Netdev List
 help / color / mirror / Atom feed
* [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler
@ 2026-11-16 23:06 Eric Joyner
  2026-11-16 23:06 ` [PATCH net-next v4 1/6] ionic: write the extra_stats invalid markers one counter at a time Eric Joyner
                   ` (5 more replies)
  0 siblings, 6 replies; 10+ messages in thread
From: Eric Joyner @ 2026-11-16 23:06 UTC (permalink / raw)
  To: netdev
  Cc: Brett Creeley, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Vadim Fedorenko, Nikhil P. Rao,
	Eric Joyner, Eric Dumazet

Add a .get_fec_stats ethtool handler, and fix the FEC reporting it
depends on.

Patch 1 writes the IONIC_STAT_INVALID markers in extra_stats one
counter at a time, so a reader that tests a single counter cannot see a
torn value.

Patch 2 adds FEC off to the supported and advertised link modes.
Firmware does not report which FEC modes a port supports, so the
supported list holds every mode an ionic device can use, and firmware
rejects a request for a mode the port cannot run.

Patches 3-5 fix ionic_get_fecparam(). fec->fec reported the constant
set of supported modes instead of the configured mode, and
fec->active_fec reported the requested mode instead of the operational
one. Patches 2, 3 and 5 have Fixes: tags, but the later patches build on
the lines they change, so they are sent to net-next with them rather
than to net. The bugs only affect what ethtool reports.

Patch 6 adds the handler. It reports statistics only while the active
FEC mode is RS.

The netdev AI review of v2, which Simon forwarded, asked whether the
per-bin local really guarantees a single read of each counter. It does
not, so v3 reads every counter with READ_ONCE() and drops the struct
copy. The port_info lifetime problem that the AI review raised on v1
and v2 is fixed in net-next by commit d90d1a768b3b ("ionic: fix
port_info use-after-free on firmware recovery"). The netdev AI review
of v3, which Jakub forwarded, found that FEC off was not listed as a
supported link mode, which the new patch 2 fixes, and that patch 4 made
old firmware report forced-FEC links as FEC off, which v4 also fixes.

Claude was used while developing this series: to investigate the v2
and v3 review questions, and to help write the code changes and commit
messages. The authors reviewed every change. The series was tested on
Pollara and Vulcano NICs.

Signed-off-by: Eric Joyner <eric.joyner@amd.com>
---
Changes in v4:
- Add patch 2, which reports FEC off as a supported link mode (netdev
  AI review of v3).
- Patch 5: with autoneg off, report the configured FEC mode as active,
  so firmware that predates status.fec_type no longer reports forced-FEC
  links as FEC off, and add a Fixes: tag (netdev AI review of v3).
- Patch 6: gate the statistics on the same active FEC mode as patch 5.
- Patch 6: explain why an RS check is enough for the histogram (netdev
  AI review of v3).
- Rebase onto current net-next.
- Link to v3: https://lore.kernel.org/r/20261104-ionic-get-fec-stats-v3-0-b261fd53d0a7@amd.com

Changes in v3:
- Read each counter with READ_ONCE() and drop the struct copy of
  extra_stats (netdev AI review of v2).
- Report statistics based on status.fec_type, the operational FEC mode,
  instead of config.fec_type, matching the active FEC mode that patch 4
  reports.
- Add patches 1-4, which fix the FEC reporting the handler depends on.
  Patches 1-4 are by Nikhil, who is also a Co-developed-by on patch 5.
- Drop Vadim's Reviewed-by from v1, since the code has changed.
- Rebase onto current net-next.
- Link to v2: https://lore.kernel.org/r/20260814-ionic-get-fec-stats-v2-1-13f0e7d2e702@amd.com

Changes in v2:
- Latch each histogram bin into a local before testing it against
  IONIC_STAT_INVALID, so the sentinel check and the store cannot be
  folded into two separate reads of the firmware-updated DMA buffer.
- Explain in the commit message why the RS FEC check gates every
  reported counter rather than just the histogram: they all come from
  the RS decoder, including fec_corrected_bits_total despite its name.
- Rebased onto current net-next.
- Link to v1: https://lore.kernel.org/r/20260731214021.15279-1-eric.joyner@amd.com

---
Eric Joyner (2):
      ionic: report FEC off as a supported link mode
      ionic: Add .get_fec_stats ethtool handler

Nikhil P. Rao (4):
      ionic: write the extra_stats invalid markers one counter at a time
      ionic: report the configured FEC mode in ethtool_fecparam.fec
      ionic: map the FEC type through a helper in ionic_get_fecparam()
      ionic: report the operational FEC mode from ethtool --show-fec

 .../net/ethernet/pensando/ionic/ionic_ethtool.c    | 137 +++++++++++++++++++--
 drivers/net/ethernet/pensando/ionic/ionic_main.c   |   8 +-
 drivers/net/ethernet/pensando/ionic/ionic_stats.h  |   2 +
 3 files changed, 132 insertions(+), 15 deletions(-)
---
base-commit: 573af801552999954caa8347e4490b04cf4763f0
change-id: 20260814-ionic-get-fec-stats-990736964908

