[Bluez-users] Multiple connections with kernel 2.4 make HCI die

[Bluez-users] Multiple connections with kernel 2.4 make HCI die

[Kevin Read]

Hello,

I know that kernel 2.4 is no longer supported, but I hope that somebody 
can give me a clue here...

I'm attempting to create a multi-threaded, multi-dongle OBEX client
appliance. When I run it on a kernel 2.6 based machine (either a x86 or
on a NSLU2, which is a big-endian ARM box), everything runs very smooth.

The same code compiled for a little endian MIPS machine running kernel
2.4.30 has problems. I cannot change the kernel version, because the
on-board Broadcom wireless chipset of the router doesn't really work yet
  under kernel 2.6 (damn closed source drivers) :(

When one bt interface does device discovery and another interface does
service discovery and the RFCOMM transfers, everything is fine. As soon
as data is sent/received over two connections simultaneously, the HCIs
begin to "die" - they don't respond to hciconfig reset/up/down commands
and all data transmissions time out, as do SDP connects. If I 'hciconfig
down' the interfaces, they don't come up again afterwards. The only
thing that helps is re-plugging the dongles. Re-loading hci-usb or the
uhci drivers is not enough, although all interfaces show as "up" again
afterwards. They still time out, though.

This hang might come after a few kb's of simultenous data was sent, or
after up to two megabyte have been transferred. It comes quicker when
two dongles transmit data simultaneously, but also happens when I have
two active RFCOMM connections on one dongle.

I have applied the changes in the 2.4.32 patch by Marcel, but I think
they only refer to module unloading. Bluez installed is version 2.32. I
already stressed the USB bus with wireless and a hard disk, to rule out
that the usb driver has some funkiness, but everything went smooth.

It would be great if somebody could help me here, I'm totally stuck. I
can give out shell access to the router, if that is of any help with
this problem.

Best regards,

Kevin



_______________________________________________
Bluez-users mailing list
Bluez-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-users

Re: [Bluez-users] Multiple connections with kernel 2.4 make HCI die

[Marcel Holtmann]

Hi Kevin,

> I know that kernel 2.4 is no longer supported, but I hope that somebody 
> can give me a clue here...
> 
> I'm attempting to create a multi-threaded, multi-dongle OBEX client
> appliance. When I run it on a kernel 2.6 based machine (either a x86 or
> on a NSLU2, which is a big-endian ARM box), everything runs very smooth.
> 
> The same code compiled for a little endian MIPS machine running kernel
> 2.4.30 has problems. I cannot change the kernel version, because the
> on-board Broadcom wireless chipset of the router doesn't really work yet
>   under kernel 2.6 (damn closed source drivers) :(

I am not buying this argument, because the bcm43xx open source driver
exists. Otherwise simply swap the mini-PCI card for an Intel one.

> When one bt interface does device discovery and another interface does
> service discovery and the RFCOMM transfers, everything is fine. As soon
> as data is sent/received over two connections simultaneously, the HCIs
> begin to "die" - they don't respond to hciconfig reset/up/down commands
> and all data transmissions time out, as do SDP connects. If I 'hciconfig
> down' the interfaces, they don't come up again afterwards. The only
> thing that helps is re-plugging the dongles. Re-loading hci-usb or the
> uhci drivers is not enough, although all interfaces show as "up" again
> afterwards. They still time out, though.
> 
> This hang might come after a few kb's of simultenous data was sent, or
> after up to two megabyte have been transferred. It comes quicker when
> two dongles transmit data simultaneously, but also happens when I have
> two active RFCOMM connections on one dongle.
> 
> I have applied the changes in the 2.4.32 patch by Marcel, but I think
> they only refer to module unloading. Bluez installed is version 2.32. I
> already stressed the USB bus with wireless and a hard disk, to rule out
> that the usb driver has some funkiness, but everything went smooth.

You can try to make sure that the hci_usb driver doesn't use any ISOC
transfers for SCO. The 2.4 USB subsystem was kinda picky about it. Other
than that you don't really have a choice except re-writing the hci_usb
driver.

Regards

Marcel




_______________________________________________
Bluez-users mailing list
Bluez-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-users

Re: [Bluez-users] Multiple connections with kernel 2.4 make HCI die

[Kevin Read]

Marcel Holtmann wrote:

Hi Marcel,

thanks for the quick answer.

