diff options
| author | Owain Ainsworth <oga@cvs.openbsd.org> | 2009-06-02 23:00:20 +0000 |
|---|---|---|
| committer | Owain Ainsworth <oga@cvs.openbsd.org> | 2009-06-02 23:00:20 +0000 |
| commit | 6e1481035fcdd30fd7cc0e2650ad2e51cabb8d9d (patch) | |
| tree | c0a17586063eb7ab1d1f02beb45592aa9705ab6c /sys/uvm/uvm_device.c | |
| parent | a35c336f9ebdea1410b999c9869e68c40354c1e3 (diff) | |
Instead of the global hash table with the terrible hashfunction and a
global lock, switch the uvm object pages to being kept in a per-object
RB_TREE. Right now this is approximately the same speed, but cleaner.
When biglock usage is reduced this will improve concurrency due to lock
contention..
ok beck@ art@. Thanks to jasper for the speed testing.
Diffstat (limited to 'sys/uvm/uvm_device.c')
| -rw-r--r-- | sys/uvm/uvm_device.c | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/sys/uvm/uvm_device.c b/sys/uvm/uvm_device.c index 1013f46d20a..a3743490068 100644 --- a/sys/uvm/uvm_device.c +++ b/sys/uvm/uvm_device.c @@ -1,4 +1,4 @@ -/* $OpenBSD: uvm_device.c,v 1.32 2009/05/12 20:49:56 oga Exp $ */ +/* $OpenBSD: uvm_device.c,v 1.33 2009/06/02 23:00:19 oga Exp $ */ /* $NetBSD: uvm_device.c,v 1.30 2000/11/25 06:27:59 chs Exp $ */ /* @@ -227,7 +227,7 @@ udv_attach(void *arg, vm_prot_t accessprot, voff_t off, vsize_t size) simple_lock_init(&udv->u_obj.vmobjlock); udv->u_obj.pgops = &uvm_deviceops; - TAILQ_INIT(&udv->u_obj.memq); + RB_INIT(&udv->u_obj.memt); udv->u_obj.uo_npages = 0; udv->u_obj.uo_refs = 1; udv->u_flags = 0; @@ -287,7 +287,7 @@ again: uobj,uobj->uo_refs,0,0); return; } - KASSERT(uobj->uo_npages == 0 && TAILQ_EMPTY(&uobj->memq)); + KASSERT(uobj->uo_npages == 0 && RB_EMPTY(&uobj->memt)); /* * is it being held? if so, wait until others are done. |