Best regards,
-- 
Eric Joyner <eric.joyner@amd.com>


^ permalink raw reply	[flat|nested] 10+ messages in thread

* [PATCH net-next v4 1/6] ionic: write the extra_stats invalid markers one counter at a time
  2026-11-16 23:06 [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler Eric Joyner
@ 2026-11-16 23:06 ` Eric Joyner
  2026-11-16 23:06 ` [PATCH net-next v4 2/6] ionic: report FEC off as a supported link mode Eric Joyner
                   ` (4 subsequent siblings)
  5 siblings, 0 replies; 10+ messages in thread
From: Eric Joyner @ 2026-11-16 23:06 UTC (permalink / raw)
  To: netdev
  Cc: Brett Creeley, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Vadim Fedorenko, Nikhil P. Rao,
	Eric Joyner, Eric Dumazet

From: "Nikhil P. Rao" <nikhil.rao@amd.com>

ionic_port_init() marks every counter in the device-written extra_stats
block invalid before asking firmware to refresh it, so that a statistic
this driver knows about but the firmware does not is reported as absent
rather than as zero.  It does that with a memset() of the whole block.

The .get_fec_stats handler added later in this series reads those
counters one at a time and tests each against IONIC_STAT_INVALID to
decide whether to report it.  memset() gives no guarantee about the
width of its stores, so such a reader can observe a counter that is
neither its old value nor the marker.

The block is entirely __le64, each member naturally aligned, so walk it
as an array and write the markers one counter at a time instead.  A
counter read on its own is then either a complete value or the marker.

Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Signed-off-by: Eric Joyner <eric.joyner@amd.com>
---
 drivers/net/ethernet/pensando/ionic/ionic_main.c  | 8 ++++++--
 drivers/net/ethernet/pensando/ionic/ionic_stats.h | 2 ++
 2 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/pensando/ionic/ionic_main.c b/drivers/net/ethernet/pensando/ionic/ionic_main.c
index 9955ab77d86d..b8d0ce597176 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_main.c
+++ b/drivers/net/ethernet/pensando/ionic/ionic_main.c
@@ -12,6 +12,7 @@
 #include "ionic.h"
 #include "ionic_bus.h"
 #include "ionic_lif.h"
+#include "ionic_stats.h"
 #include "ionic_debugfs.h"
 
 MODULE_DESCRIPTION(IONIC_DRV_DESCRIPTION);
@@ -727,7 +728,9 @@ int ionic_port_init(struct ionic *ionic)
 	struct ionic_identity *ident = &ionic->ident;
 	struct ionic_dev *idev = &ionic->idev;
 	bool allocated = false;
+	__le64 *stat;
 	size_t sz;
+	size_t i;
 	int err;
 
 	if (!idev->port_info) {
@@ -745,8 +748,9 @@ int ionic_port_init(struct ionic *ionic)
 	/* If the driver knows about more "extra stats" than the firmware,
 	 * make sure these stats are marked as invalid.
 	 */
-	memset(&idev->port_info->extra_stats, 0xff,
-	       sizeof(idev->port_info->extra_stats));
+	stat = (__le64 *)&idev->port_info->extra_stats;
+	for (i = 0; i < sizeof(idev->port_info->extra_stats) / sizeof(*stat); i++)
+		WRITE_ONCE(stat[i], IONIC_STAT_INVALID);
 
 	WRITE_ONCE(idev->link_down_count_init, false);
 
diff --git a/drivers/net/ethernet/pensando/ionic/ionic_stats.h b/drivers/net/ethernet/pensando/ionic/ionic_stats.h
index 7ed935868e84..8a0edcc86fb8 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_stats.h
+++ b/drivers/net/ethernet/pensando/ionic/ionic_stats.h
@@ -4,6 +4,8 @@
 #ifndef _IONIC_STATS_H_
 #define _IONIC_STATS_H_
 
+#include <linux/ethtool.h>
+
 #define IONIC_STAT_INVALID	(cpu_to_le64(~0ULL))
 
 #define IONIC_STAT_TO_OFFSET(type, stat_name) (offsetof(type, stat_name))

-- 
2.53.0


^ permalink raw reply related	[flat|nested] 10+ messages in thread

* [PATCH net-next v4 2/6] ionic: report FEC off as a supported link mode
  2026-11-16 23:06 [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler Eric Joyner
  2026-11-16 23:06 ` [PATCH net-next v4 1/6] ionic: write the extra_stats invalid markers one counter at a time Eric Joyner
@ 2026-11-16 23:06 ` Eric Joyner
  2026-11-17 23:10   ` netdev-bot+sashiko
  2026-11-16 23:06 ` [PATCH net-next v4 3/6] ionic: report the configured FEC mode in ethtool_fecparam.fec Eric Joyner
                   ` (3 subsequent siblings)
  5 siblings, 1 reply; 10+ messages in thread
From: Eric Joyner @ 2026-11-16 23:06 UTC (permalink / raw)
  To: netdev
  Cc: Brett Creeley, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Vadim Fedorenko, Nikhil P. Rao,
	Eric Joyner, Eric Dumazet

ionic_get_link_ksettings() lists BaseR and RS as supported FEC modes but
never lists FEC off, although ionic_set_fecparam() accepts it.  Add
FEC_NONE to the supported modes, and to the advertised modes when the
port is configured with FEC off.

