udev or kernel broken with rsdl 0.30 patch [u]

udev or kernel broken with rsdl 0.30 patch [u]

[Andreas Jellinghaus [c]]

the rsdl patch only changes the scheduler, so I really wonder
why udev/kernel emit less events when the kernel is patched
with it. maybe a race condition or other bug in this area
only exposed by the rsdl 0.30 patch?

here is my analysis so far:
udev events on 2.6.20.2 (without rsdl):
( "$ACTION - $DEVICE - $DEVNAME - $PRODUCT - $$ - $PPID")
add - /proc/bus/usb/002/016 -  - 8e6/3438/100 - 6071 - 6065
add -  - /dev/usbdev2.16_ep02 -  - 6116 - 6113
add -  - /dev/usbdev2.16_ep82 -  - 6118 - 6114
add -  - /dev/usbdev2.16_ep81 -  - 6120 - 6115

udev events on 2.6.20.2 (with rsdl): 
( "$ACTION - $DEVICE - $DEVNAME - $PRODUCT - $$ - $PPID")
add -  - /dev/usbdev2.8_ep02 -  - 5287 - 5286
add -  - /dev/usbdev2.8_ep82 -  - 5290 - 5288
add -  - /dev/usbdev2.8_ep81 -  - 5291 - 5289

the /proc event is missing. in both cases /proc/ was mounted.

syslog with udev on debugging (only the /proc event from the kernel w/o
rsdl):

Mar 15 11:19:06 localhost udevd[2622]: udev_event_run: seq 2869 forked, pid
[6065], 'add' 'usb', 0 seconds old
Mar 15 11:19:06 localhost udevd-event[6065]: wait_for_sysfs:
file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus'
appeared after 0 loops
Mar 15 11:19:06 localhost udevd-event[6065]:
run_program: '/lib/udev/openct_usb'
Mar 15 11:19:07 localhost udevd-event[6065]:
run_program: '/lib/udev/openct_usb' returned with status 0
Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed 338
bytes to socket '/org/freedesktop/hal/udev_event', 
Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe -Q
usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe'
returned with status 1
Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed -1
bytes to socket '/org/kernel/udev/monitor', 
Mar 15 11:19:07 localhost udevd-event[6065]: udev_event_run: seq 2869
finished
Mar 15 11:19:07 localhost udevd[2622]: udev_done: seq 2869, pid [6065] exit
with 1, 1 seconds old

there is something similar in syslog (on the rsdl kernel):
Mar 15 11:24:32 localhost udevd[2607]: udev_event_run: seq 2717 forked, pid
[5284], 'add' 'usb', 0 seconds old
Mar 15 11:24:32 localhost udevd-event[5284]: wait_for_sysfs:
file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus'
appeared after 0 loops
Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed 338
bytes to socket '/org/freedesktop/hal/udev_event', 
Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe -Q
usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe'
returned with status 1
Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed -1
bytes to socket '/org/kernel/udev/monitor', 
Mar 15 11:24:32 localhost udevd-event[5284]: udev_event_run: seq 2717
finished
Mar 15 11:24:32 localhost udevd[2607]: udev_done: seq 2717, pid [5284] exit
with 1, 0 seconds old

the machine and the usb device are 100% the same, except that the first
kernel is 2.6.20.2 plain and the second was patched with rsdl 0.30.
the machine is a thinkpad t40 running ubuntu edgy (093-0ubuntu18.0edgy2).
the config was unchanged. the hardware is a usb smart card reader.

also maybe someone knows: what happened to the /dev/bus/usb events?
for a while I got two events from udev (guess that was dapper / much
older kernel and udev) and thus spawned two processed - one for handling
the device file in /proc/bus/usb and one for the device file
in /dev/bus/usb. now I see I'm no longer getting events for /dev/bus/usb/
at all. why that? so if /proc/bus/usb/ is no longer mounted, my application
will fail?

the udev config is quite simple:

BUS!="usb", ACTION!="add", GOTO="openct_usb_rules_end"

..
SYSFS{bInterfaceClass}="0b", SYSFS{bInterfaceSubClass}="00",
SYSFS{bInterfaceProtocol}="00" RUN+="/lib/udev/openct_usb"
..

LABEL="openct_usb_rules_end"

.. (pcmcia stuff)

Thanks for your help.

Regards, Andreas


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch [u]

[Kay Sievers]