>> I know that kernel 2.4 is no longer supported, but I hope that somebody 
>> can give me a clue here...
>>
>> I'm attempting to create a multi-threaded, multi-dongle OBEX client
>> appliance. When I run it on a kernel 2.6 based machine (either a x86 or
>> on a NSLU2, which is a big-endian ARM box), everything runs very smooth.
>>
>> The same code compiled for a little endian MIPS machine running kernel
>> 2.4.30 has problems. I cannot change the kernel version, because the
>> on-board Broadcom wireless chipset of the router doesn't really work yet
>>   under kernel 2.6 (damn closed source drivers) :(
> 
> I am not buying this argument, because the bcm43xx open source driver
> exists. Otherwise simply swap the mini-PCI card for an Intel one.

This is slightly OT but I hope it might be of interest to some:
I have already tried to compile the bcm43xx driver in OpenWRT kamikaze. 
Master mode will only work with the devicescape-branch of bcm43xx. Some 
OpenWRT co-users patched the Kamikaze kernel up to the latest code from 
the wireless-dev git repository. Unfortunately there was no joy there. 
Master mode does not yet run correctly, although the outlook is pretty 
good, as patches integrating bcm43xx and Hostap where sent to the 
bcm43xx devs and will hopefully get integrated soon.

Unfortunately for me, I need WDS and this has not yet been coded. And 
Kamikaze kernel 2.6 USB is broken at the moment, so I couldn't test 
bluetooth operation on the 2.6 :(

I'll look into swapping the mini-PCI card, thanks for the pointer. That 
might become a great alternative once the USB problems get sorted out.

<snip>
>>
>> I have applied the changes in the 2.4.32 patch by Marcel, but I think
>> they only refer to module unloading. Bluez installed is version 2.32. I
>> already stressed the USB bus with wireless and a hard disk, to rule out
>> that the usb driver has some funkiness, but everything went smooth.
> 
> You can try to make sure that the hci_usb driver doesn't use any ISOC
> transfers for SCO. The 2.4 USB subsystem was kinda picky about it. Other
> than that you don't really have a choice except re-writing the hci_usb
> driver.

Ok, I'll rebuild a kernel with SCO support right away. Will that 
suffice, or should I comment out the ISOC stuff in hci_usb.c?

I don't think I'll manage a re-write of hci_usb :/ Waiting for solid 2.6 
usb and bcm43xx support will definitely be faster than that ;)

Regards and thanks,

Kevin


_______________________________________________
Bluez-users mailing list
Bluez-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-users

Re: [Bluez-users] Multiple connections with kernel 2.4 make HCI die

[Marcel Holtmann]

Hi Kevin,

> >> I have applied the changes in the 2.4.32 patch by Marcel, but I think
> >> they only refer to module unloading. Bluez installed is version 2.32. I
> >> already stressed the USB bus with wireless and a hard disk, to rule out
> >> that the usb driver has some funkiness, but everything went smooth.
> > 
> > You can try to make sure that the hci_usb driver doesn't use any ISOC
> > transfers for SCO. The 2.4 USB subsystem was kinda picky about it. Other
> > than that you don't really have a choice except re-writing the hci_usb
> > driver.
> 
> Ok, I'll rebuild a kernel with SCO support right away. Will that 
> suffice, or should I comment out the ISOC stuff in hci_usb.c?

no. Disable the SCO support of the hci_usb driver. With a 2.6 kernel you
can do this with a module paramter, but I am not sure if this made it
ever into the 2.4 kernel.

Regards

Marcel




_______________________________________________
Bluez-users mailing list
Bluez-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-users

Re: [Bluez-users] Multiple connections with kernel 2.4 make HCI die

[Brand, Chris]

Hi Kevin,

> no. Disable the SCO support of the hci_usb driver. With a 2.6 
> kernel you can do this with a module paramter, but I am not 
> sure if this made it ever into the 2.4 kernel.

Just to avoid you wasting your time, I'm seeing what sounds like exactly
the same thing with a UART connection rather than USB. This is with a
2.4.18 kernel and 2.25 bluez-libs. Fortunately in my system I can easily
power-cycle the dongle, which seems to be the only way to bring it back
to life.

Hopefully it goes without saying that I'm very interested to hear about
anything you find that helps :-)

Chris



_______________________________________________
Bluez-users mailing list
Bluez-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-users