Introducing Multi-Path DMA Support for mlx5 RDMA Driver

index cab228cf51c6..970c9d8473ef 100644
@@ -313,6 +313,7 @@ enum {
 	MLX5_CMD_OP_MODIFY_VHCA_STATE             = 0xb0e,
 	MLX5_CMD_OP_SYNC_CRYPTO                   = 0xb12,
 	MLX5_CMD_OP_ALLOW_OTHER_VHCA_ACCESS       = 0xb16,
+	MLX5_CMD_OPCODE_QUERY_VUID                = 0xb22,
 	MLX5_CMD_OP_MAX
 };
 

@@ -1885,7 +1886,8 @@ struct mlx5_ifc_cmd_hca_cap_bits {
 
 	u8         reserved_at_5a0[0x10];
 	u8         enhanced_cqe_compression[0x1];
-	u8         reserved_at_5b1[0x2];
+	u8         reserved_at_5b1[0x1];
+	u8         crossing_vhca_mkey[0x1];
 	u8         log_max_dek[0x5];
 	u8         reserved_at_5b8[0x4];
 	u8         mini_cqe_resp_stride_index[0x1];

@@ -1954,7 +1956,9 @@ struct mlx5_ifc_cmd_hca_cap_bits {
 	u8	   dynamic_msix_table_size[0xc];
 	u8	   reserved_at_740[0xc];
 	u8	   min_dynamic_vf_msix_table_size[0x4];
-	u8	   reserved_at_750[0x4];
+	u8	   reserved_at_750[0x2];
+	u8	   data_direct[0x1];
+	u8	   reserved_at_753[0x1];
 	u8	   max_dynamic_vf_msix_table_size[0xc];
 
 	u8         reserved_at_760[0x3];

@@ -1982,7 +1986,9 @@ struct mlx5_ifc_cmd_hca_cap_2_bits {
 	u8	   reserved_at_0[0x80];
 
 	u8         migratable[0x1];
-	u8         reserved_at_81[0x1f];
+	u8         reserved_at_81[0x11];
+	u8         query_vuid[0x1];
+	u8         reserved_at_93[0xd];
 
 	u8	   max_reformat_insert_size[0x8];
 	u8	   max_reformat_insert_offset[0x8];

@@ -4154,6 +4160,7 @@ enum {
 	MLX5_MKC_ACCESS_MODE_KSM   = 0x3,
 	MLX5_MKC_ACCESS_MODE_SW_ICM = 0x4,
 	MLX5_MKC_ACCESS_MODE_MEMIC = 0x5,
+	MLX5_MKC_ACCESS_MODE_CROSSING = 0x6,
 };
 
 struct mlx5_ifc_mkc_bits {

@@ -4196,7 +4203,10 @@ struct mlx5_ifc_mkc_bits {
 
 	u8         bsf_octword_size[0x20];
 
-	u8         reserved_at_120[0x80];
+	u8         reserved_at_120[0x60];
+
+	u8         crossing_target_vhca_id[0x10];
+	u8         reserved_at_190[0x10];
 
 	u8         translations_octword_size[0x20];
 

@@ -5124,6 +5134,36 @@ struct mlx5_ifc_query_vport_state_out_bits {
 	u8         state[0x4];
 };
 
+struct mlx5_ifc_array1024_auto_bits {
+	u8         array1024_auto[32][0x20];
+};
+
+struct mlx5_ifc_query_vuid_in_bits {
+	u8         opcode[0x10];
+	u8         uid[0x10];
+
+	u8         reserved_at_20[0x40];
+
+	u8         query_vfs_vuid[0x1];
+	u8         data_direct[0x1];
+	u8         reserved_at_62[0xe];
+	u8         vhca_id[0x10];
+};
+
+struct mlx5_ifc_query_vuid_out_bits {
+	u8        status[0x8];
+	u8        reserved_at_8[0x18];
+
+	u8        syndrome[0x20];
+
+	u8        reserved_at_40[0x1a0];
+
+	u8        reserved_at_1e0[0x10];
+	u8        num_of_entries[0x10];
+
+	struct mlx5_ifc_array1024_auto_bits vuid[];
+};
+
 enum {
 	MLX5_VPORT_STATE_OP_MOD_VNIC_VPORT  = 0x0,
 	MLX5_VPORT_STATE_OP_MOD_ESW_VPORT   = 0x1,