Firmware does not report which FEC modes a port supports, so the list
holds every mode an ionic device can use, and firmware rejects a
request for a mode the port cannot run.  BaseR and RS are already
listed on every port this way.  Off follows the same rule: AINIC
firmware rejects it, while DSC firmware accepts it.

Fixes: 4d03e00a2140 ("ionic: Add initial ethtool support")
Signed-off-by: Eric Joyner <eric.joyner@amd.com>
---
 drivers/net/ethernet/pensando/ionic/ionic_ethtool.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
index 119ac720c078..84378bef9fb8 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
+++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
@@ -298,9 +298,16 @@ static int ionic_get_link_ksettings(struct net_device *netdev,
 
 	linkmode_copy(ks->link_modes.advertising, ks->link_modes.supported);
 
+	/* Firmware does not report which FEC modes a port supports, so list
+	 * every mode an ionic device can use; firmware rejects a request for
+	 * a mode the port cannot run.
+	 */
+	ethtool_link_ksettings_add_link_mode(ks, supported, FEC_NONE);
 	ethtool_link_ksettings_add_link_mode(ks, supported, FEC_BASER);
 	ethtool_link_ksettings_add_link_mode(ks, supported, FEC_RS);
-	if (idev->port_info->config.fec_type == IONIC_PORT_FEC_TYPE_FC)
+	if (idev->port_info->config.fec_type == IONIC_PORT_FEC_TYPE_NONE)
+		ethtool_link_ksettings_add_link_mode(ks, advertising, FEC_NONE);
+	else if (idev->port_info->config.fec_type == IONIC_PORT_FEC_TYPE_FC)
 		ethtool_link_ksettings_add_link_mode(ks, advertising, FEC_BASER);
 	else if (idev->port_info->config.fec_type == IONIC_PORT_FEC_TYPE_RS)
 		ethtool_link_ksettings_add_link_mode(ks, advertising, FEC_RS);

-- 
2.53.0


^ permalink raw reply related	[flat|nested] 10+ messages in thread

* [PATCH net-next v4 3/6] ionic: report the configured FEC mode in ethtool_fecparam.fec
  2026-11-16 23:06 [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler Eric Joyner
  2026-11-16 23:06 ` [PATCH net-next v4 1/6] ionic: write the extra_stats invalid markers one counter at a time Eric Joyner
  2026-11-16 23:06 ` [PATCH net-next v4 2/6] ionic: report FEC off as a supported link mode Eric Joyner
@ 2026-11-16 23:06 ` Eric Joyner
  2026-11-16 23:06 ` [PATCH net-next v4 4/6] ionic: map the FEC type through a helper in ionic_get_fecparam() Eric Joyner
                   ` (2 subsequent siblings)
  5 siblings, 0 replies; 10+ messages in thread
From: Eric Joyner @ 2026-11-16 23:06 UTC (permalink / raw)
  To: netdev
  Cc: Brett Creeley, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Vadim Fedorenko, Nikhil P. Rao,
	Eric Joyner, Eric Dumazet

From: "Nikhil P. Rao" <nikhil.rao@amd.com>

ionic_get_fecparam() fills fec->fec with the constant
ETHTOOL_FEC_OFF | ETHTOOL_FEC_RS | ETHTOOL_FEC_BASER.  The uAPI defines
fec as the configured FEC modes, so a port configured for RS reports
all three:

  Supported/Configured FEC encodings: Off RS BaseR

where it should report RS alone.  Take fec->fec from config.fec_type.
Supported modes belong in ETHTOOL_GLINKSETTINGS, which
ionic_get_link_ksettings() already fills.

fec->active_fec still reads config.fec_type; reporting the operational
mode is a separate change.

Fixes: e95f922f4c2f ("ionic: implement ethtool set-fec")
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Signed-off-by: Eric Joyner <eric.joyner@amd.com>
---
 drivers/net/ethernet/pensando/ionic/ionic_ethtool.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
index 84378bef9fb8..13b5874fe5ad 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
+++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
@@ -434,17 +434,18 @@ static int ionic_get_fecparam(struct net_device *netdev,
 	switch (lif->ionic->idev.port_info->config.fec_type) {
 	case IONIC_PORT_FEC_TYPE_NONE:
 		fec->active_fec = ETHTOOL_FEC_OFF;
+		fec->fec = ETHTOOL_FEC_OFF;
 		break;
 	case IONIC_PORT_FEC_TYPE_RS:
 		fec->active_fec = ETHTOOL_FEC_RS;
+		fec->fec = ETHTOOL_FEC_RS;
 		break;
 	case IONIC_PORT_FEC_TYPE_FC:
 		fec->active_fec = ETHTOOL_FEC_BASER;
+		fec->fec = ETHTOOL_FEC_BASER;
 		break;
 	}
 
-	fec->fec = ETHTOOL_FEC_OFF | ETHTOOL_FEC_RS | ETHTOOL_FEC_BASER;
-
 	return 0;
 }
 

