexible.

---

v5: Removed the special casing of "0" for diversity. I hope i made the descriptions more clear and incorportated all comments concerning 802.11n. If we can agree on this i will re-send the whole series.

---
 include/linux/nl80211.h |   25 +++++++++++++++++++++++++
 include/net/cfg80211.h  |    3 +++
 net/wireless/nl80211.c  |   30 +++++++++++++++++++++++++++++-
 3 files changed, 57 insertions(+), 1 deletions(-)
include/linux/nl80211.h +25 -0
index 2c87016..8a88921 100644
@@ -731,6 +731,28 @@ enum nl80211_commands {
  *      This is used in association with @NL80211_ATTR_WIPHY_TX_POWER_SETTING
  *      for non-automatic settings.
  *
+ * @NL80211_ATTR_WIPHY_ANTENNA_TX: Bitmap of allowed antennas for transmitting.
+ *	This can be used to mask out antennas which are not attached or should
+ *	not be used for transmitting. If an antenna is not selected in this
+ *	bitmap the hardware is not allowed to transmit on this antenna.
+ *
+ *	Each bit represents one antenna, starting with antenna 1 at the first
+ *	bit. Depending on which antennas are selected in the bitmap, 802.11n
+ *	drivers can derive which chainmasks to use (if all antennas belonging to
+ *	a particular chain are disabled this chain should be disabled) and if
+ *	a chain has diversity antennas wether diversity should be used or not.
+ *	HT capabilities (STBC, TX Beamforming, Antenna selection) can be
+ *	derived from the available chains after applying the antenna mask.
+ *	Non-802.11n drivers can derive wether to use diversity or not.
+ *	Drivers may reject configurations or RX/TX mask combinations they cannot
+ *	support by returning -EINVAL.
+ *
+ * @NL80211_ATTR_WIPHY_ANTENNA_RX: Bitmap of allowed antennas for receiving.
+ *	This can be used to mask out antennas which are not attached or should
+ *	not be used for receiving. If an antenna is not selected in this bitmap
+ *	the hardware should not be configured to receive on this antenna.
+ *	For a more detailed descripton see @NL80211_ATTR_WIPHY_ANTENNA_TX.
+ *
  * @NL80211_ATTR_MAX: highest attribute number currently defined
  * @__NL80211_ATTR_AFTER_LAST: internal use
  */
@@ -891,6 +913,9 @@ enum nl80211_attrs {
 	NL80211_ATTR_WIPHY_TX_POWER_SETTING,
 	NL80211_ATTR_WIPHY_TX_POWER_LEVEL,
 
+	NL80211_ATTR_WIPHY_ANTENNA_TX,
+	NL80211_ATTR_WIPHY_ANTENNA_RX,
+
 	/* add attributes here, update the policy in nl80211.c */
 
 	__NL80211_ATTR_AFTER_LAST,
include/net/cfg80211.h +3 -0
index f68ae54..7dc0e8d 100644
@@ -1184,6 +1184,9 @@ struct cfg80211_ops {
 	int	(*set_cqm_rssi_config)(struct wiphy *wiphy,
 				       struct net_device *dev,
 				       s32 rssi_thold, u32 rssi_hyst);
+
+	int	(*set_antenna)(struct wiphy *wiphy, u8 tx_ant, u8 rx_ant);
+	int	(*get_antenna)(struct wiphy *wiphy, u8 *tx_ant, u8 *rx_ant);
 };
 
 /*
net/wireless/nl80211.c +29 -1
index cea595e..62fc5fe 100644
@@ -156,6 +156,9 @@ static const struct nla_policy nl80211_policy[NL80211_ATTR_MAX+1] = {
 
 	[NL80211_ATTR_WIPHY_TX_POWER_SETTING] = { .type = NLA_U32 },
 	[NL80211_ATTR_WIPHY_TX_POWER_LEVEL] = { .type = NLA_U32 },
+
+	[NL80211_ATTR_WIPHY_ANTENNA_TX] = { .type = NLA_U8 },
+	[NL80211_ATTR_WIPHY_ANTENNA_RX] = { .type = NLA_U8 },
 };
 
