[LEDE-DEV] [PATCH 1/4] firmware: ath10k-firmware: update QCA4019 firmware to 10.4-3.2.1-00058

[Christian Lamparter]

On Monday, November 20, 2017 11:23:18 AM CET Matthew McClintock wrote:
> On Mon, Nov 20, 2017 at 10:53 AM, Christian Lamparter
> <chunkeey at gmail.com> wrote:
> > On Monday, November 20, 2017 9:45:28 AM CET Matthew McClintock wrote:
> >> Did you get ath10k working on ipq4019?
> >
> > Yes, I managed to upstream the wifi dts changes some time ago and they
> > all have been accepted. In fact the new 4.14 does ships with those.
> >
> > https://patchwork.kernel.org/patch/9858197/
> >
> > How is your upstreaming progess going?
> >
> >> I'm having issues with the board data files? I've never messed with this.
> >
> > What do you want to know?
> >
> > I'm assuming that you are looking for a way to package the two board.bin
> > (2GHz and 5GHz) into a board-2.bin. You can just use the ath10k-bdencoder
> > tool from the <https://github.com/qca/qca-swiss-army-knife>
> >
> > Note: ath10k-bdencoder can also unpack existing board-2.bin that are
> > currently shipped with any of the the ipq-wifi packages. So the fastest way
> > would be to unpack a existing board-2.bin (this will create a .json file and
> > the individual .bins). Then you just have to replace the two board.bin files
> > and repack it (the repack process will need the created .json file again.)
> >
> > If this isn't what you are looking for, you might find it in the
> > "firmware: add custom IPQ wifi board definitions" commit message.
> >
> 
> Caveat: I'm using an older version of ath10k at the moment, but....
This is can be a big caveat. Depending on how "old" the ath10k /
compat-wireless / backports really is. On OpenWRT/older LEDE versions
QCA4019 was broken by a patch: 936-ath10k_skip_otp_check.patch.

You should definitely go for a recent version.
 
> I think it's complaining that it can't find the right board info from
> the board-2.bin file which is just multiple board files wrapped up
> together?
> 
> [  233.761660] ath10k_ahb a000000.wifi: failed to fetch board data for
> bus=ahb,vendor=0000,device=0000,subsystem-vendor=0000,subsystem-device=0000
> from ath10k/QCA4019/hw1.0/board-2.bin
The device identifcation doesn't work.

The driver is thinking it has a non-sensical "ahb pci(e)" device.
Just look at the fact that the driver is looking for "bus=ahb" but also 
"vendor=0,device=0...,subsystem...=...,..."(which is used to identify
pcie devices). If it was looking for a bmi device, the board search string
would look something like "bus=ahb,bmi-chip-id=0,bmi-board-id=16"

I can only guess what's really going on, since this is such a short extract.

But there at least two likely scenarios:
1. 936-ath10k_skip_otp_check.patch is present / ath10k is too old

2. the pre-cal-ahb-a800000.wifi.bin and pre-cal-ahb-a000000.wifi.bin
   don't get loaded or the pre-cal* data is shifted/trimmed/mangled/bad.

For 2. you should check if the firmware hotplug 11-ath10k-caldata script
is correctly extracting the pre-cal from the "ART" Partition of your device
to /lib/firmware/ath10k. I posted one of the RT-AC58U's pre-cal files on the
linux-wireless / ath10k-devel mailing-list.
If you need something to compare to, look for the cal-file attachment:
<https://www.mail-archive.com/ath10k@lists.infradead.org/msg05911.html>

(Note: Netgear used to ship the source from which these .bin files are
created by the calibration utility in thier GPL source of the R7800 / D7800)

> This is the first pass, it moves onto board.bin but still fails (the
> file is not present).
> 
> Where is a description of the which board.bin files to use for which
> radios? Where does this info come from? Somewhere in the radio or OTP?
I don't think this was actually documented anywhere officially by
Qualcomm Atheros. The best resource I currently know of, is this write-up
from Sven Eckelmann:
<http://lists.infradead.org/pipermail/ath10k/2017-January/009025.html>

followed by the response from Adrian Chadd (he used to work for QCA, but
no longer does):
<http://lists.infradead.org/pipermail/ath10k/2017-January/009104.html>

If you have further questions, your best bet is to write to the
linux-wireless and ath10k-devel ML.
 
> Interestingly, PCI is complaining too:
> 
> [  236.712965] ath10k_pci 0000:01:00.0: failed to fetch board data for
> bus=pci,bmi-chip-id=0,bmi-board-id=19 from
> ath10k/QCA9888/hw2.0/board-2.bin
> 
> Is my board just not setup right? Missing OTP id? I guess I can force
> a working radio with a board.bin file?
This "board-id=19" is currently not available in the ath10k-firmware.git:

According to the latest QCA9888 patch, it just supports
<https://github.com/kvalo/ath10k-firmware/commit/3cfabec862d4969a64ac6ddf2bfe69deb72107ac#diff-d57e9359d3fc1215cfc7c23ddb9189cb>

"New:
bus=pci,bmi-chip-id=0,bmi-board-id=24
bus=pci,bmi-chip-id=0,bmi-board-id=25
bus=pci,bmi-chip-id=0,bmi-board-id=17
bus=pci,bmi-chip-id=0,bmi-board-id=18
bus=pci,bmi-chip-id=0,bmi-board-id=23

Changed:
bus=pci,bmi-chip-id=0,bmi-board-id=16
"

I know that Henryk had similar issues with his Archer C58/C59/C60 and his
"board-id=20". I advised him to take this matter upstream, which he did:
<https://github.com/kvalo/ath10k-firmware/pull/5>

however, it has been sitting in the ath10k-firmware queue for a while now.

Note: The QSDK installs the boardData.* into the /lib/firmware/IPQ4019/hw.1/
directory (replace the IPQ4019 with QCA9888 etc...). If you have a vendor 
image, you might be able to use binwalk -e to extract the rootfs and get 
the boardData.bin from there.

However these boardData.bins are in the old "board.bin" format. You'll need
the ath10k-bdencoder tool to generate a board-2.bin that you can put into
the ipq-wifi package.

Regards,
Christian