brcmsmac: regression in transmit throughput since 3.19-rc1

brcmsmac: regression in transmit throughput since 3.19-rc1

[Arend van Spriel]

[-- Attachment #1: Type: text/plain, Size: 3661 bytes --]

Hi Johannes,

On our nightly test setups I found that STA->AP TCP iperf showed a 
regression in brcmsmac after I upgraded them from 3.18-rc5 
wireless-testing (f638577)to 3.19-rc1 wireless-testing (f9e7c6c). The 
attached pictures are historic results showing clear drop on the last run.

Has there been any report on there since 3.19-rc1 and would you have a 
pointer on where to start looking. The changes in brcmsmac look trivial 
so looking at mac80211 now. There are some changes in minstrel and 
brcmsmac is using minstrel_ht rate control.

Regards,
Arend
--8<-------------------------------------------------------------------
$ git log f638577..f9e7c6c --oneline -- net/mac80211
28a9bc6 mac80211: free management frame keys when removing station
722ddb0 mac80211: update the channel context after channel switch
d025933 mac80211: fix multicast LED blinking and counter
7e6225a mac80211: avoid using uninitialized stack data
81c4126 Merge branch 'master' of 
git://git.kernel.org/pub/scm/linux/kernel/git/l
de51f16 Merge 
git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
f027c2a mac80211: add ieee80211_tx_status_noskb
7e1cdcb mac0211: add a helper function for fixing up tx status rates
ad9dda6 mac80211: pass tx info to ieee80211_lost_packet instead of an skb
63558a6 mac80211: minstrel_ht: switch to .tx_status_noskb
fb7acfb mac80211: minstrel: switch to .tx_status_noskb
f684565 mac80211: add tx_status_noskb to rate_control_ops
9594342 mac80211: minstrel_ht: move aggregation check to .get_rate()
336004e mac80211: add more missing checks for VHT tx rates
2967e03 mac80211: copy chandef from AP vif to VLANs
98f0334 cfg80211: clean up beacon loss CQM event
40a11ca mac80211: check if channels allow 80 MHz for VHT probe requests
575f053 mac80211: disable 80+80/160 in VHT correctly
9a638dd Merge tag 'mac80211-next-for-john-2014-11-20' of 
git://git.kernel.org/pu
75769c8 mac80211: minstrel_ht: add a small optimization to 
minstrel_aggr_check
f815e2b mac80211: notify drivers on sta rate table changes
628c010 mac80211: skip legacy rate mask handling for VHT rates
8b1956f mac80211: don't allow 40MHz tx rates in case of 20MHz chandef
a344d67 mac80211: allow drivers to support NL80211_SCAN_FLAG_RANDOM_ADDR
6ea0a69 mac80211: rcu-ify scan and scheduled scan request pointers
ff5db43 mac80211: remove redundant check
b6da911 mac80211: synchronously reserve TID per station
4f9610d mac80211: add specific-queue flushing support
8a4d32f mac80211: add TDLS channel-switch Rx flow
a7a6bdd mac80211: introduce TDLS channel switch ops
5383758 mac80211: add parsing of TDLS specific IEs
c273390 mac80211: prepare TDLS mgmt code for channel-switch templates
9041c1f mac80211: track AP and peer STA TDLS chan-switch support
78632a1 cfg/mac80211: define TDLS channel switch feature bit
2cedd87 mac80211: add BSS coex IE to TDLS setup frames
f0d29cb mac80211: add supported channels IE during TDLS setup
7528ec5 mac80211: add function to create data frame template including key
4c9451e mac80211: factor out 802.11 header building code
73c4e19 mac80211: move skb info band assignment out
1277b4a mac80211: retransmit TDLS teardown packet through AP if not ACKed
24d342c mac80211: add option for setting skb flags before xmit
0395442 mac80211: refactor duplicate detection
1f7bba7 mac80211: add back support for radiotap vendor namespace data
d04b5ac cfg80211/mac80211: allow any interface to send channel switch 
notificati
2f45729 mac80211: send channel switch started notifications
127f10e mac80211: add device_timestamp to the drv_pre_channel_switch trace
000baa5 mac80211: fix order of setting ch_switch and 
drv_pre_channel_switch call
(

[-- Attachment #2: nightly NGR6300.4331 0 ch=3 {4313_09 NGR6300.4331}.png --]
[-- Type: image/png, Size: 4791 bytes --]

[-- Attachment #3: nightly NGR6300.4331 0 ch=3 {43224_09 NGR6300.4331}.png --]
[-- Type: image/png, Size: 4686 bytes --]

[-- Attachment #4: nightly NGR6300.4331 0 ch=3 {43224_21 NGR6300.4331}.png --]
[-- Type: image/png, Size: 5690 bytes --]

[-- Attachment #5: nightly NGR6300.4360 0 ch=36 {43224_09 NGR6300.4360}.png --]
[-- Type: image/png, Size: 5695 bytes --]

Re: brcmsmac: regression in transmit throughput since 3.19-rc1

[Johannes Berg]

Hi Arend,

> On our nightly test setups I found that STA->AP TCP iperf showed a 
> regression in brcmsmac after I upgraded them from 3.18-rc5 
> wireless-testing (f638577)to 3.19-rc1 wireless-testing (f9e7c6c). The 
> attached pictures are historic results showing clear drop on the last run.
> 
> Has there been any report on there since 3.19-rc1 and would you have a 
> pointer on where to start looking. The changes in brcmsmac look trivial 
> so looking at mac80211 now. There are some changes in minstrel and 
> brcmsmac is using minstrel_ht rate control.

I haven't heard of anything like that yet. No idea where to look either
- but the drop is significant enough that perhaps you can bisect?

johannes

Re: brcmsmac: regression in transmit throughput since 3.19-rc1

[Arend van Spriel]

On 01/19/15 13:52, Johannes Berg wrote:
> Hi Arend,
>
>> On our nightly test setups I found that STA->AP TCP iperf showed a
>> regression in brcmsmac after I upgraded them from 3.18-rc5
>> wireless-testing (f638577)to 3.19-rc1 wireless-testing (f9e7c6c). The
>> attached pictures are historic results showing clear drop on the last run.
>>
>> Has there been any report on there since 3.19-rc1 and would you have a
>> pointer on where to start looking. The changes in brcmsmac look trivial
>> so looking at mac80211 now. There are some changes in minstrel and
>> brcmsmac is using minstrel_ht rate control.
>
> I haven't heard of anything like that yet. No idea where to look either
> - but the drop is significant enough that perhaps you can bisect?

Bisect it is. Was hoping for some obvious change to avoid that.

Regards,
Arend