Re: data structures for use in signal handlers

[Ben Pfaff]

On Tue, Feb 9, 2021 at 5:01 PM Bruno Haible <bruno@clisp.org> wrote:
>
> Ben Pfaff wrote:
> > Building an RCU implementation isn't necessarily difficult (I have done it,
> > but the implementation isn't suitable for gnulib).
> >
> > There is a liburcu that is under LGPL v2.1: https://liburcu.org/
>
> Thanks for the pointers, Ben. Yes, RCU is the technique I had in mind.
> So: "can it actually be made to work?" -> Yes.
>
> But the implementation of liburcu uses extra signals and a helper thread of
> its own (or even a helper thread per CPU).
>
> I have to rephrase the question: can it be made to work in a straightforward
> (not necessarily scalability-optimized) way?

Signals are not needed. I haven't studied liburcu (it wasn't suitable for my
userspace use case because it only worked with GCC), so I am not sure
why it needs signals.

The userspace RCU implementation that I built in OVS (which is free but
Apache licensed) does use a thread, but not signals. The nice thing about
using a thread is that you always have something freeing memory that is
known to no longer be in use.

But I can imagine implementations that don't even use an extra thread.
For example, each thread could keep its own local list of callbacks to be
called after the next grace period. Then, whenever a thread quiesces, it
goes through its own list and executes all of the callbacks that now can
be. It could take a long time for some callbacks to be called, though,
especially if a thread goes to sleep and thus won't call any of its own
until it wakes up again.