[PATCH 8/11] x86, UV: remove BAU check for stay-busy

[PATCH 8/11] x86, UV: remove BAU check for stay-busy

[Cliff Wickman]

Remove a faulty assumption that a long running BAU request has encountered
a hardware problem and will never finish.

Numalink congestion can make a request appear to have encountered such a
problem, but it is not safe to cancel the request.  If such a cancel is
done but a reply is later received we can miss a TLB shootdown.

We depend upon the max_bau_concurrent 'throttle' to prevent the stay-busy
case from happening.

Diffed against 2.6.34 -tip

Signed-off-by: Cliff Wickman <cpw@sgi.com>
---
 arch/x86/include/asm/uv/uv_bau.h |    1 -
 arch/x86/kernel/tlb_uv.c         |   23 -----------------------
 2 files changed, 24 deletions(-)

Index: 100531.linux-tip/arch/x86/kernel/tlb_uv.c
===================================================================
--- 100531.linux-tip.orig/arch/x86/kernel/tlb_uv.c
+++ 100531.linux-tip/arch/x86/kernel/tlb_uv.c
@@ -405,12 +405,10 @@ static int uv_wait_completion(struct bau
 	unsigned long mmr;
 	unsigned long mask;
 	cycles_t ttime;
-	cycles_t timeout_time;
 	struct ptc_stats *stat = bcp->statp;
 	struct bau_control *hmaster;
 
 	hmaster = bcp->uvhub_master;
-	timeout_time = get_cycles() + bcp->timeout_interval;
 
 	/* spin on the status MMR, waiting for it to go idle */
 	while ((descriptor_status = (((unsigned long)
@@ -450,26 +448,6 @@ static int uv_wait_completion(struct bau
 			 * descriptor_status is still BUSY
 			 */
 			cpu_relax();
-			relaxes++;
-			if (relaxes >= 10000) {
-				relaxes = 0;
-				if (get_cycles() > timeout_time) {
-					quiesce_local_uvhub(hmaster);
-
-					/* single-thread the register change */
-					spin_lock(&hmaster->masks_lock);
-					mmr = uv_read_local_mmr(mmr_offset);
-					mask = 0UL;
-					mask |= (3UL < right_shift);
-					mask = ~mask;
-					mmr &= mask;
-					uv_write_local_mmr(mmr_offset, mmr);
-					spin_unlock(&hmaster->masks_lock);
-					end_uvhub_quiesce(hmaster);
-					stat->s_busy++;
-					return FLUSH_GIVEUP;
-				}
-			}
 		}
 	}
 	bcp->conseccompletes++;
@@ -1580,7 +1558,6 @@ static void uv_init_per_cpu(int nuvhubs)
 	for_each_present_cpu(cpu) {
 		bcp = &per_cpu(bau_control, cpu);
 		memset(bcp, 0, sizeof(struct bau_control));
-		spin_lock_init(&bcp->masks_lock);
 		pnode = uv_cpu_hub_info(cpu)->pnode;
 		uvhub = uv_cpu_hub_info(cpu)->numa_blade_id;
 		uvhub_mask |= (1 << uvhub);
Index: 100531.linux-tip/arch/x86/include/asm/uv/uv_bau.h
===================================================================
--- 100531.linux-tip.orig/arch/x86/include/asm/uv/uv_bau.h
+++ 100531.linux-tip/arch/x86/include/asm/uv/uv_bau.h
@@ -402,7 +402,6 @@ struct bau_control {
 	unsigned short uvhub_quiesce;
 	short socket_acknowledge_count[DEST_Q_SIZE];
 	cycles_t send_message;
-	spinlock_t masks_lock;
 	spinlock_t uvhub_lock;
 	spinlock_t queue_lock;
 	/* tunables */

[tip:x86/uv] x86, UV: Remove BAU check for stay-busy

[tip-bot for Cliff Wickman]

Commit-ID:  90cc7d944981a6d06b49bb26fde1b490e28c90e5
Gitweb:     http://git.kernel.org/tip/90cc7d944981a6d06b49bb26fde1b490e28c90e5
Author:     Cliff Wickman <cpw@sgi.com>
AuthorDate: Wed, 2 Jun 2010 16:22:02 -0500
Committer:  Ingo Molnar <mingo@elte.hu>
CommitDate: Tue, 8 Jun 2010 21:13:47 +0200

x86, UV: Remove BAU check for stay-busy

Remove a faulty assumption that a long running BAU request has
encountered a hardware problem and will never finish.