 /* policy for the attributes */
@@ -458,7 +461,6 @@ static int nl80211_send_wiphy(struct sk_buff *msg, u32 pid, u32 seq, int flags,
 		    dev->wiphy.rts_threshold);
 	NLA_PUT_U8(msg, NL80211_ATTR_WIPHY_COVERAGE_CLASS,
 		    dev->wiphy.coverage_class);
-
 	NLA_PUT_U8(msg, NL80211_ATTR_MAX_NUM_SCAN_SSIDS,
 		   dev->wiphy.max_scan_ssids);
 	NLA_PUT_U16(msg, NL80211_ATTR_MAX_SCAN_IE_LEN,
@@ -471,6 +473,16 @@ static int nl80211_send_wiphy(struct sk_buff *msg, u32 pid, u32 seq, int flags,
 	NLA_PUT_U8(msg, NL80211_ATTR_MAX_NUM_PMKIDS,
 		   dev->wiphy.max_num_pmkids);
 
+	if (dev->ops->get_antenna) {
+		u8 tx_ant = 0, rx_ant = 0;
+		int res;
+		res = dev->ops->get_antenna(&dev->wiphy, &tx_ant, &rx_ant);
+		if (!res) {
+			NLA_PUT_U8(msg, NL80211_ATTR_WIPHY_ANTENNA_TX, tx_ant);
+			NLA_PUT_U8(msg, NL80211_ATTR_WIPHY_ANTENNA_RX, rx_ant);
+		}
+	}
+
 	nl_modes = nla_nest_start(msg, NL80211_ATTR_SUPPORTED_IFTYPES);
 	if (!nl_modes)
 		goto nla_put_failure;
@@ -900,6 +912,22 @@ static int nl80211_set_wiphy(struct sk_buff *skb, struct genl_info *info)
 			goto bad_res;
 	}
 
+	if (info->attrs[NL80211_ATTR_WIPHY_ANTENNA_TX] &&
+	    info->attrs[NL80211_ATTR_WIPHY_ANTENNA_RX]) {
+		u8 tx_ant, rx_ant;
+		if (!rdev->ops->set_antenna) {
+			result = -EOPNOTSUPP;
+			goto bad_res;
+		}
+
+		tx_ant = nla_get_u8(info->attrs[NL80211_ATTR_WIPHY_ANTENNA_TX]);
+		rx_ant = nla_get_u8(info->attrs[NL80211_ATTR_WIPHY_ANTENNA_RX]);
+
+		result = rdev->ops->set_antenna(&rdev->wiphy, tx_ant, rx_ant);
+		if (result)
+			goto bad_res;
+	}
+
 	changed = 0;
 
 	if (info->attrs[NL80211_ATTR_WIPHY_RETRY_SHORT]) {
#1 lore raw
Quoted 10 lines · Bruno Randolf
include/linux/nl80211.h +2 -0
index 2c87016..8a88921 100644
@@ -731,6 +731,28 @@ enum nl80211_commands {
  *      This is used in association with @NL80211_ATTR_WIPHY_TX_POWER_SETTING
  *      for non-automatic settings.
  *
+ * @NL80211_ATTR_WIPHY_ANTENNA_TX: Bitmap of allowed antennas for transmitting.
+ *     This can be used to mask out antennas which are not attached or should

include/linux/nl80211.h enum nl80211_commands { · line 745

Quoted 10 lines
include/linux/nl80211.h +10 -0
+ *     not be used for transmitting. If an antenna is not selected in this
+ *     bitmap the hardware is not allowed to transmit on this antenna.
+ *
+ *     Each bit represents one antenna, starting with antenna 1 at the first
+ *     bit. Depending on which antennas are selected in the bitmap, 802.11n
+ *     drivers can derive which chainmasks to use (if all antennas belonging to
+ *     a particular chain are disabled this chain should be disabled) and if
+ *     a chain has diversity antennas wether diversity should be used or not.
+ *     HT capabilities (STBC, TX Beamforming, Antenna selection) can be
+ *     derived from the available chains after applying the antenna mask.

