Fwd: ZFS and nvidia driver -- panic with spontaneous reboot

[Norbert Papke]

----------  Forwarded Message  ----------

Subject: Fwd: ZFS and nvidia driver -- panic with spontaneous reboot
Date: July 17, 2008
From: Norbert Papke <fbsd-ml at scrapper.ca>
To: freebsd-test at freebsd.org

Not showing up in either -questions or -stable ...

----------  Forwarded Message  ----------

Subject: ZFS and nvidia driver -- panic with spontaneous reboot
Date: July 16, 2008
From: Norbert Papke <fbsd-ml at scrapper.ca>
To: freebsd-stable at freebsd.org

I am having a terrible time trying to get my desktop system (recent 
7.0-Stable) working with both ZFS and the nvidia proprietary driver 
(nvidia-driver-173.14.09).  The nvidia driver, of course, restricts me to 
i386.

I tried to follow the recommended practice of increasing KVA_PAGES.  With this 
change, I cannot start X -- the nvidia driver errors out.  Without the KVA 
change, I cannot increase kmem sufficiently.  It is limited to 512M.

ZFS doesn't seem to have any problems with 512M kmem on my workload.  
Unfortunately, the nvidia driver is less happy.  After some time (a couple of 
days usually), it triggers panics that cause a spontaneous reboot.  There is 
no core dump nor does the kernel debugger take control.  Sometimes details 
are logged, sometimes the details are lost.  The details that have been 
logged always look similar to this:

panic: kmem_malloc(847872): kmem_map too small: 322166784 total allocated
 KDB: stack backtrace:
 db_trace_self_wrapper(c072d320,f4663990,c052aeaa,c072b6b3,c0794e80,...) at 
db_trace_self_wrapper+0x26
 kdb_backtrace(c072b6b3,c0794e80,c073cd53,f466399c,f466399c,...) at 
kdb_backtrace+0x29
 panic(c073cd53,cf000,1333e000,c073cd4d,7d0,...) at panic+0xaa
 kmem_malloc(c185408c,cf000,102,f4663a1c,c06b1da0,...) at kmem_malloc+0x28a
 page_alloc(0,cf000,f4663a0f,102,2000000,...) at page_alloc+0x27
 uma_large_malloc(cf000,102,102,c6bd4800,c61dcc60,...) at 
uma_large_malloc+0x50
 malloc(cf000,c1067420,102,c6bd4800,c55e1a00,...) at malloc+0x88
 nv_alloc_system_pages(c55e1d00,cf,0,c6bd4858,c6bd4838,...) at 
nv_alloc_system_pages+0x5d
 nv_alloc_pages(c55e1d00,cf,0,0,0,...) at nv_alloc_pages+0x115
 _nv003436rm(c55f3c00,0,c9024000,20,cf567000,...) at _nv003436rm+0x3b

Am I correct to interpret this as simply "out of memory"?  If so, presumably 
the solution is make more kmem available.

What can I do to make ZFS and nvidia work together?  Does it work for anybody?

Cheers,

-- Norbert.

-------------------------------------------------------

-------------------------------------------------------