| Message ID | 1485893272-6893-1-git-send-email-rmanohar@qca.qualcomm.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Johannes Berg |
| Headers | show |
On Tue, 2017-01-31 at 12:07 -0800, Rajkumar Manoharan wrote: > As moving average is not considering fractional part after > certain ratio, it will stuck at the same state. For example > with current values, moving average stuck at 96 and it will > not move forward. Fortunately current threshold is matching > against 95%. If thresold is increased more than 96, mesh path > never be deactivated under worst case. Fix failure average > movement by bumping up average at stuck state. This is ... really strange to me. Can't we just actually take into account fractional parts instead? I think you should instead convert this to the EMWA helpers (see include/linux/average.h). johannes
diff --git a/net/mac80211/mesh_hwmp.c b/net/mac80211/mesh_hwmp.c index b747c9645e43..f0aa1da15bd0 100644 --- a/net/mac80211/mesh_hwmp.c +++ b/net/mac80211/mesh_hwmp.c @@ -300,6 +300,7 @@ void ieee80211s_update_metric(struct ieee80211_local *local, { struct ieee80211_tx_info *txinfo = IEEE80211_SKB_CB(skb); struct ieee80211_hdr *hdr = (struct ieee80211_hdr *) skb->data; + u32 fail_avg = sta->mesh->fail_avg; int failed; if (!ieee80211_is_data(hdr->frame_control)) @@ -308,8 +309,17 @@ void ieee80211s_update_metric(struct ieee80211_local *local, failed = !(txinfo->flags & IEEE80211_TX_STAT_ACK); /* moving average, scaled to 100 */ - sta->mesh->fail_avg = - ((80 * sta->mesh->fail_avg + 5) / 100 + 20 * failed); + fail_avg = ((80 * fail_avg + 5) / 100 + 20 * failed); + + /* bump up fail average since fractional part of average is ignored. + * Otherwise fail average always stuck at the same level and + * never moves forward. + */ + if (fail_avg && fail_avg == sta->mesh->fail_avg) + fail_avg++; + + sta->mesh->fail_avg = fail_avg; + if (sta->mesh->fail_avg > 95) mesh_plink_broken(sta); }
As moving average is not considering fractional part after certain ratio, it will stuck at the same state. For example with current values, moving average stuck at 96 and it will not move forward. Fortunately current threshold is matching against 95%. If thresold is increased more than 96, mesh path never be deactivated under worst case. Fix failure average movement by bumping up average at stuck state. Signed-off-by: Rajkumar Manoharan <rmanohar@qca.qualcomm.com> --- net/mac80211/mesh_hwmp.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-)