Numalink congestion can make a request appear to have
encountered such a problem, but it is not safe to cancel the
request.  If such a cancel is done but a reply is later received
we can miss a TLB shootdown.

We depend upon the max_bau_concurrent 'throttle' to prevent the
stay-busy case from happening.

Signed-off-by: Cliff Wickman <cpw@sgi.com>
Cc: gregkh@suse.de
LKML-Reference: <E1OJvNy-0004ad-BV@eag09.americas.sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 arch/x86/include/asm/uv/uv_bau.h |    1 -
 arch/x86/kernel/tlb_uv.c         |   23 -----------------------
 2 files changed, 0 insertions(+), 24 deletions(-)

diff --git a/arch/x86/include/asm/uv/uv_bau.h b/arch/x86/include/asm/uv/uv_bau.h
index 1c8f1e9..c19b870 100644
--- a/arch/x86/include/asm/uv/uv_bau.h
+++ b/arch/x86/include/asm/uv/uv_bau.h
@@ -402,7 +402,6 @@ struct bau_control {
 	unsigned short uvhub_quiesce;
 	short socket_acknowledge_count[DEST_Q_SIZE];
 	cycles_t send_message;
-	spinlock_t masks_lock;
 	spinlock_t uvhub_lock;
 	spinlock_t queue_lock;
 	/* tunables */
diff --git a/arch/x86/kernel/tlb_uv.c b/arch/x86/kernel/tlb_uv.c
index ab929e9..dc962b5 100644
--- a/arch/x86/kernel/tlb_uv.c
+++ b/arch/x86/kernel/tlb_uv.c
@@ -405,12 +405,10 @@ static int uv_wait_completion(struct bau_desc *bau_desc,
 	unsigned long mmr;
 	unsigned long mask;
 	cycles_t ttime;
-	cycles_t timeout_time;
 	struct ptc_stats *stat = bcp->statp;
 	struct bau_control *hmaster;
 
 	hmaster = bcp->uvhub_master;
-	timeout_time = get_cycles() + bcp->timeout_interval;
 
 	/* spin on the status MMR, waiting for it to go idle */
 	while ((descriptor_status = (((unsigned long)
@@ -450,26 +448,6 @@ static int uv_wait_completion(struct bau_desc *bau_desc,
 			 * descriptor_status is still BUSY
 			 */
 			cpu_relax();
-			relaxes++;
-			if (relaxes >= 10000) {
-				relaxes = 0;
-				if (get_cycles() > timeout_time) {
-					quiesce_local_uvhub(hmaster);
-
-					/* single-thread the register change */
-					spin_lock(&hmaster->masks_lock);
-					mmr = uv_read_local_mmr(mmr_offset);
-					mask = 0UL;
-					mask |= (3UL < right_shift);
-					mask = ~mask;
-					mmr &= mask;
-					uv_write_local_mmr(mmr_offset, mmr);
-					spin_unlock(&hmaster->masks_lock);
-					end_uvhub_quiesce(hmaster);
-					stat->s_busy++;
-					return FLUSH_GIVEUP;
-				}
-			}
 		}
 	}
 	bcp->conseccompletes++;
@@ -1580,7 +1558,6 @@ static void uv_init_per_cpu(int nuvhubs)
 	for_each_present_cpu(cpu) {
 		bcp = &per_cpu(bau_control, cpu);
 		memset(bcp, 0, sizeof(struct bau_control));
-		spin_lock_init(&bcp->masks_lock);
 		pnode = uv_cpu_hub_info(cpu)->pnode;
 		uvhub = uv_cpu_hub_info(cpu)->numa_blade_id;
 		uvhub_mask |= (1 << uvhub);