On 3/15/07, Andreas Jellinghaus [c] <aj@ciphirelabs.com> wrote:
> the rsdl patch only changes the scheduler, so I really wonder
> why udev/kernel emit less events when the kernel is patched
> with it. maybe a race condition or other bug in this area
> only exposed by the rsdl 0.30 patch?
>
> here is my analysis so far:
> udev events on 2.6.20.2 (without rsdl):
> ( "$ACTION - $DEVICE - $DEVNAME - $PRODUCT - $$ - $PPID")
> add - /proc/bus/usb/002/016 -  - 8e6/3438/100 - 6071 - 6065
> add -  - /dev/usbdev2.16_ep02 -  - 6116 - 6113
> add -  - /dev/usbdev2.16_ep82 -  - 6118 - 6114
> add -  - /dev/usbdev2.16_ep81 -  - 6120 - 6115
>
> udev events on 2.6.20.2 (with rsdl):
> ( "$ACTION - $DEVICE - $DEVNAME - $PRODUCT - $$ - $PPID")
> add -  - /dev/usbdev2.8_ep02 -  - 5287 - 5286
> add -  - /dev/usbdev2.8_ep82 -  - 5290 - 5288
> add -  - /dev/usbdev2.8_ep81 -  - 5291 - 5289
>
> the /proc event is missing. in both cases /proc/ was mounted.
>
> syslog with udev on debugging (only the /proc event from the kernel w/o
> rsdl):
>
> Mar 15 11:19:06 localhost udevd[2622]: udev_event_run: seq 2869 forked, pid
> [6065], 'add' 'usb', 0 seconds old
> Mar 15 11:19:06 localhost udevd-event[6065]: wait_for_sysfs:
> file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus'
> appeared after 0 loops
> Mar 15 11:19:06 localhost udevd-event[6065]:
> run_program: '/lib/udev/openct_usb'
> Mar 15 11:19:07 localhost udevd-event[6065]:
> run_program: '/lib/udev/openct_usb' returned with status 0
> Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed 338
> bytes to socket '/org/freedesktop/hal/udev_event',
> Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe -Q
> usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
> Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe'
> returned with status 1
> Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed -1
> bytes to socket '/org/kernel/udev/monitor',
> Mar 15 11:19:07 localhost udevd-event[6065]: udev_event_run: seq 2869
> finished
> Mar 15 11:19:07 localhost udevd[2622]: udev_done: seq 2869, pid [6065] exit
> with 1, 1 seconds old
>
> there is something similar in syslog (on the rsdl kernel):
> Mar 15 11:24:32 localhost udevd[2607]: udev_event_run: seq 2717 forked, pid
> [5284], 'add' 'usb', 0 seconds old
> Mar 15 11:24:32 localhost udevd-event[5284]: wait_for_sysfs:
> file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus'
> appeared after 0 loops
> Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed 338
> bytes to socket '/org/freedesktop/hal/udev_event',
> Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe -Q
> usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
> Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe'
> returned with status 1
> Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed -1
> bytes to socket '/org/kernel/udev/monitor',
> Mar 15 11:24:32 localhost udevd-event[5284]: udev_event_run: seq 2717
> finished
> Mar 15 11:24:32 localhost udevd[2607]: udev_done: seq 2717, pid [5284] exit
> with 1, 0 seconds old
>
> the machine and the usb device are 100% the same, except that the first
> kernel is 2.6.20.2 plain and the second was patched with rsdl 0.30.
> the machine is a thinkpad t40 running ubuntu edgy (093-0ubuntu18.0edgy2).
> the config was unchanged. the hardware is a usb smart card reader.
>
> also maybe someone knows: what happened to the /dev/bus/usb events?
> for a while I got two events from udev (guess that was dapper / much
> older kernel and udev) and thus spawned two processed - one for handling
> the device file in /proc/bus/usb and one for the device file
> in /dev/bus/usb. now I see I'm no longer getting events for /dev/bus/usb/
> at all. why that? so if /proc/bus/usb/ is no longer mounted, my application
> will fail?
>
> the udev config is quite simple:
>
> BUS!="usb", ACTION!="add", GOTO="openct_usb_rules_end"
>
> ..
> SYSFS{bInterfaceClass}="0b", SYSFS{bInterfaceSubClass}="00",
> SYSFS{bInterfaceProtocol}="00" RUN+="/lib/udev/openct_usb"
> ..
>
> LABEL="openct_usb_rules_end"
>
> .. (pcmcia stuff)

What does udevmonitor print in both cases?

Kay

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Donnerstag, 15. März 2007 16:22 schrieb Kay Sievers:
> What does udevmonitor print in both cases?

without rsdl patch / working:
UEVENT[1173977420.910024] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
UDEV  [1173977420.910024] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
UEVENT[1173977420.910092] add@/class/usb_endpoint/usbdev2.9_ep00
UDEV  [1173977420.910092] add@/class/usb_endpoint/usbdev2.9_ep00
UEVENT[1173977420.910110] 
add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
UEVENT[1173977420.910121] add@/class/usb_endpoint/usbdev2.9_ep02
UEVENT[1173977420.910133] add@/class/usb_endpoint/usbdev2.9_ep82
UEVENT[1173977420.910143] add@/class/usb_endpoint/usbdev2.9_ep81
UEVENT[1173977420.910169] add@/class/usb_device/usbdev2.9
UDEV  [1173977421.859029] add@/class/usb_endpoint/usbdev2.9_ep02
UDEV  [1173977421.940517] 
add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
UDEV  [1173977421.998001] add@/class/usb_device/usbdev2.9
UDEV  [1173977422.112676] add@/class/usb_endpoint/usbdev2.9_ep81
UDEV  [1173977422.113827] add@/class/usb_endpoint/usbdev2.9_ep82
UEVENT[1173977422.121223] remove@/class/usb_endpoint/usbdev2.9_ep02
UDEV  [1173977422.124010] remove@/class/usb_endpoint/usbdev2.9_ep02
UEVENT[1173977422.124544] remove@/class/usb_endpoint/usbdev2.9_ep82
UDEV  [1173977422.127046] remove@/class/usb_endpoint/usbdev2.9_ep82
UEVENT[1173977422.127538] remove@/class/usb_endpoint/usbdev2.9_ep81
UDEV  [1173977422.129962] remove@/class/usb_endpoint/usbdev2.9_ep81
UEVENT[1173977422.130507] add@/class/usb_endpoint/usbdev2.9_ep02
UEVENT[1173977422.140809] add@/class/usb_endpoint/usbdev2.9_ep82
UEVENT[1173977422.151652] add@/class/usb_endpoint/usbdev2.9_ep81
UDEV  [1173977423.162898] add@/class/usb_endpoint/usbdev2.9_ep02
UDEV  [1173977423.181297] add@/class/usb_endpoint/usbdev2.9_ep81
UDEV  [1173977423.194416] add@/class/usb_endpoint/usbdev2.9_ep82


with:
UEVENT[1173977613.316489] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
UDEV  [1173977613.316489] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
UEVENT[1173977613.316551] add@/class/usb_endpoint/usbdev2.9_ep00
UDEV  [1173977613.316551] add@/class/usb_endpoint/usbdev2.9_ep00
UEVENT[1173977613.361385] 
add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
UEVENT[1173977613.361411] add@/class/usb_endpoint/usbdev2.9_ep02
UEVENT[1173977613.361422] add@/class/usb_endpoint/usbdev2.9_ep82
UEVENT[1173977613.361434] add@/class/usb_endpoint/usbdev2.9_ep81
UEVENT[1173977613.361444] add@/class/usb_device/usbdev2.9
UDEV  [1173977613.380601] 
add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
UDEV  [1173977613.423356] add@/class/usb_device/usbdev2.9
UDEV  [1173977614.388135] add@/class/usb_endpoint/usbdev2.9_ep82
UDEV  [1173977614.388165] add@/class/usb_endpoint/usbdev2.9_ep02
UDEV  [1173977614.395648] add@/class/usb_endpoint/usbdev2.9_ep81

why are the endpointes removed and added in the working case?
the device is unchanged. 

do you need a --env printout? of all events?

Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Thu, 2007-03-15 at 18:03 +0100, Andreas Jellinghaus wrote:
> Am Donnerstag, 15. März 2007 16:22 schrieb Kay Sievers:
> > What does udevmonitor print in both cases?
> 
> without rsdl patch / working:
> UEVENT[1173977420.910024] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
> UDEV  [1173977420.910024] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
> UEVENT[1173977420.910092] add@/class/usb_endpoint/usbdev2.9_ep00
> UDEV  [1173977420.910092] add@/class/usb_endpoint/usbdev2.9_ep00
> UEVENT[1173977420.910110] 
> add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
> UEVENT[1173977420.910121] add@/class/usb_endpoint/usbdev2.9_ep02
> UEVENT[1173977420.910133] add@/class/usb_endpoint/usbdev2.9_ep82
> UEVENT[1173977420.910143] add@/class/usb_endpoint/usbdev2.9_ep81
> UEVENT[1173977420.910169] add@/class/usb_device/usbdev2.9
> UDEV  [1173977421.859029] add@/class/usb_endpoint/usbdev2.9_ep02
> UDEV  [1173977421.940517] 
> add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
> UDEV  [1173977421.998001] add@/class/usb_device/usbdev2.9
> UDEV  [1173977422.112676] add@/class/usb_endpoint/usbdev2.9_ep81
> UDEV  [1173977422.113827] add@/class/usb_endpoint/usbdev2.9_ep82
> UEVENT[1173977422.121223] remove@/class/usb_endpoint/usbdev2.9_ep02
> UDEV  [1173977422.124010] remove@/class/usb_endpoint/usbdev2.9_ep02
> UEVENT[1173977422.124544] remove@/class/usb_endpoint/usbdev2.9_ep82
> UDEV  [1173977422.127046] remove@/class/usb_endpoint/usbdev2.9_ep82
> UEVENT[1173977422.127538] remove@/class/usb_endpoint/usbdev2.9_ep81
> UDEV  [1173977422.129962] remove@/class/usb_endpoint/usbdev2.9_ep81
> UEVENT[1173977422.130507] add@/class/usb_endpoint/usbdev2.9_ep02
> UEVENT[1173977422.140809] add@/class/usb_endpoint/usbdev2.9_ep82
> UEVENT[1173977422.151652] add@/class/usb_endpoint/usbdev2.9_ep81
> UDEV  [1173977423.162898] add@/class/usb_endpoint/usbdev2.9_ep02
> UDEV  [1173977423.181297] add@/class/usb_endpoint/usbdev2.9_ep81
> UDEV  [1173977423.194416] add@/class/usb_endpoint/usbdev2.9_ep82
> 
> 
> with:
> UEVENT[1173977613.316489] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
> UDEV  [1173977613.316489] add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5
> UEVENT[1173977613.316551] add@/class/usb_endpoint/usbdev2.9_ep00
> UDEV  [1173977613.316551] add@/class/usb_endpoint/usbdev2.9_ep00
> UEVENT[1173977613.361385] 
> add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
> UEVENT[1173977613.361411] add@/class/usb_endpoint/usbdev2.9_ep02
> UEVENT[1173977613.361422] add@/class/usb_endpoint/usbdev2.9_ep82
> UEVENT[1173977613.361434] add@/class/usb_endpoint/usbdev2.9_ep81
> UEVENT[1173977613.361444] add@/class/usb_device/usbdev2.9
> UDEV  [1173977613.380601] 
> add@/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
> UDEV  [1173977613.423356] add@/class/usb_device/usbdev2.9
> UDEV  [1173977614.388135] add@/class/usb_endpoint/usbdev2.9_ep82
> UDEV  [1173977614.388165] add@/class/usb_endpoint/usbdev2.9_ep02
> UDEV  [1173977614.395648] add@/class/usb_endpoint/usbdev2.9_ep81
> 
> why are the endpointes removed and added in the working case?
> the device is unchanged. 

Don't know what's with the endpoints, but what device event do you miss?
It looks like it's all there in the udevmonitor output, right?

Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Donnerstag, 15. März 2007 20:57 schrieb Kay Sievers:
> > why are the endpointes removed and added in the working case?
> > the device is unchanged.
>
> Don't know what's with the endpoints, but what device event do you miss?
> It looks like it's all there in the udevmonitor output, right?

it was not a missing event. the problem is the events are the same, the
hardware is the same, the configuration is the same, and still in one
cases udev runs the openct hook, and in the other case it does not.

maybe it has trouble looking at the sysfs files or something?

if I strace udevd it always works fine. if the system has high load too.
looks like a classic race condition to me.

Regards, Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Fri, 2007-03-16 at 11:06 +0100, Andreas Jellinghaus wrote:
> Am Donnerstag, 15. März 2007 20:57 schrieb Kay Sievers:
> > > why are the endpointes removed and added in the working case?
> > > the device is unchanged.
> >
> > Don't know what's with the endpoints, but what device event do you miss?
> > It looks like it's all there in the udevmonitor output, right?
> 
> it was not a missing event.

Well, you wrote: "why udev/kernel emit less events?" :)

> the problem is the events are the same, the
> hardware is the same, the configuration is the same, and still in one
> cases udev runs the openct hook, and in the other case it does not.
> 
> maybe it has trouble looking at the sysfs files or something?
> 
> if I strace udevd it always works fine. if the system has high load too.
> looks like a classic race condition to me.

