Re: [PATCH] IB/mlx5: Fix decision to avoid using MAD_IFC command in ISSI > 0 mode

On Sun, Sep 11, 2016 at 10:15 AM, Leon Romanovsky [off-list ref] wrote:

quoted

Hi David,
Please find this UNTESTED patch. We will do formal testing during the
coming work week and will properly submit it for inclusion for 4.8.

From 9147fabc9b189e09a982de8ac30f01f04468f6ce Mon Sep 17 00:00:00 2001
From: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Date: Sun, 11 Sep 2016 10:00:27 +0300
Subject: [PATCH rdma-rc] IB/mlx5: Enable MAD_IFC commands for IB ports only

MAD_IFC command is supported only for physical function (PF) drivers
and only when physical port is IB.

The lack of check if port is IB caused to following trace to appear.

The word drivers isn't accurate. The change log doesn't say enough on
the nature of the fix. You can say
"MAD_IFC command is supported only for physical function (PF) and when
the port link type is IB, enforce that"

[    8.456327] mlx5_core 0000:03:00.0: firmware version: 12.12.780

does the FW version matters here or the bug/fix apply for all GA FWs
that support IB SRIOV and ETH (Roce)?
...
[   10.417421] mlx5_ib: Mellanox Connect-IB Infiniband driver v2.2-1 (Feb 2014)
[   10.419282] ------------[ cut here ]------------
[   10.419291] WARNING: CPU: 2 PID: 2517 at
../drivers/infiniband/core/cache.c:702
ib_cache_gid_set_default_gid+0x2f8/0x340 [ib_core]


This trace teaches us nothing.  If you really want to keep it here,
say something what the trace means

()
[   10.419386] CPU: 2 PID: 2517 Comm: modprobe Tainted: G X 4.4.19-1-default #1
[   10.419387] Hardware name: Dell Inc. PowerEdge R730xd/072T6D,
BIOS2.1.7 06/16/2016
[   10.419389]  0000000000000000 ffffffff8130d740 0000000000000000
ffffffffa04e0300
[   10.419395]  ffffffff8107c121
[   10.419400]  ffff88017bfe0000 ffff88003712b9e0 ffff88045ad905c0
[   10.419401]  0000000000000001 fffffffffffffffc ffffffffa04d8a58
0000000000000000
[   10.419406] Call Trace:
[   10.419415]  [<ffffffff81019a59>] dump_trace+0x59/0x310
[   10.419419]  [<ffffffff81019dfa>] show_stack_log_lvl+0xea/0x170
[   10.419421]  [<ffffffff8101ab81>] show_stack+0x21/0x40
[   10.419426]  [<ffffffff8130d740>] dump_stack+0x5c/0x7c
[   10.419431]  [<ffffffff8107c121>] warn_slowpath_common+0x81/0xb0
[   10.419436]  [<ffffffffa04d8a58>]
ib_cache_gid_set_default_gid+0x2f8/0x340 [ib_core]
[   10.419449]  [<ffffffffa04da2dd>] add_netdev_ips+0x9d/0xa0 [ib_core]
[   10.419456]  [<ffffffffa04da45b>] enum_all_gids_of_dev_cb+0x7b/0xb0 [ib_core]
[   10.419461]  [<ffffffffa04d641d>] ib_enum_roce_netdev+0xdd/0x100 [ib_core]
[   10.419466]  [<ffffffffa04da5ed>] roce_rescan_device+0x1d/0x20 [ib_core]
[   10.419470]  [<ffffffffa04d8cdb>] ib_cache_setup_one+0x23b/0x3d0 [ib_core]
[   10.419475]  [<ffffffffa04d606b>] ib_register_device+0x2bb/0x4f0 [ib_core]
[   10.419483]  [<ffffffffa0618bbf>] mlx5_ib_add+0xaaf/0x12e0 [mlx5_ib]
[   10.419492]  [<ffffffffa08b76c1>] mlx5_add_device+0x41/0xa0 [mlx5_core]
[   10.419498]  [<ffffffffa08b7785>] mlx5_register_interface+0x65/0xa0
[mlx5_core]
[   10.419502]  [<ffffffffa0474030>] mlx5_ib_init+0x30/0x42 [mlx5_ib]
[   10.419506]  [<ffffffff81002138>] do_one_initcall+0xc8/0x1f0
[   10.419510]  [<ffffffff811827e8>] do_init_module+0x5a/0x1d7
[   10.419514]  [<ffffffff81103536>] load_module+0x1366/0x1c50
[   10.419518]  [<ffffffff81103fd0>] SYSC_finit_module+0x70/0xa0
[   10.419523]  [<ffffffff815e126e>] entry_SYSCALL_64_fastpath+0x12/0x6d
[   10.420681] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x12/0x6d
[   10.420682] Leftover inexact backtrace:
[   10.420684] ---[ end trace fc8ccb16c9d8e28a ]---


say here what commit/s you are fixing, add Fixes: line  -- I assume
this bug is here before 4.8-rc1 so the fix needs to go anyway to
stable kernels. As we're close to rc6, its better to push the patch
for rdma-next (4.9) and later carry it back to stable.

Reported-by: David Chang <dchang-IBi9RG/b67k@public.gmane.org>
Signed-off-by: Noa Osherovich <noaos-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 drivers/infiniband/hw/mlx5/main.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/mlx5/main.c

b/drivers/infiniband/hw/mlx5/main.c
index 8150ea3..0480b64 100644

--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c

@@ -288,7 +288,9 @@ __be16 mlx5_get_roce_udp_sport(struct mlx5_ib_dev

*dev, u8 port_num,

 static int mlx5_use_mad_ifc(struct mlx5_ib_dev *dev)
 {
- return !MLX5_CAP_GEN(dev->mdev, ib_virt);
+ if (MLX5_CAP_GEN(dev->mdev, port_type) == MLX5_CAP_PORT_TYPE_IB)
+ return !MLX5_CAP_GEN(dev->mdev, ib_virt);
+ return 0;
 }

 enum {