bnx2x: Fix NULL pointer dereference in bnx2x_del_all_vlans() on some hw
authorIvan Mironov <mironov.ivan@gmail.com>
Mon, 24 Dec 2018 15:13:05 +0000 (20:13 +0500)
committerDavid S. Miller <davem@davemloft.net>
Mon, 24 Dec 2018 22:45:51 +0000 (14:45 -0800)
This happened when I tried to boot normal Fedora 29 system with latest
available kernel (from fedora rawhide, plus some unrelated custom
patches):

BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
PGD 0 P4D 0
Oops: 0010 [#1] SMP PTI
CPU: 6 PID: 1422 Comm: libvirtd Tainted: G          I       4.20.0-0.rc7.git3.hpsa2.1.fc29.x86_64 #1
Hardware name: HP ProLiant BL460c G6, BIOS I24 05/21/2018
RIP: 0010:          (null)
Code: Bad RIP value.
RSP: 0018:ffffa47ccdc9fbe0 EFLAGS: 00010246
RAX: 0000000000000000 RBX: 00000000000003e8 RCX: ffffa47ccdc9fbf8
RDX: ffffa47ccdc9fc00 RSI: ffff97d9ee7b01f8 RDI: ffff97d9f0150b80
RBP: ffff97d9f0150b80 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000003
R13: ffff97d9ef1e53e8 R14: 0000000000000009 R15: ffff97d9f0ac6730
FS:  00007f4d224ef700(0000) GS:ffff97d9fa200000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffffffffffffd6 CR3: 00000011ece52006 CR4: 00000000000206e0
Call Trace:
 ? bnx2x_chip_cleanup+0x195/0x610 [bnx2x]
 ? bnx2x_nic_unload+0x1e2/0x8f0 [bnx2x]
 ? bnx2x_reload_if_running+0x24/0x40 [bnx2x]
 ? bnx2x_set_features+0x79/0xa0 [bnx2x]
 ? __netdev_update_features+0x244/0x9e0
 ? netlink_broadcast_filtered+0x136/0x4b0
 ? netdev_update_features+0x22/0x60
 ? dev_disable_lro+0x1c/0xe0
 ? devinet_sysctl_forward+0x1c6/0x211
 ? proc_sys_call_handler+0xab/0x100
 ? __vfs_write+0x36/0x1a0
 ? rcu_read_lock_sched_held+0x79/0x80
 ? rcu_sync_lockdep_assert+0x2e/0x60
 ? __sb_start_write+0x14c/0x1b0
 ? vfs_write+0x159/0x1c0
 ? vfs_write+0xba/0x1c0
 ? ksys_write+0x52/0xc0
 ? do_syscall_64+0x60/0x1f0
 ? entry_SYSCALL_64_after_hwframe+0x49/0xbe

After some investigation I figured out that recently added cleanup code
tries to call VLAN filtering de-initialization function which exist only
for newer hardware. Corresponding function pointer is not
set (== 0) for older hardware, namely these chips:

#define CHIP_NUM_57710 0x164e
#define CHIP_NUM_57711 0x164f
#define CHIP_NUM_57711E 0x1650

And I have one of those in my test system:

Broadcom Inc. and subsidiaries NetXtreme II BCM57711E 10-Gigabit PCIe [14e4:1650]

Function bnx2x_init_vlan_mac_fp_objs() from
drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.h decides whether to
initialize relevant pointers in bnx2x_sp_objs.vlan_obj or not.

This regression was introduced after v4.20-rc7, and still exists in v4.20
release.

Fixes: 04f05230c5c13 ("bnx2x: Remove configured vlans as part of unload sequence.")
Signed-off-by: Ivan Mironov <mironov.ivan@gmail.com>
Signed-off-by: Ivan Mironov <mironov.ivan@gmail.com>
Acked-by: Sudarsana Kalluru <Sudarsana.Kalluru@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/broadcom/bnx2x/bnx2x_main.c

index b164f705709d083576e92fb4e2fd007cd3dbe9bb..3b5b47e98c73d1aa458938e586721b3ad89221b1 100644 (file)
@@ -9360,10 +9360,16 @@ void bnx2x_chip_cleanup(struct bnx2x *bp, int unload_mode, bool keep_link)
                BNX2X_ERR("Failed to schedule DEL commands for UC MACs list: %d\n",
                          rc);
 
-       /* Remove all currently configured VLANs */
-       rc = bnx2x_del_all_vlans(bp);
-       if (rc < 0)
-               BNX2X_ERR("Failed to delete all VLANs\n");
+       /* The whole *vlan_obj structure may be not initialized if VLAN
+        * filtering offload is not supported by hardware. Currently this is
+        * true for all hardware covered by CHIP_IS_E1x().
+        */
+       if (!CHIP_IS_E1x(bp)) {
+               /* Remove all currently configured VLANs */
+               rc = bnx2x_del_all_vlans(bp);
+               if (rc < 0)
+                       BNX2X_ERR("Failed to delete all VLANs\n");
+       }
 
        /* Disable LLH */
        if (!CHIP_IS_E1(bp))