From: Yishai Hadas <yishaih@xxxxxxxxxx> Sync device with CPU pages upon ODP MR registration. This reduce potential page faults down the road and improve performance. Signed-off-by: Yishai Hadas <yishaih@xxxxxxxxxx> Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx> --- drivers/infiniband/hw/mlx5/mlx5_ib.h | 5 +++++ drivers/infiniband/hw/mlx5/mr.c | 11 +++++++---- drivers/infiniband/hw/mlx5/odp.c | 21 ++++++++++++++++++++- 3 files changed, 32 insertions(+), 5 deletions(-) diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index 6ab3efb75b21..b1f2b34e5955 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -1283,6 +1283,7 @@ void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries, int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, u32 flags, struct ib_sge *sg_list, u32 num_sge); +int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable); #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev) { @@ -1304,6 +1305,10 @@ mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, { return -EOPNOTSUPP; } +static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) +{ + return -EOPNOTSUPP; +} #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ extern const struct mmu_interval_notifier_ops mlx5_mn_ops; diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index 1a82a57fc415..910120b551c5 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -1422,7 +1422,7 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, mr->umem = umem; set_mr_fields(dev, mr, npages, length, access_flags); - if (xlt_with_umr) { + if (xlt_with_umr && !(access_flags & IB_ACCESS_ON_DEMAND)) { /* * If the MR was created with reg_create then it will be * configured properly but left disabled. It is safe to go ahead @@ -1430,9 +1430,6 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, */ int update_xlt_flags = MLX5_IB_UPD_XLT_ENABLE; - if (access_flags & IB_ACCESS_ON_DEMAND) - update_xlt_flags |= MLX5_IB_UPD_XLT_ZAP; - err = mlx5_ib_update_xlt(mr, 0, ncont, page_shift, update_xlt_flags); if (err) { @@ -1452,6 +1449,12 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, dereg_mr(dev, mr); return ERR_PTR(err); } + + err = mlx5_ib_init_odp_mr(mr, xlt_with_umr); + if (err) { + dereg_mr(dev, mr); + return ERR_PTR(err); + } } return &mr->ibmr; diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index 28b7227d31bf..15fd6d224527 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -666,6 +666,7 @@ void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr) #define MLX5_PF_FLAGS_DOWNGRADE BIT(1) #define MLX5_PF_FLAGS_SNAPSHOT BIT(2) +#define MLX5_PF_FLAGS_ENABLE BIT(3) static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp, u64 user_va, size_t bcnt, u32 *bytes_mapped, u32 flags) @@ -675,6 +676,10 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp, u64 access_mask; u64 start_idx; bool fault = !(flags & MLX5_PF_FLAGS_SNAPSHOT); + u32 xlt_flags = MLX5_IB_UPD_XLT_ATOMIC; + + if (flags & MLX5_PF_FLAGS_ENABLE) + xlt_flags |= MLX5_IB_UPD_XLT_ENABLE; page_shift = odp->page_shift; start_idx = (user_va - ib_umem_start(odp)) >> page_shift; @@ -692,7 +697,7 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp, * ib_umem_odp_map_dma_and_lock already checks this. */ ret = mlx5_ib_update_xlt(mr, start_idx, np, - page_shift, MLX5_IB_UPD_XLT_ATOMIC); + page_shift, xlt_flags); mutex_unlock(&odp->umem_mutex); if (ret < 0) { @@ -827,6 +832,20 @@ static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt, flags); } +int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) +{ + u32 flags = MLX5_PF_FLAGS_SNAPSHOT; + int ret; + + if (enable) + flags |= MLX5_PF_FLAGS_ENABLE; + + ret = pagefault_real_mr(mr, to_ib_umem_odp(mr->umem), + mr->umem->address, mr->umem->length, NULL, + flags); + return ret >= 0 ? 0 : ret; +} + struct pf_frame { struct pf_frame *next; u32 key; -- 2.26.2