Re: [RFC PATCH rdma-next 08/10] RDMA/rxe: Implement flush execution in responder side

[Tom Talpey]

On 12/28/2021 3:07 AM, Li Zhijian wrote:

> +static enum resp_states process_flush(struct rxe_qp *qp,
> +				       struct rxe_pkt_info *pkt)
> +{
> +	u64 length = 0, start = qp->resp.va;
> +	u32 sel = feth_sel(pkt);
> +	u32 plt = feth_plt(pkt);
> +	struct rxe_mr *mr = qp->resp.mr;
> +
> +	if (sel == IB_EXT_SEL_MR_RANGE)
> +		length = qp->resp.length;
> +	else if (sel == IB_EXT_SEL_MR_WHOLE)
> +		length = mr->cur_map_set->length;

I noticed another issue in this. When the "Memory Region" flag is
set, the RETH:VA field in the request is ignored, in addition to
the RETH:DMALEN. Therefore, both the start and length must be set
from the map.

Is the mr->cur_map_set[0]->addr field a virtual address, or a
physical? I can't find the cur_map_set in any patch. The virtual
(mapped) address will be needed, to pass to arch_wb_cache_pmem().

Something like this?

	if (sel == IB_EXT_SEL_MR_WHOLE) {
		length = mr->cur_map_set->length;
		start = mr->cur_map_set[0]->addr;
	}

(in addition to my other review suggestions about 0-length, etc...)

Tom.