@Aissen There's som scoped areas where I need to just play with the driver a bit, and observe how the ends interact. It has been hard to focus to those areas while there's been much bigger existential issues inherited from the OOT driver (broken driver model, practically all code related to frame buffering was obselete by any modern standards and required a full rewrite to bring to vb2/dma-buf universum).
E.g. it might make sense to consider some sort of signaling messaging for the output producer so that it can go to voluntararily to sleep.
E.g. it might make sense to consider some sort of signaling messaging for the output producer so that it can go to voluntararily to sleep.
1
0
1
Jarkko Sakkinen
jarkko
vcamera is probably a better name (vs. v4l2-vcam) and v4l2-core is wrong place for the driver so for RFC review I just place it straights under drivers/media. No reason to try to be smart here...
0
1
1
Jarkko Sakkinen
jarkko
vcamera is probably a better name (vs. v4l2-vcam) and v4l2-core is wrong place for the driver so for RFC review I just place it straights under drivers/media. No reason to try to be smart here...
0
1
1
Jarkko Sakkinen
jarkko
@Aissen
1. Test program: https://codeberg.org/jarkko/video-loop-test
2. Documenting API usually makes sense as 777 + 1 change, doesn't it? ;-) Anyhow, I keep this in mind! Thanks for the comment.
I'm thinking of kselftest but for the moment out-of-tree test program is easier to work with.
1. Test program: https://codeberg.org/jarkko/video-loop-test
2. Documenting API usually makes sense as 777 + 1 change, doesn't it? ;-) Anyhow, I keep this in mind! Thanks for the comment.
I'm thinking of kselftest but for the moment out-of-tree test program is easier to work with.
1
0
1
Jarkko Sakkinen
jarkko
Edited 4 months ago
Let me present v4l2-vcam:
https://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd.git/log/?h=vcam
Vcam user driver is created by opening /dev/vcam and then issuing V4L2_VCAM_IOC_NEW. This will emit /dev/videoX for readers. Reader semantics enforces single reader limitation.
In addition:
1. Supports zero-copy operation for a shared capture and output buffer (enabled automatically when possible).
2. Uses DMA contig allocator automatically when possible (falls back to vmalloc).
3. VB2 framework used for the pipeline.
These all are BTW unresolved in the OOT driver, and should help with getting better picture and frame rates over what we can do outside of kernel.
The code will code about 777 rounds of cleanups and tweaks and then I'll send the RFC patch ;-)
https://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd.git/log/?h=vcam
Vcam user driver is created by opening /dev/vcam and then issuing V4L2_VCAM_IOC_NEW. This will emit /dev/videoX for readers. Reader semantics enforces single reader limitation.
In addition:
1. Supports zero-copy operation for a shared capture and output buffer (enabled automatically when possible).
2. Uses DMA contig allocator automatically when possible (falls back to vmalloc).
3. VB2 framework used for the pipeline.
These all are BTW unresolved in the OOT driver, and should help with getting better picture and frame rates over what we can do outside of kernel.
The code will code about 777 rounds of cleanups and tweaks and then I'll send the RFC patch ;-)
1
2
1
@jani I'm now also experimenting with zerocopy dmabuf changes and contiguous allocator. I'm worried that if I leave them out, I might be ignoring something that would become bottleneck adding them :-) I'm glad I wrote year ago a program that streams scrolling checkerboard to loopback device, as I think I can make its flow dmabuf enabled quite easily :-)
0
0
1
Jarkko Sakkinen
jarkko
@jani https://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd.git/log/?h=vcam
Not yet ready for submission but it has the basic blocks in some form: VB2 and file-descriptor based life-cycle. It requires some iterations of cleanup, simplifications, tweaks etc. but I think it is now framed correctly :-)
Not yet ready for submission but it has the basic blocks in some form: VB2 and file-descriptor based life-cycle. It requires some iterations of cleanup, simplifications, tweaks etc. but I think it is now framed correctly :-)
1
1
1
Jarkko Sakkinen
jarkko
v4l2-vcam represents a rare situation where I'm testing a patch with the host kernel (or finalizing).
1
1
0
Jarkko Sakkinen
jarkko
@puhuri @suihkulokki @antti CEO of Verge seems excited tho at LinkedIn ;-) It's a weird combination of high-profile investors, one most likely legit company and a somewhat fuzzy startup. I'm waiting for the next episode...
0
0
2
@suihkulokki There's also quite known angel investors behind Donut Lab, such as founder of F-Secure, Risto Siilasmaa. This is weird.
0
0
1
Jarkko Sakkinen
jarkko
lol, who would guessed: tony blair is in the executive board
... and trump just called brits who participated to Nato wars as "losers"
🇬🇧
... and trump just called brits who participated to Nato wars as "losers"
🇬🇧
0
0
1
@suihkulokki looking at news magic battery looks like a huge hoax :D weird that they ended up collaborating.
0
0
1
@suihkulokki OK but battery technology is still somewhat question mark, isn't it? I.e. only claims and rumours
0
0
1
@suihkulokki "AK-47 - accept no subsitutes" ;-)
https://www.topgear.com/car-news/kalashnikov-has-made-its-own-retro-ev
https://www.topgear.com/car-news/kalashnikov-has-made-its-own-retro-ev
0
0
0
@suihkulokki still the motorcycle looks savvy despite being probably fake :-) or like russian EV ....
1
0
1