I don't want to do any work myself on drivers for this, can we have cfg80211/mac80211 do this for us?

Luis

maintainer
#2 lore raw
Quoted 16 lines · Luis R. Rodriguez
Quoted 15 lines · Bruno Randolf
include/linux/nl80211.h +7 -0
index 2c87016..8a88921 100644
@@ -731,6 +731,28 @@ enum nl80211_commands {
 *      This is used in association with @NL80211_ATTR_WIPHY_TX_POWER_SETTING
 *      for non-automatic settings.
 *
+ * @NL80211_ATTR_WIPHY_ANTENNA_TX: Bitmap of allowed antennas for transmitting.
+ *     This can be used to mask out antennas which are not attached or should
+ *     not be used for transmitting. If an antenna is not selected in this
+ *     bitmap the hardware is not allowed to transmit on this antenna.
+ *
+ *     Each bit represents one antenna, starting with antenna 1 at the first
+ *     bit. Depending on which antennas are selected in the bitmap, 802.11n
Quoted 8 lines

include/linux/nl80211.h enum nl80211_commands { · line 745

Quoted 5 lines
include/linux/nl80211.h +5 -0
+ *     drivers can derive which chainmasks to use (if all antennas belonging to
+ *     a particular chain are disabled this chain should be disabled) and if
+ *     a chain has diversity antennas wether diversity should be used or not.
+ *     HT capabilities (STBC, TX Beamforming, Antenna selection) can be
+ *     derived from the available chains after applying the antenna mask.

I don't want to do any work myself on drivers for this, can we have cfg80211/mac80211 do this for us?

Also, this is a serious issue when you want to support this call -- it may not be done while you're associated since you cannot change those things properly then.

johannes

#3 lore raw
In reply to Luis R. Rodriguez
Quoted 17 lines · On Thu July 29 2010 15:12:29 you wrote:
Quoted 16 lines · Bruno Randolf
include/linux/nl80211.h +0 -0
index 2c87016..8a88921 100644
@@ -731,6 +731,28 @@ enum nl80211_commands {
 *      This is used in association with

@NL80211_ATTR_WIPHY_TX_POWER_SETTING * for non-automatic settings.
*
+ * @NL80211_ATTR_WIPHY_ANTENNA_TX: Bitmap of allowed antennas for
transmitting. + * This can be used to mask out antennas which are
not attached or should + * not be used for transmitting. If an
antenna is not selected in this + * bitmap the hardware is not
allowed to transmit on this antenna. + *
+ * Each bit represents one antenna, starting with antenna 1 at the
first + * bit. Depending on which antennas are selected in the
bitmap, 802.11n + * drivers can derive which chainmasks to use (if

Quoted 8 lines
Quoted 5 lines

all antennas belonging to + * a particular chain are disabled this
chain should be disabled) and if + * a chain has diversity antennas
wether diversity should be used or not. + * HT capabilities (STBC,
TX Beamforming, Antenna selection) can be + * derived from the
available chains after applying the antenna mask.

I don't want to do any work myself on drivers for this, can we have cfg80211/mac80211 do this for us?

is this not a separate issue from defining the API? you could have it do this for you even now, with or without the antenna API, no?

i think this should be dealt with seperately. for now let's just define an API for an antenna mask.

bruno

#4 lore raw
Quoted 27 lines · Bruno Randolf
Quoted 17 lines · On Thu July 29 2010 15:12:29 you wrote:
Quoted 16 lines · Bruno Randolf
include/linux/nl80211.h +0 -0
index 2c87016..8a88921 100644
@@ -731,6 +731,28 @@ enum nl80211_commands {
  *      This is used in association with

