network communication

[Paolo Mantegazza]

Jan Kiszka wrote:
> 
> Alexander Eisenhuth schrieb:
> >
> >
> > Jan Kiszka wrote:
> >
> >> Alexander Eisenhuth schrieb:
> >> > ...
> >>
> >>>
> >>> RTNet
> >>> =====
> >>> I've read in the archive about RTNet.  It sound exactly what I need,
> >>> but unfortunatly the chipset I'm using is not supported.
> >>>
> >>> * Can anyone say something about the dependencies to the chipset ? At
> >>> a first look there seems to be no abstract approach to "soften" the
> >>> support of new Cards.
> >>>
> >>
> >> If you only want soft real-time networking with a RTAI kernel API, you
> >> could use the RTnet-wrapper for net_rpc. It comes with current RTAI
> >> versions.
> >
> >
> > What priorities are raised in that case ? Can you loose a few words on
> > net_rpc ?
> >
> 
> I would say (without knowing everything in details) that no priorities
> are raised. All incoming and outgoing udp data is treated as any other
> linux tcp/udp/ip packets. That means everything happening in the
> RT-domain has higher priority than your udp packets - even if the
> rt-priority of that tasks are lower than you sender or receiver. And in
> linux, there is no change of priorities at all.
> 
> About net_rpc: we haven't tried it for months here at the RTS (at the
> moment we do need it). If it already works with RTnet, you get a remote
> procedure call mechanism for most RTAI functions over ethernet. Maybe
> Paolo has already made some experiments with current RTnet and net_rpc...
> 

Not yet, but will soon when we have finished a general tool we'll make
available. I assume no difficulty in that as I saw that RTNet supports
the following, and only functions used by net_rpc:

int rt_socket(int domain, int type, int protocol);
int rt_close(int sock);
int rt_bind(int sock, struct sockaddr *addr, int addrlen);
int rt_socket_callback(int sock, int (*func)(int sock, void *arg), void
*arg);
int rt_sendto(int sock, const void *msg, int len, unsigned int sflags,
struct sockaddr *to, int tolen);
int rt_recvfrom(int sock, void *msg, int len, unsigned int flags, struct
sockaddr *from, int *fromlen);

with the following:

extern int rt_socket            (int family, int type, int protocol); 
extern int rt_socket_close      (int fd);
extern int rt_socket_bind       (int fd, struct sockaddr *my_addr, int
addr_len); 
extern int rt_socket_callback   (int fd, int (*func)(int,void *), void
*arg);
extern int rt_socket_sendto     (int fd, void *buf, int len, unsigned
int flags, struct sockaddr *to, int tolen);
extern int rt_socket_recvfrom   (int fd, void *buf, int len, unsigned
int flags, struct sockaddr *from, int *fromlen);  

through the following defines:

#define rt_bind         rt_socket_bind
#define rt_close        rt_socket_close
#define rt_sendto       rt_socket_sendto
#define rt_recvfrom     rt_socket_recvfrom.

Under a separate email I'm about to send a small tarball that shows
net_rpc at work with Dominic's code generator.

Stay tuned.

Paolo.