@@ -8989,7 +9029,8 @@ struct mlx5_ifc_create_mkey_in_bits {
 
 	u8         pg_access[0x1];
 	u8         mkey_umem_valid[0x1];
-	u8         reserved_at_62[0x1e];
+	u8         data_direct[0x1];
+	u8         reserved_at_63[0x1d];
 
 	struct mlx5_ifc_mkc_bits memory_key_mkey_entry;
 

227 lines not on this page · raw message

new file mode 100644
index 000000000000..2fd2bdbe8f69
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/*
+ * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved
+ */
+
+#ifndef _MLX5_IB_DATA_DIRECT_H
+#define _MLX5_IB_DATA_DIRECT_H
+
+struct mlx5_ib_dev;
+
+struct mlx5_data_direct_dev {
+	struct device *device;
+	struct pci_dev *pdev;
+	char *vuid;
+	struct list_head list;
+};
+
+int mlx5_data_direct_ib_reg(struct mlx5_ib_dev *ibdev, char *vuid);
+void mlx5_data_direct_ib_unreg(struct mlx5_ib_dev *ibdev);
+int mlx5_data_direct_driver_register(void);
+void mlx5_data_direct_driver_unregister(void);
+
+#endif

index 6048b9ad13bb..de254cf03173 100644
@@ -48,6 +48,7 @@
 #include <rdma/mlx5_user_ioctl_verbs.h>
 #include <rdma/mlx5_user_ioctl_cmds.h>
 #include "macsec.h"
+#include "data_direct.h"
 
 #define UVERBS_MODULE_NAME mlx5_ib
 #include <rdma/uverbs_named_ioctl.h>

@@ -3866,6 +3867,7 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
 	dev->ib_dev.num_comp_vectors    = mlx5_comp_vectors_max(mdev);
 
 	mutex_init(&dev->cap_mask_mutex);
+	mutex_init(&dev->data_direct_lock);
 	INIT_LIST_HEAD(&dev->qp_list);
 	spin_lock_init(&dev->reset_flow_resource_lock);
 	xa_init(&dev->odp_mkeys);

@@ -4293,6 +4295,21 @@ static void mlx5_ib_stage_dev_notifier_cleanup(struct mlx5_ib_dev *dev)
 	mlx5_notifier_unregister(dev->mdev, &dev->mdev_events);
 }
 
+void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev,
+			      struct mlx5_data_direct_dev *dev)
+{
+	mutex_lock(&ibdev->data_direct_lock);
+	ibdev->data_direct_dev = dev;
+	mutex_unlock(&ibdev->data_direct_lock);
+}
+
+void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev)
+{
+	mutex_lock(&ibdev->data_direct_lock);
+	ibdev->data_direct_dev = NULL;
+	mutex_unlock(&ibdev->data_direct_lock);
+}
+
 void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
 		      const struct mlx5_ib_profile *profile,
 		      int stage)

@@ -4715,17 +4732,23 @@ static int __init mlx5_ib_init(void)
 	ret = mlx5r_rep_init();
 	if (ret)
 		goto rep_err;
+	ret = mlx5_data_direct_driver_register();
+	if (ret)
+		goto dd_err;
 	ret = auxiliary_driver_register(&mlx5r_mp_driver);
 	if (ret)
 		goto mp_err;
 	ret = auxiliary_driver_register(&mlx5r_driver);
 	if (ret)
 		goto drv_err;
+
 	return 0;
 
 drv_err:
 	auxiliary_driver_unregister(&mlx5r_mp_driver);
 mp_err:
+	mlx5_data_direct_driver_unregister();
+dd_err:
 	mlx5r_rep_cleanup();
 rep_err:
 	mlx5_ib_qp_event_cleanup();

