The branch main has been updated by adrian:

URL: 
https://cgit.FreeBSD.org/src/commit/?id=42f3088bd980a469dbbc3e6b568edb7ce0ab2bc3

commit 42f3088bd980a469dbbc3e6b568edb7ce0ab2bc3
Author:     Jérémie Jourdin <[email protected]>
AuthorDate: 2026-07-19 18:47:17 +0000
Commit:     Adrian Chadd <[email protected]>
CommitDate: 2026-07-19 18:47:21 +0000

    re(4): harden re_watchdog() recovery and log controller state
    
    Distinguish the two failure classes from the PR in a single log line
    (ring indices, ISR/IMR, TXCFG, interrupt mode): lost interrupt vs
    genuine DMA stall.
    
    Bail out instead of re-initializing when the controller reads back
    all-ones (fallen off the bus; reinit cannot help).
    
    Re-assert the driver's existing ASPM-disabled policy before reinit,
    since firmware/power transitions re-arming L0s/L1 is a documented
    stall trigger.
    
    Diagnostics-only on the recovered path; no fast-path change.
    
    * Field diagnostics running on an RTL8168H production fleet; the log
      format distinguishes lost-doorbell / DMA-stall / dead-controller
      without a debug build.
    
    Reviewed by:    adrian
    Differential Revision:  https://reviews.freebsd.org/D58279
    PR: kern/166724
---
 sys/dev/re/if_re.c | 61 +++++++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 58 insertions(+), 3 deletions(-)

diff --git a/sys/dev/re/if_re.c b/sys/dev/re/if_re.c
index b04dc3cbaf4a..5a3e8e68d26f 100644
--- a/sys/dev/re/if_re.c
+++ b/sys/dev/re/if_re.c
@@ -3600,6 +3600,8 @@ re_watchdog(struct rl_softc *sc)
 {
        struct epoch_tracker et;
        if_t ifp;
+       const char *imode;
+       uint32_t isr, imr, txcfg;
 
        RL_LOCK_ASSERT(sc);
 
@@ -3607,18 +3609,71 @@ re_watchdog(struct rl_softc *sc)
                return;
 
        ifp = sc->rl_ifp;
+       /*
+        * Snapshot the controller state before reclaim for the diagnostics
+        * below.  Reading RL_ISR is side-effect free: the interrupt status
+        * is cleared by writing ones, not by reading.
+        */
+       imode = (sc->rl_flags & RL_FLAG_MSIX) ? "MSI-X" :
+           (sc->rl_flags & RL_FLAG_MSI) ? "MSI" : "INTx";
+       isr = CSR_READ_2(sc, RL_ISR);
+       imr = CSR_READ_2(sc, RL_IMR);
        re_txeof(sc);
        if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) {
-               if_printf(ifp, "watchdog timeout (missed Tx interrupts) "
-                   "-- recovering\n");
+               /*
+                * The retry reclaim drained the ring: a Tx completion interrupt
+                * was lost.  Log the interrupt state so the lost-interrupt path
+                * can be diagnosed without a debug build.
+                */
+               if_printf(ifp, "watchdog timeout (missed Tx interrupts) -- "
+                   "recovering (ISR 0x%04x IMR 0x%04x %s)\n",
+                   isr, imr, imode);
                if (!if_sendq_empty(ifp))
                        re_start_locked(ifp);
                return;
        }
 
-       if_printf(ifp, "watchdog timeout\n");
+       txcfg = CSR_READ_4(sc, RL_TXCFG);
+       /*
+        * A genuine Tx stall.  Log the Tx ring position and controller
+        * state in a single line -- enough to distinguish a lost doorbell,
+        * a DMA stall, and a controller that has fallen off the bus.
+        */
+       if_printf(ifp, "watchdog timeout -- resetting (tx %d/%d cons %d "
+           "prod %d ISR 0x%04x IMR 0x%04x TXCFG 0x%08x %s)\n",
+           sc->rl_ldata.rl_tx_free, sc->rl_ldata.rl_tx_desc_cnt,
+           sc->rl_ldata.rl_tx_considx, sc->rl_ldata.rl_tx_prodidx,
+           isr, imr, txcfg, imode);
        if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 
+       /*
+        * If the controller reads back all-ones it has fallen off the bus;
+        * re_init_locked() cannot recover it, so flag it and bail rather
+        * than spinning through reset after reset.
+        */
+       if (txcfg == 0xFFFFFFFF) {
+               if_printf(ifp,
+                   "controller not responding; not reinitializing\n");
+               return;
+       }
+
+       /*
+        * ASPM is a common cause of these Tx timeouts.  Make sure L0s/L1
+        * and CLKREQ are still disabled on the link before reinitializing;
+        * firmware or a power transition may have re-armed them.
+        */
+       if (sc->rl_expcap != 0) {
+               uint16_t ctl;
+
+               ctl = pci_read_config(sc->rl_dev,
+                   sc->rl_expcap + PCIER_LINK_CTL, 2);
+               if ((ctl & (PCIEM_LINK_CTL_ECPM | PCIEM_LINK_CTL_ASPMC)) != 0) {
+                       ctl &= ~(PCIEM_LINK_CTL_ECPM | PCIEM_LINK_CTL_ASPMC);
+                       pci_write_config(sc->rl_dev,
+                           sc->rl_expcap + PCIER_LINK_CTL, ctl, 2);
+               }
+       }
+
        NET_EPOCH_ENTER(et);
        re_rxeof(sc, NULL);
        NET_EPOCH_EXIT(et);

Reply via email to