Re: b44: high ping times with wireless-dev

On Sunday 17 June 2007, Stephen Hemminger wrote:

On Sat, 16 Jun 2007 23:27:43 +0200

Maximilian Engelhardt [off-list ref] wrote:

quoted

Hello,

I recently did some test and found out something interesting about the
b44 problem I wrote earlier.

The problem is the following:
When I use my BCM4401 with the b44 driver in wireless-dev I get very high
ping times looking like this:

64 bytes from 172.30.10.1: icmp_seq=1 ttl=64 time=1863 ms
64 bytes from 172.30.10.1: icmp_seq=2 ttl=64 time=855 ms
64 bytes from 172.30.10.1: icmp_seq=3 ttl=64 time=1855 ms
64 bytes from 172.30.10.1: icmp_seq=4 ttl=64 time=855 ms
64 bytes from 172.30.10.1: icmp_seq=5 ttl=64 time=1854 ms
64 bytes from 172.30.10.1: icmp_seq=6 ttl=64 time=854 ms
64 bytes from 172.30.10.1: icmp_seq=7 ttl=64 time=1851 ms
64 bytes from 172.30.10.1: icmp_seq=8 ttl=64 time=851 ms
64 bytes from 172.30.10.1: icmp_seq=9 ttl=64 time=1851 ms
64 bytes from 172.30.10.1: icmp_seq=10 ttl=64 time=851 ms

I also found out that shortly after I boot my laptop and log into kde
ping times are not that high but start to increase very quickly:

64 bytes from 172.30.10.1: icmp_seq=53 ttl=64 time=2.19 ms
64 bytes from 172.30.10.1: icmp_seq=54 ttl=64 time=2.22 ms
64 bytes from 172.30.10.1: icmp_seq=55 ttl=64 time=2.20 ms
64 bytes from 172.30.10.1: icmp_seq=56 ttl=64 time=2.20 ms
64 bytes from 172.30.10.1: icmp_seq=57 ttl=64 time=18.6 ms
64 bytes from 172.30.10.1: icmp_seq=58 ttl=64 time=1268 ms
64 bytes from 172.30.10.1: icmp_seq=59 ttl=64 time=268 ms
64 bytes from 172.30.10.1: icmp_seq=60 ttl=64 time=1268 ms
64 bytes from 172.30.10.1: icmp_seq=61 ttl=64 time=268 ms
64 bytes from 172.30.10.1: icmp_seq=62 ttl=64 time=6.08 ms
64 bytes from 172.30.10.1: icmp_seq=63 ttl=64 time=268 ms
64 bytes from 172.30.10.1: icmp_seq=64 ttl=64 time=1264 ms
64 bytes from 172.30.10.1: icmp_seq=65 ttl=64 time=264 ms

After some time digging around I found out something really interesting.
When I play some music ping times are immediately lower. If I stop
playing music they are back to the same times as they were before.

I guess that there is a problem with interrupts so I post some
information of my system in hope it will be usefull.

maxi@koala:~$ cat /proc/interrupts
          CPU0
 0:     126317    XT-PIC-XT        timer
 1:       3600    XT-PIC-XT        i8042
 2:          0    XT-PIC-XT        cascade
 7:          1    XT-PIC-XT        parport0
 8:          1    XT-PIC-XT        rtc
 9:      17371    XT-PIC-XT        acpi
10:      13237    XT-PIC-XT        firewire_ohci, yenta, yenta,
ehci_hcd:usb1, uhci_hcd:usb3, uhci_hcd:usb4, Intel 82801DB-ICH4, Intel
82801DB-ICH4 Modem, eth0
11:      89059    XT-PIC-XT        uhci_hcd:usb2, i915@pci:0000:00:02.0
12:        632    XT-PIC-XT        i8042
14:      10354    XT-PIC-XT        libata
15:       7408    XT-PIC-XT        libata
NMI:          0
ERR:          0


[...]
ACPI: PCI Interrupt Link [LNKD] enabled at IRQ 10
ACPI: PCI Interrupt 0000:02:02.0[A] -> Link [LNKD] -> GSI 10 (level, low)
-> IRQ 10
ssb: Sonics Silicon Backplane found on PCI device 0000:02:02.0
b44.c:v2.0
eth0: Broadcom 44xx/47xx 10/100BaseT Ethernet 00:c0:9f:29:99:a7
[...]

This problem did only happen with wireless-dev (checkout this evening)
and with -mm kernels I used some time ago for testing. Currently I'm
running 2.6.22-rc4 that works perfectly fine and doesn't show that
problem.

Maxi

Can you build with APIC for uniprocessor.

I did enable CONFIG_X86_UP_APIC and CONFIG_X86_UP_IOAPIC and tried with lapic 
and apic=force but couldn't get APIC working.

There is lots of IRQ sharing, so
 - one of the other device's may be not handling shared IRQ properly.
   Try unloading firewhire modem and yenta devices.

 - IRQ might be set edge triggered which doesn't work with NAPI
	or shared IRQ.

I did build a kernel without the three mentioned above but the problem is 
still the same. I also did remove everything but eth0 on interrupt 10 so the 
only device using that interrupt is eth0 and then the card completely stopped 
working.

Maxi