@@ -4737,6 +4760,7 @@ static int __init mlx5_ib_init(void)
 
 static void __exit mlx5_ib_cleanup(void)
 {
+	mlx5_data_direct_driver_unregister();
 	auxiliary_driver_unregister(&mlx5r_driver);
 	auxiliary_driver_unregister(&mlx5r_mp_driver);
 	mlx5r_rep_cleanup();

index d5eb1b726675..b0d7d8b9e672 100644
@@ -1131,6 +1131,9 @@ struct mlx5_macsec {
 struct mlx5_ib_dev {
 	struct ib_device		ib_dev;
 	struct mlx5_core_dev		*mdev;
+	struct mlx5_data_direct_dev	*data_direct_dev;
+	/* protect accessing data_direct_dev */
+	struct mutex			data_direct_lock;
 	struct notifier_block		mdev_events;
 	int				num_ports;
 	/* serialize update of capability mask

@@ -1425,6 +1428,9 @@ int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
 				struct ib_dm_mr_attr *attr,
 				struct uverbs_attr_bundle *attrs);
+void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev,
+			      struct mlx5_data_direct_dev *dev);
+void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev);
 
 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);

index 895b62cc528d..7c08e3008927 100644
@@ -245,3 +245,24 @@ int mlx5_cmd_uar_dealloc(struct mlx5_core_dev *dev, u32 uarn, u16 uid)
 	MLX5_SET(dealloc_uar_in, in, uid, uid);
 	return mlx5_cmd_exec_in(dev, dealloc_uar, in);
 }
+
+int mlx5_cmd_query_vuid(struct mlx5_core_dev *dev, bool data_direct,
+			char *out_vuid)
+{
+	u8 out[MLX5_ST_SZ_BYTES(query_vuid_out) +
+		MLX5_ST_SZ_BYTES(array1024_auto)] = {};
+	u8 in[MLX5_ST_SZ_BYTES(query_vuid_in)] = {};
+	char *vuid;
+	int err;
+
+	MLX5_SET(query_vuid_in, in, opcode, MLX5_CMD_OPCODE_QUERY_VUID);
+	MLX5_SET(query_vuid_in, in, vhca_id, MLX5_CAP_GEN(dev, vhca_id));
+	MLX5_SET(query_vuid_in, in, data_direct, data_direct);
+	err = mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
+	if (err)
+		return err;
+
+	vuid = MLX5_ADDR_OF(query_vuid_out, out, vuid);
+	memcpy(out_vuid, vuid, MLX5_ST_SZ_BYTES(array1024_auto));
+	return 0;
+}

index e5cd31270443..e6c88b6ebd0d 100644
@@ -58,4 +58,6 @@ int mlx5_cmd_mad_ifc(struct mlx5_ib_dev *dev, const void *inb, void *outb,
 		     u16 opmod, u8 port);
 int mlx5_cmd_uar_alloc(struct mlx5_core_dev *dev, u32 *uarn, u16 uid);
 int mlx5_cmd_uar_dealloc(struct mlx5_core_dev *dev, u32 uarn, u16 uid);
+int mlx5_cmd_query_vuid(struct mlx5_core_dev *dev, bool data_direct,
+			char *out_vuid);
 #endif /* MLX5_IB_CMD_H */

114 lines not on this page · raw message

index b0d7d8b9e672..b2ebea173547 100644
@@ -835,6 +835,11 @@ struct mlx5_ib_port_resources {
 	struct work_struct pkey_change_work;
 };
 
+struct mlx5_data_direct_resources {
+	u32 pdn;
+	u32 mkey;
+};
+
 struct mlx5_ib_resources {
 	struct ib_cq	*c0;
 	struct mutex cq_lock;

@@ -1188,6 +1193,7 @@ struct mlx5_ib_dev {
 	u16 pkey_table_len;
 	u8 lag_ports;
 	struct mlx5_special_mkeys mkeys;
+	struct mlx5_data_direct_resources ddr;
 
 #ifdef CONFIG_MLX5_MACSEC
 	struct mlx5_macsec macsec;

index 39357dc2d229..726a09786547 100644
@@ -110,10 +110,12 @@ void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf)
 }
 EXPORT_SYMBOL(ib_umem_dmabuf_unmap_pages);
 
-struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device,
-					  unsigned long offset, size_t size,
-					  int fd, int access,
-					  const struct dma_buf_attach_ops *ops)
+static struct ib_umem_dmabuf *
+ib_umem_dmabuf_get_with_dma_device(struct ib_device *device,
+				   struct device *dma_device,
+				   unsigned long offset, size_t size,
+				   int fd, int access,
+				   const struct dma_buf_attach_ops *ops)
 {
 	struct dma_buf *dmabuf;
 	struct ib_umem_dmabuf *umem_dmabuf;

@@ -152,7 +154,7 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device,
 
 	umem_dmabuf->attach = dma_buf_dynamic_attach(
 					dmabuf,
-					device->dma_device,
+					dma_device,
 					ops,
 					umem_dmabuf);
 	if (IS_ERR(umem_dmabuf->attach)) {

@@ -168,6 +170,15 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device,
 	dma_buf_put(dmabuf);
 	return ret;
 }
+
+struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device,
+					  unsigned long offset, size_t size,
+					  int fd, int access,
+					  const struct dma_buf_attach_ops *ops)
+{
+	return ib_umem_dmabuf_get_with_dma_device(device, device->dma_device,
+						  offset, size, fd, access, ops);
+}
 EXPORT_SYMBOL(ib_umem_dmabuf_get);
 
 static void

@@ -184,16 +195,18 @@ static struct dma_buf_attach_ops ib_umem_dmabuf_attach_pinned_ops = {
 	.move_notify = ib_umem_dmabuf_unsupported_move_notify,
 };
 
-struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device,
-						 unsigned long offset,
-						 size_t size, int fd,
-						 int access)
+struct ib_umem_dmabuf *
+ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device,
+					  struct device *dma_device,
+					  unsigned long offset, size_t size,
+					  int fd, int access)
 {
 	struct ib_umem_dmabuf *umem_dmabuf;
 	int err;
 
-	umem_dmabuf = ib_umem_dmabuf_get(device, offset, size, fd, access,
-					 &ib_umem_dmabuf_attach_pinned_ops);
+	umem_dmabuf = ib_umem_dmabuf_get_with_dma_device(device, dma_device, offset,
+							 size, fd, access,
+							 &ib_umem_dmabuf_attach_pinned_ops);
 	if (IS_ERR(umem_dmabuf))
 		return umem_dmabuf;
 

@@ -217,6 +230,16 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device,
 	ib_umem_release(&umem_dmabuf->umem);
 	return ERR_PTR(err);
 }
+EXPORT_SYMBOL(ib_umem_dmabuf_get_pinned_with_dma_device);
+
+struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device,
+						 unsigned long offset,
+						 size_t size, int fd,
+						 int access)
+{
+	return ib_umem_dmabuf_get_pinned_with_dma_device(device, device->dma_device,
+							 offset, size, fd, access);
+}
 EXPORT_SYMBOL(ib_umem_dmabuf_get_pinned);
 
 void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf)

index 565a85044541..de05268ed632 100644
@@ -150,6 +150,11 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device,
 						 unsigned long offset,
 						 size_t size, int fd,
 						 int access);
+struct ib_umem_dmabuf *
+ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device,
+					  struct device *dma_device,
+					  unsigned long offset, size_t size,
+					  int fd, int access);
 int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf);
 void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf);
 void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf);

@@ -196,6 +201,16 @@ ib_umem_dmabuf_get_pinned(struct ib_device *device, unsigned long offset,
 {
 	return ERR_PTR(-EOPNOTSUPP);
 }
+
+static inline struct ib_umem_dmabuf *
+ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device,
+					  struct device *dma_device,
+					  unsigned long offset, size_t size,
+					  int fd, int access)
+{
+	return ERR_PTR(-EOPNOTSUPP);
+}
+
 static inline int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf)
 {
 	return -EOPNOTSUPP;

index 726a09786547..9fcd37761264 100644
@@ -23,6 +23,9 @@ int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf)
 
 	dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
 
+	if (umem_dmabuf->revoked)
+		return -EINVAL;
+
 	if (umem_dmabuf->sgt)
 		goto wait_fence;
 

@@ -242,15 +245,29 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device,
 }
 EXPORT_SYMBOL(ib_umem_dmabuf_get_pinned);
 
