Re: [PATCH net-next v08 1/9] hinic3: Add PF framework

[Jakub Kicinski <kuba@kernel.org>]

AI code review points out the following.
Please address or add a relevant comment explaining why the code 
is fine.

> diff --git a/drivers/net/ethernet/huawei/hinic3/hinic3_mbox.c b/drivers/net/ethernet/huawei/hinic3/hinic3_mbox.c
> index cf67e26acece..1cb0d88911a2 100644
> --- a/drivers/net/ethernet/huawei/hinic3/hinic3_mbox.c
> +++ b/drivers/net/ethernet/huawei/hinic3/hinic3_mbox.c
> @@ -82,10 +82,19 @@ static struct hinic3_msg_desc *get_mbox_msg_desc(struct hinic3_mbox *mbox,
>  						 enum mbox_msg_direction_type dir,
>  						 u16 src_func_id)
>  {
> +	struct hinic3_hwdev *hwdev = mbox->hwdev;
>  	struct hinic3_msg_channel *msg_ch;
>
> -	msg_ch = (src_func_id == MBOX_MGMT_FUNC_ID) ?
> -		&mbox->mgmt_msg : mbox->func_msg;
> +	if (src_func_id == MBOX_MGMT_FUNC_ID) {
> +		msg_ch = &mbox->mgmt_msg;
> +	} else if (HINIC3_IS_VF(hwdev)) {
> +		/* message from pf */
> +		msg_ch = mbox->func_msg;
> +		if (src_func_id != hinic3_pf_id_of_vf(hwdev) || !msg_ch)
> +			return NULL;
> +	} else {
> +		return NULL;
> +	}

Can this return NULL for PF when src_func_id is not MBOX_MGMT_FUNC_ID?

Looking at hinic3_mbox_func_aeqe_handler(), the caller of get_mbox_msg_desc():

    msg_desc = get_mbox_msg_desc(mbox, dir, src_func_id);
    recv_mbox_handler(mbox, header, msg_desc);

The return value is passed directly to recv_mbox_handler() without a
NULL check. Inside recv_mbox_handler():

    if (!mbox_segment_valid(mbox, msg_desc, mbox_header)) {
        msg_desc->seq_id = MBOX_SEQ_ID_MAX_VAL;

And mbox_segment_valid() dereferences msg_desc unconditionally:

    msg_desc->seq_id = seq_id;

If a PF receives a mailbox message from any source other than management
(for example, from a VF), get_mbox_msg_desc() returns NULL, and then
mbox_segment_valid() will dereference the NULL pointer causing a kernel
crash.

Should hinic3_mbox_func_aeqe_handler() check for NULL before calling
recv_mbox_handler()?