You mean, that the /proc file is not there, when you get the event with
$DEVICE? It is "broken" since it exists, you have to wait until it shows
up.
Or don't use it at all, and hook into the "usb_device" event and use the
files in /dev with the same layout, instead of the crappy usbfs files.

Thanks,
Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Freitag, 16. März 2007 11:35 schrieb Kay Sievers:
> On Fri, 2007-03-16 at 11:06 +0100, Andreas Jellinghaus wrote:
> > Am Donnerstag, 15. März 2007 20:57 schrieb Kay Sievers:
> > > > why are the endpointes removed and added in the working case?
> > > > the device is unchanged.
> > >
> > > Don't know what's with the endpoints, but what device event do you
> > > miss? It looks like it's all there in the udevmonitor output, right?
> >
> > it was not a missing event.
>
> Well, you wrote: "why udev/kernel emit less events?" :)

and debugging shows udev gets the event, but does not emit it,
i.e. doesn't run my script. in one case it runs my script,
in the other case it does not. why?

> You mean, that the /proc file is not there,

does udev look for the /proc file? I don't get a chance to do so - my script
is not even run, as udev indicates. the udev config lines are the same:
refering to BUS and some SYSFS stuff, but not to /proc, so why do you
think it has to do with /proc? also my script would work fine with both
$DEVICE - traditional /proc stuff - and $DEVNAME - was used when
introducing /dev/bus/usb stuff. but I don't get to that point, the script
isn't run at all (I checked with a echo in the first line of the script and
with udevd in debug logging mode where it logs each time it runs
something).

> Or don't use it at all, and hook into the "usb_device" event and use the
> files in /dev with the same layout, instead of the crappy usbfs files.

earlier I got those events, but these days I don't get called for them.
any idea why? (ubuntu edgy, config as posted).

guess I got them with dapper - and had to implement proper interface
claiming to prevent two processes talking to the same hardware through
different devices.

how can I debug why udev doesn't run openct on the /dev/bus/usb
device? I can see in udev log that it creates the device:

Mar 15 11:19:07 localhost udevd[2622]: udev_event_run: seq 2873 forked, pid 
[6094], 'add' 'usb_device', 1 seconds old
Mar 15 11:19:07 localhost udevd-event[6094]: 
run_program: 'usb_device_name --export usbdev2.16'
Mar 15 11:19:07 localhost udevd-event[6094]: 
run_program: '/lib/udev/usb_device_name' (stdout) 'USB_BUS\02'
Mar 15 11:19:07 localhost udevd-event[6094]: 
run_program: '/lib/udev/usb_device_name' (stdout) 'USB_DEV\x016'
Mar 15 11:19:07 localhost udevd-event[6094]: 
run_program: '/lib/udev/usb_device_name' returned with status 0
Mar 15 11:19:07 localhost udevd-event[6094]: udev_rules_get_name: rule 
applied, 'usbdev2.16' becomes 'bus/usb/002/016'
Mar 15 11:19:07 localhost udevd-event[6094]: udev_db_get_device: no db file to 
read /dev/.udev/db/class@usb_device@usbdev2.16: No such file or directory
Mar 15 11:19:07 localhost udevd-event[6094]: udev_node_add: creating device 
node '/dev/bus/usb/002/016', major = '189', minor = '143', mode = '0664', uid 
= '0', gid = '0'
Mar 15 11:19:07 localhost udevd-event[6094]: pass_env_to_socket: passed 305 
bytes to socket '/org/freedesktop/hal/udev_event', 
Mar 15 11:19:07 localhost udevd-event[6094]: pass_env_to_socket: passed -1 
bytes to socket '/org/kernel/udev/monitor', 
Mar 15 11:19:07 localhost udevd-event[6094]: udev_event_run: seq 2873 finished
Mar 15 11:19:07 localhost udevd[2622]: udev_done: seq 2873, pid [6094] exit 
with 0, 1 seconds old

see, the device is created and udev is run. but my openct rule:
BUS!="usb", ACTION!="add", GOTO="openct_usb_rules_end"
SYSFS{bInterfaceClass}="0b", SYSFS{bInterfaceSubClass}="00", 
SYSFS{bInterfaceProtocol}="00" RUN+="/lib/udev/openct_usb"

is not run. why?

or does udev create these new /dev/bus/usb devices are /proc/bus/usb 
replacement, but does not allow to match simple usb config values like
interface class on them? that would be a huge step back, so I hope it
isn't the case.

Thanks for your help.

Regards, Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Fri, 2007-03-16 at 11:59 +0100, Andreas Jellinghaus wrote:
> Am Freitag, 16. März 2007 11:35 schrieb Kay Sievers:
> > On Fri, 2007-03-16 at 11:06 +0100, Andreas Jellinghaus wrote:
> > > Am Donnerstag, 15. März 2007 20:57 schrieb Kay Sievers:
> > > > > why are the endpointes removed and added in the working case?
> > > > > the device is unchanged.
> > > >
> > > > Don't know what's with the endpoints, but what device event do you
> > > > miss? It looks like it's all there in the udevmonitor output, right?
> > >
> > > it was not a missing event.
> >
> > Well, you wrote: "why udev/kernel emit less events?" :)
> 
> and debugging shows udev gets the event, but does not emit it,

What shows that?

> i.e. doesn't run my script. in one case it runs my script,
> in the other case it does not. why?

In both cases in your log, there are no events for the usb-device
itself: /devices/pci0000:00/0000:00:1d.1/usb2/2-1
Maybe you missed to copy a few lines?

> > You mean, that the /proc file is not there,
> 
> does udev look for the /proc file?

