From: Maciej Fijalkowski <maciej.fijalkowski@xxxxxxxxx>
The approach of having XDP queue per CPU regardless of user's setting
exposed a hidden bug that could occur in case when Rx queue count differ
from Tx queue count. Currently vsi->txq_map's size is equal to the
doubled vsi->alloc_txq, which is not correct due to the fact that XDP
rings were previously based on the Rx queue count. Below splat can be
seen when ethtool -L is used and XDP rings are configured:
[ 682.875339] BUG: kernel NULL pointer dereference, address: 000000000000000f
[ 682.883403] #PF: supervisor read access in kernel mode
[ 682.889345] #PF: error_code(0x0000) - not-present page
[ 682.895289] PGD 0 P4D 0
[ 682.898218] Oops: 0000 [#1] PREEMPT SMP PTI
[ 682.903055] CPU: 42 PID: 2878 Comm: ethtool Tainted: G OE 5.15.0-rc5+ #1
[ 682.912214] Hardware name: Intel Corp. GRANTLEY/GRANTLEY, BIOS GRRFCRB1.86B.0276.D07.1605190235 05/19/2016
[ 682.923380] RIP: 0010:devres_remove+0x44/0x130
[ 682.928527] Code: 49 89 f4 55 48 89 fd 4c 89 ff 53 48 83 ec 10 e8 92 b9 49 00 48 8b 9d a8 02 00 00 48 8d 8d a0 02 00 00 49 89 c2 48 39 cb 74 0f <4c> 3b 63 10 74 25 48 8b 5b 08 48 39 cb 75 f1 4c 89 ff 4c 89 d6 e8
[ 682.950237] RSP: 0018:ffffc90006a679f0 EFLAGS: 00010002
[ 682.956285] RAX: 0000000000000286 RBX: ffffffffffffffff RCX: ffff88908343a370
[ 682.964538] RDX: 0000000000000001 RSI: ffffffff81690d60 RDI: 0000000000000000
[ 682.972789] RBP: ffff88908343a0d0 R08: 0000000000000000 R09: 0000000000000000
[ 682.981040] R10: 0000000000000286 R11: 3fffffffffffffff R12: ffffffff81690d60
[ 682.989282] R13: ffffffff81690a00 R14: ffff8890819807a8 R15: ffff88908343a36c
[ 682.997535] FS: 00007f08c7bfa740(0000) GS:ffff88a03fd00000(0000) knlGS:0000000000000000
[ 683.006910] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 683.013557] CR2: 000000000000000f CR3: 0000001080a66003 CR4: 00000000003706e0
[ 683.021819] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 683.030075] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 683.038336] Call Trace:
[ 683.041167] devm_kfree+0x33/0x50
[ 683.045004] ice_vsi_free_arrays+0x5e/0xc0 [ice]
[ 683.050380] ice_vsi_rebuild+0x4c8/0x750 [ice]
[ 683.055543] ice_vsi_recfg_qs+0x9a/0x110 [ice]
[ 683.060697] ice_set_channels+0x14f/0x290 [ice]
[ 683.065962] ethnl_set_channels+0x333/0x3f0
[ 683.070807] genl_family_rcv_msg_doit+0xea/0x150
[ 683.076152] genl_rcv_msg+0xde/0x1d0
[ 683.080289] ? channels_prepare_data+0x60/0x60
[ 683.085432] ? genl_get_cmd+0xd0/0xd0
[ 683.089667] netlink_rcv_skb+0x50/0xf0
[ 683.094006] genl_rcv+0x24/0x40
[ 683.097638] netlink_unicast+0x239/0x340
[ 683.102177] netlink_sendmsg+0x22e/0x470
[ 683.106717] sock_sendmsg+0x5e/0x60
[ 683.110756] __sys_sendto+0xee/0x150
[ 683.114894] ? handle_mm_fault+0xd0/0x2a0
[ 683.119535] ? do_user_addr_fault+0x1f3/0x690
[ 683.134173] __x64_sys_sendto+0x25/0x30
[ 683.148231] do_syscall_64+0x3b/0xc0
[ 683.161992] entry_SYSCALL_64_after_hwframe+0x44/0xae
Fix this by taking into account the value that num_possible_cpus()
yields in addition to vsi->alloc_txq instead of doubling the latter.
Fixes: efc2214b6047 ("ice: Add support for XDP")
Fixes: 22bf877e528f ("ice: introduce XDP_TX fallback path")
Reviewed-by: Alexander Lobakin <alexandr.lobakin@xxxxxxxxx>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@xxxxxxxxx>
Tested-by: Kiran Bhandare <kiranx.bhandare@xxxxxxxxx>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@xxxxxxxxx>
---
drivers/net/ethernet/intel/ice/ice_lib.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/intel/ice/ice_lib.c b/drivers/net/ethernet/intel/ice/ice_lib.c
index 40562600a8cf..09a3297cd63c 100644
--- a/drivers/net/ethernet/intel/ice/ice_lib.c
+++ b/drivers/net/ethernet/intel/ice/ice_lib.c
@@ -89,8 +89,13 @@ static int ice_vsi_alloc_arrays(struct ice_vsi *vsi)
if (!vsi->rx_rings)
goto err_rings;
- /* XDP will have vsi->alloc_txq Tx queues as well, so double the size */
- vsi->txq_map = devm_kcalloc(dev, (2 * vsi->alloc_txq),
+ /* txq_map needs to have enough space to track both Tx (stack) rings
+ * and XDP rings; at this point vsi->num_xdp_txq might not be set,
+ * so use num_possible_cpus() as we want to always provide XDP ring
+ * per CPU, regardless of queue count settings from user that might
+ * have come from ethtool's set_channels() callback;
+ */
+ vsi->txq_map = devm_kcalloc(dev, (vsi->alloc_txq + num_possible_cpus()),
sizeof(*vsi->txq_map), GFP_KERNEL);
if (!vsi->txq_map)
--
2.31.1
