On Tue, Dec 26, 2017 at 5:23 PM, Ryota Ozaki <ozak...@netbsd.org> wrote:
> On Tue, Dec 26, 2017 at 4:50 PM, Tom Ivar Helbekkmo
> <t...@hamartun.priv.no> wrote:
>> Ryota Ozaki <ozak...@netbsd.org> writes:
>>
>>> One possible fix has been committed.
>>>
>>> Can you update the source code and try a new kernel?
>>
>> Will do.
>
> Thanks.
>
>
>>
>> Meanwhile, before I got around to building a kernel with debug options
>> enabled, I had another hang. Got it into DDB successfully, but then I
>> managed, while looking for a way to switch between CPUs (the man page is
>> wrong in this respect), to get DDB to look at something it shouldn't
>> have, so it just said "fatal pag", and that was that.
>>
>> I did get a backtrace of CPU 0, though, and it looked interesting:
>>
>> pool_catchup()
>> pool_get()
>> pool_cache_get_slow()
>> pool_cache_get_paddr()
>> m_get()
>> m_gethdr()
>> wm_add_rxbuf()
>> wm_rxeof()
>> wm_intr_legacy()
>> intr_biglock_wrapper()
>> Xintr_ioapic_level2()
>> --- interrupt ---
>> Xspllower()
>> uvm_km_kmem_alloc()
>> pool_page_alloc()
>> pool_grow()
>> pool_catchup()
>> pool_get()
>> pool_cache_get_slow()
>> pool_cache_get_paddr()
>> m_get()
>> m_gethdr()
>> tcp_output()
>> tcp_send_wrapper()
>> sosend()
>> soo_write()
>> dofilewrite()
>> sys_write()
>> syscall()
>> --- syscall (number 4) ---
>
> Looks the below infinite loop is happening?
>
> I think we need to summon a pool expert.
>
> ozaki-r
>
>
> (Copied and modified the diagram from PR 52858)
>
> [lwp #1]
> |
> [pool_grow with PR_NOWAIT
> [set PR_GROWING and PR_GROWINGNOWAIT
> [mutex_exit(&pp->pr_lock)
> |
> (interrupted)
>
> [intr #1]
> |
> [pool_catchup
> [pool_grow with PR_NOWAIT
> [see PR_GROWING and PR_GROWINGNOWAIT are set
> [return ERESTART
> [repeat pool_grow in pool_catchup...
I think the below patch fixes the above issue, but probably
there is a better solution.
ozaki-r
diff --git a/sys/kern/subr_pool.c b/sys/kern/subr_pool.c
index 3e6c9225482..e4be8bf0682 100644
--- a/sys/kern/subr_pool.c
+++ b/sys/kern/subr_pool.c
@@ -1247,6 +1247,7 @@ static int
pool_catchup(struct pool *pp)
{
int error = 0;
+ int s = splvm();
while (POOL_NEEDS_CATCHUP(pp)) {
error = pool_grow(pp, PR_NOWAIT);
@@ -1256,6 +1257,7 @@ pool_catchup(struct pool *pp)
break;
}
}
+ splx(s);
return error;
}
