Date: Sat, 14 Dec 2002 18:09:01 -0800 (PST) From: Matthew Dillon <dillon@apollo.backplane.com> To: "Brian F. Feldman" <green@FreeBSD.ORG>, Jake Burkholder <jake@locore.ca>, "Brian F. Feldman" <green@FreeBSD.ORG>, John Baldwin <jhb@FreeBSD.ORG>, Kris Kennaway <kris@obsecurity.org>, current@FreeBSD.ORG, alpha@FreeBSD.ORG Subject: Patch #3 Re: UMA panic under load Message-ID: <200212150209.gBF291Zf078019@apollo.backplane.com> References: <200212150121.gBF1L15m014304@green.bikeshed.org> <200212150134.gBF1Yit5060312@apollo.backplane.com> <200212150141.gBF1fHJc060370@apollo.backplane.com> <200212150143.gBF1hERJ060388@apollo.backplane.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Whoop. Ok, here's a new patch. I think this covers all the cases.
I've done some testing and it appears to do the right thing, please
look it over (the last patch had type-o's and didn't cover the correct
cases).
-Matt
Index: kern/kern_exit.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/kern_exit.c,v
retrieving revision 1.187
diff -u -r1.187 kern_exit.c
--- kern/kern_exit.c 10 Dec 2002 02:33:44 -0000 1.187
+++ kern/kern_exit.c 15 Dec 2002 01:45:15 -0000
@@ -287,7 +287,15 @@
* Need to do this early enough that we can still sleep.
* Can't free the entire vmspace as the kernel stack
* may be mapped within that space also.
+ *
+ * Processes sharing the same vmspace may exit in one order, and
+ * get cleaned up by vmspace_exit() in a different order. The
+ * last exiting process to reach this point releases as much of
+ * the environment as it can, and the last process cleaned up
+ * by vmspace_exit() (which decrements exitingcnt) cleans up the
+ * remainder.
*/
+ ++vm->vm_exitingcnt;
if (--vm->vm_refcnt == 0) {
if (vm->vm_shm)
shmexit(p);
@@ -297,7 +305,6 @@
vm_page_unlock_queues();
(void) vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
vm_map_max(&vm->vm_map));
- vm->vm_freer = p;
}
sx_xlock(&proctree_lock);
Index: vm/vm_map.c
===================================================================
RCS file: /home/ncvs/src/sys/vm/vm_map.c,v
retrieving revision 1.273
diff -u -r1.273 vm_map.c
--- vm/vm_map.c 1 Dec 2002 18:57:56 -0000 1.273
+++ vm/vm_map.c 15 Dec 2002 02:05:13 -0000
@@ -258,7 +258,7 @@
vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
vm->vm_refcnt = 1;
vm->vm_shm = NULL;
- vm->vm_freer = NULL;
+ vm->vm_exitingcnt = 0;
return (vm);
}
@@ -304,7 +304,7 @@
if (vm->vm_refcnt == 0)
panic("vmspace_free: attempt to free already freed vmspace");
- if (--vm->vm_refcnt == 0)
+ if (--vm->vm_refcnt == 0 && vm->vm_exitingcnt == 0)
vmspace_dofree(vm);
}
@@ -314,11 +314,22 @@
struct vmspace *vm;
GIANT_REQUIRED;
- if (p == p->p_vmspace->vm_freer) {
- vm = p->p_vmspace;
- p->p_vmspace = NULL;
+ vm = p->p_vmspace;
+ p->p_vmspace = NULL;
+
+ /*
+ * cleanup by parent process wait()ing on exiting child. vm_refcnt
+ * may not be 0 (e.g. fork() and child exits without exec()ing).
+ * exitingcnt may increment above 0 and drop back down to zero
+ * several times while vm_refcnt is held non-zero. vm_refcnt
+ * may also increment above 0 and drop back down to zero several
+ * times while vm_exitingcnt is held non-zero.
+ *
+ * The last wait on the exiting child's vmspace will clean up
+ * the remainder of the vmspace.
+ */
+ if (--vm->vm_exitingcnt == 0 && vm->vm_refcnt == 0)
vmspace_dofree(vm);
- }
}
/*
Index: vm/vm_map.h
===================================================================
RCS file: /home/ncvs/src/sys/vm/vm_map.h,v
retrieving revision 1.92
diff -u -r1.92 vm_map.h
--- vm/vm_map.h 22 Sep 2002 04:33:43 -0000 1.92
+++ vm/vm_map.h 15 Dec 2002 01:47:49 -0000
@@ -218,8 +218,8 @@
caddr_t vm_taddr; /* (c) user virtual address of text */
caddr_t vm_daddr; /* (c) user virtual address of data */
caddr_t vm_maxsaddr; /* user VA at max stack growth */
-#define vm_endcopy vm_freer
- struct proc *vm_freer; /* vm freed on whose behalf */
+#define vm_endcopy vm_exitingcnt
+ int vm_exitingcnt; /* several processes zombied in exit1 */
};
#ifdef _KERNEL
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200212150209.gBF291Zf078019>