Date: Sun, 17 Feb 2002 00:29:22 -0800 (PST) From: Matthew Dillon <dillon@apollo.backplane.com> To: Julian Elischer <julian@elischer.org> Cc: Alfred Perlstein <bright@mu.org>, Poul-Henning Kamp <phk@critter.freebsd.dk>, arch@FreeBSD.ORG, jhb@FreeBSD.ORG, peter@wemm.org, jake@locore.ca Subject: Re: gettimeofday() and crhold()/crfree() (was Re: gettimeofday() Message-ID: <200202170829.g1H8TME68268@apollo.backplane.com> References: <Pine.BSF.4.21.0202162354060.39539-100000@InterJet.elischer.org>
next in thread | previous in thread | raw e-mail | index | archive | help
:either way, could you do you timing tests on a NON-INVARIANT
:kernel to judge the difference this makes?
:In the mean while, John, is it really necessary to have this there?
:I can remove it (and the KASSERTS) in a flash if you'll let me...
Sure, here they are... actually fairly significant, I didn't expect
complete removal of invariants to have this much of an effect. I fear
someone has done something with invariants that's eating more cpu then
it should :-(
As a bonus, I've also included the patch set I am using against the
current -current which identifies the KASSERTs and INVARIANTS I
removed to make your code work as advertised with INVARIANTS turned on.
This patch set is fairly well tested.
I did not include the patch for removing Giant from gettimeofday()
(I'll be comitting the gettimeofday giant removal patch on sunday).
I'll also re-run the buildworld test with invariants turned off and
post a followup to my buildworld posting tomorrow.
-Matt
(matt's mutex pool adjustments and julian's initial commit now part of
current)
CURRENT 2xCPU SMP BUILD and removal of Giant for gettimeofday()
INVARIANTS TURNED OFF GLOBALLY.
one TG running: 576089/sec
two TGs running: 429141/sec per TG
CURRENT 2xCPU SMP BUILD and ucred invariants blasted to bits
and removal of Giant for gettimeofday().
INVARIANTS TURNED ON GLOBALLY
one TG running: 462478/sec
two TGs running: 361962/sec per TG
Index: i386/i386/trap.c
===================================================================
RCS file: /home/ncvs/src/sys/i386/i386/trap.c,v
retrieving revision 1.212
diff -u -r1.212 trap.c
--- i386/i386/trap.c 17 Feb 2002 01:09:54 -0000 1.212
+++ i386/i386/trap.c 17 Feb 2002 08:19:55 -0000
@@ -255,7 +255,9 @@
sticks = td->td_kse->ke_sticks;
td->td_frame = &frame;
+#if 0
KASSERT(td->td_ucred == NULL, ("already have a ucred"));
+#endif
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
@@ -643,7 +645,7 @@
userret(td, &frame, sticks);
mtx_assert(&Giant, MA_NOTOWNED);
userout:
-#ifdef INVARIANTS
+#if 0
mtx_lock(&Giant);
crfree(td->td_ucred);
mtx_unlock(&Giant);
@@ -954,7 +956,9 @@
sticks = td->td_kse->ke_sticks;
td->td_frame = &frame;
+#if 0
KASSERT(td->td_ucred == NULL, ("already have a ucred"));
+#endif
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
params = (caddr_t)frame.tf_esp + sizeof(int);
@@ -1099,7 +1103,7 @@
*/
STOPEVENT(p, S_SCX, code);
-#ifdef INVARIANTS
+#if 0
mtx_lock(&Giant);
crfree(td->td_ucred);
mtx_unlock(&Giant);
Index: kern/kern_fork.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/kern_fork.c,v
retrieving revision 1.131
diff -u -r1.131 kern_fork.c
--- kern/kern_fork.c 17 Feb 2002 01:09:55 -0000 1.131
+++ kern/kern_fork.c 17 Feb 2002 08:19:55 -0000
@@ -799,7 +799,7 @@
kthread_exit(0);
}
PROC_UNLOCK(p);
-#ifdef INVARIANTS
+#if 0
mtx_lock(&Giant);
crfree(td->td_ucred);
mtx_unlock(&Giant);
Index: kern/subr_trap.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/subr_trap.c,v
retrieving revision 1.208
diff -u -r1.208 subr_trap.c
--- kern/subr_trap.c 17 Feb 2002 01:09:55 -0000 1.208
+++ kern/subr_trap.c 17 Feb 2002 08:19:57 -0000
@@ -131,7 +131,9 @@
#endif
KASSERT(TRAPF_USERMODE(framep), ("ast in kernel mode"));
+#if 0
KASSERT(td->td_ucred == NULL, ("leaked ucred"));
+#endif
#ifdef WITNESS
if (witness_list(td))
panic("Returning to user mode with mutex(s) held");
@@ -187,7 +189,7 @@
}
userret(td, framep, sticks);
-#ifdef INVARIANTS
+#if 0
mtx_lock(&Giant);
crfree(td->td_ucred);
mtx_unlock(&Giant);
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-arch" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200202170829.g1H8TME68268>