No, that will never happen. :)

> I don't get a chance to do so - my script
> is not even run, as udev indicates. the udev config lines are the same:
> refering to BUS and some SYSFS stuff, but not to /proc, so why do you
> think it has to do with /proc? also my script would work fine with both
> $DEVICE - traditional /proc stuff - and $DEVNAME - was used when
> introducing /dev/bus/usb stuff. but I don't get to that point, the script
> isn't run at all (I checked with a echo in the first line of the script and
> with udevd in debug logging mode where it logs each time it runs
> something).
> 
> > Or don't use it at all, and hook into the "usb_device" event and use the
> > files in /dev with the same layout, instead of the crappy usbfs files.
> 
> earlier I got those events, but these days I don't get called for them.
> any idea why? (ubuntu edgy, config as posted).

No idea, I get all usual events just fine here.

> guess I got them with dapper

Oh, can you please talk about kernel versions, and not distro nicknames?
I have no clue what they mean.

> and had to implement proper interface
> claiming to prevent two processes talking to the same hardware through
> different devices.
> 
> how can I debug why udev doesn't run openct on the /dev/bus/usb
> device? I can see in udev log that it creates the device:

udevmonitor, if you see the event from the kernel, it's udev problem, if
not, it's a kernel problem.

> Mar 15 11:19:07 localhost udevd[2622]: udev_event_run: seq 2873 forked, pid 
> [6094], 'add' 'usb_device', 1 seconds old
> Mar 15 11:19:07 localhost udevd-event[6094]: 
> run_program: 'usb_device_name --export usbdev2.16'
> Mar 15 11:19:07 localhost udevd-event[6094]: 
> run_program: '/lib/udev/usb_device_name' (stdout) 'USB_BUS\02'
> Mar 15 11:19:07 localhost udevd-event[6094]: 
> run_program: '/lib/udev/usb_device_name' (stdout) 'USB_DEV\x016'
> Mar 15 11:19:07 localhost udevd-event[6094]: 
> run_program: '/lib/udev/usb_device_name' returned with status 0
> Mar 15 11:19:07 localhost udevd-event[6094]: udev_rules_get_name: rule 
> applied, 'usbdev2.16' becomes 'bus/usb/002/016'
> Mar 15 11:19:07 localhost udevd-event[6094]: udev_db_get_device: no db file to 
> read /dev/.udev/db/class@usb_device@usbdev2.16: No such file or directory
> Mar 15 11:19:07 localhost udevd-event[6094]: udev_node_add: creating device 
> node '/dev/bus/usb/002/016', major = '189', minor = '143', mode = '0664', uid 
> = '0', gid = '0'
> Mar 15 11:19:07 localhost udevd-event[6094]: pass_env_to_socket: passed 305 
> bytes to socket '/org/freedesktop/hal/udev_event', 
> Mar 15 11:19:07 localhost udevd-event[6094]: pass_env_to_socket: passed -1 
> bytes to socket '/org/kernel/udev/monitor', 
> Mar 15 11:19:07 localhost udevd-event[6094]: udev_event_run: seq 2873 finished
> Mar 15 11:19:07 localhost udevd[2622]: udev_done: seq 2873, pid [6094] exit 
> with 0, 1 seconds old
> 
> see, the device is created and udev is run. but my openct rule:
> BUS!="usb", ACTION!="add", GOTO="openct_usb_rules_end"
> SYSFS{bInterfaceClass}="0b", SYSFS{bInterfaceSubClass}="00", 
> SYSFS{bInterfaceProtocol}="00" RUN+="/lib/udev/openct_usb"
> 
> is not run. why?
> 
> or does udev create these new /dev/bus/usb devices are /proc/bus/usb 
> replacement, but does not allow to match simple usb config values like
> interface class on them? that would be a huge step back, so I hope it
> isn't the case.

Right, that never worked. The usb-interfaces are childs of the
usb-device in /sys/devices. The usb_device class-devices (the ones with
the /dev/bus node) are also just childs of the usb-device and therefore
can't access the (sibling) interface devices.

With the current usb driver core, you can't match the /dev device-node
to interfaces, only to the device.

There is a patch pending (now that the new driver core allows us to do
that), that moves the device-node directly to the /sys/devices
usb-device, which should solve all the issues in the future.

Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Freitag, 16. März 2007 12:34 schrieb Kay Sievers:
> > and debugging shows udev gets the event, but does not emit it,
>
> What shows that?
>
> > i.e. doesn't run my script. in one case it runs my script,
> > in the other case it does not. why?
>
> In both cases in your log, there are no events for the usb-device
> itself: /devices/pci0000:00/0000:00:1d.1/usb2/2-1
> Maybe you missed to copy a few lines?

ok, here are again the criticial lines from syslog messages: with 2.6.20.2:
Mar 15 11:19:06 localhost udevd[2622]: udev_event_run: seq 2869 forked, pid 
[6065], 'add' 'usb', 0 seconds old
Mar 15 11:19:06 localhost udevd-event[6065]: wait_for_sysfs: 
file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus' 
appeared after 0 loops
Mar 15 11:19:06 localhost udevd-event[6065]: 
run_program: '/lib/udev/openct_usb'
Mar 15 11:19:07 localhost udevd-event[6065]: 
run_program: '/lib/udev/openct_usb' returned with status 0
Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed 338 
bytes to socket '/org/freedesktop/hal/udev_event', 
Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe -Q 
usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe' 
returned with status 1
Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed -1 
bytes to socket '/org/kernel/udev/monitor', 
Mar 15 11:19:07 localhost udevd-event[6065]: udev_event_run: seq 2869 finished
Mar 15 11:19:07 localhost udevd[2622]: udev_done: seq 2869, pid [6065] exit 
with 1, 1 seconds old

and if I add rsdl:
Mar 15 11:24:32 localhost udevd[2607]: udev_event_run: seq 2717 forked, pid 
[5284], 'add' 'usb', 0 seconds old
Mar 15 11:24:32 localhost udevd-event[5284]: wait_for_sysfs: 
file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus' 
appeared after 0 loops
Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed 338 
bytes to socket '/org/freedesktop/hal/udev_event', 
Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe -Q 
usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe' 
returned with status 1
Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed -1 
bytes to socket '/org/kernel/udev/monitor', 
Mar 15 11:24:32 localhost udevd-event[5284]: udev_event_run: seq 2717 finished
Mar 15 11:24:32 localhost udevd[2607]: udev_done: seq 2717, pid [5284] exit 
with 1, 0 seconds old

same  hardware, same config, same kernel (except for rsdl patch), and suddenly
it does not spawn openct. why? how can I debug this? if I strace, it works 
fine again, so that didn't help.