-void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf)
+void ib_umem_dmabuf_revoke(struct ib_umem_dmabuf *umem_dmabuf)
 {
 	struct dma_buf *dmabuf = umem_dmabuf->attach->dmabuf;
 
 	dma_resv_lock(dmabuf->resv, NULL);
+	if (umem_dmabuf->revoked)
+		goto end;
 	ib_umem_dmabuf_unmap_pages(umem_dmabuf);
-	if (umem_dmabuf->pinned)
+	if (umem_dmabuf->pinned) {
 		dma_buf_unpin(umem_dmabuf->attach);
+		umem_dmabuf->pinned = 0;
+	}
+	umem_dmabuf->revoked = 1;
+end:
 	dma_resv_unlock(dmabuf->resv);
+}
+EXPORT_SYMBOL(ib_umem_dmabuf_revoke);
+
+void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf)
+{
+	struct dma_buf *dmabuf = umem_dmabuf->attach->dmabuf;
+
+	ib_umem_dmabuf_revoke(umem_dmabuf);
 
 	dma_buf_detach(dmabuf, umem_dmabuf->attach);
 	dma_buf_put(dmabuf);

index de05268ed632..7dc7b1cc71b5 100644
@@ -38,6 +38,7 @@ struct ib_umem_dmabuf {
 	unsigned long last_sg_trim;
 	void *private;
 	u8 pinned : 1;
+	u8 revoked : 1;
 };
 
 static inline struct ib_umem_dmabuf *to_ib_umem_dmabuf(struct ib_umem *umem)

@@ -158,6 +159,7 @@ ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device,
 int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf);
 void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf);
 void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf);
+void ib_umem_dmabuf_revoke(struct ib_umem_dmabuf *umem_dmabuf);
 
 #else /* CONFIG_INFINIBAND_USER_MEM */
 

@@ -217,6 +219,7 @@ static inline int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf)
 }
 static inline void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf) { }
 static inline void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf) { }
+static inline void ib_umem_dmabuf_revoke(struct ib_umem_dmabuf *umem_dmabuf) {}
 
 #endif /* CONFIG_INFINIBAND_USER_MEM */
 #endif /* IB_UMEM_H */

index 03e1db5d1e8c..7ebc7bd3caae 100644
@@ -239,7 +239,7 @@ static int UVERBS_HANDLER(UVERBS_METHOD_REG_DMABUF_MR)(
 
 	mr = pd->device->ops.reg_user_mr_dmabuf(pd, offset, length, iova, fd,
 						access_flags,
-						&attrs->driver_udata);
+						attrs);
 	if (IS_ERR(mr))
 		return PTR_ERR(mr);
 

index 7c757351a016..43a68e7de02a 100644
@@ -4122,7 +4122,8 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length,
 
 struct ib_mr *bnxt_re_reg_user_mr_dmabuf(struct ib_pd *ib_pd, u64 start,
 					 u64 length, u64 virt_addr, int fd,
-					 int mr_access_flags, struct ib_udata *udata)
+					 int mr_access_flags,
+					 struct uverbs_attr_bundle *attrs)
 {
 	struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd);
 	struct bnxt_re_dev *rdev = pd->rdev;

index e98cb1717338..3ddeda312376 100644
@@ -242,7 +242,7 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 struct ib_mr *bnxt_re_reg_user_mr_dmabuf(struct ib_pd *ib_pd, u64 start,
 					 u64 length, u64 virt_addr,
 					 int fd, int mr_access_flags,
-					 struct ib_udata *udata);
+					 struct uverbs_attr_bundle *attrs);
 int bnxt_re_alloc_ucontext(struct ib_ucontext *ctx, struct ib_udata *udata);
 void bnxt_re_dealloc_ucontext(struct ib_ucontext *context);
 int bnxt_re_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);

index e580e087e9da..d7fc9d5eeefd 100644
@@ -168,7 +168,7 @@ struct ib_mr *efa_reg_mr(struct ib_pd *ibpd, u64 start, u64 length,
 struct ib_mr *efa_reg_user_mr_dmabuf(struct ib_pd *ibpd, u64 start,
 				     u64 length, u64 virt_addr,
 				     int fd, int access_flags,
-				     struct ib_udata *udata);
+				     struct uverbs_attr_bundle *attrs);
 int efa_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
 int efa_get_port_immutable(struct ib_device *ibdev, u32 port_num,
 			   struct ib_port_immutable *immutable);

