ecc.diff

commit af2393a5280b7c7f4346d866315b7814a845d878 (HEAD -> main)
Author: Kevin Bowling <[email protected]>
Date:   Wed Sep 25 22:54:34 2024 -0700

    e1000: Add PCH ECC Error Handling and Stats

diff --git a/sys/dev/e1000/e1000_hw.h b/sys/dev/e1000/e1000_hw.h
index f17877f3e463..26553f4b7d09 100644
--- a/sys/dev/e1000/e1000_hw.h
+++ b/sys/dev/e1000/e1000_hw.h
@@ -650,6 +650,8 @@ struct e1000_hw_stats {
        u64 o2bspc;
        u64 b2ospc;
        u64 b2ogprc;
+       u64 pbeccsts_c;
+       u64 pbeccsts_uc;
 };
 
 struct e1000_vf_stats {
diff --git a/sys/dev/e1000/if_em.c b/sys/dev/e1000/if_em.c
index ad12171487f6..69b0342f20cb 100644
--- a/sys/dev/e1000/if_em.c
+++ b/sys/dev/e1000/if_em.c
@@ -319,6 +319,7 @@ static void em_if_vlan_filter_enable(struct e1000_softc *);
 static void    em_if_vlan_filter_disable(struct e1000_softc *);
 static void    em_if_vlan_filter_write(struct e1000_softc *);
 static void    em_setup_vlan_hw_support(if_ctx_t ctx);
+static int em_ecc_inject_error(SYSCTL_HANDLER_ARGS);
 static int     em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS);
 static void    em_print_nvm_info(struct e1000_softc *);
 static void    em_fw_version_locked(if_ctx_t);
@@ -1451,7 +1452,7 @@ em_intr(void *arg)
 {
        struct e1000_softc *sc = arg;
        if_ctx_t ctx = sc->ctx;
-       u32 reg_icr;
+       u32 reg_icr, pbeccsts;
 
        reg_icr = E1000_READ_REG(&sc->hw, E1000_ICR);
 
@@ -1479,6 +1480,28 @@ em_intr(void *arg)
         */
        IFDI_INTR_DISABLE(ctx);
 
+       if (sc->hw.mac.type >= e1000_pch_lpt && sc->hw.mac.type <= igb_mac_min) {
+               if (reg_icr & E1000_ICR_ECCER) {
+                       /* Uncorrectable ECC Error */
+
+                       pbeccsts = E1000_READ_REG(&sc->hw, E1000_PBECCSTS);
+
+                       sc->stats.pbeccsts_c +=
+                           pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
+                       sc->stats.pbeccsts_uc += (pbeccsts &
+                           E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
+                           E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
+
+                       /*
+                        * Let admin reset the interface - we can't grab the STATE_LOCK
+                        * mutex in this interrupt filter.
+                        */
+                       sc->ecc_intr_reset = 1;
+
+                       return (FILTER_SCHEDULE_THREAD);
+               }
+       }
+
        /* Link status change */
        if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))
                em_handle_link(ctx);
@@ -1838,6 +1861,18 @@ em_if_update_admin_status(if_ctx_t ctx)
        u32 link_check, thstat, ctrl;
        bool automasked = false;
 
+       if (sc->ecc_intr_reset) {
+               device_printf(dev, "Uncorrectable ECC Error, requesting reset.");
+               sc->ecc_intr_reset = 0;
+               iflib_request_reset(ctx);
+               /*
+                * We need to go through admin again because update_admin_status is
+                * done after checking the context state..
+                */
+               iflib_admin_intr_deferred(ctx);
+               return;
+       }
+
        link_check = thstat = ctrl = 0;
        /* Get the cached link value or read phy for real */
        switch (hw->phy.media_type) {
@@ -3743,6 +3778,11 @@ em_if_intr_enable(if_ctx_t ctx)
                E1000_WRITE_REG(hw, EM_EIAC, sc->ims);
                ims_mask |= sc->ims;
        }
+
+       if (hw->mac.type >= e1000_pch_lpt && hw->mac.type < igb_mac_min) {
+               ims_mask |= E1000_IMS_ECCER;
+       }
+
        E1000_WRITE_REG(hw, E1000_IMS, ims_mask);
        E1000_WRITE_FLUSH(hw);
 }
@@ -4262,6 +4302,7 @@ static void
 em_update_stats_counters(struct e1000_softc *sc)
 {
        u64 prev_xoffrxc = sc->stats.xoffrxc;
+       u32 pbeccsts;
 
