On Thu, 2024-03-07 at 09:27 +0000, David Woodhouse wrote: > > How about this? It's fugly as hell, it puts PREEMPT_RT knowledge into > the code to make the local trylock function *conditionally* disable > interrupts. I hate it, but let's start with is it even *correct*? Oh but wait... The only reason we use read_lock_irqsave() in the first place is because there was code which runs in interrupt context which takes the same lock. But Paul's patch is *changing* that, so the in-interrupt code will now use read_trylock() instead, and can't deadlock. It always have to have a fallback to a slow path anyway, for when the cache isn't valid. So I think we can just drop the irqsave from *all* use of the locks in question. More coffee required...
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
