[dpdk-dev] [PATCH] net/mlx5: fix port dead lock when port shutdown

Yongseok Koh yskoh at mellanox.com
Thu May 24 20:33:50 CEST 2018


> On May 24, 2018, at 11:29 AM, Yongseok Koh <yskoh at mellanox.com> wrote:
> 
>> 
>> On May 24, 2018, at 8:20 AM, Xueming Li <xuemingl at mellanox.com> wrote:
>> 
>> This patch add check to avoid init device MR data twcie which caused
>> priv data linked to itself in list.
> 
> A typo.
> 
>> Fixes: 974f1e7ef146 ("net/mlx5: add new memory region support")
>> Cc: yskoh at mellanox.com
>> 
>> Signed-off-by: Xueming Li <xuemingl at mellanox.com>
>> ---
>> drivers/net/mlx5/mlx5_ethdev.c | 2 ++
>> drivers/net/mlx5/mlx5_mr.c     | 1 +
>> 2 files changed, 3 insertions(+)
>> 
>> diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
>> index c52ec6fb7..7ffd17ff3 100644
>> --- a/drivers/net/mlx5/mlx5_ethdev.c
>> +++ b/drivers/net/mlx5/mlx5_ethdev.c
>> @@ -398,6 +398,8 @@ mlx5_dev_configure(struct rte_eth_dev *dev)
>> 	 * deadlock. If it overflows, lookup should be done by searching MR list
>> 	 * linearly, which is slow.
>> 	 */
>> +	if (priv->mr.cache.table)
>> +		return 0;
> 
> Thanks, nice catch!
> 
> Adrien submitted a fix [1] for the similar issue. Adding device to the memory event
> callback list is moved to mlx5_pci_probe() as it should be done once. So, how about
> moving the rest (mlx5_mr_btree_init()) to mlx5_pci_probe() as well? I mean, instead
> of adding this check, you can move the btree_init() above the line where Adrien moved
> LIST_INSERT_HEAD() to. You'll need to rebase on Adrien's patch and add a comment to
> Shahaf there's dependency.
> 
> 
> [1]
> net/mlx5: fix crash when configure is not called
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fdpdk.org%2Fdev%2Fpatchwork%2Fpatch%2F40404%2F&data=02%7C01%7Cyskoh%40mellanox.com%7Ca731aadfe6834a7a435f08d5c1a45fcf%7Ca652971c7d2e4d9ba6a4d149256f461b%7C0%7C0%7C636627834121490391&sdata=D8HjhDMzMfRucGCTSWCieRkNmptGXLJHo8it0Py%2F3gY%3D&reserved=0

And you might want to write the same patch for mlx4.

Thanks,
Yongseok

>> 	if (mlx5_mr_btree_init(&priv->mr.cache, MLX5_MR_BTREE_CACHE_N * 2,
>> 			       dev->device->numa_node)) {
>> 		/* rte_errno is already set. */
>> diff --git a/drivers/net/mlx5/mlx5_mr.c b/drivers/net/mlx5/mlx5_mr.c
>> index abb1f5179..08105a443 100644
>> --- a/drivers/net/mlx5/mlx5_mr.c
>> +++ b/drivers/net/mlx5/mlx5_mr.c
>> @@ -191,6 +191,7 @@ mlx5_mr_btree_init(struct mlx5_mr_btree *bt, int n, int socket)
>> 		rte_errno = EINVAL;
>> 		return -rte_errno;
>> 	}
>> +	assert(!bt->table && !bt->size);
>> 	memset(bt, 0, sizeof(*bt));
>> 	bt->table = rte_calloc_socket("B-tree table",
>> 				      n, sizeof(struct mlx5_mr_cache),
>> -- 
>> 2.13.3



More information about the dev mailing list