        if(sc->hw.phy.media_type == e1000_media_type_copper ||
           (E1000_READ_REG(&sc->hw, E1000_STATUS) & E1000_STATUS_LU)) {
@@ -4358,6 +4399,15 @@ em_update_stats_counters(struct e1000_softc *sc)
                sc->stats.tsctfc +=
                E1000_READ_REG(&sc->hw, E1000_TSCTFC);
        }
+
+       /* PCH Packet Buffer ECC errors */
+       if (sc->hw.mac.type >= e1000_pch_lpt && sc->hw.mac.type < igb_mac_min) {
+               pbeccsts = E1000_READ_REG(&sc->hw, E1000_PBECCSTS);
+               sc->stats.pbeccsts_c += pbeccsts & E1000_PBECCSTS_CORR_ERR_CNT_MASK;
+               sc->stats.pbeccsts_uc += (pbeccsts &
+                   E1000_PBECCSTS_UNCORR_ERR_CNT_MASK) >>
+                   E1000_PBECCSTS_UNCORR_ERR_CNT_SHIFT;
+       }
 }
 
 static uint64_t
@@ -4427,8 +4477,8 @@ em_add_hw_stats(struct e1000_softc *sc)
        struct sysctl_oid_list *child = SYSCTL_CHILDREN(tree);
        struct e1000_hw_stats *stats = &sc->stats;
 
-       struct sysctl_oid *stat_node, *queue_node, *int_node;
-       struct sysctl_oid_list *stat_list, *queue_list, *int_list;
+       struct sysctl_oid *stat_node, *queue_node, *int_node, *ecc_node;
+       struct sysctl_oid_list *stat_list, *queue_list, *int_list, *ecc_list;
 
 #define QUEUE_NAME_LEN 32
        char namebuf[QUEUE_NAME_LEN];
@@ -4708,6 +4758,25 @@ em_add_hw_stats(struct e1000_softc *sc)
        SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_overrun",
                        CTLFLAG_RD, &sc->stats.icrxoc,
                        "Interrupt Cause Receiver Overrun Count");
+
+       if (sc->hw.mac.type >= e1000_pch_lpt && sc->hw.mac.type < igb_mac_min) {
+               ecc_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "ecc_stats",
+                       CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "ECC Statistics");
+               ecc_list = SYSCTL_CHILDREN(ecc_node);
+
+               SYSCTL_ADD_UQUAD(ctx, ecc_list, OID_AUTO, "corrected_errors",
+                               CTLFLAG_RD, &sc->stats.pbeccsts_c,
+                               "Packet Buffer ECC Corrected Errors");
+
+               SYSCTL_ADD_UQUAD(ctx, ecc_list, OID_AUTO, "uncorrected_errors",
+                               CTLFLAG_RD, &sc->stats.pbeccsts_uc,
+                               "Packet Buffer ECC Uncorrected Errors");
+
+               SYSCTL_ADD_PROC(ctx, ecc_list, OID_AUTO, "inject_error",
+                               CTLTYPE_INT | CTLFLAG_RW, sc, 0,
+                               em_ecc_inject_error, "I", "Inject ECC Error");
+       }
+
 }
 
 static void
@@ -4833,6 +4902,27 @@ em_sysctl_print_fw_version(SYSCTL_HANDLER_ARGS)
  *  32 words, stuff that matters is in that extent.
  *
  **********************************************************************/
+static int
+em_ecc_inject_error(SYSCTL_HANDLER_ARGS)
+{
+       struct e1000_softc *sc = (struct e1000_softc *)arg1;
+       int error, result;
+
+       result = -1;
+       error = sysctl_handle_int(oidp, &result, 0, req);
+
+       if (error || !req->newptr)
+               return (error);
+
+       if (result == 1) {
+               device_printf(sc->dev, "Injecting ECC Error\n");
+               E1000_WRITE_REG(&sc->hw, 0x01010, ((1 << 24) | (1 << 23)));
+               E1000_WRITE_REG(&sc->hw, 0x01004, 0xFF01000F);
+       }
+
+       return (error);
+}
+
 static int
 em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS)
 {
diff --git a/sys/dev/e1000/if_em.h b/sys/dev/e1000/if_em.h
index 244762e8ed0d..c725722d8b1c 100644
--- a/sys/dev/e1000/if_em.h
+++ b/sys/dev/e1000/if_em.h
@@ -507,6 +507,7 @@ struct e1000_softc {
        u32                     dmac;
        int                     link_mask;
        int                     tso_automasked;
+       int                     ecc_intr_reset;
 
        u64                     que_mask;