[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
kmem_malloc crash with 4.8
- Subject: kmem_malloc crash with 4.8
- From: freebsd-stable-local at be-well.no-ip.com (Lowell Gilbert)
- Date: Tue Apr 8 12:38:07 2003
Lamont Lucas <lamont_(_at_)_cluepon_(_dot_)_com> writes:
> Hi. I've recently borrowed a SuperMicro 6013P-8 from a vendor
> for the purposes of testing it to see if FreeBSD and these machines will
> make a good replacements for our current flock of 11 Netra T1s running
> solaris. Things are going very well with the tests, with the glaring
> exception of the following crash:
> While doing some significant disk operations: du, deleting a different
> part of the tree and finally running "sync" the machine locked hard
> with the following errors:
> panic: kmem_malloc(4096): kmem_map too small: 230162432 total allocated
> mp_lock = 03000001; cpuid = 3; lapic.id = 07000000
> boot() called on cpu#3
> Then "Synching disks" but no futher output is printed, nor is any
> further progress made.
> This machine was installed off of 4.8-release and was rebuilt under
> -stable as of april 4th, late in the afternoon. The kernel conf file
> is attached, but biggest changes I made were to enable SMP as well as
> hyperthreading. It has 2 gigs of ram and 2 2.4 ghz xeon processors
> capable of hyperthreading. The drive is a single <SEAGATE ST336753LC 0003>
> controlled by a AIC7902 Ultra 320 scsi adapter.
> I saw the previous discussion about the adaptec U320 controllers and
> scott long's advice about "lower[ing] the tag depth to 32" using
> camcontrol tags. I'm currently planning on removing hyperthread
> support and rerunning some of my tests to see if I can reproduce the
> error. I'm also going to try some bonnie tests. If I can
> consistantly reproduce the error I'll report back, but I'd appreciate
> knowing what additional debug info I can give back. I'm not clear
> what would cause this type of error.
> Attached are dmesg as well as the kernel config file.
> Can anybody recommend any other steps to try and eliminate or fix
> this problem?
You might want to reduce the maxusers value...