Re: PLEASE TEST: IPI deadlock avoidance patch

On Mon, Aug 23, 2004 at 02:16:25PM -0400, John Baldwin wrote:
> On Sunday 22 August 2004 03:05 pm, Doug White wrote:
> > If you have a reasonably fast i386 or amd64 multiprocessor and/or
> > hyperthreading machine and are experiencing reproducible hangs during -j
> > buildwords and other highly parallel operations, please try this patch:
> > http://people.freebsd.org/~dwhite/smp_rv_mtx.patch
..
> Looks good to me.  I also think this might help the KSE + SMP hangs.  Only 
> suggestion I might have is to rename smp_rv_mtx to smp_ipi_mtx and "SMP IPI 
> lock" or some such.

Like this?

Index: amd64/amd64/mp_machdep.c
===================================================================
RCS file: /home/ncvs/src/sys/amd64/amd64/mp_machdep.c,v
retrieving revision 1.243
diff -u -r1.243 mp_machdep.c
--- amd64/amd64/mp_machdep.c	23 Aug 2004 21:39:28 -0000	1.243
+++ amd64/amd64/mp_machdep.c	24 Aug 2004 18:31:48 -0000
_at__at_ -708,7 +708,7 _at__at_
 	ncpu = mp_ncpus - 1;	/* does not shootdown self */
 	if (ncpu < 1)
 		return;		/* no other cpus */
-	mtx_assert(&smp_rv_mtx, MA_OWNED);
+	mtx_assert(&smp_ipi_mtx, MA_OWNED);
 	smp_tlb_addr1 = addr1;
 	smp_tlb_addr2 = addr2;
 	atomic_store_rel_int(&smp_tlb_wait, 0);
_at__at_ -794,7 +794,7 _at__at_
 		if (ncpu < 1)
 			return;
 	}
-	mtx_assert(&smp_rv_mtx, MA_OWNED);
+	mtx_assert(&smp_ipi_mtx, MA_OWNED);
 	smp_tlb_addr1 = addr1;
 	smp_tlb_addr2 = addr2;
 	atomic_store_rel_int(&smp_tlb_wait, 0);
Index: amd64/amd64/pmap.c
===================================================================
RCS file: /home/ncvs/src/sys/amd64/amd64/pmap.c,v
retrieving revision 1.493
diff -u -r1.493 pmap.c
--- amd64/amd64/pmap.c	24 Aug 2004 00:17:52 -0000	1.493
+++ amd64/amd64/pmap.c	24 Aug 2004 18:31:48 -0000
_at__at_ -631,7 +631,7 _at__at_
 	if (smp_started) {
 		if (!(read_rflags() & PSL_I))
 			panic("%s: interrupts disabled", __func__);
-		mtx_lock_spin(&smp_rv_mtx);
+		mtx_lock_spin(&smp_ipi_mtx);
 	} else
 		critical_enter();
 	/*
_at__at_ -652,7 +652,7 _at__at_
 			smp_masked_invlpg(pmap->pm_active & other_cpus, va);
 	}
 	if (smp_started)
-		mtx_unlock_spin(&smp_rv_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
 	else
 		critical_exit();
 }
_at__at_ -667,7 +667,7 _at__at_
 	if (smp_started) {
 		if (!(read_rflags() & PSL_I))
 			panic("%s: interrupts disabled", __func__);
-		mtx_lock_spin(&smp_rv_mtx);
+		mtx_lock_spin(&smp_ipi_mtx);
 	} else
 		critical_enter();
 	/*
_at__at_ -691,7 +691,7 _at__at_
 			    sva, eva);
 	}
 	if (smp_started)
-		mtx_unlock_spin(&smp_rv_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
 	else
 		critical_exit();
 }
_at__at_ -705,7 +705,7 _at__at_
 	if (smp_started) {
 		if (!(read_rflags() & PSL_I))
 			panic("%s: interrupts disabled", __func__);
-		mtx_lock_spin(&smp_rv_mtx);
+		mtx_lock_spin(&smp_ipi_mtx);
 	} else
 		critical_enter();
 	/*
_at__at_ -726,7 +726,7 _at__at_
 			smp_masked_invltlb(pmap->pm_active & other_cpus);
 	}
 	if (smp_started)
-		mtx_unlock_spin(&smp_rv_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
 	else
 		critical_exit();
 }
Index: i386/i386/mp_machdep.c
===================================================================
RCS file: /home/ncvs/src/sys/i386/i386/mp_machdep.c,v
retrieving revision 1.236
diff -u -r1.236 mp_machdep.c
--- i386/i386/mp_machdep.c	23 Aug 2004 21:39:28 -0000	1.236
+++ i386/i386/mp_machdep.c	24 Aug 2004 18:31:48 -0000
_at__at_ -910,7 +910,7 _at__at_
 	ncpu = mp_ncpus - 1;	/* does not shootdown self */
 	if (ncpu < 1)
 		return;		/* no other cpus */
-	mtx_assert(&smp_rv_mtx, MA_OWNED);
+	mtx_assert(&smp_ipi_mtx, MA_OWNED);
 	smp_tlb_addr1 = addr1;
 	smp_tlb_addr2 = addr2;
 	atomic_store_rel_int(&smp_tlb_wait, 0);
_at__at_ -996,7 +996,7 _at__at_
 		if (ncpu < 1)
 			return;
 	}
-	mtx_assert(&smp_rv_mtx, MA_OWNED);
+	mtx_assert(&smp_ipi_mtx, MA_OWNED);
 	smp_tlb_addr1 = addr1;
 	smp_tlb_addr2 = addr2;
 	atomic_store_rel_int(&smp_tlb_wait, 0);
Index: i386/i386/pmap.c
===================================================================
RCS file: /home/ncvs/src/sys/i386/i386/pmap.c,v
retrieving revision 1.495
diff -u -r1.495 pmap.c
--- i386/i386/pmap.c	23 Aug 2004 21:39:28 -0000	1.495
+++ i386/i386/pmap.c	24 Aug 2004 18:31:48 -0000
_at__at_ -582,7 +582,7 _at__at_
 	if (smp_started) {
 		if (!(read_eflags() & PSL_I))
 			panic("%s: interrupts disabled", __func__);
-		mtx_lock_spin(&smp_rv_mtx);
+		mtx_lock_spin(&smp_ipi_mtx);
 	} else
 		critical_enter();
 	/*
_at__at_ -603,7 +603,7 _at__at_
 			smp_masked_invlpg(pmap->pm_active & other_cpus, va);
 	}
 	if (smp_started)
-		mtx_unlock_spin(&smp_rv_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
 	else
 		critical_exit();
 }
_at__at_ -618,7 +618,7 _at__at_
 	if (smp_started) {
 		if (!(read_eflags() & PSL_I))
 			panic("%s: interrupts disabled", __func__);
-		mtx_lock_spin(&smp_rv_mtx);
+		mtx_lock_spin(&smp_ipi_mtx);
 	} else
 		critical_enter();
 	/*
_at__at_ -642,7 +642,7 _at__at_
 			    sva, eva);
 	}
 	if (smp_started)
-		mtx_unlock_spin(&smp_rv_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
 	else
 		critical_exit();
 }
_at__at_ -656,7 +656,7 _at__at_
 	if (smp_started) {
 		if (!(read_eflags() & PSL_I))
 			panic("%s: interrupts disabled", __func__);
-		mtx_lock_spin(&smp_rv_mtx);
+		mtx_lock_spin(&smp_ipi_mtx);
 	} else
 		critical_enter();
 	/*
_at__at_ -677,7 +677,7 _at__at_
 			smp_masked_invltlb(pmap->pm_active & other_cpus);
 	}
 	if (smp_started)
-		mtx_unlock_spin(&smp_rv_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
 	else
 		critical_exit();
 }
_at__at_ -1301,7 +1301,7 _at__at_
 	while ((mask = pmap->pm_active) != 0) {
 		spins = 50000000;
 		mask = mask & -mask;	/* Find least significant set bit */
-		mtx_lock_spin(&smp_rv_mtx);
+		mtx_lock_spin(&smp_ipi_mtx);
 #ifdef PAE
 		lazyptd = vtophys(pmap->pm_pdpt);
 #else
_at__at_ -1321,7 +1321,7 _at__at_
 					break;
 			}
 		}
-		mtx_unlock_spin(&smp_rv_mtx);
+		mtx_unlock_spin(&smp_ipi_mtx);
 		if (spins == 0)
 			printf("pmap_lazyfix: spun for 50000000\n");
 	}
Index: kern/subr_smp.c
===================================================================
RCS file: /home/ncvs/src/sys/kern/subr_smp.c,v
retrieving revision 1.189
diff -u -r1.189 subr_smp.c
--- kern/subr_smp.c	23 Aug 2004 21:39:29 -0000	1.189
+++ kern/subr_smp.c	24 Aug 2004 18:31:48 -0000
_at__at_ -109,7 +109,7 _at__at_
  * functions trigger at once and cause multiple CPUs to busywait with
  * interrupts disabled. 
  */
-struct mtx smp_rv_mtx;
+struct mtx smp_ipi_mtx;
 
