hostap interrupt problems, maintainers unresponsive - "wifi0: interrupt delivery does not seem to work"

hostap interrupt problems, maintainers unresponsive - "wifi0: interrupt delivery does not seem to work"

[Chip Salzenberg]

I've sent this problem description to the upstream maintainers three
times and received no reply, let alone illumination.  Thus:

Hostap 0.3.9 under kernel 2.6.12 works fine with the D-Link DWL-650.

Hostap 0.4.4-kernel, included with kernel 2.6.14, does not work; nor
do versions 0.3.9 nor 0.4.1 compiled separately against 2.6.14.  There
seems to be a problem with interrupt delivery.  Soon after the module
is installed, keystrokes and all other interrupt-driven activity pause
periodically for a LONG time (on the order of five seconds).
Keystrokes will often auto-repeat because the key-raise interrupt gets
delayed and the kernel thinks the key is still down.

Here's a log excerpt:

Oct 29 16:34:11 tytlal kernel: usb 2-1: new full speed USB device using uhci_hcd and address 3
Oct 29 16:34:11 tytlal kernel: usb 2-1: configuration #1 chosen from 2 choices
Oct 29 16:34:11 tytlal kernel: cdc_acm 2-1:1.0: ttyACM0: USB ACM device
Oct 29 16:34:23 tytlal kernel: hostap_cs: 0.4.4-kernel (Jouni Malinen <jkmaline@cc.hut.fi>)
Oct 29 16:34:23 tytlal kernel: hostap_cs: Registered netdevice wifi0
Oct 29 16:34:23 tytlal kernel: hostap_cs: index 0x01: Vcc 3.3, irq 3, io 0x4100-0x413f
Oct 29 16:34:23 tytlal kernel: wifi0: NIC: id=0x800c v1.0.0
Oct 29 16:34:23 tytlal kernel: wifi0: PRI: id=0x15 v1.1.1
Oct 29 16:34:23 tytlal kernel: wifi0: STA: id=0x1f v1.8.4
Oct 29 16:34:26 tytlal kernel: wifi0: Interrupt, but SWSUPPORT0 does not match: 0000 != 8A32 - card removed?
Oct 29 16:34:26 tytlal kernel: hostap_cs: wifi0: resetting card
Oct 29 16:34:42 tytlal pppd[6511]: tcsetattr: No such device or address (line 1010)
Oct 29 16:34:42 tytlal pppd[6511]: Exit.
Oct 29 16:35:57 tytlal kernel: wifi0: interrupt delivery does not seem to work
Oct 29 16:35:57 tytlal kernel: wifi0: cnfAuthentication setting to 0x1 failed
Oct 29 16:37:03 tytlal kernel: hostap_cs: Registered netdevice wifi0
Oct 29 16:37:03 tytlal kernel: hostap_cs: index 0x01: Vcc 3.3, irq 3, io 0x4100-0x413f
Oct 29 16:37:03 tytlal kernel: wifi0: NIC: id=0x800c v1.0.0
Oct 29 16:37:03 tytlal kernel: wifi0: PRI: id=0x15 v1.1.1
Oct 29 16:37:03 tytlal kernel: wifi0: STA: id=0x1f v1.8.4
Oct 29 16:37:05 tytlal kernel: wifi0: Interrupt, but SWSUPPORT0 does not match: 0000 != 8A32 - card removed?
Oct 29 16:37:05 tytlal kernel: hostap_cs: wifi0: resetting card
Oct 29 16:37:10 tytlal kernel: hostap_cs: wifi0: resetting card
Oct 29 16:37:16 tytlal kernel: wifi0: Interrupt, but SWSUPPORT0 does not match: 0000 != 8A32 - card removed?
Oct 29 16:37:16 tytlal kernel: hostap_cs: wifi0: resetting card

-- 
Chip Salzenberg <chip@pobox.com>

Re: hostap interrupt problems, maintainers unresponsive - "wifi0: interrupt delivery does not seem to work"

[Jouni Malinen]

On Wed, Nov 02, 2005 at 09:46:39AM -0800, Chip Salzenberg wrote:

> I've sent this problem description to the upstream maintainers three
> times and received no reply, let alone illumination.  Thus:

You may need to re-adjust your believes that everyone is working 24/7
just waiting for your questions and providing immediate solutions for
whatever you may come up with. Let's just say that some of your messages
on the Host AP mailing list were not exactly best way of getting people
interested in using any time to help you.

