Jürgen Keil wrote:
kmdb ::memstat showed that almost 100% of memory was in use by the
kernel, so most likely there is a memory leak somewhere.
Shot in the dark here, but maybe 6786794?
could be. The bug's submit date matches
the date when I first observed the problem (IIRC) ...
If you have a dump it can easily be verified but first see below
No dumps, that's the problem.
I did use $<systemdump from kmdb, but
somehow that failed half way through the dump.
(disk driver in use for the dump was nv_sata)
When it happened a second time I tried to collect some
facts directly from kmdb before trying to get the dump,
but kmdb got some internal error and all I could do was
reset the machine...
That box is transferring lots of backup data over
tcp, and is also used as an nfs client.
The above bug only happens if you are using a
protocol that does not have it's own flow control like UDP/ICMP etc.
Most of the data should be transferred over TCP (nfs & amanda).
If you provide a crash dump I can look at it and verify if you are
seeing the same problem, or you can easily do so by verifying if the
socket is UDP socket , it has lots of data in it's queue and if it's
in a drain list or not (conn_idl != NULL).
Sorry, no dump is available.
It would be helpful to know which cache is using up
the memory.
IIRC it was the kmem_va_4096 cache...
So it most likely not networking related. How about dumping the log and
finding out who is allocating the memory.
> ::walk kmem_va_4096
ffffff01e4000000
ffffff01e4001000
<............>
> ffffff01e4000000::whatis -b
ffffff01e4000000 is ffffff01e4000000+0, bufctl ffffff01e3de1be8 freed from
anon_cache
> ffffff01e3de1be8::bufctl -v
ADDR BUFADDR TIMESTAMP THREAD
CACHE LASTLOG CONTENTS
ffffff01e3de1be8 ffffff01e4000000 20d4e25747 ffffff01d2d2e780
ffffff01ceb596a0 ffffff01c6b10d00 ffffff01cb3db0c0
kmem_cache_free_debug+0x12f
kmem_cache_free+0x98
anon_decref+0x122
anon_free+0x9f
segvn_free+0x1db
seg_free+0x3c
segvn_unmap+0xec1
as_free+0x117
relvm+0x116
proc_exit+0x4b0
exit+0x15
rexit+0x1c
Rao.
_______________________________________________
opensolaris-code mailing list
opensolaris-code@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
