[Bug 201245] panic: spin lock held too long on sparc64

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=201245

--- Comment #8 from Marius Strobl <marius at FreeBSD.org> ---
You may give the following patch a try:
https://people.freebsd.org/~marius/sparc64_ipi_mtx.diff

Note that the issue at hand here is that the hardware architectures of v210,
v240 and v440 are flawed in that they don't include a globally accessible time
counter, just per-cores ones which aren't synchronized and therefore drift.
Given how time keeping in FreeBSD works and although a few ways for working
around that situation exist, all of these approaches come with some not really
desirable side-effects and most are insane implementation-wise. Some changes in
MI code over time also revealed some bugs in the current path taken in that
they made race-conditions more likely to be hit when running on one of these
models. That's also why subtle changes e. g. in network settings, differences
of the generated code caused by commits in unrelated kernel parts etc. may
avoid running into the panic encountered, but in fact these are red herrings.

-- 
You are receiving this mail because:
You are the assignee for the bug.