Re: [PATCH for-next 1/2] RDMA/hns: Support query information of functions from FW

[Jason Gunthorpe <jgg@nvidia.com>]

On Wed, Mar 24, 2021 at 03:07:47AM +0000, liweihang wrote:
> On 2021/3/24 9:55, liweihang wrote:
> > On 2021/3/24 3:56, Jason Gunthorpe wrote:
> >> On Fri, Mar 12, 2021 at 05:48:26PM +0800, Weihang Li wrote:
> >>
> >>> +static int hns_roce_query_func_info(struct hns_roce_dev *hr_dev)
> >>> +{
> >>> +	struct hns_roce_pf_func_info *resp;
> >>> +	struct hns_roce_cmq_desc desc;
> >>> +	int ret;
> >>> +
> >>> +	if (hr_dev->pci_dev->revision < PCI_REVISION_ID_HIP09)
> >>> +		return 0;
> >>> +
> >>> +	hns_roce_cmq_setup_basic_desc(&desc, HNS_ROCE_OPC_QUERY_FUNC_INFO,
> >>> +				      true);
> >>> +	ret = hns_roce_cmq_send(hr_dev, &desc, 1);
> >>> +	if (ret)
> >>> +		return ret;
> >>> +
> >>> +	resp = (struct hns_roce_pf_func_info *)desc.data;
> >>
> >> WTF is this cast?
> >>
> >> struct hns_roce_cmq_desc {
> >>         __le16 opcode;
> >>         __le16 flag;
> >>         __le16 retval;
> >>         __le16 rsv;
> >>         __le32 data[6];
> >> };
> >>
> >> Casting __le32 to a pointer is wrong
> >>
> >> Jason
> >>
> > 
> > Hi Jason
> > 
> > desc.data is the address of array 'data[6]', it is got from the firmware, we
> > cast it to 'struct hns_roce_pf_func_info *' to parse its contents. I think this
> > is a cast from '__le32 *' to 'struct hns_roce_pf_func_info *'.
> > 
> > Thanks
> > Weihang
> > 
> > 
> 
> Jason, do you mean the current code is not written correctly? Do you have any
> suggestions for achieving our purpose?

Use a union

Jason