> Oh, can you please talk about kernel versions, and not distro nicknames?
> I have no clue what they mean.

ok, sorry. lets skip the question why I don't get events for /dev any more.

> Right, that never worked. The usb-interfaces are childs of the
> usb-device in /sys/devices. The usb_device class-devices (the ones with
> the /dev/bus node) are also just childs of the usb-device and therefore
> can't access the (sibling) interface devices.
>
> With the current usb driver core, you can't match the /dev device-node
> to interfaces, only to the device.
>
> There is a patch pending (now that the new driver core allows us to do
> that), that moves the device-node directly to the /sys/devices
> usb-device, which should solve all the issues in the future.

ok. so until that patch is in the kernel (or udev? or both?), /dev/bus/usb
will not work for udev users as /proc/bus/usb replacement, since the
rules to match the attributes of the usb device won't work. right?

is there some page documenting the compatiblity issues with udev
and the kernel? this might be interesting for other people as well.

Thanks for your help.

Regards, Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Fri, 2007-03-16 at 12:45 +0100, Andreas Jellinghaus wrote:
> Am Freitag, 16. März 2007 12:34 schrieb Kay Sievers:
> > > and debugging shows udev gets the event, but does not emit it,
> >
> > What shows that?
> >
> > > i.e. doesn't run my script. in one case it runs my script,
> > > in the other case it does not. why?
> >
> > In both cases in your log, there are no events for the usb-device
> > itself: /devices/pci0000:00/0000:00:1d.1/usb2/2-1
> > Maybe you missed to copy a few lines?
> 
> ok, here are again the criticial lines from syslog messages: with 2.6.20.2:
> Mar 15 11:19:06 localhost udevd[2622]: udev_event_run: seq 2869 forked, pid 
> [6065], 'add' 'usb', 0 seconds old
> Mar 15 11:19:06 localhost udevd-event[6065]: wait_for_sysfs: 
> file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus' 
> appeared after 0 loops
> Mar 15 11:19:06 localhost udevd-event[6065]: 
> run_program: '/lib/udev/openct_usb'
> Mar 15 11:19:07 localhost udevd-event[6065]: 
> run_program: '/lib/udev/openct_usb' returned with status 0
> Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed 338 
> bytes to socket '/org/freedesktop/hal/udev_event', 
> Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe -Q 
> usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
> Mar 15 11:19:07 localhost udevd-event[6065]: run_program: '/sbin/modprobe' 
> returned with status 1
> Mar 15 11:19:07 localhost udevd-event[6065]: pass_env_to_socket: passed -1 
> bytes to socket '/org/kernel/udev/monitor', 
> Mar 15 11:19:07 localhost udevd-event[6065]: udev_event_run: seq 2869 finished
> Mar 15 11:19:07 localhost udevd[2622]: udev_done: seq 2869, pid [6065] exit 
> with 1, 1 seconds old
> 
> and if I add rsdl:
> Mar 15 11:24:32 localhost udevd[2607]: udev_event_run: seq 2717 forked, pid 
> [5284], 'add' 'usb', 0 seconds old
> Mar 15 11:24:32 localhost udevd-event[5284]: wait_for_sysfs: 
> file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bus' 
> appeared after 0 loops
> Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed 338 
> bytes to socket '/org/freedesktop/hal/udev_event', 
> Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe -Q 
> usb:v08E6p3438d0100dc00dsc00dp00ic0Bisc00ip00'
> Mar 15 11:24:32 localhost udevd-event[5284]: run_program: '/sbin/modprobe' 
> returned with status 1
> Mar 15 11:24:32 localhost udevd-event[5284]: pass_env_to_socket: passed -1 
> bytes to socket '/org/kernel/udev/monitor', 
> Mar 15 11:24:32 localhost udevd-event[5284]: udev_event_run: seq 2717 finished
> Mar 15 11:24:32 localhost udevd[2607]: udev_done: seq 2717, pid [5284] exit 
> with 1, 0 seconds old

If you run:
  udevtest /devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
it works, right?

It that works, it seems the sysfs files are not created at the time you
want to read them, and the attribute creation time needs to be fixed in
the kernel.

> same  hardware, same config, same kernel (except for rsdl patch), and suddenly
> it does not spawn openct. why? how can I debug this? if I strace, it works 
> fine again, so that didn't help.
> 
> > Oh, can you please talk about kernel versions, and not distro nicknames?
> > I have no clue what they mean.
> 
> ok, sorry. lets skip the question why I don't get events for /dev any more.
> 
> > Right, that never worked. The usb-interfaces are childs of the
> > usb-device in /sys/devices. The usb_device class-devices (the ones with
> > the /dev/bus node) are also just childs of the usb-device and therefore
> > can't access the (sibling) interface devices.
> >
> > With the current usb driver core, you can't match the /dev device-node
> > to interfaces, only to the device.
> >
> > There is a patch pending (now that the new driver core allows us to do
> > that), that moves the device-node directly to the /sys/devices
> > usb-device, which should solve all the issues in the future.
> 
> ok. so until that patch is in the kernel (or udev? or both?),

A patch to the kernel, and a rule to udev, to create the nodes
in /dev/bus/usb, from the event of the real usb-device in /sys/devices
instead of the current class device.

> /dev/bus/usb
> will not work for udev users as /proc/bus/usb replacement, since the
> rules to match the attributes of the usb device won't work. right?

That works fine, to match device properties. but not to match interface
properties of that device. It only works with HAL, because it has a
weird hack to look at the sibling device.

