Re: [PATCH v3 4/8] x86/mem-sharing: copy GADDR based shared guest areas

[Tamas K Lengyel <tamas@xxxxxxxxxxxxx>]

> @@ -1974,7 +2046,10 @@ int mem_sharing_fork_reset(struct domain
>
>   state:
>      if ( reset_state )
> +    {
>          rc = copy_settings(d, pd);
> +        /* TBD: What to do here with -ERESTART? */

Ideally we could avoid hitting code-paths that are restartable during fork reset since it gets called from vm_event replies that have no concept of handling errors. If we start having errors like this we would just have to drop the vm_event reply optimization and issue a standalone fork reset hypercall every time which isn't a big deal, it's just slower. My preference would actually be that after the initial forking is performed a local copy of the parent's state is maintained for the fork to reset to so there would be no case of hitting an error like this. It would also allow us in the future to unpause the parent for example..

Tamas