Andriy Gapon wrote:
> on 23/02/2010 12:03 Санек Гурин said the following:
>> FreeBSD freeze
>>
>> After update few weeks ago freebsd began to freeze from time to time. Can't
>> find any reason in logs or other way. System may halt after an hour of work,
>> or may work few days. Before halt it may respond to shell commands very-very
>> slow....
>>
>> MB: MSI P45 NEO3-FR. BIOS ATA configured as AHCI. JMicron ATA controller is
>> disabled.
>>
>> % uname -a
>> FreeBSD wk.kiev.ua 8.0-STABLE FreeBSD 8.0-STABLE #3 r204163: Sun Feb 21
>> 10:07:59 EET 2010 [email protected]:/usr/obj/usr/src/sys/GNRDBG amd64
>>
>> Kernel Config: http://xdevs.com/wk/kernel_config.txt
>> dmesg output: http://xdevs.com/wk/dmesg.txt
>>
>>
>> panic: _mtx_lock_sleep: recursed on non-recursive mutex ATA state lock @
>> /usr/sys/dev/ata/ata-all.c:334
>>
>> cpuid = 1
>> KDB: stack backtrace:
>> db_trace_self_wrapper() at 0xffffffff801e05c5a = db_trace_self_wrapper+0x2a
>> panic() at 0xffffffff8058d692 = panic+0x182
>> _mtx_lock_sleep() at 0xffffffff8057ec52 = _mtx_lock_sleep+0x152
>> _mtx_lock_flags() at 0xffffffff8057ed41 = _mtx_lock_flags+exe1
>> ata_reinit() at 0xffffffff802723f3 = ata_reinit+0xb3
>> ata_conn_event() at 0xffffffff80272b7e = ata_conn_event+0xe3
>> taskqueue_run() at 0xffffffff805c9e01 = taskqueue_run+0x91
>> taskqueue_thread_loop() at 0xffffffff805c9f8f = taskqueue_thread_loop+0x3f
>> fork_exit() at 0xffffffff80564f3a = fork_exit+0x12a
>> fork_trampoline() at 0xffffffff8085b84e = fork_trampoline+0xe
>> --- trap 0, rip = 0, rsp = 0xffffff80000b9d30, rbp = 0 ---
>> Uptime: 19s
>> Cannot dump. Device not defined or unavailable.
>> Automatic reboot in 15 seconds - press a key on the console to abort
>> VOP_STRATEGY: bp in not locked but should be
>
> I believe that this a (well known?) bug in ata driver: state_mtx is not
> initialized as recursive, but is used as such.
> E.g. it is locked in ata_conn_event and then ata_reinit locks it again.
> Of course, an external condition is needed to trigger ata_conn_event in the
> first
> place.
Looks like my fault. Attached patch should fix that. But if you need
really working hot-plug (and many other tasty things) - look to the new
CAM-based ATA implementation.
--
Alexander Motin
--- ata-all.c.prev 2010-02-23 09:17:04.000000000 +0200
+++ ata-all.c 2010-02-23 16:23:49.000000000 +0200
@@ -289,15 +289,13 @@ static void
ata_conn_event(void *context, int dummy)
{
device_t dev = (device_t)context;
- struct ata_channel *ch = device_get_softc(dev);
#ifdef ATA_CAM
+ struct ata_channel *ch = device_get_softc(dev);
union ccb *ccb;
-#endif
mtx_lock(&ch->state_mtx);
ata_reinit(dev);
mtx_unlock(&ch->state_mtx);
-#ifdef ATA_CAM
if ((ccb = xpt_alloc_ccb()) == NULL)
return;
if (xpt_create_path(&ccb->ccb_h.path, NULL,
@@ -307,6 +305,8 @@ ata_conn_event(void *context, int dummy)
return;
}
xpt_rescan(ccb);
+#else
+ ata_reinit(dev);
#endif
}
_______________________________________________
[email protected] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-stable
To unsubscribe, send any mail to "[email protected]"