index b1e0a1b7c59d..cc13415ff7e7 100644
@@ -1684,14 +1684,14 @@ static int efa_register_mr(struct ib_pd *ibpd, struct efa_mr *mr, u64 start,
 struct ib_mr *efa_reg_user_mr_dmabuf(struct ib_pd *ibpd, u64 start,
 				     u64 length, u64 virt_addr,
 				     int fd, int access_flags,
-				     struct ib_udata *udata)
+				     struct uverbs_attr_bundle *attrs)
 {
 	struct efa_dev *dev = to_edev(ibpd->device);
 	struct ib_umem_dmabuf *umem_dmabuf;
 	struct efa_mr *mr;
 	int err;
 
-	mr = efa_alloc_mr(ibpd, access_flags, udata);
+	mr = efa_alloc_mr(ibpd, access_flags, &attrs->driver_udata);
 	if (IS_ERR(mr)) {
 		err = PTR_ERR(mr);
 		goto err_out;

index fc0ce35da14e..6a107decb704 100644
@@ -3085,7 +3085,7 @@ static struct ib_mr *irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len,
 static struct ib_mr *irdma_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
 					      u64 len, u64 virt,
 					      int fd, int access,
-					      struct ib_udata *udata)
+					      struct uverbs_attr_bundle *attrs)
 {
 	struct irdma_device *iwdev = to_iwdev(pd->device);
 	struct ib_umem_dmabuf *umem_dmabuf;

index b2ebea173547..e915a62da49c 100644
@@ -1354,7 +1354,7 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
 					 u64 length, u64 virt_addr,
 					 int fd, int access_flags,
-					 struct ib_udata *udata);
+					 struct uverbs_attr_bundle *attrs);
 int mlx5_ib_advise_mr(struct ib_pd *pd,
 		      enum ib_uverbs_advise_mr_advice advice,
 		      u32 flags,

index 98bd8eaa393e..1dfd9124bdd1 100644
@@ -1513,7 +1513,7 @@ static struct dma_buf_attach_ops mlx5_ib_dmabuf_attach_ops = {
 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset,
 					 u64 length, u64 virt_addr,
 					 int fd, int access_flags,
-					 struct ib_udata *udata)
+					 struct uverbs_attr_bundle *attrs)
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 	struct mlx5_ib_mr *mr = NULL;

index 6c5712ae559d..a1dcf812d787 100644
@@ -2476,7 +2476,7 @@ struct ib_device_ops {
 	struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
 					    u64 length, u64 virt_addr, int fd,
 					    int mr_access_flags,
-					    struct ib_udata *udata);
+					    struct uverbs_attr_bundle *attrs);
 	struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
 				       u64 length, u64 virt_addr,
 				       int mr_access_flags, struct ib_pd *pd,

index fc0562f07249..b85ad3c0bfa1 100644
@@ -3490,6 +3490,7 @@ static int mlx5_ib_data_direct_init(struct mlx5_ib_dev *dev)
 	if (ret)
 		return ret;
 
+	INIT_LIST_HEAD(&dev->data_direct_mr_list);
 	ret = mlx5_data_direct_ib_reg(dev, vuid);
 	if (ret)
 		mlx5_ib_free_data_direct_resources(dev);

@@ -3882,6 +3883,14 @@ ADD_UVERBS_ATTRIBUTES_SIMPLE(
 				   dump_fill_mkey),
 		UA_MANDATORY));
 
+ADD_UVERBS_ATTRIBUTES_SIMPLE(
+	mlx5_ib_reg_dmabuf_mr,
+	UVERBS_OBJECT_MR,
+	UVERBS_METHOD_REG_DMABUF_MR,
+	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS,
+			     enum mlx5_ib_uapi_reg_dmabuf_flags,
+			     UA_OPTIONAL));
+
 static const struct uapi_definition mlx5_ib_defs[] = {
 	UAPI_DEF_CHAIN(mlx5_ib_devx_defs),
 	UAPI_DEF_CHAIN(mlx5_ib_flow_defs),

