]> www.infradead.org Git - users/jedix/linux-maple.git/commitdiff
net: enetc: read TSN capabilities from port register, not SI
authorVladimir Oltean <vladimir.oltean@nxp.com>
Mon, 25 Nov 2024 09:07:18 +0000 (17:07 +0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 29 Nov 2024 12:52:04 +0000 (12:52 +0000)
Configuring TSN (Qbv, Qbu, PSFP) capabilities requires access to port
registers, which are available to the PSI but not the VSI.

Yet, the SI port capability register 0 (PSICAPR0), exposed to both PSIs
and VSIs, presents the same capabilities to the VF as to the PF, thus
leading the VF driver into thinking it can configure these features.

In the case of ENETC_SI_F_QBU, having it set in the VF leads to a crash:

root@ls1028ardb:~# tc qdisc add dev eno0vf0 parent root handle 100: \
mqprio num_tc 4 map 0 0 1 1 2 2 3 3 queues 1@0 1@1 1@2 1@3 hw 1
[  187.290775] Unable to handle kernel paging request at virtual address 0000000000001f00
[  187.424831] pc : enetc_mm_commit_preemptible_tcs+0x1c4/0x400
[  187.430518] lr : enetc_mm_commit_preemptible_tcs+0x30c/0x400
[  187.511140] Call trace:
[  187.513588]  enetc_mm_commit_preemptible_tcs+0x1c4/0x400
[  187.518918]  enetc_setup_tc_mqprio+0x180/0x214
[  187.523374]  enetc_vf_setup_tc+0x1c/0x30
[  187.527306]  mqprio_enable_offload+0x144/0x178
[  187.531766]  mqprio_init+0x3ec/0x668
[  187.535351]  qdisc_create+0x15c/0x488
[  187.539023]  tc_modify_qdisc+0x398/0x73c
[  187.542958]  rtnetlink_rcv_msg+0x128/0x378
[  187.547064]  netlink_rcv_skb+0x60/0x130
[  187.550910]  rtnetlink_rcv+0x18/0x24
[  187.554492]  netlink_unicast+0x300/0x36c
[  187.558425]  netlink_sendmsg+0x1a8/0x420
[  187.606759] ---[ end trace 0000000000000000 ]---

while the other TSN features in the VF are harmless, because the
net_device_ops used for the VF driver do not expose entry points for
these other features.

These capability bits are in the process of being defeatured from the SI
registers. We should read them from the port capability register, where
they are also present, and which is naturally only exposed to the PF.

The change to blame (relevant for stable backports) is the one where
this started being a problem, aka when the kernel started to crash due
to the wrong capability seen by the VF driver.

Fixes: 827145392a4a ("net: enetc: only commit preemptible TCs to hardware when MM TX is active")
Reported-by: Wei Fang <wei.fang@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/freescale/enetc/enetc.c
drivers/net/ethernet/freescale/enetc/enetc_hw.h
drivers/net/ethernet/freescale/enetc/enetc_pf.c

index 35634c516e266e69812dff35fc8312d895062cfd..bece220535a1a4873f41832df554393169dc10bf 100644 (file)
@@ -1756,15 +1756,6 @@ void enetc_get_si_caps(struct enetc_si *si)
                rss = enetc_rd(hw, ENETC_SIRSSCAPR);
                si->num_rss = ENETC_SIRSSCAPR_GET_NUM_RSS(rss);
        }
-
-       if (val & ENETC_SIPCAPR0_QBV)
-               si->hw_features |= ENETC_SI_F_QBV;
-
-       if (val & ENETC_SIPCAPR0_QBU)
-               si->hw_features |= ENETC_SI_F_QBU;
-
-       if (val & ENETC_SIPCAPR0_PSFP)
-               si->hw_features |= ENETC_SI_F_PSFP;
 }
 EXPORT_SYMBOL_GPL(enetc_get_si_caps);
 
index 7c3285584f8a5bcc6f8640ba60ba0b97a8604649..55ba949230ffd097d894046d37ad369e303ba44c 100644 (file)
 #define ENETC_SICTR0   0x18
 #define ENETC_SICTR1   0x1c
 #define ENETC_SIPCAPR0 0x20
-#define ENETC_SIPCAPR0_PSFP    BIT(9)
 #define ENETC_SIPCAPR0_RSS     BIT(8)
-#define ENETC_SIPCAPR0_QBV     BIT(4)
-#define ENETC_SIPCAPR0_QBU     BIT(3)
 #define ENETC_SIPCAPR0_RFS     BIT(2)
 #define ENETC_SIPCAPR1 0x24
 #define ENETC_SITGTGR  0x30
@@ -194,6 +191,9 @@ enum enetc_bdr_type {TX, RX};
 #define ENETC_PCAPR0           0x0900
 #define ENETC_PCAPR0_RXBDR(val)        ((val) >> 24)
 #define ENETC_PCAPR0_TXBDR(val)        (((val) >> 16) & 0xff)
+#define ENETC_PCAPR0_PSFP      BIT(9)
+#define ENETC_PCAPR0_QBV       BIT(4)
+#define ENETC_PCAPR0_QBU       BIT(3)
 #define ENETC_PCAPR1           0x0904
 #define ENETC_PSICFGR0(n)      (0x0940 + (n) * 0xc)  /* n = SI index */
 #define ENETC_PSICFGR0_SET_TXBDR(val)  ((val) & 0xff)
index c47b4a743d93b97493db4cb393ca6a57f9bfdb1e..203862ec111427b857c920d135cc764b4e6f1c55 100644 (file)
@@ -409,6 +409,23 @@ static void enetc_port_assign_rfs_entries(struct enetc_si *si)
        enetc_port_wr(hw, ENETC_PRFSMR, ENETC_PRFSMR_RFSE);
 }
 
+static void enetc_port_get_caps(struct enetc_si *si)
+{
+       struct enetc_hw *hw = &si->hw;
+       u32 val;
+
+       val = enetc_port_rd(hw, ENETC_PCAPR0);
+
+       if (val & ENETC_PCAPR0_QBV)
+               si->hw_features |= ENETC_SI_F_QBV;
+
+       if (val & ENETC_PCAPR0_QBU)
+               si->hw_features |= ENETC_SI_F_QBU;
+
+       if (val & ENETC_PCAPR0_PSFP)
+               si->hw_features |= ENETC_SI_F_PSFP;
+}
+
 static void enetc_port_si_configure(struct enetc_si *si)
 {
        struct enetc_pf *pf = enetc_si_priv(si);
@@ -416,6 +433,8 @@ static void enetc_port_si_configure(struct enetc_si *si)
        int num_rings, i;
        u32 val;
 
+       enetc_port_get_caps(si);
+
        val = enetc_port_rd(hw, ENETC_PCAPR0);
        num_rings = min(ENETC_PCAPR0_RXBDR(val), ENETC_PCAPR0_TXBDR(val));