From: Yishai Hadas <yishaih@xxxxxxxxxx>
Extend advice MR to support non faulting mode, this improves performance
by eliminating page faults and bring only the existing CPU pages.
Signed-off-by: Yishai Hadas <yishaih@xxxxxxxxxx>
Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx>
---
drivers/infiniband/hw/mlx5/mr.c | 3 ++-
drivers/infiniband/hw/mlx5/odp.c | 7 ++++++-
include/uapi/rdma/ib_user_ioctl_verbs.h | 1 +
3 files changed, 9 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
index eac869340158..1a82a57fc415 100644
--- a/drivers/infiniband/hw/mlx5/mr.c
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -1312,7 +1312,8 @@ int mlx5_ib_advise_mr(struct ib_pd *pd,
struct uverbs_attr_bundle *attrs)
{
if (advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH &&
- advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE)
+ advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE &&
+ advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT)
return -EOPNOTSUPP;
return mlx5_ib_advise_mr_prefetch(pd, advice, flags,
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 5bd5e19d76a2..28b7227d31bf 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -665,6 +665,7 @@ 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)
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)
@@ -673,6 +674,7 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp,
bool downgrade = flags & MLX5_PF_FLAGS_DOWNGRADE;
u64 access_mask;
u64 start_idx;
+ bool fault = !(flags & MLX5_PF_FLAGS_SNAPSHOT);
page_shift = odp->page_shift;
start_idx = (user_va - ib_umem_start(odp)) >> page_shift;
@@ -681,7 +683,7 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp,
if (odp->umem.writable && !downgrade)
access_mask |= ODP_WRITE_ALLOWED_BIT;
- np = ib_umem_odp_map_dma_and_lock(odp, user_va, bcnt, access_mask, true);
+ np = ib_umem_odp_map_dma_and_lock(odp, user_va, bcnt, access_mask, fault);
if (np < 0)
return np;
@@ -1851,6 +1853,9 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH)
pf_flags |= MLX5_PF_FLAGS_DOWNGRADE;
+ if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT)
+ pf_flags |= MLX5_PF_FLAGS_SNAPSHOT;
+
if (flags & IB_UVERBS_ADVISE_MR_FLAG_FLUSH)
return mlx5_ib_prefetch_sg_list(pd, advice, pf_flags, sg_list,
num_sge);
diff --git a/include/uapi/rdma/ib_user_ioctl_verbs.h b/include/uapi/rdma/ib_user_ioctl_verbs.h
index cfea82acfe57..22483799cd07 100644
--- a/include/uapi/rdma/ib_user_ioctl_verbs.h
+++ b/include/uapi/rdma/ib_user_ioctl_verbs.h
@@ -208,6 +208,7 @@ enum ib_uverbs_read_counters_flags {
enum ib_uverbs_advise_mr_advice {
IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH,
IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE,
+ IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT,
};
enum ib_uverbs_advise_mr_flag {
--
2.26.2
