[PATCH net-next 0/2] cxgb4: move stats fetched from firmware to debugfs

[PATCH net-next 0/2] cxgb4: move stats fetched from firmware to debugfs

[Rahul Lakkireddy]

Some stats are fetched via slow firmware mailbox, which can cause
packet drops under heavy load. So, this series removes these stats
from ethtool -S and expose them via debugfs.

Patch 1 removes stats fetched via firmware from ethtool -S.
Patch 2 exposes stats removed in Patch 1 via debugfs.

Thanks,
Rahul

Rahul Lakkireddy (2):
  cxgb4: remove stats fetched from firmware
  cxgb4: expose stats fetched from firmware via debugfs

 drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c | 164 +++++++++++++++++++++
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_ethtool.c | 133 -----------------
 2 files changed, 164 insertions(+), 133 deletions(-)

-- 
2.14.1

[PATCH net-next 1/2] cxgb4: remove stats fetched from firmware

[Rahul Lakkireddy]

When running ethtool -S, some stats are requested from firmware.
Since getting these stats via firmware mailbox is slow, some packets
get dropped under heavy load while running ethtool -S.

So, remove these stats from ethtool -S.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_ethtool.c | 133 ---------------------
 1 file changed, 133 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ethtool.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ethtool.c
index ddb8b9eba6bf..a14a290a56ee 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ethtool.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_ethtool.c
@@ -115,42 +115,10 @@ static char adapter_stats_strings[][ETH_GSTRING_LEN] = {
 	"db_drop                ",
 	"db_full                ",
 	"db_empty               ",
-	"tcp_ipv4_out_rsts      ",
-	"tcp_ipv4_in_segs       ",
-	"tcp_ipv4_out_segs      ",
-	"tcp_ipv4_retrans_segs  ",
-	"tcp_ipv6_out_rsts      ",
-	"tcp_ipv6_in_segs       ",
-	"tcp_ipv6_out_segs      ",
-	"tcp_ipv6_retrans_segs  ",
-	"usm_ddp_frames         ",
-	"usm_ddp_octets         ",
-	"usm_ddp_drops          ",
-	"rdma_no_rqe_mod_defer  ",
-	"rdma_no_rqe_pkt_defer  ",
-	"tp_err_ofld_no_neigh   ",
-	"tp_err_ofld_cong_defer ",
 	"write_coal_success     ",
 	"write_coal_fail        ",
 };
 
-static char channel_stats_strings[][ETH_GSTRING_LEN] = {
-	"--------Channel--------- ",
-	"tp_cpl_requests        ",
-	"tp_cpl_responses       ",
-	"tp_mac_in_errs         ",
-	"tp_hdr_in_errs         ",
-	"tp_tcp_in_errs         ",
-	"tp_tcp6_in_errs        ",
-	"tp_tnl_cong_drops      ",
-	"tp_tnl_tx_drops        ",
-	"tp_ofld_vlan_drops     ",
-	"tp_ofld_chan_drops     ",
-	"fcoe_octets_ddp        ",
-	"fcoe_frames_ddp        ",
-	"fcoe_frames_drop       ",
-};
-
 static char loopback_stats_strings[][ETH_GSTRING_LEN] = {
 	"-------Loopback----------- ",
 	"octets_ok              ",
@@ -187,7 +155,6 @@ static int get_sset_count(struct net_device *dev, int sset)
 	case ETH_SS_STATS:
 		return ARRAY_SIZE(stats_strings) +
 		       ARRAY_SIZE(adapter_stats_strings) +
-		       ARRAY_SIZE(channel_stats_strings) +
 		       ARRAY_SIZE(loopback_stats_strings);
 	case ETH_SS_PRIV_FLAGS:
 		return ARRAY_SIZE(cxgb4_priv_flags_strings);
@@ -252,9 +219,6 @@ static void get_strings(struct net_device *dev, u32 stringset, u8 *data)
 		memcpy(data, adapter_stats_strings,
 		       sizeof(adapter_stats_strings));
 		data += sizeof(adapter_stats_strings);
-		memcpy(data, channel_stats_strings,
-		       sizeof(channel_stats_strings));
-		data += sizeof(channel_stats_strings);
 		memcpy(data, loopback_stats_strings,
 		       sizeof(loopback_stats_strings));
 	} else if (stringset == ETH_SS_PRIV_FLAGS) {
@@ -280,41 +244,10 @@ struct adapter_stats {
 	u64 db_drop;
 	u64 db_full;
 	u64 db_empty;
-	u64 tcp_v4_out_rsts;
-	u64 tcp_v4_in_segs;
-	u64 tcp_v4_out_segs;
-	u64 tcp_v4_retrans_segs;
-	u64 tcp_v6_out_rsts;
-	u64 tcp_v6_in_segs;
-	u64 tcp_v6_out_segs;
-	u64 tcp_v6_retrans_segs;
-	u64 frames;
-	u64 octets;
-	u64 drops;
-	u64 rqe_dfr_mod;
-	u64 rqe_dfr_pkt;
-	u64 ofld_no_neigh;
-	u64 ofld_cong_defer;
 	u64 wc_success;
 	u64 wc_fail;
 };
 
