Re: [PATCH] hurd: Implement device memory mapping

[Samuel Thibault]

Samuel Thibault, le dim. 09 janv. 2022 01:02:57 +0100, a ecrit:
> Joan Lledó via Bug reports for the GNU Hurd, le mer. 05 janv. 2022 13:08:01 
> +0100, a ecrit:
> > @@ -523,18 +637,18 @@ pci_system_hurd_create(void)
> >  
> >      pci_sys->num_devices = 0;
> >  
> > -    if ((err = get_privileged_ports (NULL, &device_master)) || 
> > (device_master == MACH_PORT_NULL)) {
> > -        pci_system_cleanup();
> > -        return err;
> > -    }
> > -
> > -    err = device_open (device_master, D_READ|D_WRITE, "pci", &pci_port);
> > -    if (!err) {
> > -        root = file_name_lookup_under (pci_port, ".", O_DIRECTORY | O_RDWR 
> > | O_EXEC, 0);
> > -    }
> > -
> > -    if (!root) {
> > -        root = file_name_lookup (_SERVERS_BUS_PCI, O_RDWR, 0);
> > +    if ((err = get_privileged_ports (NULL, &device_master))
> > +            || (device_master == MACH_PORT_NULL)) {
> > +        root = file_name_lookup (_SERVERS_BUS_PCI, O_RDONLY, 0);
> > +    } else {
> > +        err = device_open (device_master, D_READ, "pci", &pci_port);
> > +        mach_port_deallocate (mach_task_self (), device_master);
> > +        if (!err) {
> > +            root = file_name_lookup_under (pci_port, ".",
> > +                                           O_DIRECTORY | O_RDONLY | 
> > O_EXEC, 0);
> > +            device_close (pci_port);
> > +            mach_port_deallocate (mach_task_self (), pci_port);
> > +        }
> 
> Err, this is breaking everything when pci-arbiter is not running as a
> bootstrap translator. In that case e.g. netdde succeeds getting the
> privileged port, and thus tries to open pci, but fails, and thus the
> whole thing aborts. What problem the older approach had?

I uploaded a version with that ordering fixed back to debian-ports, to
be seen on mirrors within about 6h.

Samuel