> is there some page documenting the compatiblity issues with udev
> and the kernel? this might be interesting for other people as well.

As usual, unfortunately no there is not documentation besides what the
distros are doing with it.

Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Freitag, 16. März 2007 13:08 schrieb Kay Sievers:
> If you run:
>   udevtest /devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
> it works, right?

yes, I booted both kernels once more and tested this, and the output is 
exactly the same, and it runs openct_usb in both cases.

> It that works, it seems the sysfs files are not created at the time you
> want to read them, and the attribute creation time needs to be fixed in
> the kernel.

ok. that seems likely, with rsdl the kernel has different timings and that 
seems to trigger the problem.

what exactly is "wait_for_sysfs"? I'm asking because it shows up in the udevd
log right before openct is started / should be started. I thought it would 
take care of problems like this? or does it only check the basedir, and not
the subdirs / files in the dirs to be present or something like that?


> > ok. so until that patch is in the kernel (or udev? or both?),
>
> A patch to the kernel, and a rule to udev, to create the nodes
> in /dev/bus/usb, from the event of the real usb-device in /sys/devices
> instead of the current class device.
>
> > /dev/bus/usb
> > will not work for udev users as /proc/bus/usb replacement, since the
> > rules to match the attributes of the usb device won't work. right?
>
> That works fine, to match device properties. but not to match interface
> properties of that device. It only works with HAL, because it has a
> weird hack to look at the sibling device.
>
> > is there some page documenting the compatiblity issues with udev
> > and the kernel? this might be interesting for other people as well.
>
> As usual, unfortunately no there is not documentation besides what the
> distros are doing with it.

the kernel has the requirement document which documents which version
of udev is needed. it could be a good start to put this as list on the net / 
which kernel needs which minimum udev.

I ran into udev/kernel incompatibilities at least once (but my fault, didn't
check that document properly).

Thanks,

Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Fri, 2007-03-16 at 13:32 +0100, Andreas Jellinghaus wrote:
> Am Freitag, 16. März 2007 13:08 schrieb Kay Sievers:
> > If you run:
> >   udevtest /devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0
> > it works, right?
> 
> yes, I booted both kernels once more and tested this, and the output is 
> exactly the same, and it runs openct_usb in both cases.
> 
> > It that works, it seems the sysfs files are not created at the time you
> > want to read them, and the attribute creation time needs to be fixed in
> > the kernel.
> 
> ok. that seems likely, with rsdl the kernel has different timings and that 
> seems to trigger the problem.

Yeah, if it's the case, it should be fixed in the kernel, to create the
attributes before the device is announced to userspace.

> what exactly is "wait_for_sysfs"? I'm asking because it shows up in the udevd
> log right before openct is started / should be started. I thought it would 
> take care of problems like this? or does it only check the basedir, and not
> the subdirs / files in the dirs to be present or something like that?

It waits for what you ask it to wait for. This may work:
  SUBSYSTEM="usb", KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceProtocol"

Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Freitag, 16. März 2007 13:48 schrieb Kay Sievers:
> > what exactly is "wait_for_sysfs"? I'm asking because it shows up in the
> > udevd log right before openct is started / should be started. I thought
> > it would take care of problems like this? or does it only check the
> > basedir, and not the subdirs / files in the dirs to be present or
> > something like that?
>
> It waits for what you ask it to wait for. This may work:
>   SUBSYSTEM="usb", KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceProtocol"

I tried it like this:
BUS!="usb", ACTION!="add", GOTO="openct_usb_rules_end"

KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceClass"
KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceSubClass"
KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceProtocol"

SYSFS{bInterfaceClass}="0b", SYSFS{bInterfaceSubClass}="00", 
SYSFS{bInterfaceProtocol}="00" RUN+="/lib/udev/openct_usb"

LABEL="openct_usb_rules_end"

i.e. with extra lines in the config file. is that correct? now I have a 50/50
chance that openct_usb run (I plugged in&out 15 times for testing).
and I get lots of warnings during boot because some usb hardware
doesn't have these attributes. is there a way to restrict them to 
everything that has a usb device file?

Thanks, Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Freitag, 16. März 2007 14:35 schrieb Andreas Jellinghaus:
> now I have a 50/50 chance that openct_usb is run 

that was with usbd debug logging. one reboot later, it didn't work
once (normal system / no debugging for usbd). so it didn't help.
but maybe my config was wrong?

Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Fri, 2007-03-16 at 14:35 +0100, Andreas Jellinghaus wrote:
> Am Freitag, 16. März 2007 13:48 schrieb Kay Sievers:
> > > what exactly is "wait_for_sysfs"? I'm asking because it shows up in the
> > > udevd log right before openct is started / should be started. I thought
> > > it would take care of problems like this? or does it only check the
> > > basedir, and not the subdirs / files in the dirs to be present or
> > > something like that?
> >
> > It waits for what you ask it to wait for. This may work:
> >   SUBSYSTEM="usb", KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceProtocol"
> 
> I tried it like this:
> BUS!="usb", ACTION!="add", GOTO="openct_usb_rules_end"

What do you mean here? You only skip the rules if it's !usb _and_ !add.
So any other subsystem will enter here. It should probably be:
  ACTION=!"add", GOTO="openct_usb_rules_end"
  SUBSYSTEM=!"usb", GOTO="openct_usb_rules_end"

And don't use BUS=, it will search all devices in the devpath, but you
need only want to see real usb-device events, not other classes of
devices that are childs of any usb-device.

> KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceClass"
> KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceSubClass"
> KERNEL="[0-9]*:*", WAIT_FOR_SYSFS="bInterfaceProtocol"

You only need bInterfaceProtocol, it's the last file created by the
kernel.

