Re: [PATCH] Cast kernel server call arguments to the correct type when mach_port_t and ipc_port_t are used interchangeably

[Flavio Cruz]

On Thu, Dec 31, 2015 at 03:11:44PM +0100, Justus Winter wrote:
> Quoting Flavio Cruz (2015-12-30 18:03:58)
> > This fixes a warning when compiling the mig generated file
> > gnumach/kern/mach4.server.c. Here, an ipc_port_t* is passed to a
> > function that expects a mach_port_t* (In0P->object). Since mig uses
> > mach_port_t and ipc_port_t interchangeably in kernel interfaces, we need
> > to use casts to make GCC happy.
> 
> Are you thinking of this issue:
> 
>   kern/mach4.server.c: In function ‘_Xmemory_object_create_proxy’:
>   kern/mach4.server.c:691:73: warning: passing argument 3 of 
> ‘memory_object_create_proxy’ from incompatible pointer type 
> [-Wincompatible-pointer-types]
>   OutP->RetCode = memory_object_create_proxy(task, In0P->max_protection, 
> In0P->object, In0P->objectType.msgt_number, In1P->offset, 
> In1P->offsetType.msgt_number, In2P->start, In2P->startType.msgt_number, 
> In3P->len, In3P->lenType.msgt_number, &OutP->proxy);
>   kern/mach4.server.c:585:29: note: expected ‘memory_object_array_t {aka long 
> unsigned int *}’ but argument is of type ‘struct ipc_port **’
>   mig_external kern_return_t memory_object_create_proxy
> 
> ?

Yes.

> 
> I've been wondering about that, but didn't took a closer look.  I
> always assumed that this is due to MatchMakers poor type support.

In this case it's related to the way mig handles ipc_port_t objects and
mach_port_t names internally.

> 
> Cheers,
> Justus

Cheers
Flavio