tags 919632 - upstream
thanks
Dear Ben, thanks a lot for all your time and effort!
tl;dr: Please don't revert. I've found the culprit. See my recommended
bug fixing procedure near the bottom of this message.
Zitat von Ben Hutchings <b...@decadent.org.uk>:
It looked like the problem in these three bugs was that
newer versions of the driver that preferred firmware-6.bin were
incompatible with firmware-5.bin even though they tried to use it.
Very close. This bug is in fact dependent on the driver version. I'm
very sorry for not having picked up that clue before, and instead
havin tried the stretch linux-image 4.9.0 all the time. Your latest
E-Mail led me to the correct hunch. I'm not certain about the other
bug reports right now, but your bugfix might take care of all of them.
Since the older driver version also fails, I suppose I should revert
the previous upstream change to firmware-5.bin.
No, please don't do that (quite yet), for three reasons:
1. This very change might well have solved some of the other bugs you
tagged solved.
2. The mere addition of ath10k/QCA9377/hw1.0/firmware-6.bin definitely
didn't worsen (at least) my situation.
3. But it improved it quite a lot: I just switched linux-image from
stretch 4.9.0-9-amd64 to stretch-backports 4.19.0-0.bpo.4-amd64, and
*snap* that made firmware-atheros_20190502-1 work for me, (like a
charm IDK quite yet, but at least it doesn't instantly crash any more)
successfully using firmware-6.bin. \o/ Would you like to see my now
working dmesg, or any other test results? (I'll need a few days before
I return to broadband land, :-/ and also I've got a lot of AppArmor
mess to clear up.)
My '3.' makes it look to me like the bug is really a missing
dependency in firmware-atheros (since 20180518-1) on the specific
kernel version when the firmware API or the ath10k_pci driver changed;
somewhere >> 4.9.0 and <= 4.19.0. (Which one, and exactly for which
drivers IDK, and I won't be able to run a lot of tests on that for a
few days from now, so please bear with me.)
I strongly recommend the following procedure to fix this bug:
* Leave this bug open for now.
* Leave its severity and title (at least somewhat) the way they are,
so that other users can easily find the cause.
* Remove its 'upstream' tag. (hereby done, I hope)
* Leave the firmware blobs the way they are in 20190502-1.
* Find out their required minimum driver/kernel version. (possibly
with my future help)
* Add the missing dependency firmware-atheros -> linux-image
${version}, and maybe document that in the package description, and/or
in firmware-.*/README.Debian, and/or wherever you deem appropriate. At
least the dependency seems important to me for all the present and
future backports users.
* Close this bug :D
(BTW: Although ath10k_pci is compiled as a module rather than as a
built-in, reading
https://www.kernel.org/doc/html/latest/driver-api/firmware/direct-fs-lookup.html#firmware-and-initramfs leads me to believe that it might make sense to include the firmware blob(s) into the initrd along with the module(s). That's not the default behavior at the moment. I strongly expect this to have already been discussed somewhere within debian. @all: Would somebody please point me anywhere near that
reasoning?)
Again, thanks a lot Ben for all your time and effort! Cheers
Kevin
