On Sun, May 27, 2012 at 3:50 PM, Ludovic Rousseau
<[email protected]> wrote:
> 2012/5/27 Martin Vogt <[email protected]>:
>> Thanks it works. No pattern "C" currently, but
>> now the stresstest shows, that the segfaults happen more often.
>
> Do you still have the endless loop of pattern "B" or it is also fixed?
I think its fixed, at least I had it no more.
(Maybe I had it once but I'm unsure about it, so I would say
its fixed)
But currently I look into the segfault, because its so easy to
reproduce :)

>> (Here I have an "uptime" of ~20secs until pcscd segfaults.)
>
> Only 20 seconds. We are far from high availability :-)
Yes ;)
Additionally I found out that the USB script triggers a BUG() in
the kernel, but it does not happen very often.

> I have not yet fixed this bug.
> And yes. It is a bit more complex to fix. I don't know yet how I will fix it.

I looked at the code and added a few printfs.
The sequence seems to be:
HOTPLUG UDEV calls RFRemoveReader -s
[....]
00000010 readerfactory.c:1009:RFUnInitializeReader() Attempting
shutdown of Aladdin eToken PRO USB 72K Java [Main Interface] 00 00.
00000013 ifdhandler.c:244:IFDHCloseChannel()
usb:0529/0620:libudev:0:/dev/bus/usb/002/018 (lun: 0)
00000021 ccid_usb.c:649:WriteUSB() write failed (2/18): -4 No such device
RFUnInitializeReader-1
00000026 readerfactory.c:880:RFUnloadReader() Unloading reader driver.

In between this call starts a new ContexThread:

00000105 winscard_svc.c:315:ContextThread() Received command:
DISCONNECT from client 9

HOTPLU UDEV ends call RFRemoveReader -e

Then I have:

00073910 winscard.c:796:SCardDisconnect() Lock released
00000010 winscard.c:820:SCardDisconnect() Active Contexts: 0
00000006 winscard.c:821:SCardDisconnect() dwDisposition: 2
Segmentation fault

Maybe the hotplug thread can deliver Add/Remove over the SVCServiceRunLoop?
(As far as I understand it, the hotplug mechanism works independently,
and this may be the problem?)

regards,

Martin
_______________________________________________
Muscle mailing list
[email protected]
http://lists.drizzle.com/mailman/listinfo/muscle

Reply via email to