@NL80211_ATTR_WIPHY_TX_POWER_SETTING *      for non-automatic settings. *
+ * @NL80211_ATTR_WIPHY_ANTENNA_TX: Bitmap of allowed antennas for transmitting. + *     This can be used to mask out antennas which are not attached or should + *     not be used for transmitting. If an antenna is not selected in this + *     bitmap the hardware is not allowed to transmit on this antenna. + *
+ *     Each bit represents one antenna, starting with antenna 1 at the first + *     bit. Depending on which antennas are selected in the bitmap, 802.11n + *     drivers can derive which chainmasks to use (if

Quoted 8 lines
Quoted 5 lines

all antennas belonging to + *     a particular chain are disabled this chain should be disabled) and if + *     a chain has diversity antennas wether diversity should be used or not. + *     HT capabilities (STBC, TX Beamforming, Antenna selection) can be + *     derived from the available chains after applying the antenna mask.

I don't want to do any work myself on drivers for this, can we have cfg80211/mac80211 do this for us?

Quoted 2 lines

is this not a separate issue from defining the API? you could have it do this for you even now, with or without the antenna API, no?

It depends on the consumer and at what level we think the consumer should be doing whatever checks or changes it should. For starters I would expect at least a check for association within cfg80211 which disallows hw config changes. Then, we would need to address changes on the hw config, and caching the original values and determine whether or not we want to differentiate them on userspace to make it clear to the user that the hw config is just tweaked right now but the real capability is different than the current setting.

Quoted 2 lines

i think this should be dealt with seperately. for now let's just define an API for an antenna mask.

This needs to be dealt with in cfg80211 if we want to allow cfg80211 drivers to use this and can deal with sanity checks in cfg80211. If mac80211 and other cfg80211 drivers will be the consumers then the change would be in mac80211 but a separate patch would be required.

Luis

#5 lore raw
Quoted 10 lines · On Fri July 30 2010 00:04:49 Luis R. Rodriguez wrote:
Quoted 5 lines
Quoted 2 lines

I don't want to do any work myself on drivers for this, can we have cfg80211/mac80211 do this for us?

is this not a separate issue from defining the API? you could have it do this for you even now, with or without the antenna API, no?

It depends on the consumer and at what level we think the consumer should be doing whatever checks or changes it should. For starters I would expect at least a check for association within cfg80211 which disallows hw config changes.

for ath5k at least, it's o.k. to change antenna configuration during association.

if 802.11n devices cannot change antenna setup during association, the driver could simply reject the configuration. but i can see that it can make sense to handle some of this in cfg80211 or mac80211. it should be possible to add that on top of my patches.

i think you are expecting to much. for now i just want to define an API which
1) can be used right now and 2) is flexible enough to support 802.11n devices. as i don't work with 802.11n devices right now, please don't expect me to solve all 802.11n related antenna problems. as long as the API is flexible enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

bruno

#6 lore raw
Quoted 7 lines · Bruno Randolf
Quoted 6 lines · On Fri July 30 2010 00:04:49 Luis R. Rodriguez wrote:
Quoted 5 lines
Quoted 2 lines

I don't want to do any work myself on drivers for this, can we have cfg80211/mac80211 do this for us?

is this not a separate issue from defining the API? you could have it do this for you even now, with or without the antenna API, no?

Quoted 7 lines
Quoted 4 lines

It depends on the consumer and at what level we think the consumer should be doing whatever checks or changes it should. For starters I would expect at least a check for association within cfg80211 which disallows hw config changes.

for ath5k at least, it's o.k. to change antenna configuration during association.

Then do it through debugfs.

Quoted 4 lines

if 802.11n devices cannot change antenna setup during association, the driver could simply reject the configuration. but i can see that it can make sense to handle some of this in cfg80211 or mac80211. it should be possible to add that on top of my patches.

I would expect the respective changes within your patches, not after.

Quoted 1 line

i think you are expecting to much.

