patm, idt, ipfw - next adentures

[Harti Brandt]

On Fri, 3 Oct 2003, Vincent Jardin wrote:

VJ>Hi,
VJ>
VJ>I do not see the following error message that you had yesterday ?
VJ>
VJ>Oct  2 03:18:45 ordos /kernel: proatm0: i=101, status=00080008
VJ>Oct  2 03:18:45 ordos /kernel: proatm0: not enough large buffers.
VJ>
VJ>Are you using exactly the same configuration ? Or how long do you have to wait
VJ>before getting the "not enough large buffers" error ?
VJ>
VJ>Regards,
VJ>  Vincent
VJ>
VJ>
VJ>> This is without ipfw (I removed ipfw from kernel).
VJ>
VJ>> proatm0: <PROSUM PROATM adapter> port 0xd400-0xd4ff mem
VJ>> 0xfb000000-0xfb3fffff,0xf8800000-0xf8800fff
VJ>> proatm0: ATM card is PROATM-155 or IDT, PHY=PM5346 or IDT77155
VJ>> proatm0: 128K words of RAM up to 4096 PVC
VJ>> proatm0: MAC address 00:c0:fd:00:7b:a0, HWrev=6
VJ>[...]
VJ>> proatm0: proatm_openvcc: 0/141, PCR=15000, SCR=15000
VJ>> proatm0: txopen UBR connection for 0/142
VJ>> proatm0: proatm_openvcc: 0/142, PCR=15000, SCR=15000
VJ>> proatm0: txopen UBR connection for 0/143
VJ>[...]
VJ>> proatm0 BUFFER STATUS: small=277/510, large=304/510.
VJ>> proatm0 BUFFER STATUS: small=256/510, large=283/510.

This suggests that you have increased bufsmall and buflarge to 510,
correct? In this case I think the resupply algorithm will leak mbufs. I
think the leak shouldn't happen when you set both variables not larger
than 278.

 VJ>
VJ>Oct  2 03:18:45 ordos /kernel: proatm0: i=101, status=00080008
VJ>Oct  2 03:18:45 ordos /kernel: proatm0: not enough large buffers.

As I said in another mail, this may happen if you're assembling PDUs on a
lot of VCs simultaneously (when the number of VCs approaches the limit on
free buffers (or half that number?)).

harti
-- 
harti brandt,
http://www.fokus.fraunhofer.de/research/cc/cats/employees/hartmut.brandt/private
brandt at fokus.fraunhofer.de, harti at freebsd.org