-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

> yea, that's what I was trying to say in my second message yesterday.
>
> I've attached a diff that eliminates the crash from the reproduction 
> case
> given in the PR, and it passes all the libpthread conformance tests on 
> sparc.
> I'm not sure that it's correct to skip the rwindow_save() in all the 
> contexts
> that sa_upcall_getstate() is called, but I suspect it is.

First results -I couldn't reproduce a kernel panic anymore but 
processes don't seem to wake up after getting paged in again.
Symptoms: I started a couple of fairly memory-hungry, multithreaded X11 
apps, then started a compile job in an xterm ( kde stuff - makes the 
c++ compiler eat copious amounts of memory ). Left the whole thing 
alone for a while and then tried to bring a window of one of the big 
X11 apps to front ( mysqlcc in this case ) - lots of paging activity 
and then nothing. The process seems to hang, at least the window 
content doesn't get redrawn. Another app ( gcvs ) got a bus error and 
dumped core instead but I'm not sure that's really related.
Trying to kill the hung process led to ~100% CPU usage, all kernel 
time, from mysqlcc.

So, whatever you're doing - you seem to be on the right way :)

have fun
Michael
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (Darwin)

iQEVAwUBQfhzsMpnzkX8Yg2nAQJfvwf/T227qRGOv1M8boLvkvJuPH9BE9WwYaG/
GOeFdz0apKOzelzlHnP9YZj1mkbMshz7zQe4JeP6HLxgGQ9zSJlPN89DrksHb/Mx
xCoaOobTinhe5aBuLas8HrCDVWdn9djXDzop1KqrYOBd2FMUpowqA6kkzHk+enQl
ZUyGL4Md8Xt66fs2gJv5+FYVtJ1iaRE3Yf04wng/F/Qu/xw/zqQZxrC3q+kZ+N+N
HPUR7D+n5Z6rjjk+ve6hjFMIhjzysEJFfHsimQ/9McLwwIKucbzlLrMhUywmOyW9
0lXZcP8iIl7sMP1SRWkaM2EHn4lowERG/PKEmkTtJTHIrQOWihG28Q==
=OhJK
-----END PGP SIGNATURE-----