| Message ID | 20260928210552.13266-1-hagan@amazon.com |
|---|---|
| State | Under Review |
| Delegated to: | Anthony Nguyen |
| Headers | show |
| Series | [iwl-net,v3] e1000e: restore jumbo config after DMoff exit | expand |
> -----Original Message----- > From: Craig I. Hagan <hagan@amazon.com> > Sent: Monday, September 28, 2026 11:06 PM > To: Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Kitszel, > Przemyslaw <przemyslaw.kitszel@intel.com> > Cc: Andrew Lunn <andrew+netdev@lunn.ch>; David S . Miller > <davem@davemloft.net>; Eric Dumazet <edumazet@google.com>; Jakub > Kicinski <kuba@kernel.org>; Paolo Abeni <pabeni@redhat.com>; intel- > wired-lan@lists.osuosl.org; netdev@vger.kernel.org; linux- > kernel@vger.kernel.org; stable@vger.kernel.org; Fabio Sabai > <sabfabio@amazon.com>; Matt Vollrath <tactii@gmail.com> > Subject: [PATCH iwl-net v3] e1000e: restore jumbo config after DMoff > exit > > On an I219 (pch_spt+) at jumbo MTU, after a link partner reset (e.g. a > switch reboot) the interface can come back up unable to pass traffic > until it is manually reset. Observed on I219-LM (8086:15b7) at MTU > 9000. > > When the MAC exits the DMoff state on link-up, the watchdog resets > only the PHY. A bare PHY reset does not re-run the jumbo-frame > workaround (e1000_lv_jumbo_workaround_ich8lan), so the jumbo > configuration is left unapplied and traffic stalls. This was seen with > NIC runtime PM disabled; if runtime PM were enabled, the device should > go through a full reset on resume, avoiding this issue. > > Request a full reset (FLAG_RESTART_NOW) instead of the bare PHY reset > when the MAC exits DMoff, so the normal reset path re-applies the > jumbo-frame configuration. > > Fixes: def4ec6dce39 ("e1000e: PCIm function state support") > Cc: stable@vger.kernel.org > Reported-by: Fabio Sabai <sabfabio@amazon.com> > Closes: https://github.com/systemd/systemd/issues/25925 > Suggested-by: Matt Vollrath <tactii@gmail.com> > Reviewed-by: Matt Vollrath <tactii@gmail.com> > Tested-by: Matt Vollrath <tactii@gmail.com> > Assisted-by: LLM > Signed-off-by: Craig I. Hagan <hagan@amazon.com> > --- > v3: Correct the Fixes: tag to def4ec6dce39 ("e1000e: PCIm function > state support"), which introduced the DMoff-exit bare PHY reset; > 2e7256f12cdb only added a pre-condition. This lets stable reach > the affected trees (v5.3+, including 5.4/5.10). Add Reviewed-by/ > Tested-by from Matt Vollrath. > v2: Rework per review: request a full reset on DMoff exit so the > jumbo-frame workaround is re-applied (root cause identified by > Matt Vollrath); drop the diagnostic e_info. v1: > https://lore.kernel.org/netdev/20260923235832.473-1- > hagan@amazon.com/ > > drivers/net/ethernet/intel/e1000e/netdev.c | 8 +++++--- > 1 file changed, 5 insertions(+), 3 deletions(-) > > diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c > b/drivers/net/ethernet/intel/e1000e/netdev.c > index 844f31ab3..df7da86a8 100644 > --- a/drivers/net/ethernet/intel/e1000e/netdev.c > +++ b/drivers/net/ethernet/intel/e1000e/netdev.c > @@ -5264,9 +5264,11 @@ static void e1000_watchdog_task(struct > work_struct *work) > usleep_range(10000, 20000); > pcim_state = er32(STATUS); > > - /* Checking if MAC exited DMoff > state */ > - if (!(pcim_state & > E1000_STATUS_PCIM_STATE)) > - e1000_phy_hw_reset(&adapter- > >hw); > + /* MAC exited DMoff; do a full > reset. */ > + if (!(pcim_state & > E1000_STATUS_PCIM_STATE)) { > + adapter->flags |= > FLAG_RESTART_NOW; > + goto link_up; > + } > } > } > > -- > 2.47.3 Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c index 844f31ab3..df7da86a8 100644 --- a/drivers/net/ethernet/intel/e1000e/netdev.c +++ b/drivers/net/ethernet/intel/e1000e/netdev.c @@ -5264,9 +5264,11 @@ static void e1000_watchdog_task(struct work_struct *work) usleep_range(10000, 20000); pcim_state = er32(STATUS); - /* Checking if MAC exited DMoff state */ - if (!(pcim_state & E1000_STATUS_PCIM_STATE)) - e1000_phy_hw_reset(&adapter->hw); + /* MAC exited DMoff; do a full reset. */ + if (!(pcim_state & E1000_STATUS_PCIM_STATE)) { + adapter->flags |= FLAG_RESTART_NOW; + goto link_up; + } } }
On an I219 (pch_spt+) at jumbo MTU, after a link partner reset (e.g. a switch reboot) the interface can come back up unable to pass traffic until it is manually reset. Observed on I219-LM (8086:15b7) at MTU 9000. When the MAC exits the DMoff state on link-up, the watchdog resets only the PHY. A bare PHY reset does not re-run the jumbo-frame workaround (e1000_lv_jumbo_workaround_ich8lan), so the jumbo configuration is left unapplied and traffic stalls. This was seen with NIC runtime PM disabled; if runtime PM were enabled, the device should go through a full reset on resume, avoiding this issue. Request a full reset (FLAG_RESTART_NOW) instead of the bare PHY reset when the MAC exits DMoff, so the normal reset path re-applies the jumbo-frame configuration. Fixes: def4ec6dce39 ("e1000e: PCIm function state support") Cc: stable@vger.kernel.org Reported-by: Fabio Sabai <sabfabio@amazon.com> Closes: https://github.com/systemd/systemd/issues/25925 Suggested-by: Matt Vollrath <tactii@gmail.com> Reviewed-by: Matt Vollrath <tactii@gmail.com> Tested-by: Matt Vollrath <tactii@gmail.com> Assisted-by: LLM Signed-off-by: Craig I. Hagan <hagan@amazon.com> --- v3: Correct the Fixes: tag to def4ec6dce39 ("e1000e: PCIm function state support"), which introduced the DMoff-exit bare PHY reset; 2e7256f12cdb only added a pre-condition. This lets stable reach the affected trees (v5.3+, including 5.4/5.10). Add Reviewed-by/ Tested-by from Matt Vollrath. v2: Rework per review: request a full reset on DMoff exit so the jumbo-frame workaround is re-applied (root cause identified by Matt Vollrath); drop the diagnostic e_info. v1: https://lore.kernel.org/netdev/20260923235832.473-1-hagan@amazon.com/ drivers/net/ethernet/intel/e1000e/netdev.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)