From: Steve Best <sbest@redhat.com>
Date: Thu, 4 Feb 2010 22:11:09 -0500
Subject: [net] cxgb3: add memory barriers
Message-id: <20100204220400.11927.32233.sendpatchset@squad5-lp1.lab.bos.redhat.com>
Patchwork-id: 23135
O-Subject: [PATCH RHEL5.5 BZ561957] cxgb3: add memory barriers
Bugzilla: 561957
RH-Acked-by: David S. Miller <davem@redhat.com>
RH-Acked-by: Neil Horman <nhorman@redhat.com>
RH-Acked-by: Andy Gospodarek <gospo@redhat.com>

RHBZ#:
======
https://bugzilla.redhat.com/show_bug.cgi?id=561957

Description:
============
Add memory barriers to fix crashes observed on newest PowerPC platforms.
The HW and driver state of the receive rings were getting out of sync.

RHEL Version Found:
===================
RHEL 5.5

kABI Status:
============
No symbols were harmed.

Brew:
=====
http://brewweb.devel.redhat.com/brew/taskinfo?taskID=2242234

Upstream Status:
================
http://lkml.indiana.edu/hypermail/linux/kernel/1002.0/00312.html
upstream accepted here
http://patchwork.ozlabs.org/patch/44225/

Test Status:
============
Tested by IBM test team using HTX
---Steps to Reproduce---
(1)Set up HTX on both of lpars on P7.
(2)Set up HTX stress level to 5, then start HTX runs.
(3)After about 10 hours, one of system hits EEH error.

Signed-off-by: Jarod Wilson <jarod@redhat.com>

diff --git a/drivers/net/cxgb3/sge.c b/drivers/net/cxgb3/sge.c
index 6d8a3bd..8a38915 100644
--- a/drivers/net/cxgb3/sge.c
+++ b/drivers/net/cxgb3/sge.c
@@ -480,6 +480,7 @@ static inline void ring_fl_db(struct adapter *adap, struct sge_fl *q)
 {
 	if (q->pend_cred >= q->credits / 4) {
 		q->pend_cred = 0;
+		wmb();
 		t3_write_reg(adap, A_SG_KDOORBELL, V_EGRCNTX(q->cntxt_id));
 	}
 }
@@ -2281,11 +2282,14 @@ static int process_responses(struct adapter *adap, struct sge_qset *qs,
 	while (likely(budget_left && is_new_response(r, q))) {
 		int packet_complete, eth, ethpad = 2, lro = qs->lro_enabled;
 		struct sk_buff *skb = NULL;
-		u32 len, flags = ntohl(r->flags);
-		__be32 rss_hi = *(const __be32 *)r,
-		       rss_lo = r->rss_hdr.rss_hash_val;
+		u32 len, flags;
+		__be32 rss_hi, rss_lo;
 
+		rmb();
 		eth = r->rss_hdr.opcode == CPL_RX_PKT;
+		rss_hi = *(const __be32 *)r,
+		rss_lo = r->rss_hdr.rss_hash_val;
+		flags = ntohl(r->flags);
 
 		if (unlikely(flags & F_RSPD_ASYNC_NOTIF)) {
 			skb = alloc_skb(AN_PKT_SIZE, GFP_ATOMIC);
@@ -2500,7 +2504,10 @@ static int process_pure_responses(struct adapter *adap, struct sge_qset *qs,
 			refill_rspq(adap, q, q->credits);
 			q->credits = 0;
 		}
-	} while (is_new_response(r, q) && is_pure_response(r));
+		if (!is_new_response(r, q))
+			break;
+		rmb();
+	} while (is_new_response(r, q));
 
 	if (sleeping)
 		check_ring_db(adap, qs, sleeping);
@@ -2534,6 +2541,7 @@ static inline int handle_responses(struct adapter *adap, struct sge_rspq *q)
 
 	if (!is_new_response(r, q))
 		return -1;
+	rmb();
 	if (is_pure_response(r) && process_pure_responses(adap, qs, r) == 0) {
 		t3_write_reg(adap, A_SG_GTS, V_RSPQ(q->cntxt_id) |
 			     V_NEWTIMER(q->holdoff_tmr) | V_NEWINDEX(q->cidx));