At some point in time, I decided it was a good idea to attend three conferences
in September. This was a busy but fairly productive month.
Early September was Open Source Summit and Linux Plumbers Conference. Open
Source Summit is the renamed LinuxCon. Matthew Garrett gave a talk about
signing binaries with the Integrity Measurment Architecture (IMA) subsystem.
This is a subsystem that ties hashes to the TPM module available on many
modern systems. Two of the kernel Outreachy interns gave a talk on their
projects involving the radix trees. I always enjoy listening to Outreachy
interns talk about their projects and it sounded like they made good progress
on improving the radix tree. Dawn Foster gave a talk on collaboration in
kernel mailing lists. This was focusing on how kernel developers collaborate
via e-mail. There were graphs showing the network via e-mail, which mostly
showed that there are a handful of people who tend to e-mail each other
frequently. This work was part of a PhD program so I look forward to seeing
future work. Sarah Sharp gave a talk on code of conducts and enforcement.
My favorite take away from this talk was thinking of code of conduct violations
in terms of a threat model and what they mean for the well-being of your
participants. I was also on the kernel panel with several other developers.
This is usually a chance for people to come and ask questions about whatever
kernel stuff. I didn’t put my foot in my mouth so I consider it a success.
I was on the planning committee for Linux Plumbers so I ended up doing a bunch
of behinds the scenes work in addition to the hallway track and going to the
occasional session. Jon Corbet gave a talk about the kernel’s limits to
growth. He gave a version of this talk at Linaro Connect and I gave some
about this previously. Most of my opinions there still stand. Grant Likely
held a BoF about the upcoming ARM EBBR. If you’ve been following the arm64
server space, you may have heard of SBBR which is a boot specification for
arm64 servers. The EBBR is something similar for embedded devices. As a
Fedora maintainer, I’m happy to see this moving forward to make booting arm64
SBCs easier. There was a discussion about contiguous memory allocation for
DMA. Some hardware vendors have discovered they get better performance if
they use a contiguous block of memory despite possibly having an IOMMU. The
proposal was to use the
MAP_CONTIG with mmap to get appropriate memory. There
wasn’t a conclusion and discussion is ongoing on the mailing list.
The next week was XDC. This is nominally a graphics conference and was out
of my typical scope. My primary purpose was to present on Ion.
This was a one-track conference so it was nice to not
have to make hard choices. Since I’m not a graphics developer, many of the
details of some talks went over my head. The discussion about the
Intel Graphics CI
was useful to hear. Intel has put a lot of effort in terms of
machines and tests to graphics testing. A big takeaway from me was the need
to add tests slowly to make sure that bugs can actually be fixed when
they are found. This sounds obvious but without doing this, CI becomes noisy
and is worthless.
The final week was Linaro Connect. My primary reason for attending was (again)
to talk about Ion but as always I attended some talks in between doing my
regular work. I went to session about Secure Boot on ARM with an update of
“not much has changed”. There was a session about the Linaro common kernels
and how those are maintained. The Linaro kernels once contained a large
number of backports but there’s been a lot of progress made towards getting
code upstream. These days, there’s more focus on testing and validating stable
kernels upstream. Linaro should be providing automated testing to stable
updates in the near future. There was a cross distro BoF which mostly touched
on some discussion with toolchains because everything is pretty okay in
distro land! Illyan Malchev from Google gave a keynote about Project Treble
and announced that LTS kernels are now 6 years instead of 2. I talked
about some of these Android announcements and the move to a 6 year cycle
is a very good thing. This will make it easier to give updates without
having to rebase to a brand new kernel.
During these several weeks, I talked a lot about Ion. There’s been some good
progress this past year towards moving Ion out of mainline. I removed a bunch
of code and greatly simplified the ABI. The biggest issue keeping Ion in
staging is making sure the ABI is stable. There was agreement that we could
look to move to a split ion (/dev/ion0, /dev/ion1 etc.) to better restrict
heap access. I got some feedback that the existing 32-bit flags field may
not be enough for expanding use cases so I’m going to look to utilize some of
the existing fields we have for padding. There was a session on Ion integration
challenges which was very useful to me. One of the hardest parts of doing Ion
work has been finding a direction. Hearing what problems people are actually
having makes this easier. Some of the problems are because I took the obvious
(read slow) approach at things like cache maintenance. I tried to encourage
people to submit patches for the features that are mission so hopefully I’ll
see those coming in the near future.
This past weekend was SeaGL, a local Seattle conference.
Because I’m local, it was easy for me to attend. Unfortunately three weeks
of conferencing caught up with me so I didn’t attend as many parts of it as
I would have liked. I attended the keynotes Friday and Saturday. Both Nithya
Ruff and Rikki Endsley gave fantastic talks which were recorded. I gave a talk
on Saturday afternoon called “Creating Fresh Kernels”. This was a variation
of the talk I gave at DevConf last year, focusing more on the choices
distributions might make with their kernels instead of just why Fedora is
awesome. I was fairly happy with how it went; the room was full and everyone
asked good questions. I’d like to present again next year if I can come up
with a good topic.
I have a little bit of downtime before Open Source Summit Europe/Kernel Summit
in a few weeks. We’ll see if I can follow up on some of the discussions in