Re: [PATCH x3] hurd: pci arbiter bootstrap

[Damien Zammit]

On 3/2/21 8:53 am, Samuel Thibault wrote:
> Damien Zammit, le mar. 02 févr. 2021 18:29:42 +1100, a ecrit:
>> This is a work in progress, I would like some comments on how I can
>> make the arbiter start up without a / filesystem when it needs to
>> expose the pci interface netfs on a real node.
> 
> That should already be handled by libmachdev's machdev_trivfs_init?  The
> disk translator, in the meanwhile, can device_open("pci") on the device
> master port that it got from pci-arbiter.

I'm not sure what you mean, in pci-arbiter's main we have:

  /* Start the PCI system: NB: pciaccess will choose x86 first and take lock */
  err = pci_system_init ();
  if (err)
    error (1, err, "Starting the PCI system");

  /* Create the PCI filesystem */
  err = init_file_system (netfs_startup (bootstrap, O_READ), fs);

After machdev_trivfs_init is called, bootstrap is set to MACH_PORT_NULL.
So, how is netfs_startup supposed to get a non-null bootstrap port and what 
port should this be?
Do we need to modify machdev_trivfs_init to return some kind of port back to 
"bootstrap"?

Cheers,
Damien