-- 
2.53.0


^ permalink raw reply related	[flat|nested] 10+ messages in thread

* [PATCH net-next v4 4/6] ionic: map the FEC type through a helper in ionic_get_fecparam()
  2026-11-16 23:06 [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler Eric Joyner
                   ` (2 preceding siblings ...)
  2026-11-16 23:06 ` [PATCH net-next v4 3/6] ionic: report the configured FEC mode in ethtool_fecparam.fec Eric Joyner
@ 2026-11-16 23:06 ` Eric Joyner
  2026-11-17 23:10   ` netdev-bot+sashiko
  2026-11-16 23:06 ` [PATCH net-next v4 5/6] ionic: report the operational FEC mode from ethtool --show-fec Eric Joyner
  2026-11-16 23:06 ` [PATCH net-next v4 6/6] ionic: Add .get_fec_stats ethtool handler Eric Joyner
  5 siblings, 1 reply; 10+ messages in thread
From: Eric Joyner @ 2026-11-16 23:06 UTC (permalink / raw)
  To: netdev
  Cc: Brett Creeley, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Vadim Fedorenko, Nikhil P. Rao,
	Eric Joyner, Eric Dumazet

From: "Nikhil P. Rao" <nikhil.rao@amd.com>

ionic_get_fecparam() maps config.fec_type onto the ethtool FEC values
with a switch that assigns both fec->active_fec and fec->fec in each
arm.  A later patch gives the two fields distinct sources, which would
mean duplicating the switch.

Move the mapping into ionic_fec_type_to_ethtool() and call it once per
field.

No functional change.

Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Signed-off-by: Eric Joyner <eric.joyner@amd.com>
---
 .../net/ethernet/pensando/ionic/ionic_ethtool.c    | 34 +++++++++++++---------
 1 file changed, 20 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
index 13b5874fe5ad..8f0f1585116b 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
+++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
@@ -426,25 +426,31 @@ static int ionic_set_pauseparam(struct net_device *netdev,
 	return 0;
 }
 
-static int ionic_get_fecparam(struct net_device *netdev,
-			      struct ethtool_fecparam *fec)
+static u32 ionic_fec_type_to_ethtool(u8 fec_type)
 {
-	struct ionic_lif *lif = netdev_priv(netdev);
-
-	switch (lif->ionic->idev.port_info->config.fec_type) {
+	switch (fec_type) {
 	case IONIC_PORT_FEC_TYPE_NONE:
-		fec->active_fec = ETHTOOL_FEC_OFF;
-		fec->fec = ETHTOOL_FEC_OFF;
-		break;
+		return ETHTOOL_FEC_OFF;
 	case IONIC_PORT_FEC_TYPE_RS:
-		fec->active_fec = ETHTOOL_FEC_RS;
-		fec->fec = ETHTOOL_FEC_RS;
-		break;
+		return ETHTOOL_FEC_RS;
 	case IONIC_PORT_FEC_TYPE_FC:
-		fec->active_fec = ETHTOOL_FEC_BASER;
-		fec->fec = ETHTOOL_FEC_BASER;
-		break;
+		return ETHTOOL_FEC_BASER;
+	default:
+		return 0;
 	}
+}
+
+static int ionic_get_fecparam(struct net_device *netdev,
+			      struct ethtool_fecparam *fec)
+{
+	struct ionic_port_info *port_info;
+	struct ionic_lif *lif;
+
+	lif = netdev_priv(netdev);
+	port_info = lif->ionic->idev.port_info;
+
+	fec->active_fec = ionic_fec_type_to_ethtool(port_info->config.fec_type);
+	fec->fec = ionic_fec_type_to_ethtool(port_info->config.fec_type);
 
 	return 0;
 }

-- 
2.53.0


^ permalink raw reply related	[flat|nested] 10+ messages in thread

* [PATCH net-next v4 5/6] ionic: report the operational FEC mode from ethtool --show-fec
  2026-11-16 23:06 [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler Eric Joyner
                   ` (3 preceding siblings ...)
  2026-11-16 23:06 ` [PATCH net-next v4 4/6] ionic: map the FEC type through a helper in ionic_get_fecparam() Eric Joyner
@ 2026-11-16 23:06 ` Eric Joyner
  2026-11-17 23:10   ` netdev-bot+sashiko
  2026-11-16 23:06 ` [PATCH net-next v4 6/6] ionic: Add .get_fec_stats ethtool handler Eric Joyner
  5 siblings, 1 reply; 10+ messages in thread
From: Eric Joyner @ 2026-11-16 23:06 UTC (permalink / raw)
  To: netdev
  Cc: Brett Creeley, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Vadim Fedorenko, Nikhil P. Rao,
	Eric Joyner, Eric Dumazet

From: "Nikhil P. Rao" <nikhil.rao@amd.com>

ionic_get_fecparam() fills fec->active_fec from config.fec_type, the mode
the host requested, which on an autoneg link is whatever was last
programmed while the link negotiates something else.  On an autoneg
link, read it from status.fec_type, where firmware reports the
operational mode.

status.fec_type was carved out of the reserved area by commit
c4e7a75a096c ("ionic: updates to ionic FW api description"), and
firmware older than that leaves the byte zero, which reads as FEC off.
With autoneg disabled the port runs the configured mode, so keep
reporting config.fec_type there; only autoneg links on such old
firmware report FEC off.

