git: 138a5dafba31 - main - vfs: trylock vnode requeue

From: Mateusz Guzik <mjg_at_FreeBSD.org>
Date: Sat, 25 Mar 2023 13:42:34 UTC
The branch main has been updated by mjg:

URL: https://cgit.FreeBSD.org/src/commit/?id=138a5dafba312ff39ce0eefdbe34de95519e600d

commit 138a5dafba312ff39ce0eefdbe34de95519e600d
Author:     Mateusz Guzik <mjg@FreeBSD.org>
AuthorDate: 2023-03-21 04:23:15 +0000
Commit:     Mateusz Guzik <mjg@FreeBSD.org>
CommitDate: 2023-03-25 13:42:27 +0000

    vfs: trylock vnode requeue
    
    The quasi-LRU still gets in the way for example when doing an
    incremental bzImage build, with vnode_list lock being at the
    top of the profile. Further damage control the problem by trylocking.
    
    Note the entire mechanism desperately wants to be reaped out in favor
    of something(tm) which both scales in a multicore setting and provides
    sensible replacement policy.
    
    With this change everything vfs almost disappears from the on CPU
    flamegraph, what is left is tons of contention in the VM.
---
 sys/kern/vfs_subr.c | 26 +++++++++++++++++---------
 1 file changed, 17 insertions(+), 9 deletions(-)

diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
index 72be478e310d..48aea3b6c8ff 100644
--- a/sys/kern/vfs_subr.c
+++ b/sys/kern/vfs_subr.c
@@ -3521,17 +3521,25 @@ vdbatch_process(struct vdbatch *vd)
 	MPASS(curthread->td_pinned > 0);
 	MPASS(vd->index == VDBATCH_SIZE);
 
-	mtx_lock(&vnode_list_mtx);
 	critical_enter();
-	for (i = 0; i < VDBATCH_SIZE; i++) {
-		vp = vd->tab[i];
-		TAILQ_REMOVE(&vnode_list, vp, v_vnodelist);
-		TAILQ_INSERT_TAIL(&vnode_list, vp, v_vnodelist);
-		MPASS(vp->v_dbatchcpu != NOCPU);
-		vp->v_dbatchcpu = NOCPU;
+	if (mtx_trylock(&vnode_list_mtx)) {
+		for (i = 0; i < VDBATCH_SIZE; i++) {
+			vp = vd->tab[i];
+			vd->tab[i] = NULL;
+			TAILQ_REMOVE(&vnode_list, vp, v_vnodelist);
+			TAILQ_INSERT_TAIL(&vnode_list, vp, v_vnodelist);
+			MPASS(vp->v_dbatchcpu != NOCPU);
+			vp->v_dbatchcpu = NOCPU;
+		}
+		mtx_unlock(&vnode_list_mtx);
+	} else {
+		for (i = 0; i < VDBATCH_SIZE; i++) {
+			vp = vd->tab[i];
+			vd->tab[i] = NULL;
+			MPASS(vp->v_dbatchcpu != NOCPU);
+			vp->v_dbatchcpu = NOCPU;
+		}
 	}
-	mtx_unlock(&vnode_list_mtx);
-	bzero(vd->tab, sizeof(vd->tab));
 	vd->index = 0;
 	critical_exit();
 }