Re: [PATCH v36 21/24] x86/vdso: Implement a vDSO for Intel SGX enclave call

[Jarkko Sakkinen]

On Wed, Aug 19, 2020 at 09:33:45AM -0400, Nathaniel McCallum wrote:
> On Tue, Aug 18, 2020 at 12:44 PM Jarkko Sakkinen
> <jarkko.sakkinen@xxxxxxxxxxxxxxx> wrote:
> >
> > On Tue, Aug 18, 2020 at 11:15:32AM -0400, Nathaniel McCallum wrote:
> > > That seems like overkill to me. I'm just asking for one additional mov
> > > instruction. :)
> >
> > I started to consider eBPF since the complexity and constraints of the
> > callback look like an overkill and without doubt will be a burden to
> > maintain.
> 
> That feels to me like more complexity just to move the existing
> complexity from one place to another.

My thinking was that there is like two parts on AEX handler:

A. You have a set of data and some application dependent logic.
   Right now this part is written in assembly most of the time.
B. You have code that decide how to continue based on that logic.

I was thinking that you could have the logic manipulating data (the
existing context structure we have can be passed directly to the BPF
state machine) patchable with a BPF program.

The logic would be executed in ring-0. vDSO would contain part B.

It is better to at least make some rational conclusions about this
because it seems to be the trend [1].

On the other hand the history with callbacks has not been a history
of victory as far as it comes to all the possible issues with them.
Like think about signals for example...

[1] https://lwn.net/Articles/813261/

/Jarkko