Fixes: e95f922f4c2f ("ionic: implement ethtool set-fec")
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Signed-off-by: Eric Joyner <eric.joyner@amd.com>
---
 drivers/net/ethernet/pensando/ionic/ionic_ethtool.c | 18 ++++++++++++++++--
 1 file changed, 16 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
index 8f0f1585116b..ac728b8aa780 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
+++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
@@ -440,6 +440,18 @@ static u32 ionic_fec_type_to_ethtool(u8 fec_type)
 	}
 }
 
+/* Firmware that predates status.fec_type leaves it zero, which reads as
+ * FEC off.  With autoneg disabled the port runs the configured mode, so
+ * only trust status.fec_type on an autoneg link.
+ */
+static u8 ionic_active_fec_type(const struct ionic_port_info *port_info)
+{
+	if (READ_ONCE(port_info->config.an_enable))
+		return READ_ONCE(port_info->status.fec_type);
+
+	return READ_ONCE(port_info->config.fec_type);
+}
+
 static int ionic_get_fecparam(struct net_device *netdev,
 			      struct ethtool_fecparam *fec)
 {
@@ -449,8 +461,10 @@ static int ionic_get_fecparam(struct net_device *netdev,
 	lif = netdev_priv(netdev);
 	port_info = lif->ionic->idev.port_info;
 
-	fec->active_fec = ionic_fec_type_to_ethtool(port_info->config.fec_type);
-	fec->fec = ionic_fec_type_to_ethtool(port_info->config.fec_type);
+	fec->active_fec =
+		ionic_fec_type_to_ethtool(ionic_active_fec_type(port_info));
+	fec->fec =
+		ionic_fec_type_to_ethtool(READ_ONCE(port_info->config.fec_type));
 
 	return 0;
 }

-- 
2.53.0


^ permalink raw reply related	[flat|nested] 10+ messages in thread

* [PATCH net-next v4 6/6] ionic: Add .get_fec_stats ethtool handler
  2026-11-16 23:06 [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler Eric Joyner
                   ` (4 preceding siblings ...)
  2026-11-16 23:06 ` [PATCH net-next v4 5/6] ionic: report the operational FEC mode from ethtool --show-fec Eric Joyner
@ 2026-11-16 23:06 ` Eric Joyner
  5 siblings, 0 replies; 10+ messages in thread
From: Eric Joyner @ 2026-11-16 23:06 UTC (permalink / raw)
  To: netdev
  Cc: Brett Creeley, Andrew Lunn, David S. Miller, Jakub Kicinski,
	Paolo Abeni, Simon Horman, Vadim Fedorenko, Nikhil P. Rao,
	Eric Joyner, Eric Dumazet

Reports FEC statistics totals and an 802.3ck FEC histogram. Per-lane
counts currently aren't supported. Every counter reported here comes
from the RS decoder, including fec_corrected_bits_total despite its
name, and the only histogram format firmware produces is the 16-bin one
from RS(544,514), so nothing is reported unless RS is the active FEC
mode. The only devices whose firmware reports the histogram do not
support RS(528,514), so RS on those devices is always RS(544,514).

These are physical port counters, so virtual functions are skipped the
same way ionic_get_link_ext_stats() skips them, rather than reporting the
port's counters as if they belonged to the VF.

The reporting of these statistics is gated by DEV_CAP_EXTRA_STATS and
checks for IONIC_STAT_INVALID, since only the newest devices support
reporting all of these stats. Older devices can only report some of the
statistics or not at all, and so the output will properly exclude those
unsupported statistics.

The firmware writes these counters by DMA while they are being read.
Use READ_ONCE() to ensure that the IONIC_STAT_INVALID test and the store
see the same value. Counters are still read one at a time, so the set
reported can span more than one firmware update.

Co-developed-by: Nikhil P. Rao <nikhil.rao@amd.com>
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Signed-off-by: Eric Joyner <eric.joyner@amd.com>
---
 .../net/ethernet/pensando/ionic/ionic_ethtool.c    | 83 ++++++++++++++++++++++
 1 file changed, 83 insertions(+)

diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
index ac728b8aa780..b4835cc52cc7 100644
--- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
+++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
@@ -469,6 +469,88 @@ static int ionic_get_fecparam(struct net_device *netdev,
 	return 0;
 }
 
