Hi Claude
Thank you very much for the details.
They are much appreciated.
I'll be taking a look at the 5272 code I have and see if we cant improve
that too.
Oh that everyone would explain their patches so well.
Best Regards
Phil WIlshire
Claude wrote:
Hi again,
Marco Cavallini wrote:
>Hi Claude,
>I had similar problems on a MCF5282, which seems being solved with
>modifications to cache setting code, I'd like to try your patch as soon
>as I have some spare time (I hope this week).
>Feel free to contact me privately to discuss your patch and other things
>I've done to work this board properly.
I had already disabled the CACHE completely but FEC problem persisted.
In order to let people having older driver versions I'd like to make a
summary of the changes here:
First, since FEC driver only processes buffers holding entire packets,
it is not necessary to enable
interrupts for RXB and TXB events (rx buffer and tx buffer), but only
for RXF and TXF events (rx and tx
frames). This may produce unnecessary interrupts to CPU. Interrupts are
enabled in
fec_enet_init() and fec_restart() functions. If you take a look to the
patch, only necessary int's are enabled:
fecp->fec_imask = (FEC_ENET_TXF | /*FEC_ENET_TXB |*/
FEC_ENET_RXF /*| FEC_ENET_RXB */ | FEC_ENET_MII);
Commented lines have been left for convenience. They should be removed,
however, to clean up code.
On the other hand, other changes affected the interrupt service routine.
In my opinion, it's fine that, in order to process
all packets waiting in the queue, the ring should be entirely processed
when an interrupt is raised when receiving a packet
(see the loop in fec_enet_rx(), which is called from the ISR). But I
don't agree with making a loop in the ISR to handle all
different events (tx, rx and mii events) that are waiting, since they
should be acknowledge in a different manner.
The old FEC driver does the following, in the ISR:
while ("pending interrupts"){
acknowledge all interrupts
if RX event, then process RX event
if TX event, then process TX event
if MII event, then process MII event
}
The fixed code (the patch's one) does this, instead:
if ("pending interrupts") {
if RX event, then process RX event and then acknowledge RX event
(clearing the corresponding flag)
if TX event, then process TX event and then acknowledge TX event
(clearing the corresponding flag)
if MII event, then process MII event and then acknowledge MII event
(clearing the corresponding flag)
}
I saw that the first solution produced, after some time, a lost of
synchronization when processing the ring. It seems that
acknowledging the RX event before processing the ring makes the FEC to
allow entering more packets till produce a ring overflow.
Finally, the "pending interrupts" condition above, was implemented as
follows (in the old FEC driver):
((int_events = fecp->ievent) != 0)
The fixed code does:
((int_events = (fecp->ievent & (FEC_ENET_RXF | FEC_ENET_TXF |
FEC_ENET_MII))) != 0)
This avoids the ISR (after receiving a RX,TX or MII event) to keep
running if other flag in the fecp->ievent register is raised for
any reason (error flag, for example). However, the replacement of
"while" for "if" implicitly avoids it.
Well, I hope this solution could help someone. At least it has helped me!!
Claude
_______________________________________________
uClinux-dev mailing list
[email protected]
http://mailman.uclinux.org/mailman/listinfo/uclinux-dev
This message was resent by [email protected]
To unsubscribe see:
http://mailman.uclinux.org/mailman/options/uclinux-dev
_______________________________________________
uClinux-dev mailing list
[email protected]
http://mailman.uclinux.org/mailman/listinfo/uclinux-dev
This message was resent by [email protected]
To unsubscribe see:
http://mailman.uclinux.org/mailman/options/uclinux-dev
