> Here's a new diff taking a different approach. Keeping tracing off until > all secondary CPUs have booted solves the issue of accessing curcpu() > too early. Another issue was then discovered, curproc can be NULL before > the idle thread tied the current CPU has started. Currently running with > this diff applied on my laptop (MP) and positive results from Greg. The > diff will be further exercised in the actual syzkaller setup before > committing.
Thanks Anton. This diff is running now on https://syzkaller.appspot.com/#openbsd as openbsd/ci-openbsd-multicore. Looking great so far.
