LWN: Comments on "Restartable sequences restarted"

Restartable sequences restarted

bpearlmutter — Thu, 19 Oct 2023 11:23:42 +0000

I'd be concerned that this kind of mechanism can be used to get inappropriate access to data, by monitoring the number of restarts while setting things up so data you shouldn't have access to alters the probability of getting restarted.

Restartable sequences restarted... with "frightening code"

nix — Wed, 31 Aug 2016 19:41:23 +0000

This is not kernel code. This is *userspace* code. i.e. this is something the *users* of this feature have to do (or, more likely, use this very code, migrated to a proper library, to do).

re-registration

corbet — Mon, 29 Aug 2016 09:50:43 +0000

I guess I didn't make that clear enough. The point is to let unrelated libraries all use this facility without stepping on each other's toes. The kernel can only handle one such structure per thread, but, by allowing nested registration and through the use of the __rseq_abi trick, independent users can all get along.

re-registration

lkurusa — Mon, 29 Aug 2016 03:00:02 +0000

What is the rationale behind letting people re-register their struct rseq? Isn't this something akin to recursive locks?

Restartable sequences restarted... with "frightening code"

pr1268 — Thu, 25 Aug 2016 02:55:07 +0000

the rather frightening assembly-in-CPP code found here and here.

When our esteemed editor starts describing prospective kernel code as "frightening", then I think it's time to really be scared.

In other words, don't try the self-tests without bracing yourself for possible blindness when browsing the code. ;-)

Regardless, this looks like a useful feature. Hopefully it gains enough traction to make it in...