Re: Patch for i/o permission control in GNU Mach (was: I/O permission control in OSKit-Mach)

[Samuel Thibault]

Hi,

Thomas Schwinge, le Wed 25 Apr 2007 18:45:44 +0200, a écrit :
> diff -N i386/i386/io_perm.c
> --- /dev/null   1 Jan 1970 00:00:00 -0000
> +++ i386/i386/io_perm.c 25 Apr 2007 16:36:32 -0000
...
> +kern_return_t
> +i386_io_perm_create (ipc_port_t master_port, io_port_t from, io_port_t to,
> +                  io_perm_t *new)
> +{
> +  /* TODO.  Check for limits [0;IOPB_MAX)?  */

Yes and no. io_port_t is an integer 16, and ports are from 0 to 0xffff
too.  I don't know precisely how signedness is handled here, but it
should be.

> +      else if (iopb)
> +     {
> +       target_task->machine.iopb = iopb;
> +       io_bitmap_init (iopb);
> +     }
> +      else
> +     /* TODO.  Need some unlocking here?  */
> +     return KERN_RESOURCE_SHORTAGE;

Yes.

> diff -N i386/i386/io_perm.h
> --- /dev/null   1 Jan 1970 00:00:00 -0000
> +++ i386/i386/io_perm.h 25 Apr 2007 16:36:32 -0000
...
> +/* The highest possible I/O port.  ISA bus allows ports 0..3ff, but
> +   accelerator cards are funky.  */
> +#define      IOPB_MAX        0xFFFF  

And we need PCI too nowadays.

> Index: i386/i386/ktss.h
> ===================================================================
> RCS file: /cvsroot/hurd/gnumach/i386/i386/Attic/ktss.h,v
> retrieving revision 1.2
> diff -u -p -r1.2 ktss.h
> --- i386/i386/ktss.h  5 Apr 2001 06:39:20 -0000       1.2
> +++ i386/i386/ktss.h  25 Apr 2007 16:36:32 -0000
...
> +/* Try to release as much memory from the machine specific data in
> +   task TASK.  */
> +void
> +machine_task_collect (task_t task)
> +{
> +  if (task->machine.iopb_size == 0 && task->machine.iopb)
> +    {
> +      zfree (machine_task_iopb_zone, (vm_offset_t) task->machine.iopb);
> +      task->machine.iopb = 0;
> +    }
> +}

Isn't there a lock needed here?

Else the grounds seem fine.  Did you already test it on real hardware?

Samuel