@@ -3891,6 +3900,7 @@ static const struct uapi_definition mlx5_ib_defs[] = {
 	UAPI_DEF_CHAIN(mlx5_ib_create_cq_defs),
 
 	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_DEVICE, &mlx5_ib_query_context),
+	UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_MR, &mlx5_ib_reg_dmabuf_mr),
 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_VAR,
 				UAPI_DEF_IS_OBJ_SUPPORTED(var_is_supported)),
 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_UAR),

@@ -4396,6 +4406,7 @@ void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev,
 void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev)
 {
 	mutex_lock(&ibdev->data_direct_lock);
+	mlx5_ib_revoke_data_direct_mrs(ibdev);
 	ibdev->data_direct_dev = NULL;
 	mutex_unlock(&ibdev->data_direct_lock);
 }

index e915a62da49c..be83a4d91a34 100644
@@ -682,6 +682,8 @@ struct mlx5_ib_mr {
 	struct mlx5_ib_mkey mmkey;
 
 	struct ib_umem *umem;
+	/* The mr is data direct related */
+	u8 data_direct :1;
 
 	union {
 		/* Used only by kernel MRs (umem == NULL) */

@@ -719,6 +721,10 @@ struct mlx5_ib_mr {
 			} odp_destroy;
 			struct ib_odp_counters odp_stats;
 			bool is_odp_implicit;
+			/* The affilated data direct crossed mr */
+			struct mlx5_ib_mr *dd_crossed_mr;
+			struct list_head dd_node;
+			u8 revoked :1;
 		};
 	};
 };

@@ -1169,6 +1175,7 @@ struct mlx5_ib_dev {
 	/* protect resources needed as part of reset flow */
 	spinlock_t		reset_flow_resource_lock;
 	struct list_head	qp_list;
+	struct list_head data_direct_mr_list;
 	/* Array with num_ports elements */
 	struct mlx5_ib_port	*port;
 	struct mlx5_sq_bfreg	bfreg;

@@ -1437,6 +1444,7 @@ struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
 void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev,
 			      struct mlx5_data_direct_dev *dev);
 void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev);
+void mlx5_ib_revoke_data_direct_mrs(struct mlx5_ib_dev *dev);
 
 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);

456 lines not on this page · raw message

index a524181f34df..44a3428ea342 100644
@@ -710,7 +710,10 @@ static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt,
 		ib_umem_dmabuf_unmap_pages(umem_dmabuf);
 		err = -EINVAL;
 	} else {
-		err = mlx5r_umr_update_mr_pas(mr, xlt_flags);
+		if (mr->data_direct)
+			err = mlx5r_umr_update_data_direct_ksm_pas(mr, xlt_flags);
+		else
+			err = mlx5r_umr_update_mr_pas(mr, xlt_flags);
 	}
 	dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
 

140 lines not on this page · raw message

index 5f734dc72bef..4a02c9b5aad8 100644
@@ -95,6 +95,7 @@ int mlx5r_umr_revoke_mr(struct mlx5_ib_mr *mr);
 int mlx5r_umr_rereg_pd_access(struct mlx5_ib_mr *mr, struct ib_pd *pd,
 			      int access_flags);
 int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags);
+int mlx5r_umr_update_data_direct_ksm_pas(struct mlx5_ib_mr *mr, unsigned int flags);
 int mlx5r_umr_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages,
 			 int page_shift, int flags);
 

index 5b74d6534899..106276a4cce7 100644
@@ -274,6 +274,10 @@ enum mlx5_ib_create_cq_attrs {
 	MLX5_IB_ATTR_CREATE_CQ_UAR_INDEX = UVERBS_ID_DRIVER_NS_WITH_UHW,
 };
 
+enum mlx5_ib_reg_dmabuf_mr_attrs {
+	MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS = (1U << UVERBS_ID_NS_SHIFT),
+};
+
 #define MLX5_IB_DW_MATCH_PARAM 0xA0
 
 struct mlx5_ib_match_params {

index 3189c7f08d17..7c233df475e7 100644
@@ -54,6 +54,10 @@ enum mlx5_ib_uapi_flow_action_packet_reformat_type {
 	MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL = 0x3,
 };
 
