Date: Sat, 21 Aug 2004 18:59:39 -0400 From: Brian Fundakowski Feldman <green@FreeBSD.org> To: Alan Cox <alc@FreeBSD.org> Cc: cvs-all@FreeBSD.org Subject: Re: cvs commit: src/sys/vm vm_fault.c Message-ID: <20040821225939.GA784@green.homeunix.org> In-Reply-To: <200408211920.i7LJKLlA088634@repoman.freebsd.org> References: <200408211920.i7LJKLlA088634@repoman.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Aug 21, 2004 at 07:20:21PM +0000, Alan Cox wrote:
> alc 2004-08-21 19:20:21 UTC
>
> FreeBSD src repository
>
> Modified files:
> sys/vm vm_fault.c
> Log:
> Further reduce the use of Giant by vm_fault(): Giant is held only when
> manipulating a vnode, e.g., calling vput(). This reduces contention for
> Giant during many copy-on-write faults, resulting in some additional
> speedup on SMPs.
>
> Note: debug_mpsafevm must be enabled for this optimization to take effect.
This is very broken. See included first attempt at fixing it without
regard for actually trying to reimplement debug.mpsafenet for vnodes.
Index: vm_fault.c
===================================================================
RCS file: /usr/ncvs/src/sys/vm/vm_fault.c,v
retrieving revision 1.193
diff -u -r1.193 vm_fault.c
--- vm_fault.c 21 Aug 2004 19:20:21 -0000 1.193
+++ vm_fault.c 21 Aug 2004 22:31:57 -0000
@@ -172,8 +172,6 @@
mtx_unlock(&Giant);
fs->vp = NULL;
}
- if (dealloc)
- VM_UNLOCK_GIANT();
}
#define unlock_things(fs) _unlock_things(fs, 0)
@@ -231,11 +229,13 @@
* search.
*/
fs.map = map;
+ mtx_lock(&Giant);
result = vm_map_lookup(&fs.map, vaddr, fault_type, &fs.entry,
&fs.first_object, &fs.first_pindex, &prot, &wired);
if (result != KERN_SUCCESS) {
if (result != KERN_PROTECTION_FAILURE ||
(fault_flags & VM_FAULT_WIRE_MASK) != VM_FAULT_USER_WIRE) {
+ mtx_unlock(&Giant);
if (growstack && result == KERN_INVALID_ADDRESS &&
map != kernel_map && curproc != NULL) {
result = vm_map_growstack(curproc, vaddr);
@@ -257,8 +257,10 @@
result = vm_map_lookup(&fs.map, vaddr,
VM_PROT_READ|VM_PROT_WRITE|VM_PROT_OVERRIDE_WRITE,
&fs.entry, &fs.first_object, &fs.first_pindex, &prot, &wired);
- if (result != KERN_SUCCESS)
+ if (result != KERN_SUCCESS) {
+ mtx_unlock(&Giant);
return (result);
+ }
/*
* If we don't COW now, on a user wire, the user will never
@@ -291,7 +293,6 @@
*
* XXX vnode_pager_lock() can block without releasing the map lock.
*/
- mtx_lock(&Giant);
VM_OBJECT_LOCK(fs.first_object);
vm_object_reference_locked(fs.first_object);
fs.vp = vnode_pager_lock(fs.first_object);
@@ -368,7 +369,6 @@
if (!vm_page_sleep_if_busy(fs.m, TRUE, "vmpfw"))
vm_page_unlock_queues();
atomic_add_int(&cnt.v_intrans, 1);
- VM_UNLOCK_GIANT();
vm_object_deallocate(fs.first_object);
goto RetryFault;
}
--
Brian Fundakowski Feldman \'[ FreeBSD ]''''''''''\
<> green@FreeBSD.org \ The Power to Serve! \
Opinions expressed are my own. \,,,,,,,,,,,,,,,,,,,,,,\
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040821225939.GA784>