I refuse to accept patches which can potentially create bogus bug reports or loose crap implementations in drivers. Wext had a lot of this and I am glad we haven't yet allowed for this crap with nl80211. To allow your patch to go through as-is without addressing all the points made would be enabling driver developers to get real sloppy.

Quoted 2 lines

for now i just want to define an API which
1) can be used right now and 2) is flexible enough to support 802.11n devices.

Use debugfs or consider a simple API for legacy for nl80211.

Quoted 2 lines

as i don't work with 802.11n devices right now, please don't expect me to solve all 802.11n related antenna problems.

I'm not, the point I was trying to make is that solving this for legacy and that for 802.11n this needs more thought and work.

Quoted 3 lines

as long as the API is flexible
enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

If you want really flexible stuff without addressing serious considerations before introducing a new kernel API then use debugfs.

Luis

#7 lore raw
Quoted 2 lines · Luis R. Rodriguez

I'm not, the point I was trying to make is that solving this for legacy and that for 802.11n this needs more thought and work.

I need this for 802.11n as well, and I believe the API is good enough for that.

Quoted 6 lines
Quoted 3 lines

as long as the API is flexible
enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

If you want really flexible stuff without addressing serious considerations before introducing a new kernel API then use debugfs.

I'd like to see this in nl80211. Is there any specific concern left that I haven't addressed already? If so, please point me at it...

- Felix

#8 lore raw
Quoted 5 lines · Felix Fietkau
Quoted 2 lines · Luis R. Rodriguez

I'm not, the point I was trying to make is that solving this for legacy and that for 802.11n this needs more thought and work.

I need this for 802.11n as well, and I believe the API is good enough for that.

Sure.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

as long as the API is flexible
enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

If you want really flexible stuff without addressing serious considerations before introducing a new kernel API then use debugfs.

I'd like to see this in nl80211. Is there any specific concern left that I haven't addressed already? If so, please point me at it...

The code that deals with not accepting changes unless we're not associated, and that also caches the old real hw config vs the modified one.

Luis

#9 lore raw
Quoted 13 lines · Luis R. Rodriguez
Quoted 5 lines · Felix Fietkau
Quoted 2 lines · Luis R. Rodriguez

I'm not, the point I was trying to make is that solving this for legacy and that for 802.11n this needs more thought and work.

I need this for 802.11n as well, and I believe the API is good enough for that.

Sure.

Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

as long as the API is flexible
enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

If you want really flexible stuff without addressing serious considerations before introducing a new kernel API then use debugfs.

I'd like to see this in nl80211. Is there any specific concern left that I haven't addressed already? If so, please point me at it...

The code that deals with not accepting changes unless we're not associated, and that also caches the old real hw config vs the modified one.

This is not really an API issue, this is more of an implementation aspect. I think it's up to the driver to reject changes that it cannot handle at the moment.
For legacy, changing settings is fine in any state (even with assoc). For 802.11n, changes that affect the chainmask should be rejected while the interface is up. That way we don't run into weird cached config vs hw config issues, and also keep a sane state for the HT capabilities.

- Felix

#10 lore raw
Quoted 19 lines · Felix Fietkau
Quoted 18 lines · Luis R. Rodriguez
Quoted 5 lines · Felix Fietkau
Quoted 2 lines · Luis R. Rodriguez

I'm not, the point I was trying to make is that solving this for legacy and that for 802.11n this needs more thought and work.

I need this for 802.11n as well, and I believe the API is good enough for that.

Sure.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

as long as the API is flexible
enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

If you want really flexible stuff without addressing serious considerations before introducing a new kernel API then use debugfs.

I'd like to see this in nl80211. Is there any specific concern left that I haven't addressed already? If so, please point me at it...

Quoted 6 lines
Quoted 3 lines

The code that deals with not accepting changes unless we're not associated, and that also caches the old real hw config vs the modified one.

This is not really an API issue, this is more of an implementation aspect. I think it's up to the driver to reject changes that it cannot handle at the moment.

Huh,why not deal with this on cfg80211 and/or mac80211?

