Re: [PATCH] machdev,pci,rump: Rework bootstrap

[Samuel Thibault]

We're getting close!

Damien Zammit, le dim. 14 mars 2021 22:23:25 +1100, a ecrit:
> except for a strange anomaly with:
> 
> cat /proc/5/stat /proc/6/stat

Possibly this is a missing proc_child call in startup's
launch_core_servers. I'd say that instead of explicit proc_child calls,
launch_core_servers should just go over the list of essential tasks.

That being said, proc and procfs should be resilient to missing that
proc_child call.

> @@ -278,16 +311,28 @@ trivfs_S_fsys_init (struct trivfs_control *fsys,
>    /* Install the bootstrap port on /dev/something so users
>     * can still access the bootstrapped device */
> -  if (bootstrapped && devnode)
> +  if (bootstrapping && devnode)
>      {
>        right = ports_get_send_right (&control->pi);
>        install_as_translator (right);
>        control->underlying = underlying;
> +      essential_task ();

I'd say split the if: we want to call essential_task even if !devnode.

>  int
>  machdev_trivfs_init(mach_port_t bootstrap_resume_task, const char *name, 
> const char *path, mach_port_t *bootstrap)
>  {
> +  mach_port_t mybootstrap = MACH_PORT_NULL;
> +  task_t parent_task;
>    port_bucket = ports_create_bucket ();
>    trivfs_cntl_class = ports_create_class (trivfs_clean_cntl, 0);
>    trivfs_protid_class = ports_create_class (trivfs_clean_protid, 0);
>    trivfs_create_control (MACH_PORT_NULL, trivfs_cntl_class, port_bucket,
>                           trivfs_protid_class, 0, &control);
>  
> +  *bootstrap = MACH_PORT_NULL;
> +
> +  task_get_bootstrap_port (mach_task_self (), &mybootstrap);
> +  if (mybootstrap)
> +    {
> +      *bootstrap = mybootstrap;
> +      fsys_getpriv (*bootstrap, &_hurd_host_priv, &_hurd_device_master, 
> &parent_task);
> +    }
> +
>    if (bootstrap_resume_task != MACH_PORT_NULL)
>      {
>        if (path)
>       devnode = strdup(path);
>        resume_bootstrap_server(bootstrap_resume_task, name);
> -      *bootstrap = ports_get_send_right (&control->pi);
> +      if (*bootstrap == MACH_PORT_NULL)
> +        *bootstrap = ports_get_send_right (&control->pi);

I don't think this makes sense: either we have a NULL bootstrap, or
we have one which is not us, but not this: why a control port on
ourself?  Does it actually pose any problem that bootstrap is returned
MACH_PORT_NULL? That'd probably what needs to be fixed rather that
papering over it here.

Samuel