netdev - [net 06/13] ice: Fix bugs in control queue processing

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20180823191503.15804-7-jeffrey.t.kirsher@intel.com>
Date:   Thu, 23 Aug 2018 12:14:56 -0700
From:   Jeff Kirsher <jeffrey.t.kirsher@...el.com>
To:     davem@...emloft.net
Cc:     Anirudh Venkataramanan <anirudh.venkataramanan@...el.com>,
        netdev@...r.kernel.org, nhorman@...hat.com, sassmann@...hat.com,
        jogreene@...hat.com, Jeff Kirsher <jeffrey.t.kirsher@...el.com>
Subject: [net 06/13] ice: Fix bugs in control queue processing

From: Anirudh Venkataramanan <anirudh.venkataramanan@...el.com>

This patch is a consolidation of multiple bug fixes for control queue
processing.

1)  In ice_clean_adminq_subtask() remove unnecessary reads/writes to
    registers. The bits PFINT_FW_CTL, PFINT_MBX_CTL and PFINT_SB_CTL
    are not set when an interrupt arrives, which means that clearing them
    again can be omitted.

2)  Get an accurate value in "pending" by re-reading the control queue
    head register from the hardware.

3)  Fix a corner case involving lost control queue messages by checking
    for new control messages (using ice_ctrlq_pending) before exiting the
    cleanup routine.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@...el.com>
Tested-by: Tony Brelinski <tonyx.brelinski@...el.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@...el.com>
---
 drivers/net/ethernet/intel/ice/ice_controlq.c |  5 +++-
 drivers/net/ethernet/intel/ice/ice_main.c     | 26 ++++++++++++++++---
 2 files changed, 26 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/intel/ice/ice_controlq.c b/drivers/net/ethernet/intel/ice/ice_controlq.c
index c064416080e7..62be72fdc8f3 100644
--- a/drivers/net/ethernet/intel/ice/ice_controlq.c
+++ b/drivers/net/ethernet/intel/ice/ice_controlq.c
@@ -1065,8 +1065,11 @@ ice_clean_rq_elem(struct ice_hw *hw, struct ice_ctl_q_info *cq,
 
 clean_rq_elem_out:
 	/* Set pending if needed, unlock and return */
-	if (pending)
+	if (pending) {
+		/* re-read HW head to calculate actual pending messages */
+		ntu = (u16)(rd32(hw, cq->rq.head) & cq->rq.head_mask);
 		*pending = (u16)((ntc > ntu ? cq->rq.count : 0) + (ntu - ntc));
+	}
 clean_rq_elem_err:
 	mutex_unlock(&cq->rq_lock);
 
diff --git a/drivers/net/ethernet/intel/ice/ice_main.c b/drivers/net/ethernet/intel/ice/ice_main.c
index 7d65e0ed3588..f3ba4f76b6cb 100644
--- a/drivers/net/ethernet/intel/ice/ice_main.c
+++ b/drivers/net/ethernet/intel/ice/ice_main.c
@@ -916,6 +916,21 @@ static int __ice_clean_ctrlq(struct ice_pf *pf, enum ice_ctl_q q_type)
 	return pending && (i == ICE_DFLT_IRQ_WORK);
 }
 
+/**
+ * ice_ctrlq_pending - check if there is a difference between ntc and ntu
+ * @hw: pointer to hardware info
+ * @cq: control queue information
+ *
+ * returns true if there are pending messages in a queue, false if there aren't
+ */
+static bool ice_ctrlq_pending(struct ice_hw *hw, struct ice_ctl_q_info *cq)
+{
+	u16 ntu;
+
+	ntu = (u16)(rd32(hw, cq->rq.head) & cq->rq.head_mask);
+	return cq->rq.next_to_clean != ntu;
+}
+
 /**
  * ice_clean_adminq_subtask - clean the AdminQ rings
  * @pf: board private structure
@@ -923,7 +938,6 @@ static int __ice_clean_ctrlq(struct ice_pf *pf, enum ice_ctl_q q_type)
 static void ice_clean_adminq_subtask(struct ice_pf *pf)
 {
 	struct ice_hw *hw = &pf->hw;
-	u32 val;
 
 	if (!test_bit(__ICE_ADMINQ_EVENT_PENDING, pf->state))
 		return;
@@ -933,9 +947,13 @@ static void ice_clean_adminq_subtask(struct ice_pf *pf)
 
 	clear_bit(__ICE_ADMINQ_EVENT_PENDING, pf->state);
 
-	/* re-enable Admin queue interrupt causes */
-	val = rd32(hw, PFINT_FW_CTL);
-	wr32(hw, PFINT_FW_CTL, (val | PFINT_FW_CTL_CAUSE_ENA_M));
+	/* There might be a situation where new messages arrive to a control
+	 * queue between processing the last message and clearing the
+	 * EVENT_PENDING bit. So before exiting, check queue head again (using
+	 * ice_ctrlq_pending) and process new messages if any.
+	 */
+	if (ice_ctrlq_pending(hw, &hw->adminq))
+		__ice_clean_ctrlq(pf, ICE_CTL_Q_ADMIN);
 
 	ice_flush(hw);
 }
-- 
2.17.1