On Tue, Dec 21, 2010 at 10:15 AM, Zefir Kurtisi
<zefir.kurtisi@xxxxxxxxxxx> wrote:
>
>
> Signed-off-by: Zefir Kurtisi <zefir.kurtisi@xxxxxxxxxxx>
> ---
> Âinclude/net/cfg80211.h | Â 41 +++++++++
> Âinclude/net/dfs.h   Â| Â100 +++++++++++++++++++++
> Â2 files changed, 141 insertions(+), 0 deletions(-)
> Âcreate mode 100644 include/net/dfs.h
>
> diff --git a/include/net/cfg80211.h b/include/net/cfg80211.h
> index 03b3bae..c3ace0e 100644
> --- a/include/net/cfg80211.h
> +++ b/include/net/cfg80211.h
Why not just keep this on dfs.h?
> @@ -1820,6 +1820,47 @@ struct ieee80211_rate *
> Âieee80211_get_response_rate(struct ieee80211_supported_band *sband,
> Â Â Â Â Â Â Â Â Â Â Â Â Â Âu32 basic_rates, int bitrate);
>
> +
> +/**
> + * DFS pattern detector interface
> + */
> +
> +/**
> + * ieee80211_add_radar_pulse - add a pulse detected by HW to detector
> + *
> + * @freq: channel frequency in [MHz]
> + * @ts: time stamp in [us]
Is this the best known resolution we are aware hardware can use?
> + * @rssi: rssi value for the deteced pulse
RSSI values are vendor defined values, for Atheros it happens to be in
dBm above the noise floor, and the value is measured during the
preamble and PLCP; i.e. with the initial 4us of detection. If we can
assume other radar pulses from other vendors will match this same
definition then we can share the pulse rssi information, otherwise we
can't.
Ideally I wish we could use instead RCPI but I am not sure if AR9003 /
AR9280 can support a translation of the RSSI value to RCPI, IIRC we
could not, but there was some goals in newer hardware to make this
happen, so maybe AR9003 can.. we'll have to check.
> + * @width: pulse width in [us]
> + *
I suppose us is a reasonable value to use, but do we want a higher
resolution just in case for the future?
> + * Each pulse the HW identified as radar is fed into the detector via this
> + * function. The three values for ts, rssi and width are those relevant for
> + * pattern matching, while freq is required to map the pulse to the correct
> + * DFS channel.
> + * No value is returned, assuming the HW does not need to know about the result
> + * of pattern matching. The further processing of matches is done at mac layer.
> + */
> +extern void ieee80211_add_radar_pulse(u16 freq, u64 ts, u8 rssi, u8 width);
Maybe we should just keep pulses specific to the drivers and only let
them inform us of the actual radar signals determinations. Seems
overkill to share pulse information, or at least pointless if we
cannot sure or have to address a lot to try to share. Unless of course
we can share RCPI based DFS radar pulse recipes.
> +/**
> + * ieee80211_radar_detected - notify mac that DFS radar was detected
> + *
> + * @freq: channel frequency in [MHz]
> + *
> + * This function is used to inform the mac that a DFS radar was detected on
> + * the given channel frequency. It might be called from the DFS pattern
> + * detector or from device drivers that do DFS detection in HW.
> + *
> + * It is meant to be the central hook that initiates all subsequent actions that
> + * need to be performed after the detection, including
> + * Â- put channel to Unavailable list (or adjust channel state periods
> + * Â Âin case channel was already not on Available list)
> + * Â- everything else required to select new channel and initiate
> + * Â Âchannel switch
> + */
> +extern void ieee80211_radar_detected(u16 freq);
So drivers and mac80211 would call this right?
> +
> Â/*
> Â* Radiotap parsing functions -- for controlled injection support
> Â*
> diff --git a/include/net/dfs.h b/include/net/dfs.h
> new file mode 100644
> index 0000000..1dccc10
> --- /dev/null
> +++ b/include/net/dfs.h
> @@ -0,0 +1,100 @@
> +#ifndef DFS_H
> +#define DFS_H
> +/*
> + * Copyright 2010, Neratec Solutions AG, <zefir.kurtisi@xxxxxxxxxxx>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + */
> +
> +/**
> + * DOC: Introduction
> + *
> + * DFS radar detector interface
> + *
> + * This is a proposal for a common DFS pattern detector interface.