+enum mlx5_ib_uapi_reg_dmabuf_flags {
+	MLX5_IB_UAPI_REG_DMABUF_ACCESS_DATA_DIRECT = 1 << 0,
+};
+
 struct mlx5_ib_uapi_devx_async_cmd_hdr {
 	__aligned_u64	wr_id;
 	__u8		out_data[];

index bbfcce3bdc84..ffeb1e1a1538 100644
@@ -10,6 +10,7 @@
 #include <linux/mlx5/eswitch.h>
 #include <linux/mlx5/vport.h>
 #include "mlx5_ib.h"
+#include "data_direct.h"
 
 #define UVERBS_MODULE_NAME mlx5_ib
 #include <rdma/uverbs_named_ioctl.h>

@@ -183,6 +184,50 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_QUERY_PORT)(
 					     sizeof(info));
 }
 
+static int UVERBS_HANDLER(MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH)(
+	struct uverbs_attr_bundle *attrs)
+{
+	struct mlx5_data_direct_dev *data_direct_dev;
+	struct mlx5_ib_ucontext *c;
+	struct mlx5_ib_dev *dev;
+	int out_len = uverbs_attr_get_len(attrs,
+			MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH);
+	u32 dev_path_len;
+	char *dev_path;
+	int ret;
+
+	c = to_mucontext(ib_uverbs_get_ucontext(attrs));
+	if (IS_ERR(c))
+		return PTR_ERR(c);
+	dev = to_mdev(c->ibucontext.device);
+	mutex_lock(&dev->data_direct_lock);
+	data_direct_dev = dev->data_direct_dev;
+	if (!data_direct_dev) {
+		ret = -ENODEV;
+		goto end;
+	}
+
+	dev_path = kobject_get_path(&data_direct_dev->device->kobj, GFP_KERNEL);
+	if (!dev_path) {
+		ret = -ENOMEM;
+		goto end;
+	}
+
+	dev_path_len = strlen(dev_path) + 1;
+	if (dev_path_len > out_len) {
+		ret = -ENOSPC;
+		goto end;
+	}
+
+	ret = uverbs_copy_to(attrs, MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH, dev_path,
+			     dev_path_len);
+	kfree(dev_path);
+
+end:
+	mutex_unlock(&dev->data_direct_lock);
+	return ret;
+}
+
 DECLARE_UVERBS_NAMED_METHOD(
 	MLX5_IB_METHOD_QUERY_PORT,
 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_QUERY_PORT_PORT_NUM,

@@ -193,9 +238,17 @@ DECLARE_UVERBS_NAMED_METHOD(
 				   reg_c0),
 		UA_MANDATORY));
 
+DECLARE_UVERBS_NAMED_METHOD(
+	MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH,
+	UVERBS_ATTR_PTR_OUT(
+		MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH,
+		UVERBS_ATTR_MIN_SIZE(0),
+		UA_MANDATORY));
+
 ADD_UVERBS_METHODS(mlx5_ib_device,
 		   UVERBS_OBJECT_DEVICE,
-		   &UVERBS_METHOD(MLX5_IB_METHOD_QUERY_PORT));
+		   &UVERBS_METHOD(MLX5_IB_METHOD_QUERY_PORT),
+		   &UVERBS_METHOD(MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH));
 
 DECLARE_UVERBS_NAMED_METHOD(
 	MLX5_IB_METHOD_PD_QUERY,

index 106276a4cce7..fd2e4a3a56b3 100644
@@ -348,6 +348,7 @@ enum mlx5_ib_pd_methods {
 
 enum mlx5_ib_device_methods {
 	MLX5_IB_METHOD_QUERY_PORT = (1U << UVERBS_ID_NS_SHIFT),
+	MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH,
 };
 
 enum mlx5_ib_query_port_attrs {

@@ -355,4 +356,8 @@ enum mlx5_ib_query_port_attrs {
 	MLX5_IB_ATTR_QUERY_PORT,
 };
 
+enum mlx5_ib_get_data_direct_sysfs_path_attrs {
+	MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH = (1U << UVERBS_ID_NS_SHIFT),
+};
+
 #endif