Re: [PATCH v2 for-next 4/5] RDMA/hns: Optimize base address table config flow for qp buffer

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Mar 06, 2020 at 10:44:53AM +0800, Weihang Li wrote:
> From: Xi Wang <wangxi11@xxxxxxxxxx>
>
> Currently, before the qp is created, a page size needs to be calculated for
> the base address table to store all base addresses in the mtr. As a result,
> the parameter configuration of the mtr is complex. So integrate the process
> of calculating the base table page size into the hem related interface to
> simplify the process of using mtr.
>
> Signed-off-by: Xi Wang <wangxi11@xxxxxxxxxx>
> Signed-off-by: Weihang Li <liweihang@xxxxxxxxxx>
> ---
>  drivers/infiniband/hw/hns/hns_roce_device.h |  4 ---
>  drivers/infiniband/hw/hns/hns_roce_hem.c    | 16 +++++++----
>  drivers/infiniband/hw/hns/hns_roce_qp.c     | 42 +++++++----------------------
>  3 files changed, 21 insertions(+), 41 deletions(-)
>
> diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
> index b6ae12d..f6b3cf6 100644
> --- a/drivers/infiniband/hw/hns/hns_roce_device.h
> +++ b/drivers/infiniband/hw/hns/hns_roce_device.h
> @@ -669,10 +669,6 @@ struct hns_roce_qp {
>  	struct ib_umem		*umem;
>  	struct hns_roce_mtt	mtt;
>  	struct hns_roce_mtr	mtr;
> -
> -	/* this define must less than HNS_ROCE_MAX_BT_REGION */
> -#define HNS_ROCE_WQE_REGION_MAX	 3
> -	struct hns_roce_buf_region regions[HNS_ROCE_WQE_REGION_MAX];
>  	int                     wqe_bt_pg_shift;
>
>  	u32			buff_size;
> diff --git a/drivers/infiniband/hw/hns/hns_roce_hem.c b/drivers/infiniband/hw/hns/hns_roce_hem.c
> index e822157..8380d71 100644
> --- a/drivers/infiniband/hw/hns/hns_roce_hem.c
> +++ b/drivers/infiniband/hw/hns/hns_roce_hem.c
> @@ -1383,6 +1383,7 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
>  	void *cpu_base;
>  	u64 phy_base;
>  	int ret = 0;
> +	int ba_num;
>  	int offset;
>  	int total;
>  	int step;
> @@ -1393,12 +1394,16 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
>  	if (root_hem)
>  		return 0;
>
> +	ba_num = hns_roce_hem_list_calc_root_ba(regions, region_cnt, unit);
> +	if (ba_num < 1)
> +		return -ENOMEM;
> +
>  	INIT_LIST_HEAD(&temp_root);
> -	total = r->offset;
> +	offset = r->offset;
>  	/* indicate to last region */
>  	r = &regions[region_cnt - 1];
> -	root_hem = hem_list_alloc_item(hr_dev, total, r->offset + r->count - 1,
> -				       unit, true, 0);
> +	root_hem = hem_list_alloc_item(hr_dev, offset, r->offset + r->count - 1,
> +				       ba_num, true, 0);
>  	if (!root_hem)
>  		return -ENOMEM;
>  	list_add(&root_hem->list, &temp_root);
> @@ -1410,7 +1415,7 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
>  		INIT_LIST_HEAD(&temp_list[i]);
>
>  	total = 0;
> -	for (i = 0; i < region_cnt && total < unit; i++) {
> +	for (i = 0; i < region_cnt && total < ba_num; i++) {
>  		r = &regions[i];
>  		if (!r->count)
>  			continue;
> @@ -1443,7 +1448,8 @@ static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev,
>  			/* if exist mid bt, link L1 to L0 */
>  			list_for_each_entry_safe(hem, temp_hem,
>  					  &hem_list->mid_bt[i][1], list) {
> -				offset = hem->start / step * BA_BYTE_LEN;
> +				offset = (hem->start - r->offset) / step *
> +					  BA_BYTE_LEN;
>  				hem_list_link_bt(hr_dev, cpu_base + offset,
>  						 hem->dma_addr);
>  				total++;
> diff --git a/drivers/infiniband/hw/hns/hns_roce_qp.c b/drivers/infiniband/hw/hns/hns_roce_qp.c
> index c2ea489..1c5de2b 100644
> --- a/drivers/infiniband/hw/hns/hns_roce_qp.c
> +++ b/drivers/infiniband/hw/hns/hns_roce_qp.c
> @@ -579,30 +579,6 @@ static int split_wqe_buf_region(struct hns_roce_dev *hr_dev,
>  	return region_cnt;
>  }
>
> -static int calc_wqe_bt_page_shift(struct hns_roce_dev *hr_dev,
> -				  struct hns_roce_buf_region *regions,
> -				  int region_cnt)
> -{
> -	int bt_pg_shift;
> -	int ba_num;
> -	int ret;
> -
> -	bt_pg_shift = PAGE_SHIFT + hr_dev->caps.mtt_ba_pg_sz;
> -
> -	/* all root ba entries must in one bt page */
> -	do {
> -		ba_num = (1 << bt_pg_shift) / BA_BYTE_LEN;
> -		ret = hns_roce_hem_list_calc_root_ba(regions, region_cnt,
> -						     ba_num);
> -		if (ret <= ba_num)
> -			break;
> -
> -		bt_pg_shift++;
> -	} while (ret > ba_num);
> -
> -	return bt_pg_shift - PAGE_SHIFT;
> -}
> -
>  static int set_extend_sge_param(struct hns_roce_dev *hr_dev,
>  				struct hns_roce_qp *hr_qp)
>  {
> @@ -768,7 +744,10 @@ static void free_rq_inline_buf(struct hns_roce_qp *hr_qp)
>  static int map_wqe_buf(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp,
>  		       u32 page_shift, bool is_user)
>  {
> -	dma_addr_t *buf_list[ARRAY_SIZE(hr_qp->regions)] = { NULL };
> +/* WQE buffer include 3 parts: SQ, extend SGE and RQ. */
> +#define HNS_ROCE_WQE_REGION_MAX	 3
> +	struct hns_roce_buf_region regions[HNS_ROCE_WQE_REGION_MAX] = {};
> +	dma_addr_t *buf_list[HNS_ROCE_WQE_REGION_MAX] = { NULL };

Nitpick, NULL is not needed.

Thanks



[Index of Archives]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Photo]     [Yosemite News]     [Yosemite Photos]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux