Date: Wed, 21 Jan 2015 21:04:51 -0500 From: Ryan Stone <rysto32@gmail.com> To: Konstantin Belousov <kostikbel@gmail.com> Cc: "freebsd-hackers@freebsd.org" <freebsd-hackers@freebsd.org> Subject: Re: Sleeping thread held mutex in vm_pageout_oom() Message-ID: <CAFMmRNz6Dof%2B95zyekUFEbmcqeFSqXr96Y7BUZpqJ9z0XoWVOA@mail.gmail.com> In-Reply-To: <20150120083212.GC42409@kib.kiev.ua> References: <CAFMmRNxz252HMWWBmRf=Z69zh2_w9cD5e1AZGeizyagKezm2Hw@mail.gmail.com> <20150120083212.GC42409@kib.kiev.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
Thanks for the patch. I tried it out this afternoon and got the
following panic:
panic: PHOLD of exiting process^M^M
cpuid = 9^M^M
curthread = pagedaemon/pagedaemon (8/100154)^M^M
cpu_ticks = 979566002403^M^M
KDB: stack backtrace:^M^M
db_trace_self_wrapper() at 0xffffffff801e672a = db_trace_self_wrapper+0x2a^M^M
panic() at 0xffffffff80480818 = panic+0x228^M^M
vm_pageout_oom() at 0xffffffff80669a9b = vm_pageout_oom+0x58b^M^M
vm_pageout() at 0xffffffff8066aa35 = vm_pageout+0x975^M^M
fork_exit() at 0xffffffff80454ada = fork_exit+0x12a^M^M
fork_trampoline() at 0xffffffff8067b37e = fork_trampoline+0xe^M^M
I believe that we need to check for P_WEXIT before considering p as a
candidate to be killed:
diff --git a/sys/vm/vm_pageout.c b/sys/vm/vm_pageout.c
index ca9d7f9..64d4277 100644
--- a/sys/vm/vm_pageout.c
+++ b/sys/vm/vm_pageout.c
@@ -1516,13 +1516,13 @@ vm_pageout_oom(int shortage)
FOREACH_PROC_IN_SYSTEM(p) {
int breakout;
- if (PROC_TRYLOCK(p) == 0)
- continue;
+ PROC_LOCK(p);
/*
* If this is a system, protected or killed process, skip it.
*/
if (p->p_state != PRS_NORMAL ||
- (p->p_flag & (P_INEXEC | P_PROTECTED | P_SYSTEM)) ||
+ (p->p_flag &
+ (P_INEXEC | P_PROTECTED | P_SYSTEM | P_WEXIT)) ||
(p->p_pid == 1) || P_KILLED(p) ||
((p->p_pid < 48) && (swap_pager_avail != 0))) {
PROC_UNLOCK(p);
@@ -1557,11 +1557,14 @@ vm_pageout_oom(int shortage)
PROC_UNLOCK(p);
continue;
}
+ _PHOLD(p);
if (!vm_map_trylock_read(&vm->vm_map)) {
- vmspace_free(vm);
+ _PRELE(p);
PROC_UNLOCK(p);
+ vmspace_free(vm);
continue;
}
+ PROC_UNLOCK(p);
size = vmspace_swap_count(vm);
vm_map_unlock_read(&vm->vm_map);
if (shortage == VM_OOM_MEM)
@@ -1573,16 +1576,18 @@ vm_pageout_oom(int shortage)
*/
if (size > bigsize) {
if (bigproc != NULL)
- PROC_UNLOCK(bigproc);
+ PRELE(bigproc);
bigproc = p;
bigsize = size;
} else
- PROC_UNLOCK(p);
+ PRELE(p);
}
sx_sunlock(&allproc_lock);
if (bigproc != NULL) {
+ PROC_LOCK(bigproc);
killproc(bigproc, "out of swap space");
sched_nice(bigproc, PRIO_MIN);
+ _PRELE(bigproc);
PROC_UNLOCK(bigproc);
wakeup(&vm_cnt.v_free_count);
}
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAFMmRNz6Dof%2B95zyekUFEbmcqeFSqXr96Y7BUZpqJ9z0XoWVOA>