Hi,
here's a patch that reduces the execution time for "unslipping" the
received data considerably. The patch does NOT support packets with CRC!
This would require a little bit of additional work, I think that it
would be better to rewrite the crc routine to calculate the whole packet
checksum at once for better performance. I have not done that because
the CBT100C doesn't use crc's.
Please ignore the __fastram attribute, it's for uClinux on my specific
board only. Functions with this attribute are placed into the cpu's
internal 32-bit ram for faster execution.

Regards,
Erwin

Am Don, den 28.10.2004 schrieb Marcel Holtmann um 13:47:
> Hi Erwin,
> 
> > > > I have cross-compiled bluez for uClinux-2.4.20 for ARM7, with the bluez
> > > > kernel patch applied, and bluez-utils 2.10. I'm using the Conceptronic
> > > > CBT100C PCMCIA card on the system. The card is initialized with
> > > > "hciattach ttyS2 bcsp". Basically, the system works fine, but I see
> > > > occasional overrun errors on the UART, maybe because the ARM7 is running
> > > > with 40MHz only. When I transmit larger packets (l2ping -s 300 ...), the
> > > > number of overrun errors increases, and sometimes the system locks up.
> > > > I'm getting an endless sequence of: 
> > > > 
> > > > bcsp_pkt_cull: Peer acked invalid packet
> > > > bcsp_handle_le_pkt: Found a LE sync pkt, card has reset
> > > > bcsp_pkt_cull: Peer acked invalid packet
> > > > bcsp_handle_le_pkt: Found a LE sync pkt, card has reset
> > > > bcsp_pkt_cull: Peer acked invalid packet
> > > > ...
> > > 
> > > I am not a BCSP or uClinux expert so I think that I can't help you here,
> > > but from my testing with the BCSP part of the hci_uart driver I never
> > > saw any problems like this.
> > > 
> > After a lot of debugging, I found out that the serial irq routine calls
> > tty_flip_buffer_push with the serial interrupt disabled. The processing
> > of the input buffer in bcsp_recv (hci_bcsp.c) was too slow so that the
> > characters that were buffered in the FIFO during that time caused an
> > overrun (There is no RTS/CTS flowcontrol). I have done modifiations to
> > hci_bcsp.c so that the processing is done much faster, and the problem
> > seems to have gone. 
> 
> please provide us a patch with your modifications.
> 
> Regards
> 
> Marcel