 /*
  * Let the MD SMP code initialize mp_maxid very early if it can.
_at__at_ -135,7 +135,7 _at__at_
 		return;
 	}
 
-	mtx_init(&smp_rv_mtx, "smp rendezvous", NULL, MTX_SPIN);
+	mtx_init(&smp_ipi_mtx, "smp rendezvous", NULL, MTX_SPIN);
 	cpu_mp_start();
 	printf("FreeBSD/SMP: Multiprocessor System Detected: %d CPUs\n",
 	    mp_ncpus);
_at__at_ -331,7 +331,7 _at__at_
 	}
 		
 	/* obtain rendezvous lock */
-	mtx_lock_spin(&smp_rv_mtx);
+	mtx_lock_spin(&smp_ipi_mtx);
 
 	/* set static function pointers */
 	smp_rv_setup_func = setup_func;
_at__at_ -348,7 +348,7 _at__at_
 	smp_rendezvous_action();
 
 	/* release lock */
-	mtx_unlock_spin(&smp_rv_mtx);
+	mtx_unlock_spin(&smp_ipi_mtx);
 }
 #else /* !SMP */
 
Index: sys/smp.h
===================================================================
RCS file: /home/ncvs/src/sys/sys/smp.h,v
retrieving revision 1.78
diff -u -r1.78 smp.h
--- sys/smp.h	23 Aug 2004 21:39:29 -0000	1.78
+++ sys/smp.h	24 Aug 2004 18:31:48 -0000
_at__at_ -95,7 +95,7 _at__at_
 int	restart_cpus(cpumask_t);
 int	stop_cpus(cpumask_t);
 void	smp_rendezvous_action(void);
-extern	struct mtx smp_rv_mtx;
+extern	struct mtx smp_ipi_mtx;
 #endif /* SMP */
 void	smp_rendezvous(void (*)(void *), 
 		       void (*)(void *),