> + *
> + * It should be used by devices that are able to detect radar pulses and need
> + * pattern matching (as defined by ETSI, FCC, JP regulatories).
> + *
> + * An instance of the proposed DFS handler is supposed to exist during an
> + * endpoint's lifetime within mac80211. WLAN devices should report the radar
> + * pulses they detect during their uptime, the DFS handler aggregates them and
> + * keeps track of channel states (as defined by regulatories).
> + *
> + * On channel state changes it notifies mac80211 to initiate all required
> + * processing.
> + */
> +
> +
> +/* TODO: move those to more common place */
> +enum dfs_domain {
> + Â Â Â DFS_INVALID_DOMAIN Â Â Â= 0, Â Â/* Uninitialized dfs domain */
> + Â Â Â DFS_FCC_DOMAIN Â Â Â Â Â= 1, Â Â/* FCC dfs domain */
> + Â Â Â DFS_ETSI_DOMAIN Â Â Â Â = 2, Â Â/* ETSI dfs domain */
> + Â Â Â DFS_JP_DOMAIN Â Â Â Â Â = 3, Â Â/* Japan dfs domain */
> +};
You can use the values I posted on my series for this.
> +
> +/* TODO: move dfs_state to more common place */
> +enum channel_dfs_flags {
> + Â Â Â CHANNEL_INVALID Â Â Â Â = 0x00,
Do we really need INVALID ? Please document each of these.
> + Â Â Â CHANNEL_UNAVAILABLE Â Â = 0x01,
> + Â Â Â CHANNEL_USABLE Â Â Â Â Â= 0x02,
> + Â Â Â CHANNEL_AVAILABLE Â Â Â = 0x04,
> + Â Â Â CHANNEL_OPERATING Â Â Â = 0x08,
> +};
> +
> +/**
> + * struct pulse_event - events fed to the dfs handler
> + *
> + * @ts: absolute time stamp for start of pulse in us (e.g. as TSF)
> + * @freq: channel frequency in [MHz]
> + * @rssi: rssi value for the given pulse
> + * @width: pulse width for given pulse in [us]
> + *
> + */
> +struct pulse_event {
> + Â Â Â u64 ts;
> + Â Â Â u16 freq;
> + Â Â Â u8 Ârssi;
> + Â Â Â u8 Âwidth;
> +};
Yeah, this seems hardware specific, so likely not something we can
share I think.
> +/**
> + * struct dfs_handler - DFS handler pseudo-OO interface
> + *
> + * @exit: terminate DFS handler and release all resources
> + * @add_pulse: add given pulse event to detector lines
> + * Â Â Â Â Â Â returns 1 if added event triggered a pattern match
> + * @data: private instance data
> + *
> + * To easily attach pulse detectors implementing different types matching
> + * algorithms, a pseudo-OO design approach was taken with a tiny interface is
> + * chosen.
> + */
Not sure I get what this is for.
> +struct dfs_handler {
> + Â Â Â /* VFT */
> + Â Â Â void (*exit)(struct dfs_handler *_this);
> + Â Â Â int (*add_pulse)(struct dfs_handler *_this, struct pulse_event *event);
> +
> + Â Â Â /* private data */
> + Â Â Â struct dfs_data *data;
> +};
> +
> +/**
> + * dfs_handler_init - DFS handler constructor
> + *
> + * @dfs_domain: DFS domain to detect radar patterns for
> + *
> + * A DFS handler instance is allocated via this constructor.
> + * On success the pointer to the fully initialized handler is returned that
> + * can be fed with radar pulses during its lifetime. Allocated resources are
> + * released upon calling the destructor.
> + *
> + * On failure NULL is returned.
> + */
> +struct dfs_handler *dfs_handler_init(enum dfs_domain dfs_domain);
Same here.
Luis
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
