New blog post introducing and demoing retsnoop, a powerful and ergonomic tool for debugging confusing kernel errors and learning kernel internals through tracing. https://t.co/MLeggqXb1X #linux#tracing
I am pleased to announce that our paper "An In-Kernel Solution Based on #XDP for #5G#UPF: Design, Prototype and Performance Evaluation" has been accepted 1st joint International Workshop on Network Programmability and Automation.
Computing Performance: On the Horizon, @usenix#LISA21 video & slides are now online. The current state, and my predictions, for CPUs, memory, disks, and more. https://t.co/8wWXZ1pJim https://t.co/GO2TQ0b56L
BPF Internals, @usenix#LISA21 video & slides are now online. From bpftrace to machine code, showing how it all works. https://t.co/NB4qhk9zx5 https://t.co/UT8V4HdPbS #eBPF
Having to deal with bpf_printk() recently for some fancier debug output I realized there are a bunch of interesting aspects worth sharing, especially taking into account various kernel versions. This resulted in https://t.co/wutjqgxXF4 Hope it will be helpful to someone.
New kernel post! This time: /proc/*/mem and how it writes to unwritable mem.
Key takeaway: By walking page tables in software, the kernel can access userspace mem without directly dereferencing pointers it gets from userspace. (No WP/SMAP bits involved). https://t.co/U9IOeyHIc5
Removing receive side bottlenecks in network performance with eBPF and CPUMAP...aka really ensuring flows are spread across cpus: https://t.co/5UaHsEriXH
I wrote an article about BPF and Linux perf 🎉
The article includes step-by-step debugging with flame graphs, and many nitty-gritty details that tie themselves together at the end (i promise!).
#bpf#ebpf#linux
https://t.co/e1m5qiiDcF
Prediction: exokernels era begins thanks to eBPF
Interesting use of eBPF to reduce the access latency to NVMe (half of the overhead is due to the kernel storage path usually) 👇
https://t.co/Ao5kj8a5yi
Meet bperf: Sharing hardware PMCs with BPF.
https://t.co/08nZl7bzum
2nd use of BPF skels in perf, first one was to use BPF in perf to count events in BPF programs.
Again, thanks to @SongLiuBraving!
#perf_tools#bpf#linux
gRPC server in kernel with eBPF; three times faster than go, two than Rust. Implemented minimum features just for benchmarking. eBPF (sockhash) gives hook for data arriving on TCP/UDP sockets, useful to implement server software? Or too restrictive? https://t.co/qRBK1AOA7j
E2fsprogs 1.46.0 is out:
"[…] now supports the fast_commit […] which improves the latency of the fsync(2) system call […]
now supports the stable_inodes […]
both file system
encryption and the casefold feature enabled […]
https://t.co/lSESuIxddQ #linux#kernel#ext4