Re: [LPC] pkeys ABI

On Wed, 2026-06-17 at 14:57 +0200, Kevin Brodsky wrote:
> Hi,
> 
> I am considering submitting a Refereed presentation proposal for LPC,
> but I would first like to check if it makes sense to everyone in
> To/Cc.
> 
> The pkeys ABI is heavily underspecified when it comes to asynchronous
> kernel interactions, such as signal delivery [1], rseq [2], io_uring
> worker threads, etc. Depending on the situation, we might reset the
> pkey register to "allow all", or "allow acess to pkey 0 only", or use
> the interrupted context's register (even though it is arguably
> unrelated). It isn't always consistent across architectures either. 
> 
> As Thomas put it, "that's all broken". LPC feels like the right
> occasion to summarise where we are, and explore potential solutions.
> Unfortunately this topic isn't an obvious fit for any MC, so a
> Refereed presentation seems like the least bad option.

That's not necessarily true, especially if you want a group discussion.
If all of your attendees are going to plumbers anyway you might
consider a BoF ... or possibly a workshop like the Nova GPU people did
last year if you need more than an hour.  The downside of BoF or
Workshop is that you have to run your own A/V connections (we'll have
at least USB speaker/Mic's to provide you at reception and might have
one of the owl camera things)

Regards,

James