Quoted 1 line

For legacy, changing settings is fine in any state (even with assoc).

Sure.

Quoted 3 lines

For 802.11n, changes that affect the chainmask should be rejected while the interface is up. That way we don't run into weird cached config vs hw config issues, and also keep a sane state for the HT capabilities.

Sure, I just did not see any code for this in these patches. My point about the hw config vs fake/mod'd is if we'd expose the mod'd config changes to userspace or if we'd keep them internal to cfg80211. How would this be dealt with?

Luis

#11 lore raw
Quoted 20 lines · Luis R. Rodriguez
Quoted 19 lines · Felix Fietkau
Quoted 18 lines · Luis R. Rodriguez
Quoted 5 lines · Felix Fietkau
Quoted 2 lines · Luis R. Rodriguez

I'm not, the point I was trying to make is that solving this for legacy and that for 802.11n this needs more thought and work.

I need this for 802.11n as well, and I believe the API is good enough for that.

Sure.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

as long as the API is flexible
enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

If you want really flexible stuff without addressing serious considerations before introducing a new kernel API then use debugfs.

I'd like to see this in nl80211. Is there any specific concern left that I haven't addressed already? If so, please point me at it...

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

The code that deals with not accepting changes unless we're not associated, and that also caches the old real hw config vs the modified one.

This is not really an API issue, this is more of an implementation aspect. I think it's up to the driver to reject changes that it cannot handle at the moment.

Huh,why not deal with this on cfg80211 and/or mac80211?

Right now, the drivers calculate this. We could make a helper function in cfg80211 at some point, but we're talking about really small chunks of code.

Quoted 8 lines
Quoted 3 lines

For 802.11n, changes that affect the chainmask should be rejected while the interface is up. That way we don't run into weird cached config vs hw config issues, and also keep a sane state for the HT capabilities.

Sure, I just did not see any code for this in these patches. My point about the hw config vs fake/mod'd is if we'd expose the mod'd config changes to userspace or if we'd keep them internal to cfg80211. How would this be dealt with?

Right now, cfg80211 doesn't know enough to handle this stuff on its own, so let's handle it in the driver completely on the first iteration. The patches do not need any changes for this right now.

- Felix

#12 lore raw
Quoted 37 lines · Felix Fietkau
Quoted 20 lines · Luis R. Rodriguez
Quoted 19 lines · Felix Fietkau
Quoted 18 lines · Luis R. Rodriguez
Quoted 5 lines · Felix Fietkau
Quoted 2 lines · Luis R. Rodriguez

I'm not, the point I was trying to make is that solving this for legacy and that for 802.11n this needs more thought and work.

I need this for 802.11n as well, and I believe the API is good enough for that.

Sure.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

as long as the API is flexible
enough i think it can be added later as 802.11n devices are going to accept antenna configuration.

If you want really flexible stuff without addressing serious considerations before introducing a new kernel API then use debugfs.

I'd like to see this in nl80211. Is there any specific concern left that I haven't addressed already? If so, please point me at it...

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

The code that deals with not accepting changes unless we're not associated, and that also caches the old real hw config vs the modified one.

This is not really an API issue, this is more of an implementation aspect. I think it's up to the driver to reject changes that it cannot handle at the moment.

Huh,why not deal with this on cfg80211 and/or mac80211?

Right now, the drivers calculate this. We could make a helper function in cfg80211 at some point, but we're talking about really small chunks of code.

Quoted 4 lines
Quoted 3 lines

For 802.11n, changes that affect the chainmask should be rejected while the interface is up. That way we don't run into weird cached config vs hw config issues, and also keep a sane state for the HT capabilities.

Quoted 7 lines
Quoted 4 lines

Sure, I just did not see any code for this in these patches. My point about the hw config vs fake/mod'd is if we'd expose the mod'd config changes to userspace or if we'd keep them internal to cfg80211. How would this be dealt with?

