Call for Testers: Convert watchdog spinlock into a sleepable mutex in if_ndis

Paul B. Mahol onemda at
Sun Jun 8 01:55:12 UTC 2008

On 6/7/08, Coleman Kane <cokane at> wrote:
> Hi,
> I've got another patch to if_ndis and I'd like to know if any NDIS users
> out there can test it for me and give me some feedback. I've converted
> the ndis_spinlock in if_ndis into a normal sleepable mutex, and would
> like to get some testers. The idea here is to eliminate an unnecessary
> "spinning" in the kernel, where it would be preferable to sleep (for CPU
> usage, interactivity, etc...).
> You can download it here:
>       *
> --
> Coleman Kane

I did not found any stability regressions in my environment with linked patch.

But I noticed   new   LOR (it appears as soon as radio of device is turned on):

lock order reversal:
 1st 0xc401726c ndis0 (network driver) @
 2nd 0xc09b3e74 HAL preemption lock (HAL lock) @
KDB: stack backtrace:
at db_trace_self_wrapper+0x26
kdb_backtrace(c070adf6,c09b3e74,c09b0a51,c09b0a48,c09b09f8,...) at
witness_checkorder(c09b3e74,9,c09b09f8,1a7,c04f7364,...) at
_mtx_lock_flags(c09b3e74,0,c09b09f8,1a7,c457ebd0,...) at _mtx_lock_flags+0xbc
KfRaiseIrql(2,c457ebd0,c3985c48,c09aacdd,c3d45354,...) at KfRaiseIrql+0x63
KfAcquireSpinLock(c3d45354,c094888a,c4017200,c3d6cc40,c4017200,...) at
IoQueueWorkItem(c457ebd0,c3d6cc40,0,c4017200,c07040ef,...) at
ndis_tick(c4017200,0,c070699c,166,c077e594,...) at ndis_tick+0x143
softclock(c077e560,0,c0701ca6,4d4,c3cda6ec,...) at softclock+0x24a
ithread_loop(c3cdb2e0,c3985d38,c0701a09,324,c3c99a70,...) at ithread_loop+0x1b5
fork_exit(c04e74f0,c3cdb2e0,c3985d38) at fork_exit+0xb8
fork_trampoline() at fork_trampoline+0x8
--- trap 0, eip = 0, esp = 0xc3985d70, ebp = 0 ---

More information about the freebsd-current mailing list