+#define IONIC_FEC_STAT(dst, src)			\
+	do {						\
+		__le64 __val = READ_ONCE(src);		\
+							\
+		if (__val != IONIC_STAT_INVALID)	\
+			(dst) = le64_to_cpu(__val);	\
+	} while (0)
+
+static const struct ethtool_fec_hist_range ionic_fec_hist_ranges[] = {
+	{ 0, 0},
+	{ 1, 1},
+	{ 2, 2},
+	{ 3, 3},
+	{ 4, 4},
+	{ 5, 5},
+	{ 6, 6},
+	{ 7, 7},
+	{ 8, 8},
+	{ 9, 9},
+	{ 10, 10},
+	{ 11, 11},
+	{ 12, 12},
+	{ 13, 13},
+	{ 14, 14},
+	{ 15, 15},
+	{ 0, 0},
+};
+
+static void
+ionic_fill_fec_hist(const struct ionic_port_extra_stats *port_extra_stats,
+		    struct ethtool_fec_hist *hist)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(port_extra_stats->fec_codeword_error_bin); i++) {
+		__le64 val = READ_ONCE(port_extra_stats->fec_codeword_error_bin[i]);
+
+		if (val == IONIC_STAT_INVALID)
+			return;
+
+		hist->values[i].sum = le64_to_cpu(val);
+	}
+
+	hist->ranges = ionic_fec_hist_ranges;
+}
+
+static void ionic_get_fec_stats(struct net_device *netdev,
+				struct ethtool_fec_stats *fec_stats,
+				struct ethtool_fec_hist *hist)
+{
+	struct ionic_lif *lif = netdev_priv(netdev);
+	struct ionic_port_info *port_info;
+
+	if (lif->ionic->pdev->is_virtfn)
+		return;
+
+	if (test_bit(IONIC_LIF_F_FW_RESET, lif->state))
+		return;
+
+	if (!(lif->ionic->ident.dev.capabilities &
+	      cpu_to_le64(IONIC_DEV_CAP_EXTRA_STATS)))
+		return;
+
+	port_info = lif->ionic->idev.port_info;
+	if (!port_info) {
+		netdev_err_once(netdev, "port_info not initialized\n");
+		return;
+	}
+
+	if (ionic_active_fec_type(port_info) != IONIC_PORT_FEC_TYPE_RS)
+		return;
+
+	IONIC_FEC_STAT(fec_stats->corrected_blocks.total,
+		       port_info->extra_stats.rsfec_correctable_blocks);
+	IONIC_FEC_STAT(fec_stats->uncorrectable_blocks.total,
+		       port_info->extra_stats.rsfec_uncorrectable_blocks);
+	IONIC_FEC_STAT(fec_stats->corrected_bits.total,
+		       port_info->extra_stats.fec_corrected_bits_total);
+
+	ionic_fill_fec_hist(&port_info->extra_stats, hist);
+}
+
 static int ionic_set_fecparam(struct net_device *netdev,
 			      struct ethtool_fecparam *fec)
 {
@@ -1203,6 +1285,7 @@ static const struct ethtool_ops ionic_ethtool_ops = {
 	.get_module_eeprom_by_page	= ionic_get_module_eeprom_by_page,
 	.get_pauseparam		= ionic_get_pauseparam,
 	.set_pauseparam		= ionic_set_pauseparam,
+	.get_fec_stats		= ionic_get_fec_stats,
 	.get_fecparam		= ionic_get_fecparam,
 	.set_fecparam		= ionic_set_fecparam,
 	.get_ts_info		= ionic_get_ts_info,

-- 
2.53.0


^ permalink raw reply related	[flat|nested] 10+ messages in thread

* Re: [PATCH net-next v4 2/6] ionic: report FEC off as a supported link mode
  2026-11-16 23:06 ` [PATCH net-next v4 2/6] ionic: report FEC off as a supported link mode Eric Joyner
@ 2026-11-17 23:10   ` netdev-bot+sashiko
  0 siblings, 0 replies; 10+ messages in thread
From: netdev-bot+sashiko @ 2026-11-17 23:10 UTC (permalink / raw)
  To: eric.joyner
  Cc: netdev, brett.creeley, andrew+netdev, davem, kuba, pabeni, horms,
	vadim.fedorenko, nikhil.rao, edumazet

Thank you for your contribution! Sashiko AI review found 1 potential
issue(s) to consider.

Critical: 0 · High: 0 · Medium: 1 · Low: 0

If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).

- [Medium] ionic_get_link_ksettings()
  (drivers/net/ethernet/pensando/ionic/ionic_ethtool.c) now sets…

--

ionic: report FEC off as a supported link mode

This patch adds FEC_NONE to the supported link modes that
ionic_get_link_ksettings() reports. It also adds FEC_NONE to the
advertised modes when the port is configured with FEC off. According to
the commit message, firmware does not report per-port FEC capabilities,
so every mode any ionic device can use is listed, and firmware rejects
the ones a given port cannot run.

> diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> index 119ac720c0787..84378bef9fb8b 100644
> --- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> +++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> @@ -298,9 +298,16 @@ static int ionic_get_link_ksettings(struct net_device *netdev,
>  
>  	linkmode_copy(ks->link_modes.advertising, ks->link_modes.supported);
>  
> +	/* Firmware does not report which FEC modes a port supports, so list
> +	 * every mode an ionic device can use; firmware rejects a request for
> +	 * a mode the port cannot run.
> +	 */
> +	ethtool_link_ksettings_add_link_mode(ks, supported, FEC_NONE);

[Severity: Medium]
Is the claim in this comment accurate? The firmware ABI in ionic_if.h
seems to define a per-port list of supported FEC types:

drivers/net/ethernet/pensando/ionic/ionic_if.h:
 * @fec_type:       supported fec types
 ...
union ionic_port_identity {
	struct {
		...
		u8     fec_type[4];
		u8     pause_type[2];
		u8     loopback_mode[2];
		__le32 speeds[16];
		...

ionic_port_identify() in ionic_main.c already copies this structure into
ionic->ident.port with memcpy_fromio(). However, nothing in the driver
reads ident.port.fec_type[].

The commit message says AINIC firmware rejects FEC off. With this change,
ethtool on an AINIC port will show "None" as a supported FEC mode. A
request for FEC off then reaches firmware through ionic_set_fecparam():

	if (fec_type != lif->ionic->idev.port_info->config.fec_type) {
		mutex_lock(&lif->ionic->dev_cmd_lock);
		ionic_dev_cmd_port_fec(&lif->ionic->idev, fec_type);
		ret = ionic_dev_cmd_wait(lif->ionic, DEVCMD_TIMEOUT);

and the firmware error comes back to userspace.