Right now, cfg80211 doesn't know enough to handle this stuff on its own, so let's handle it in the driver completely on the first iteration. The patches do not need any changes for this right now.

I'd prefer that code to be written rather then let this be defined as API now and let drivers deal with this differently. But that's me, I'm not the maintainer, I just will not deal with bug reports dealing with this and I'll assign them to you guys if this gets through. Still think its crap and should just go through debugfs until all the code mentioned does exist.

Luis

#13 lore raw
Quoted 10 lines · Luis R. Rodriguez
Quoted 8 lines · Felix Fietkau
Quoted 4 lines · Luis R. Rodriguez

Sure, I just did not see any code for this in these patches. My point about the hw config vs fake/mod'd is if we'd expose the mod'd config changes to userspace or if we'd keep them internal to cfg80211. How would this be dealt with?

Right now, cfg80211 doesn't know enough to handle this stuff on its own, so let's handle it in the driver completely on the first iteration. The patches do not need any changes for this right now.

Quoted 6 lines

I'd prefer that code to be written rather then let this be defined as API now and let drivers deal with this differently. But that's me, I'm not the maintainer, I just will not deal with bug reports dealing with this and I'll assign them to you guys if this gets through. Still think its crap and should just go through debugfs until all the code mentioned does exist.

Sorry, but WTF? There's two parts to this: API visible to user space, and the internal API for handling changes.
So you're suggesting to reject the user space API, because of missing parts in the internal API (which we can change any time) that will only be used for drivers that this series doesn't even contain any code for?? Am I confused here, or does this seem rather strange?

- Felix

#14 lore raw
Quoted 17 lines · Felix Fietkau
Quoted 10 lines · Luis R. Rodriguez
Quoted 8 lines · Felix Fietkau
Quoted 4 lines · Luis R. Rodriguez

Sure, I just did not see any code for this in these patches. My point about the hw config vs fake/mod'd is if we'd expose the mod'd config changes to userspace or if we'd keep them internal to cfg80211. How would this be dealt with?

Right now, cfg80211 doesn't know enough to handle this stuff on its own, so let's handle it in the driver completely on the first iteration. The patches do not need any changes for this right now.

Quoted 6 lines

I'd prefer that code to be written rather then let this be defined as API now and let drivers deal with this differently. But that's me, I'm not the maintainer, I just will not deal with bug reports dealing with this and I'll assign them to you guys if this gets through. Still think its crap and should just go through debugfs until all the code mentioned does exist.

Quoted 6 lines

Sorry, but WTF? There's two parts to this: API visible to user space, and the internal API for handling changes.
So you're suggesting to reject the user space API, because of missing parts in the internal API (which we can change any time) that will only be used for drivers that this series doesn't even contain any code for?? Am I confused here, or does this seem rather strange?

If the current patches are accepted it means anyone *can* submit patches for an 802.11n driver and expect it to be accepted. Hence why I was asking for this to be defined as a legacy API only, if that is the only purpose right now.

Luis

#15 lore raw
Quoted 25 lines · Luis R. Rodriguez
Quoted 17 lines · Felix Fietkau
Quoted 10 lines · Luis R. Rodriguez
Quoted 8 lines · Felix Fietkau
Quoted 4 lines · Luis R. Rodriguez

Sure, I just did not see any code for this in these patches. My point about the hw config vs fake/mod'd is if we'd expose the mod'd config changes to userspace or if we'd keep them internal to cfg80211. How would this be dealt with?

Right now, cfg80211 doesn't know enough to handle this stuff on its own, so let's handle it in the driver completely on the first iteration. The patches do not need any changes for this right now.

Quoted 6 lines

I'd prefer that code to be written rather then let this be defined as API now and let drivers deal with this differently. But that's me, I'm not the maintainer, I just will not deal with bug reports dealing with this and I'll assign them to you guys if this gets through. Still think its crap and should just go through debugfs until all the code mentioned does exist.