-struct channel_stats {
-	u64 cpl_req;
-	u64 cpl_rsp;
-	u64 mac_in_errs;
-	u64 hdr_in_errs;
-	u64 tcp_in_errs;
-	u64 tcp6_in_errs;
-	u64 tnl_cong_drops;
-	u64 tnl_tx_drops;
-	u64 ofld_vlan_drops;
-	u64 ofld_chan_drops;
-	u64 octets_ddp;
-	u64 frames_ddp;
-	u64 frames_drop;
-};
-
 static void collect_sge_port_stats(const struct adapter *adap,
 				   const struct port_info *p,
 				   struct queue_port_stats *s)
@@ -337,45 +270,14 @@ static void collect_sge_port_stats(const struct adapter *adap,
 
 static void collect_adapter_stats(struct adapter *adap, struct adapter_stats *s)
 {
-	struct tp_tcp_stats v4, v6;
-	struct tp_rdma_stats rdma_stats;
-	struct tp_err_stats err_stats;
-	struct tp_usm_stats usm_stats;
 	u64 val1, val2;
 
 	memset(s, 0, sizeof(*s));
 
-	spin_lock(&adap->stats_lock);
-	t4_tp_get_tcp_stats(adap, &v4, &v6, false);
-	t4_tp_get_rdma_stats(adap, &rdma_stats, false);
-	t4_get_usm_stats(adap, &usm_stats, false);
-	t4_tp_get_err_stats(adap, &err_stats, false);
-	spin_unlock(&adap->stats_lock);
-
 	s->db_drop = adap->db_stats.db_drop;
 	s->db_full = adap->db_stats.db_full;
 	s->db_empty = adap->db_stats.db_empty;
 
-	s->tcp_v4_out_rsts = v4.tcp_out_rsts;
-	s->tcp_v4_in_segs = v4.tcp_in_segs;
-	s->tcp_v4_out_segs = v4.tcp_out_segs;
-	s->tcp_v4_retrans_segs = v4.tcp_retrans_segs;
-	s->tcp_v6_out_rsts = v6.tcp_out_rsts;
-	s->tcp_v6_in_segs = v6.tcp_in_segs;
-	s->tcp_v6_out_segs = v6.tcp_out_segs;
-	s->tcp_v6_retrans_segs = v6.tcp_retrans_segs;
-
-	if (is_offload(adap)) {
-		s->frames = usm_stats.frames;
-		s->octets = usm_stats.octets;
-		s->drops = usm_stats.drops;
-		s->rqe_dfr_mod = rdma_stats.rqe_dfr_mod;
-		s->rqe_dfr_pkt = rdma_stats.rqe_dfr_pkt;
-	}
-
-	s->ofld_no_neigh = err_stats.ofld_no_neigh;
-	s->ofld_cong_defer = err_stats.ofld_cong_defer;
-
 	if (!is_t4(adap->params.chip)) {
 		int v;
 
@@ -389,36 +291,6 @@ static void collect_adapter_stats(struct adapter *adap, struct adapter_stats *s)
 	}
 }
 
-static void collect_channel_stats(struct adapter *adap, struct channel_stats *s,
-				  u8 i)
-{
-	struct tp_cpl_stats cpl_stats;
-	struct tp_err_stats err_stats;
-	struct tp_fcoe_stats fcoe_stats;
-
-	memset(s, 0, sizeof(*s));
-
-	spin_lock(&adap->stats_lock);
-	t4_tp_get_cpl_stats(adap, &cpl_stats, false);
-	t4_tp_get_err_stats(adap, &err_stats, false);
-	t4_get_fcoe_stats(adap, i, &fcoe_stats, false);
-	spin_unlock(&adap->stats_lock);
-
-	s->cpl_req = cpl_stats.req[i];
-	s->cpl_rsp = cpl_stats.rsp[i];
-	s->mac_in_errs = err_stats.mac_in_errs[i];
-	s->hdr_in_errs = err_stats.hdr_in_errs[i];
-	s->tcp_in_errs = err_stats.tcp_in_errs[i];
-	s->tcp6_in_errs = err_stats.tcp6_in_errs[i];
-	s->tnl_cong_drops = err_stats.tnl_cong_drops[i];
-	s->tnl_tx_drops = err_stats.tnl_tx_drops[i];
-	s->ofld_vlan_drops = err_stats.ofld_vlan_drops[i];
-	s->ofld_chan_drops = err_stats.ofld_chan_drops[i];
-	s->octets_ddp = fcoe_stats.octets_ddp;
-	s->frames_ddp = fcoe_stats.frames_ddp;
-	s->frames_drop = fcoe_stats.frames_drop;
-}
-
 static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
 		      u64 *data)
 {
@@ -438,11 +310,6 @@ static void get_stats(struct net_device *dev, struct ethtool_stats *stats,
 	collect_adapter_stats(adapter, (struct adapter_stats *)data);
 	data += sizeof(struct adapter_stats) / sizeof(u64);
 
-	*data++ = (u64)pi->port_id;
-	collect_channel_stats(adapter, (struct channel_stats *)data,
-			      pi->port_id);
-	data += sizeof(struct channel_stats) / sizeof(u64);
-
 	*data++ = (u64)pi->port_id;
 	memset(&s, 0, sizeof(s));
 	t4_get_lb_stats(adapter, pi->port_id, &s);
-- 
2.14.1

[PATCH net-next 2/2] cxgb4: expose stats fetched from firmware via debugfs

[Rahul Lakkireddy]

Expose stats obtained from firmware via debugfs. These stats can't
be part of ethtool -S because the slow firmware mailbox can cause
packet drops under heavy load.

Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
---
 drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c | 164 +++++++++++++++++++++
 1 file changed, 164 insertions(+)

diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c
index 516c883e1613..511606fd1b20 100644
--- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c
+++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_debugfs.c
@@ -2962,6 +2962,169 @@ static const struct file_operations chcr_stats_debugfs_fops = {
         .llseek  = seq_lseek,
         .release = single_release,
 };
+
+#define PRINT_ADAP_STATS(string, value) \
+	seq_printf(seq, "%-25s %-20llu\n", (string), \
+		   (unsigned long long)(value))
+
+#define PRINT_CH_STATS(string, value) \
+do { \
+	seq_printf(seq, "%-25s ", (string)); \
+	for (i = 0; i < adap->params.arch.nchan; i++) \
+		seq_printf(seq, "%-20llu ", \
+			   (unsigned long long)stats.value[i]); \
+	seq_printf(seq, "\n"); \
+} while (0)
+
+#define PRINT_CH_STATS2(string, value) \
+do { \
+	seq_printf(seq, "%-25s ", (string)); \
+	for (i = 0; i < adap->params.arch.nchan; i++) \
+		seq_printf(seq, "%-20llu ", \
+			   (unsigned long long)stats[i].value); \
+	seq_printf(seq, "\n"); \
+} while (0)
+
+static void show_tcp_stats(struct seq_file *seq)
+{
+	struct adapter *adap = seq->private;
+	struct tp_tcp_stats v4, v6;
+
+	spin_lock(&adap->stats_lock);
+	t4_tp_get_tcp_stats(adap, &v4, &v6, false);
+	spin_unlock(&adap->stats_lock);
+
+	PRINT_ADAP_STATS("tcp_ipv4_out_rsts:", v4.tcp_out_rsts);
+	PRINT_ADAP_STATS("tcp_ipv4_in_segs:", v4.tcp_in_segs);
+	PRINT_ADAP_STATS("tcp_ipv4_out_segs:", v4.tcp_out_segs);
+	PRINT_ADAP_STATS("tcp_ipv4_retrans_segs:", v4.tcp_retrans_segs);
+	PRINT_ADAP_STATS("tcp_ipv6_out_rsts:", v6.tcp_out_rsts);
+	PRINT_ADAP_STATS("tcp_ipv6_in_segs:", v6.tcp_in_segs);
+	PRINT_ADAP_STATS("tcp_ipv6_out_segs:", v6.tcp_out_segs);
+	PRINT_ADAP_STATS("tcp_ipv6_retrans_segs:", v6.tcp_retrans_segs);
+}
+
+static void show_ddp_stats(struct seq_file *seq)
+{
+	struct adapter *adap = seq->private;
+	struct tp_usm_stats stats;
+
+	spin_lock(&adap->stats_lock);
+	t4_get_usm_stats(adap, &stats, false);
+	spin_unlock(&adap->stats_lock);
+
+	PRINT_ADAP_STATS("usm_ddp_frames:", stats.frames);
+	PRINT_ADAP_STATS("usm_ddp_octets:", stats.octets);
+	PRINT_ADAP_STATS("usm_ddp_drops:", stats.drops);
+}
+
+static void show_rdma_stats(struct seq_file *seq)
+{
+	struct adapter *adap = seq->private;
+	struct tp_rdma_stats stats;
+
+	spin_lock(&adap->stats_lock);
+	t4_tp_get_rdma_stats(adap, &stats, false);
+	spin_unlock(&adap->stats_lock);
+
+	PRINT_ADAP_STATS("rdma_no_rqe_mod_defer:", stats.rqe_dfr_mod);
+	PRINT_ADAP_STATS("rdma_no_rqe_pkt_defer:", stats.rqe_dfr_pkt);
+}
+
+static void show_tp_err_adapter_stats(struct seq_file *seq)
+{
+	struct adapter *adap = seq->private;
+	struct tp_err_stats stats;
+
+	spin_lock(&adap->stats_lock);
+	t4_tp_get_err_stats(adap, &stats, false);
+	spin_unlock(&adap->stats_lock);
+
+	PRINT_ADAP_STATS("tp_err_ofld_no_neigh:", stats.ofld_no_neigh);
+	PRINT_ADAP_STATS("tp_err_ofld_cong_defer:", stats.ofld_cong_defer);
+}
+
+static void show_cpl_stats(struct seq_file *seq)
+{
+	struct adapter *adap = seq->private;
+	struct tp_cpl_stats stats;
+	u8 i;
+
+	spin_lock(&adap->stats_lock);
+	t4_tp_get_cpl_stats(adap, &stats, false);
+	spin_unlock(&adap->stats_lock);
+
+	PRINT_CH_STATS("tp_cpl_requests:", req);
+	PRINT_CH_STATS("tp_cpl_responses:", rsp);
+}
+
+static void show_tp_err_channel_stats(struct seq_file *seq)
+{
+	struct adapter *adap = seq->private;
+	struct tp_err_stats stats;
+	u8 i;
+
+	spin_lock(&adap->stats_lock);
+	t4_tp_get_err_stats(adap, &stats, false);
+	spin_unlock(&adap->stats_lock);
+
+	PRINT_CH_STATS("tp_mac_in_errs:", mac_in_errs);
+	PRINT_CH_STATS("tp_hdr_in_errs:", hdr_in_errs);
+	PRINT_CH_STATS("tp_tcp_in_errs:", tcp_in_errs);
+	PRINT_CH_STATS("tp_tcp6_in_errs:", tcp6_in_errs);
+	PRINT_CH_STATS("tp_tnl_cong_drops:", tnl_cong_drops);
+	PRINT_CH_STATS("tp_tnl_tx_drops:", tnl_tx_drops);
+	PRINT_CH_STATS("tp_ofld_vlan_drops:", ofld_vlan_drops);
+	PRINT_CH_STATS("tp_ofld_chan_drops:", ofld_chan_drops);
+}
+
+static void show_fcoe_stats(struct seq_file *seq)
+{
+	struct adapter *adap = seq->private;
+	struct tp_fcoe_stats stats[NCHAN];
+	u8 i;
+
+	spin_lock(&adap->stats_lock);
+	for (i = 0; i < adap->params.arch.nchan; i++)
+		t4_get_fcoe_stats(adap, i, &stats[i], false);
+	spin_unlock(&adap->stats_lock);
+
+	PRINT_CH_STATS2("fcoe_octets_ddp", octets_ddp);
+	PRINT_CH_STATS2("fcoe_frames_ddp", frames_ddp);
+	PRINT_CH_STATS2("fcoe_frames_drop", frames_drop);
+}
+
+#undef PRINT_CH_STATS2
+#undef PRINT_CH_STATS
+#undef PRINT_ADAP_STATS
+
+static int tp_stats_show(struct seq_file *seq, void *v)
+{
+	struct adapter *adap = seq->private;
+
+	seq_puts(seq, "\n--------Adapter Stats--------\n");
+	show_tcp_stats(seq);
+	show_ddp_stats(seq);
+	show_rdma_stats(seq);
+	show_tp_err_adapter_stats(seq);
+
+	seq_puts(seq, "\n-------- Channel Stats --------\n");
+	if (adap->params.arch.nchan == NCHAN)
+		seq_printf(seq, "%-25s %-20s %-20s %-20s %-20s\n",
+			   " ", "channel 0", "channel 1",
+			   "channel 2", "channel 3");
+	else
+		seq_printf(seq, "%-25s %-20s %-20s\n",
+			   " ", "channel 0", "channel 1");
+	show_cpl_stats(seq);
+	show_tp_err_channel_stats(seq);
+	show_fcoe_stats(seq);
+
+	return 0;
+}
+
+DEFINE_SIMPLE_DEBUGFS_FILE(tp_stats);
+
 /* Add an array of Debug FS files.
  */
 void add_debugfs_files(struct adapter *adap,
@@ -3038,6 +3201,7 @@ int t4_setup_debugfs(struct adapter *adap)
 		{ "blocked_fl", &blocked_fl_fops, 0600, 0 },
 		{ "meminfo", &meminfo_fops, 0400, 0 },
 		{ "crypto", &chcr_stats_debugfs_fops, 0400, 0 },
+		{ "tp_stats", &tp_stats_debugfs_fops, 0400, 0 },
 	};
 
 	/* Debug FS nodes common to all T5 and later adapters.
-- 
2.14.1

Re: [PATCH net-next 0/2] cxgb4: move stats fetched from firmware to debugfs

[David Miller]

From: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Date: Mon,  9 Jul 2018 21:42:45 +0530

> Some stats are fetched via slow firmware mailbox, which can cause
> packet drops under heavy load. So, this series removes these stats
> from ethtool -S and expose them via debugfs.
> 
> Patch 1 removes stats fetched via firmware from ethtool -S.
> Patch 2 exposes stats removed in Patch 1 via debugfs.

Series applied, thanks.