If firmware fills identity.fec_type[], could the supported FEC set be
built from it instead of extending the hardcoded list?

If firmware does not fill it, should the comment say that the field
exists but is not populated? As written, the comment contradicts the
documented ABI.

There is also another way to tell the parts apart. The driver already
reads dev_info.asic_type in ionic_init_devinfo(), and
ionic_doorbell_wa() uses it for per-ASIC behaviour. Could asic_type be
used to leave FEC_NONE out on parts whose firmware rejects FEC off?

The later patches in this series do not appear to change this. At the
end of the series, FEC_NONE is still added to supported unconditionally,
and ident.port.fec_type[] is still never read.

>  	ethtool_link_ksettings_add_link_mode(ks, supported, FEC_BASER);
>  	ethtool_link_ksettings_add_link_mode(ks, supported, FEC_RS);

[ ... ]

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261116-ionic-get-fec-stats-v4-0-d25cae9f1e87%40amd.com

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH net-next v4 4/6] ionic: map the FEC type through a helper in ionic_get_fecparam()
  2026-11-16 23:06 ` [PATCH net-next v4 4/6] ionic: map the FEC type through a helper in ionic_get_fecparam() Eric Joyner
@ 2026-11-17 23:10   ` netdev-bot+sashiko
  0 siblings, 0 replies; 10+ messages in thread
From: netdev-bot+sashiko @ 2026-11-17 23:10 UTC (permalink / raw)
  To: eric.joyner
  Cc: netdev, brett.creeley, andrew+netdev, davem, kuba, pabeni, horms,
	vadim.fedorenko, nikhil.rao, edumazet

Thank you for your contribution! Sashiko AI review found more severe
issue(s) in other patches of this series. This mail is only an FYI, in case
you want to address the nit picks below when you respin.

Low priority / nit picks for your consideration:

- [Low] In drivers/net/ethernet/pensando/ionic/ionic_ethtool.c, the
  refactor changes ionic_get_fecparam() so that it reads the device-…

--

ionic: map the FEC type through a helper in ionic_get_fecparam()

This patch moves the mapping from config.fec_type to the ethtool FEC
values out of ionic_get_fecparam() and into a new helper,
ionic_fec_type_to_ethtool(). The helper is then called once for
fec->active_fec and once for fec->fec. The commit message says there
is no functional change.

> diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> index 13b5874fe5ad4..8f0f1585116bc 100644
> --- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> +++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c

[ ... ]

> +static int ionic_get_fecparam(struct net_device *netdev,
> +			      struct ethtool_fecparam *fec)
> +{
> +	struct ionic_port_info *port_info;
> +	struct ionic_lif *lif;
> +
> +	lif = netdev_priv(netdev);
> +	port_info = lif->ionic->idev.port_info;
> +
> +	fec->active_fec = ionic_fec_type_to_ethtool(port_info->config.fec_type);
> +	fec->fec = ionic_fec_type_to_ethtool(port_info->config.fec_type);

[Severity: Low]
Could active_fec and fec end up built from two different values of
port_info->config.fec_type here?

The old code loaded config.fec_type once, as the switch expression, and
set both fields from that one value. The new code does two separate
plain loads, one per ionic_fec_type_to_ethtool() call. There is no
local snapshot, no READ_ONCE() and no lock.

port_info is coherent DMA memory shared with the device:

ionic_main.c:ionic_port_init() {
    ...
	idev->port_info = dma_alloc_coherent(ionic->dev,
				     idev->port_info_sz,
				     &idev->port_info_pa,
				     GFP_KERNEL);
    ...
}

Its address is passed to firmware in ionic_dev_cmd_port_init():

	.port_init.info_pa = cpu_to_le64(idev->port_info_pa),

>From that point, firmware writes config.fec_type. RTNL does not
serialize those writes.

If firmware updates the field between the two loads, ethtool --show-fec
could report a configured/active pair that never existed together. The
compiler cannot merge the two loads either: with -fno-strict-aliasing,
the u32 store to fec->active_fec may alias the u8 source.

Given this, is "No functional change" in the commit message quite
accurate?

The later patch in this series, "ionic: report the operational FEC mode
from ethtool --show-fec", rewrites these lines but keeps the double
read. When autoneg is off, ionic_active_fec_type() returns
READ_ONCE(port_info->config.fec_type), and ionic_get_fecparam() then
reads READ_ONCE(port_info->config.fec_type) again for fec->fec.

Would it be simpler to read config.fec_type once into a local u8 and
use that for both fields?

>  
>  	return 0;
>  }

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261116-ionic-get-fec-stats-v4-0-d25cae9f1e87%40amd.com

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH net-next v4 5/6] ionic: report the operational FEC mode from ethtool --show-fec
  2026-11-16 23:06 ` [PATCH net-next v4 5/6] ionic: report the operational FEC mode from ethtool --show-fec Eric Joyner
@ 2026-11-17 23:10   ` netdev-bot+sashiko
  0 siblings, 0 replies; 10+ messages in thread
From: netdev-bot+sashiko @ 2026-11-17 23:10 UTC (permalink / raw)
  To: eric.joyner
  Cc: netdev, brett.creeley, andrew+netdev, davem, kuba, pabeni, horms,
	vadim.fedorenko, nikhil.rao, edumazet

Thank you for your contribution! Sashiko AI review found 1 potential
issue(s) to consider.

