mchehab: when will you merge 5.10 into our master?
hverkuil: will do right now
pushing it
this is a short week due to ELCE
btw, I may not be able to join our today's meeting
I intend to watch some topics today
OK. Nothing much happened during the merge window, so I think we can skip it.
ok. Well, I should be here at the usual time
so, if are there anything for discussions, that should be ok
I believe there is one pending PR for 5.11 from me. I plan to start plowing through the pile of patches tomorrow and next week. I am very much behind :-(
I'll try to help you during this merge window in order to try cleaning up the queues
there is the 103+ patch sereis from sailus that need review
didn't have time for that yet
sailus: I'll try to reserve a day for reviewing next week, after Nov, 3
(probably I'll review it at Nov, 4 - if nothing urgent pops up)
I think it is mostly a lot of relatively small patches, I can't remember seeing anything terribly complicated in the past 3-4 weeks.
mchehab: thanks!
those small patches are the ones that usually takes most of the time ;-)
i mean, most of them are easy to review - but there tend to accumulate with time
s/there/they/
I used to reserve a couple of days on every merge window just to clean them up
haven't been done this since ~June last year, as I had several changes that prevented me to do tha
tha -> that
In file included from <built-in>:2:
In file included from ././include/linux/compiler_types.h:69:
./include/linux/compiler-clang.h:11:3: error: Sorry, your version of Clang is too old - please use 10.0.1 or newer.
# error Sorry, your version of Clang is too old - please use 10.0.1 or newer.
  ^
funny warning on today's build with clang at the builder robot
crap! it seems that there's no clang v10 backports package for Debian buster
ok, llvm.org provide packages for it
I'll try rebuilding it, with clang-11
the only question I have for today is if someone could have a look at the vb2 userspace ABI breakage :-)
mchehab: Ack. Obrigado!
pinchartl: what ABI breakage? the one from Daniel's series?
(which btw will also require a fix for VB1)
no, let me dig up the patch
"[PATCH] media: videobuf2: Fix length check for single plane dmabuf queueing"
that broke userspace by returning an error when queuing a dmabuf with v4l2_buffer.length = 0
before that, the length was taken from the dmabuf within the kernel
that seems something that require urgent fix
it was merged in v5.8 I think
I only discovered it in v5.9
I guess most distros are still at v5.8
I believe there's an equivalent breakage for multiplane buffers that's even older
(the newest ones, I mean)
for now, I think we should just revert the patch that caused the breakage on 5.8
there's a mail thread in response to this patch, maybe you can reply to it ?
sure... reading the messages first...
the proposed patch works
it doesn't fix the issue that the initial patch tried to address
but that could be tackled separately
yes. well, if we have a fix that would solve both, that would be better...
but we should at least send the ABI/API breakage fix asap
anything else could be done in separate, for 5.10
(or 5.11?)
with backport to stable I suppose ?
yes
do you know what commit caused the regression? there's no Fixes: tag on his patch
https://github.com/torvalds/linux/commit/8a75ffb81
found it
hverkuil: any comments about that? it sounds that you wrote the patch which caused the regression
mchehab: will look at that today or tomorrow.
it's very high in my todo list
ok
just reverting 961d3b27a2aff52dda0b97d35085a743a27c2f46 would be a quick solution for the regression
building media with clang-11 is OK: https://builder.linuxtv.org/job/media_tree_clang/212/console
no warnings
(W=0)
I'll switch the script to use W=1
W=1 with clang report a few issues