> SYSFS{bInterfaceClass}="0b", SYSFS{bInterfaceSubClass}="00", 
> SYSFS{bInterfaceProtocol}="00" RUN+="/lib/udev/openct_usb"
> 
> LABEL="openct_usb_rules_end"
> 
> i.e. with extra lines in the config file. is that correct? now I have a 50/50
> chance that openct_usb run (I plugged in&out 15 times for testing).
> and I get lots of warnings during boot because some usb hardware
> doesn't have these attributes. is there a way to restrict them to 
> everything that has a usb device file?

It's probably because you match all sorts of devices which aren't usb.

Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Ok, thanks for the udev rules fixes.

but with all rules fixes I still have the issue: openct not started with the 
rsdl patched kernel. 

My guess is the temp. success was only because udevd debugging and the
wrong rules slowed udevd down and thus evaded the race problem.

Please let me know if I can be of any further help, like testing more config 
options or testing patches or anything like that. 

Regards, Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Fri, 2007-03-16 at 15:39 +0100, Andreas Jellinghaus wrote:
> Ok, thanks for the udev rules fixes.
> 
> but with all rules fixes I still have the issue: openct not started with the 
> rsdl patched kernel. 
> 
> My guess is the temp. success was only because udevd debugging and the
> wrong rules slowed udevd down and thus evaded the race problem.
> 
> Please let me know if I can be of any further help, like testing more config 
> options or testing patches or anything like that. 

Hmm, if it's a timing issue we should be able to fix it with
WAIT_FOR_SYSFS. What does:
  udevtest /devices/pci0000:00/0000:00:1d.2/usb4/4-1/4-1.3/4-1.3:1.0

for the interface (the one with the ":") say to WAIT_FOR_SYSFS? It finds
the attributes, and says it would run your program?

It does not work on device plug-in, does it work if you write "add" to
the "uevent"-file of the device in sysfs?

Does adding PROGRAM="/bin/sleep 1" make a difference?

Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Andreas Jellinghaus]

Am Freitag, 16. März 2007 15:59 schrieb Kay Sievers:
> Hmm, if it's a timing issue we should be able to fix it with
> WAIT_FOR_SYSFS. What does:
>   udevtest /devices/pci0000:00/0000:00:1d.2/usb4/4-1/4-1.3/4-1.3:1.0

udevtest /devices/pci0000\:00/0000\:00\:1d.1/usb2/2-1/2-1.5/2-1.5\:1.0 
main: looking at 
device '/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0' from 
subsystem 'usb'
wait_for_sysfs: 
file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bInterfaceProtocol' 
appeared after 0 loops
main: run: '/lib/udev/openct_usb'
main: run: 'socket:/org/freedesktop/hal/udev_event'
main: run: 'socket:/org/kernel/udev/monitor'

looks ok to me.

> for the interface (the one with the ":") say to WAIT_FOR_SYSFS? It finds
> the attributes, and says it would run your program?

yes, udevtest is always fine.

> It does not work on device plug-in, does it work if you write "add" to
> the "uevent"-file of the device in sysfs?

that works as well.

> Does adding PROGRAM="/bin/sleep 1" make a difference?

that solved the problem. small delay, but now the
plugin works all the time. will add that to openct.
(or rather remove the "sleep 1" I already have in my
script to "wait" for the device file and add the sleep 1 as
udev rule instead).

btw: if this get fixed so the "sleep 1" becomes obsolete one day - is there 
some rule etc. to wait for the /proc/bus/usb or /dev/bus/usb devices, so I
won't need an arbitrary sleep for that either?

Thanks for your help.

Regards, Andreas

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: udev or kernel broken with rsdl 0.30 patch

[Kay Sievers]

On Fri, 2007-03-16 at 16:20 +0100, Andreas Jellinghaus wrote:
> Am Freitag, 16. März 2007 15:59 schrieb Kay Sievers:
> > Hmm, if it's a timing issue we should be able to fix it with
> > WAIT_FOR_SYSFS. What does:
> >   udevtest /devices/pci0000:00/0000:00:1d.2/usb4/4-1/4-1.3/4-1.3:1.0
> 
> udevtest /devices/pci0000\:00/0000\:00\:1d.1/usb2/2-1/2-1.5/2-1.5\:1.0 
> main: looking at 
> device '/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0' from 
> subsystem 'usb'
> wait_for_sysfs: 
> file '/sys/devices/pci0000:00/0000:00:1d.1/usb2/2-1/2-1.5/2-1.5:1.0/bInterfaceProtocol' 
> appeared after 0 loops
> main: run: '/lib/udev/openct_usb'
> main: run: 'socket:/org/freedesktop/hal/udev_event'
> main: run: 'socket:/org/kernel/udev/monitor'
> 
> looks ok to me.
> 
> > for the interface (the one with the ":") say to WAIT_FOR_SYSFS? It finds
> > the attributes, and says it would run your program?
> 
> yes, udevtest is always fine.
> 
> > It does not work on device plug-in, does it work if you write "add" to
> > the "uevent"-file of the device in sysfs?
> 
> that works as well.
> 
> > Does adding PROGRAM="/bin/sleep 1" make a difference?
> 
> that solved the problem.

Weird, WAIT_FOR_SYSFS should have catched that, but maybe the file is
there, and you just can't open it that time, because the device refuses
the access.

> small delay, but now the
> plugin works all the time. will add that to openct.
> (or rather remove the "sleep 1" I already have in my
> script to "wait" for the device file and add the sleep 1 as
> udev rule instead).
> 
> btw: if this get fixed so the "sleep 1" becomes obsolete one day - is there 
> some rule etc. to wait for the /proc/bus/usb or /dev/bus/usb devices, so I
> won't need an arbitrary sleep for that either?

If we get a device node at the real usb-device in /sys/devices, the
interface-event is guaranteed to come after the device-event, and the
device-file for the usb-device will always be created before udev calls
any RUN program.

Udev should be able to IMPORT{parent}="DEVNAME" to the interface event,
os that the device file name is available at the interface events too.

For /proc there will be no fix ever I guess, we can't use it on modern
desktops anyway, because is doesn't support ACL's which is the default
way to grant access to devices to untrusted users.

Kay


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel