From: Avinash JD <avinash.dayanand@intel.com>
Date: Thu, 7 May 2020 17:41:00 -0700
Subject: ice: Don't reset and rebuild for Tx timeout on PFC enabled queue
Patch-mainline: v5.8-rc1
Git-commit: 610ed0e93e7dd107afc2398d54de6b6fe9513f8d
References: jsc#SLE-12878
When there's a Tx timeout for a queue which belongs to a PFC enabled TC,
then it's not because the queue is hung but because PFC is in action.
In PFC, peer sends a pause frame for a specified period of time when its
buffer threshold is exceeded (due to congestion). Netdev on the other
hand checks if ACK is received within a specified time for a TX packet, if
not, it'll invoke the tx_timeout routine.
Signed-off-by: Avinash JD <avinash.dayanand@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
drivers/net/ethernet/intel/ice/ice_dcb_lib.c | 58 ++++++++++++++++++++++++
drivers/net/ethernet/intel/ice/ice_dcb_lib.h | 23 +++++++++
drivers/net/ethernet/intel/ice/ice_hw_autogen.h | 1
drivers/net/ethernet/intel/ice/ice_main.c | 10 ++++
drivers/net/ethernet/intel/ice/ice_type.h | 2
5 files changed, 94 insertions(+)
--- a/drivers/net/ethernet/intel/ice/ice_dcb_lib.c
+++ b/drivers/net/ethernet/intel/ice/ice_dcb_lib.c
@@ -63,6 +63,64 @@ u8 ice_dcb_get_ena_tc(struct ice_dcbx_cf
}
/**
+ * ice_is_pfc_causing_hung_q
+ * @pf: pointer to PF structure
+ * @txqueue: Tx queue which is supposedly hung queue
+ *
+ * find if PFC is causing the hung queue, if yes return true else false
+ */
+bool ice_is_pfc_causing_hung_q(struct ice_pf *pf, unsigned int txqueue)
+{
+ u8 num_tcs = 0, i, tc, up_mapped_tc, up_in_tc = 0;
+ u64 ref_prio_xoff[ICE_MAX_UP];
+ struct ice_vsi *vsi;
+ u32 up2tc;
+
+ vsi = ice_get_main_vsi(pf);
+ if (!vsi)
+ return false;
+
+ ice_for_each_traffic_class(i)
+ if (vsi->tc_cfg.ena_tc & BIT(i))
+ num_tcs++;
+
+ /* first find out the TC to which the hung queue belongs to */
+ for (tc = 0; tc < num_tcs - 1; tc++)
+ if (ice_find_q_in_range(vsi->tc_cfg.tc_info[tc].qoffset,
+ vsi->tc_cfg.tc_info[tc + 1].qoffset,
+ txqueue))
+ break;
+
+ /* Build a bit map of all UPs associated to the suspect hung queue TC,
+ * so that we check for its counter increment.
+ */
+ up2tc = rd32(&pf->hw, PRTDCB_TUP2TC);
+ for (i = 0; i < ICE_MAX_UP; i++) {
+ up_mapped_tc = (up2tc >> (i * 3)) & 0x7;
+ if (up_mapped_tc == tc)
+ up_in_tc |= BIT(i);
+ }
+
+ /* Now that we figured out that hung queue is PFC enabled, still the
+ * Tx timeout can be legitimate. So to make sure Tx timeout is
+ * absolutely caused by PFC storm, check if the counters are
+ * incrementing.
+ */
+ for (i = 0; i < ICE_MAX_UP; i++)
+ if (up_in_tc & BIT(i))
+ ref_prio_xoff[i] = pf->stats.priority_xoff_rx[i];
+
+ ice_update_dcb_stats(pf);
+
+ for (i = 0; i < ICE_MAX_UP; i++)
+ if (up_in_tc & BIT(i))
+ if (pf->stats.priority_xoff_rx[i] > ref_prio_xoff[i])
+ return true;
+
+ return false;
+}
+
+/**
* ice_dcb_get_mode - gets the DCB mode
* @port_info: pointer to port info structure
* @host: if set it's HOST if not it's MANAGED
--- a/drivers/net/ethernet/intel/ice/ice_dcb_lib.h
+++ b/drivers/net/ethernet/intel/ice/ice_dcb_lib.h
@@ -17,6 +17,8 @@
void ice_dcb_rebuild(struct ice_pf *pf);
u8 ice_dcb_get_ena_tc(struct ice_dcbx_cfg *dcbcfg);
u8 ice_dcb_get_num_tc(struct ice_dcbx_cfg *dcbcfg);
+void ice_vsi_set_dcb_tc_cfg(struct ice_vsi *vsi);
+bool ice_is_pfc_causing_hung_q(struct ice_pf *pf, unsigned int txqueue);
u8 ice_dcb_get_tc(struct ice_vsi *vsi, int queue_index);
int
ice_pf_dcb_cfg(struct ice_pf *pf, struct ice_dcbx_cfg *new_cfg, bool locked);
@@ -32,6 +34,20 @@ void
ice_dcb_process_lldp_set_mib_change(struct ice_pf *pf,
struct ice_rq_event_info *event);
void ice_vsi_cfg_netdev_tc(struct ice_vsi *vsi, u8 ena_tc);
+
+/**
+ * ice_find_q_in_range
+ * @low: start of queue range for a TC i.e. offset of TC
+ * @high: start of queue for next TC
+ * @tx_q: hung_queue/tx_queue
+ *
+ * finds if queue 'tx_q' falls between the two offsets of any given TC
+ */
+static inline bool ice_find_q_in_range(u16 low, u16 high, unsigned int tx_q)
+{
+ return (tx_q >= low) && (tx_q < high);
+}
+
static inline void
ice_set_cgd_num(struct ice_tlan_ctx *tlan_ctx, struct ice_ring *ring)
{
@@ -79,6 +95,13 @@ ice_tx_prepare_vlan_flags_dcb(struct ice
return 0;
}
+static inline bool
+ice_is_pfc_causing_hung_q(struct ice_pf __always_unused *pf,
+ unsigned int __always_unused txqueue)
+{
+ return false;
+}
+
#define ice_update_dcb_stats(pf) do {} while (0)
#define ice_pf_dcb_recfg(pf) do {} while (0)
#define ice_vsi_cfg_dcb_rings(vsi) do {} while (0)
--- a/drivers/net/ethernet/intel/ice/ice_hw_autogen.h
+++ b/drivers/net/ethernet/intel/ice/ice_hw_autogen.h
@@ -58,6 +58,7 @@
#define PRTDCB_GENS 0x00083020
#define PRTDCB_GENS_DCBX_STATUS_S 0
#define PRTDCB_GENS_DCBX_STATUS_M ICE_M(0x7, 0)
+#define PRTDCB_TUP2TC 0x001D26C0 /* Reset Source: CORER */
#define GL_PREEXT_L2_PMASK0(_i) (0x0020F0FC + ((_i) * 4))
#define GL_PREEXT_L2_PMASK1(_i) (0x0020F108 + ((_i) * 4))
#define GLFLXP_RXDID_FLX_WRD_0(_i) (0x0045c800 + ((_i) * 4))
--- a/drivers/net/ethernet/intel/ice/ice_main.c
+++ b/drivers/net/ethernet/intel/ice/ice_main.c
@@ -5115,6 +5115,16 @@ static void ice_tx_timeout(struct net_de
pf->tx_timeout_count++;
+ /* Check if PFC is enabled for the TC to which the queue belongs
+ * to. If yes then Tx timeout is not caused by a hung queue, no
+ * need to reset and rebuild
+ */
+ if (ice_is_pfc_causing_hung_q(pf, txqueue)) {
+ dev_info(ice_pf_to_dev(pf), "Fake Tx hang detected on queue %u, timeout caused by PFC storm\n",
+ txqueue);
+ return;
+ }
+
/* now that we have an index, find the tx_ring struct */
for (i = 0; i < vsi->num_txq; i++)
if (vsi->tx_rings[i] && vsi->tx_rings[i]->desc)
--- a/drivers/net/ethernet/intel/ice/ice_type.h
+++ b/drivers/net/ethernet/intel/ice/ice_type.h
@@ -607,6 +607,8 @@ struct ice_eth_stats {
u64 tx_errors; /* tepc */
};
+#define ICE_MAX_UP 8
+
/* Statistics collected by the MAC */
struct ice_hw_port_stats {
/* eth stats collected by the port */