Critical: 0 · High: 0 · Medium: 1 · Low: 0

If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).

- [Medium] The commit says old firmware "leaves the byte zero", so an
  autoneg link on that firmware reports FEC off.

--

ionic: report the operational FEC mode from ethtool --show-fec

On autoneg links, ionic_get_fecparam() now fills active_fec from
status.fec_type, the operational mode reported by firmware. Without
autoneg it keeps using config.fec_type. The commit message says firmware
that predates status.fec_type leaves that byte zero, so on such firmware
an autoneg link reports FEC off.

> diff --git a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> index 8f0f1585116bc..ac728b8aa780e 100644
> --- a/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> +++ b/drivers/net/ethernet/pensando/ionic/ionic_ethtool.c
> @@ -440,6 +440,18 @@ static u32 ionic_fec_type_to_ethtool(u8 fec_type)
>  	}
>  }
>  
> +/* Firmware that predates status.fec_type leaves it zero, which reads as
> + * FEC off.  With autoneg disabled the port runs the configured mode, so
> + * only trust status.fec_type on an autoneg link.
> + */
> +static u8 ionic_active_fec_type(const struct ionic_port_info *port_info)
> +{
> +	if (READ_ONCE(port_info->config.an_enable))
> +		return READ_ONCE(port_info->status.fec_type);

[Severity: Medium]
Is status.fec_type actually guaranteed to be zero on older firmware once
the device has been through a firmware reset?

The commit message says:

    firmware older than that leaves the byte zero, which reads as FEC off.

The driver only zeroes that byte when it first allocates the port_info DMA
buffer. ionic_lif_handle_fw_up() calls ionic_port_init() again. That
function allocates only when idev->port_info is NULL, so after a reset it
reuses the old buffer:

drivers/net/ethernet/pensando/ionic/ionic_main.c:ionic_port_init() {
    ...
	if (!idev->port_info) {
		idev->port_info_sz = ALIGN(sizeof(*idev->port_info), PAGE_SIZE);
		idev->port_info = dma_alloc_coherent(ionic->dev,
    ...
	stat = (__le64 *)&idev->port_info->extra_stats;
	for (i = 0; i < sizeof(idev->port_info->extra_stats) / sizeof(*stat); i++)
		WRITE_ONCE(stat[i], IONIC_STAT_INVALID);
    ...
}

Before PORT_INIT, only extra_stats is reset, to cover fields the firmware
does not know about. Nothing clears status.fec_type. The buffer is freed
only by ionic_port_reset(), which runs from probe and remove.

Say a newer firmware has written IONIC_PORT_FEC_TYPE_RS into
status.fec_type. The device is then reset into firmware older than
c4e7a75a096c, for example by downgrading with devlink flash and
FW_CONTROL ACTIVATE, or by falling back to another image. In that case,
can ionic_get_fecparam() keep reporting the stale RS value as active_fec
on autoneg links until the driver is reloaded? That would hold even if the
link negotiated a different mode or no FEC.

This depends on the old firmware not writing the former reserved byte.
That hasn't been confirmed, but the extra_stats invalidation suggests
firmware writes only the fields it knows about.

Would it make sense to clear status.fec_type, or the whole status block,
in ionic_port_init() before PORT_INIT, the same way extra_stats is
handled?

The later commit "ionic: Add .get_fec_stats ethtool handler" also uses
ionic_active_fec_type(). That path is gated on IONIC_DEV_CAP_EXTRA_STATS,
so old firmware does not reach it, and the rest of the series does not
change this.

> +
> +	return READ_ONCE(port_info->config.fec_type);
> +}
> +

[ ... ]

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261116-ionic-get-fec-stats-v4-0-d25cae9f1e87%40amd.com

^ permalink raw reply	[flat|nested] 10+ messages in thread

end of thread, other threads:[~2026-11-17 23:10 UTC | newest]

Thread overview: 10+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16 23:06 [PATCH net-next v4 0/6] ionic: FEC reporting fixes and .get_fec_stats handler Eric Joyner
2026-11-16 23:06 ` [PATCH net-next v4 1/6] ionic: write the extra_stats invalid markers one counter at a time Eric Joyner
2026-11-16 23:06 ` [PATCH net-next v4 2/6] ionic: report FEC off as a supported link mode Eric Joyner
2026-11-17 23:10   ` netdev-bot+sashiko
2026-11-16 23:06 ` [PATCH net-next v4 3/6] ionic: report the configured FEC mode in ethtool_fecparam.fec Eric Joyner
2026-11-16 23:06 ` [PATCH net-next v4 4/6] ionic: map the FEC type through a helper in ionic_get_fecparam() Eric Joyner
2026-11-17 23:10   ` netdev-bot+sashiko
2026-11-16 23:06 ` [PATCH net-next v4 5/6] ionic: report the operational FEC mode from ethtool --show-fec Eric Joyner
2026-11-17 23:10   ` netdev-bot+sashiko
2026-11-16 23:06 ` [PATCH net-next v4 6/6] ionic: Add .get_fec_stats ethtool handler Eric Joyner

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox