Hello
tech-kern,
I am Shivam Panwar, a fourth-year Electrical Engineering student (Dual Degree) at IIT Bombay. I am interested in the GSoC 2026 project, "Real Asynchronous I/O".
I have been working on systems programming for the past couple of years, primarily in C, C++, and Rust. The most relevant thing I have built is a deterministic record-and-replay runtime for multithreaded C programs that uses io_uring as its async I/O backend, the core problem it solves (blocking calls destroying concurrency) is the same one this project addresses at the kernel level.
NetBSD setup:
I have installed NetBSD 10.1 on QEMU, downloaded the current NetBSD source tree, and built and installed a custom kernel using build.sh. I have read through sys/kern/sys_aio.c, ufs/ffs/ffs_vnops.c, the struct buf definition and biodone/biowait pair, the UVM vslock interface, and the softint(9) API to understand the current service pool implementation and where the blocking actually happens.
I have already reached out to
Taylor Campbell directly with a detailed proposal covering the full LWP-less callback architecture I am planning to implement. I am also sending this to tech-kern in case anyone else has thoughts or pointers to known failure cases in the current AIO implementation.
Thanks,
Shivam Panwar
IIT Bombay