Quoted 6 lines

Sorry, but WTF? There's two parts to this: API visible to user space, and the internal API for handling changes.
So you're suggesting to reject the user space API, because of missing parts in the internal API (which we can change any time) that will only be used for drivers that this series doesn't even contain any code for?? Am I confused here, or does this seem rather strange?

Quoted 4 lines

If the current patches are accepted it means anyone *can* submit patches for an 802.11n driver and expect it to be accepted. Hence why I was asking for this to be defined as a legacy API only, if that is the only purpose right now.

This discussion is getting quite redundant. I would like to ask you again, what the specific concern purely from a user space API point of view is.

You asked how run time changes should be handled, I made a proposal for that (simply not accepting them at run time on 802.11n hardware if they affect the chainmask), and then you started complaining about what should be in cfg80211 and what should be in the drivers. What does that have to do with the public API?

The limitations wrt. chainmask changes at run time are known, and they aren't really hardware specific, so you won't see some random driver suddenly implementing some hypothetical crazy scheme that nobody expects. This has nothing to do with code being in cfg80211 vs code being in the driver.

It's really simple: in AP mode, hostapd needs to have the HT stuff set when it starts up, in STA mode, it needs to be set at least at assoc time. Since this is affected by the chainmask, we can't just change stuff at a random point in time, but this is expected.

So the lowest common denominator that we can use for all 802.11n hardware in all modes is to just refuse changes unless the interface is down. Then we use the antenna mask to calculate the chainmask (an inherently driver specific aspect, at least at the moment), recalculate the HT capabilities based on the chainmask (driver has to do this at init time already, needs very little refactoring), and then just apply the new settings to the hardware (again, driver specific, will probably be deferred to the interface bringup sequence).

So aside from the meaning of the actual value, which we need to make easier to understand in the user space code, the only thing that users need to be aware of is this:

Legacy hardware => run time changes possible (actually optional) 802.11n hardware => need to bring down the interface before changing
the antenna settings

What else is left that we need to figure out? What would we need the temporary debugfs crap for?

- Felix

#16 lore raw
Quoted 2 lines · Felix Fietkau

802.11n hardware => need to bring down the interface before changing the antenna settings

The current patch does not deal with the 802.11n case you mentioned.

Luis

#17 lore raw
Quoted 5 lines · Luis R. Rodriguez
Quoted 2 lines · Felix Fietkau

802.11n hardware => need to bring down the interface before changing
the antenna settings

The current patch does not deal with the 802.11n case you mentioned.

And why should it? The series does not add antenna control support to any specific 802.11n driver.

- Felix

#18 lore raw
Quoted 8 lines · Felix Fietkau
Quoted 5 lines · Luis R. Rodriguez
Quoted 2 lines · Felix Fietkau

802.11n hardware => need to bring down the interface before changing the antenna settings

The current patch does not deal with the 802.11n case you mentioned.

And why should it? The series does not add antenna control support to any specific 802.11n driver.

Because it documents support for 802.11n. You should not expect someone writing their ops for an 802.11n driver to fill in the 802.11n gaps for you unless you mark your API as being only for legacy.

Luis

maintainer
#19 lore raw
In reply to Felix Fietkau
Quoted 3 lines · Felix Fietkau

Legacy hardware => run time changes possible (actually optional) 802.11n hardware => need to bring down the interface before changing
the antenna settings

I think I'd prefer if all hw just required the device to be stopped. Otherwise things either become awkward when you have, say. software diversity control, or they become unpredictable quite, no?

johannes

#20 lore raw
Quoted 9 lines · Johannes Berg
Quoted 3 lines · Felix Fietkau

Legacy hardware => run time changes possible (actually optional) 802.11n hardware => need to bring down the interface before changing
the antenna settings

I think I'd prefer if all hw just required the device to be stopped. Otherwise things either become awkward when you have, say. software diversity control, or they become unpredictable quite, no?

If the driver does something like that, then the antenna control callback should reset that state as well. But then again, I don't mind if legacy hw requires the interface to be down as well.

- Felix