On 1/23/2019 3:44 PM, Jason Gunthorpe wrote:
> On Sun, Jan 20, 2019 at 02:27:13AM +0100, Nicholas Mc Guire wrote:
>> The kmalloc is called with | __GFP_NOFAIL so there is no point in
>> checking the return value - it either returns valid storage or it would
>> hang/terminate there. But it is not possible to say if the use of
>> __GFP_NOFAIL is really needed and the check should be removed or
>> vice-versa (use of __GFP_NOFAIL should be only in exceptional
>> cases as I understand it and alloc_srq_queue() is called in quite
>> a few places)
>> In either way it would need fixing.
>>
>> Signed-off-by: Nicholas Mc Guire <hofrat@xxxxxxxxx>
>> Fixes: 6a0b6174d35a ("rdma/cxgb4: Add support for kernel mode SRQ's")
>> ---
>
> As per steve's remarkes I revised this to the below and applied it to
> for-next
>
>>From 4b2d4262ee2ea58df867de1928bf208795344432 Mon Sep 17 00:00:00 2001
> From: Jason Gunthorpe <jgg@xxxxxxxxxxxx>
> Date: Sun, 20 Jan 2019 02:27:13 +0100
> Subject: [PATCH] RDMA/iw_cxgb4: Drop __GFP_NOFAIL
>
> There is no reason for this __GFP_NOFAIL, none of the other routines in
> this file use it, and there is an error unwind here. NOFAIL should be
> reserved for special cases, not used by network drivers.
>
> Fixes: 6a0b6174d35a ("rdma/cxgb4: Add support for kernel mode SRQ's")
> Reported-by: Nicholas Mc Guire <hofrat@xxxxxxxxx>
> Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxxxx>
> ---
> drivers/infiniband/hw/cxgb4/qp.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
> index 03f4c66c265946..c00a4114412694 100644
> --- a/drivers/infiniband/hw/cxgb4/qp.c
> +++ b/drivers/infiniband/hw/cxgb4/qp.c
> @@ -2597,7 +2597,7 @@ static int alloc_srq_queue(struct c4iw_srq *srq, struct c4iw_dev_ucontext *uctx,
> /* build fw_ri_res_wr */
> wr_len = sizeof(*res_wr) + sizeof(*res);
>
> - skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL);
> + skb = alloc_skb(wr_len, GFP_KERNEL);
> if (!skb)
> goto err_free_queue;
> set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
>
Thanks Jason!
