Profile cover photo
Profile photo
Ben Greear
Ben Greear's posts

Post has attachment
I am pleased to announce the latest beta-9 release of the 10.4 ath10k CT firmware for the 9984, 9980, 9888 and similar chipsets. This release fixes a regression related to EAPOL handling I had introduced several releases ago and fixes a few tx-hang related bugs in upstream code that I found while testing 64 station vdevs. It seems to work well with 64 station vdevs now, at least if you have a supplicant that waits to get the tx-status of the EAPOL 4/4 before setting the key.

Post has attachment
Anyone using CT ath10k firmware should consider this supplicant patch. It significantly improves the odds of getting a wave-2 NIC (and likely wave-1 as well) to properly 4-way authenticate. Upstream QCA ath10k firmware has EAPOL hacks that make this at least less important, though those EAPOL hacks were previously shown to break .11r in our testing. Maybe more recent QCA firmware no longer has the bugs we originally debugged, however.

Post has attachment
I have uploaded new CT ath10k firmware for the various chipsets supported by CT firmware. There are some significant bug-fixes in this release, including a long-term bug that caused the channel to be configured improperly sometimes. This bug was especially easy to trigger on 9888 chipset NICs with multiple station vdevs, and seems to have affected the small-sized 4x4 Compex 9984 NIC as well. I'll update LEDE soon, but in the meantime you can download the latest beta images from my page:

Post has attachment
I am proud to announce new ath10k beta firmware for all chipsets we support. The 10.4 firmware for QCA9984, QCA9888, QCA9886, and QCA9980 has a fix for a nasty memory corruption due to careless use of void*.

Both firmware version have a fix for passing invalid objects to some beacon and WDS related code. I removed several uses of void* in order to find and fix these problems as well as decrease chance of it happening again.

In NIC news, we tried out the Zcomax AC-924 (CUS239 based NIC with 9984 chipset) and it worked well for us. Thanks to them for fast lead time, small quantity purchase support, and friendly service.

I also updated the ath10k driver for 4.7 and 4.9 kernels to be able to load the Compex WLE650V5-18 NIC more reliably, and fixed a regression in the 4.9 kernel that could cause kernel deadlocks.

I updated the release notes for each firmware, so please see the respective pages for more details:

Post has attachment
What luck! Someone reported a peculiar and reproducible bug with my ath10k CT firmware for 988x and 9887 NICs. There was an un-initialized variable that made blockack not work sometimes, and that of course killed performance. The latest beta builds should have this fixed. Anyone who has been seeing occasional (very) slow throughput should give this a try.

Post has attachment
Here's an update with regard to ath10k firmware and IBSS mode. I have been doing some work for a user that has a 9887 network card and wants to run IBSS networks using batman meshing. The OS is a recent LEDE with the ath10k-ct firmware and ath10k-ct driver. It mostly would run fine for me, but sometimes the connection would drop, and then pick back up a minute or so later.

The first thing I tried was to use an 'nrcc' firmware variant that also enabled htt-mgt and compiled out the firmware connection keepalive logic. This improved the stability, probably because the kernel's keep-alive logic is better than the firmware's keepalive logic. Without encryption, LEDE doesn't run supplicant, so I guess it is just the kernel stack dealing with keepalive issues. The 'nrcc' part means that rate-ctrl caching is disabled, which uses less resources on the host, but uses more firmware RAM. So, I used the 'fwcfg' option in the ath10k-ct driver to set vdevs = 4, peers =80 so that the firmware could load.

I still see some link dropouts, so next I tried configuring psk2 (WPA2) authentication. This means supplicant loads and manages the network. I have not seen a dropout in 10+ minutes once WPA was enabled. I suspect that if LEDE were convinced to use suplicant for OPEN, stability might also increase.

Latest beta-19 firmware images are uploaded, and web page has been updated to mention some 'fwcfg' notes for nrcc variants, and a new htt-mgt, trimmed, community firmware image (firmware-2-ct-htt-mgt-nrcc-community.bin ) is linked.

I have just uploaded the latest beta-19 10.1 CT firmware. This fixes an issue with EAPOL 4/4 in some cases, and also fixes a problem where the NIC tried to transmit with STBC when configured as a 1x1 device (like 9887 chips). This STBC bug is why 9887 CT ath10k firmware would not work in /AC mode.

Post has attachment
Isaac spent some quality time using the isolation chamber to do some performance testing with LANforge and ath10k.

First, here is LANforge acting as AP and station devices, using QCA9880 3x3 hardware:

And, he also created some test rigs using LANforge with QCA9984 wifi NIC and 10G Ethernet for connectivity. We are not sure why upload is so much less than download, but that seems to be a general issue with ath10k for some time.

In case you read closely, 'Armageddon' is a modified version of pktgen.


Post has attachment
I have promoted the ath10k CT firmware beta images to final. For the 9880 and related wave-1 chips, this means 802.1Q VLAN support, lots of backport patches from upstream 10.2 firmware, improvements to rate-control and assorted bug fixes. For the 9984 and related wave-2 NICs, this means increased stability as well as updated to a fairly recent upstream 10.4 code base. More complete release notes on found on the respective pages for the firmware images.

Post has attachment
I am pleased to announce the latest ath10k CT firmware. Both 10.1 (9880, 9982, 9887 and related chipsets) and 10.4 (9886, 9888, 9980, 9984) have been updated. The 10.1 has some small fixes and extra debugging code to help diagnose some crashes seen in the field. The 10.4 has a fix for EAPOL 4/4 sending (previous code would just not send the 4/4 often).
Wait while more posts are being loaded