I happen to work on Host AP driver mainly during weekends and you left
gracious Mon-Thu (or well, Mon-Tue based on this message) time to react
before starting to complain loudly that no one is helping with your
problem. Actually, I did answer your message as soon as I read it on
Friday.

> Hostap 0.3.9 under kernel 2.6.12 works fine with the D-Link DWL-650.

Can you test whether the current development version (0.4.6) works with
2.6.12? What about 2.6.13?

> Hostap 0.4.4-kernel, included with kernel 2.6.14, does not work; nor
> do versions 0.3.9 nor 0.4.1 compiled separately against 2.6.14.  There
> seems to be a problem with interrupt delivery.  Soon after the module
> is installed, keystrokes and all other interrupt-driven activity pause
> periodically for a LONG time (on the order of five seconds).
> Keystrokes will often auto-repeat because the key-raise interrupt gets
> delayed and the kernel thinks the key is still down.

I cannot reproduce this in my test setup. Linux 2.6.14 with the Host AP
driver works on my IBM ThinkPad T43 with hostap_cs and the same station
firmware version that you seem to be using (1.8.4). Did you need to do
something with the card before seeing this issue apart from plugging it
in? I was able to associate with WPA without seeing any problems.

Are you using the new hotplug mechanism to initialize PC Card modules?
Can you please send full dmesg output including everything from the
kernel startup to the point when hostap_cs starts reporting problems?

-- 
Jouni Malinen                                            PGP id EFC895FA

Re: hostap interrupt problems, maintainers unresponsive - "wifi0: interrupt delivery does not seem to work"

[Jan Harkes]

On Wed, Nov 02, 2005 at 09:46:39AM -0800, Chip Salzenberg wrote:
> Hostap 0.4.4-kernel, included with kernel 2.6.14, does not work; nor
> do versions 0.3.9 nor 0.4.1 compiled separately against 2.6.14.  There
> seems to be a problem with interrupt delivery.  Soon after the module
> is installed, keystrokes and all other interrupt-driven activity pause
> periodically for a LONG time (on the order of five seconds).

I am seeing similar interrupt problems, but I don't have hostap (or even
a wireless network card) on my machine. For me a clear indication is the
serial port overrun errors even though that port only gets a handful of
interrupts per second from an attached GPS receiver. Under X, keyboard
and mouse are occasionally very jerky and unreponsive.

I just rebooted with 'acpi=noirq', and I'm not seeing the serial port
overrun errors anymore, it seems to have fixed, or at least mitigated
the problem a bit. But I'm logged into the machine remotely at the
moment, so I can't be sure if it really fixed the unresponsive
keyboard/mouse issues. There are a bunch of differences in the kernel
log, but one that was quite noticable was the estimated CPU speed and
that the IO-APIC seems to be intialized differently.

booting 2.6.14
--------------
Nov 10 09:49:23 thegate kernel: 247 MHz processor.

booting 2.6.14 with acpi=noirq
------------------------------
Nov 10 12:36:08 thegate kernel: Detected 2807.302 MHz processor.


It actually looks like there is a whole bunch of interesting information
from the early boot that got scrolled out of the printk buffer by the
time klog dumps everything to kern.log. I guess I should rebuild my
kernel with a larger ringbuffer.

Jan

Re: hostap interrupt problems, maintainers unresponsive - "wifi0: interrupt delivery does not seem to work"

[Jan Harkes]

On Thu, Nov 10, 2005 at 12:59:22PM -0500, Jan Harkes wrote:
> I am seeing similar interrupt problems, but I don't have hostap (or even
> a wireless network card) on my machine. For me a clear indication is the
> serial port overrun errors even though that port only gets a handful of
> interrupts per second from an attached GPS receiver. Under X, keyboard
> and mouse are occasionally very jerky and unreponsive.
> 
> I just rebooted with 'acpi=noirq', and I'm not seeing the serial port
> overrun errors anymore, it seems to have fixed, or at least mitigated
> the problem a bit. But I'm logged into the machine remotely at the
> moment, so I can't be sure if it really fixed the unresponsive
> keyboard/mouse issues.

Interestingly, noirq didn't actually help. The main difference seems to
be that DMA is not automatically enabled for my drives with the 2.6.14
kernel, and I'm getting permission denied errors when trying to enable
it with hdparm.

Seems like piix is coming up too late and the ide device is already
claimed by a generic driver that doesn't do dma. I'm building ide/piix
into the kernel now to see if that helps.

Jan