Delay mlx5_ib internal resources allocations

index c5b0e004fdd5..ec4031e851e6 100644
@@ -1841,7 +1841,7 @@ static int set_ucontext_resp(struct ib_ucontext *uctx,
 	}
 
 	resp->qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp);
-	if (dev->wc_support)
+	if (mlx5_wc_support_get(dev->mdev))
 		resp->bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev,
 						      log_bf_reg_size);
 	resp->cache_line_size = cache_line_size();

@@ -2368,7 +2368,7 @@ static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vm
 	switch (command) {
 	case MLX5_IB_MMAP_WC_PAGE:
 	case MLX5_IB_MMAP_ALLOC_WC:
-		if (!dev->wc_support)
+		if (!mlx5_wc_support_get(dev->mdev))
 			return -EPERM;
 		fallthrough;
 	case MLX5_IB_MMAP_NC_PAGE:

@@ -3643,7 +3643,7 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_UAR_OBJ_ALLOC)(
 	    alloc_type != MLX5_IB_UAPI_UAR_ALLOC_TYPE_NC)
 		return -EOPNOTSUPP;
 
-	if (!to_mdev(c->ibucontext.device)->wc_support &&
+	if (!mlx5_wc_support_get(to_mdev(c->ibucontext.device)->mdev) &&
 	    alloc_type == MLX5_IB_UAPI_UAR_ALLOC_TYPE_BF)
 		return -EOPNOTSUPP;
 

@@ -3797,18 +3797,6 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
 	return err;
 }
 
-static int mlx5_ib_enable_driver(struct ib_device *dev)
-{
-	struct mlx5_ib_dev *mdev = to_mdev(dev);
-	int ret;
-
-	ret = mlx5_ib_test_wc(mdev);
-	mlx5_ib_dbg(mdev, "Write-Combining %s",
-		    mdev->wc_support ? "supported" : "not supported");
-
-	return ret;
-}
-
 static const struct ib_device_ops mlx5_ib_dev_ops = {
 	.owner = THIS_MODULE,
 	.driver_id = RDMA_DRIVER_MLX5,

@@ -3839,7 +3827,6 @@ static const struct ib_device_ops mlx5_ib_dev_ops = {
 	.drain_rq = mlx5_ib_drain_rq,
 	.drain_sq = mlx5_ib_drain_sq,
 	.device_group = &mlx5_attr_group,
-	.enable_driver = mlx5_ib_enable_driver,
 	.get_dev_fw_str = get_dev_fw_str,
 	.get_dma_mr = mlx5_ib_get_dma_mr,
 	.get_link_layer = mlx5_ib_port_link_layer,

209 lines not on this page · raw message

index 652e1e4ca033..a74e6f5798f6 100644
@@ -341,7 +341,6 @@ struct mlx5_ib_flow_db {
  * rely on the range reserved for that use in the ib_qp_create_flags enum.
  */
 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
-#define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
 
 struct wr_list {
 	u16	opcode;

@@ -1123,7 +1122,6 @@ struct mlx5_ib_dev {
 	bool				ib_active;
 	u8				is_rep:1;
 	u8				lag_active:1;
-	u8				wc_support:1;
 	u8				fill_delay;
 	struct umr_common		umrc;
 	/* sync used page count stats

@@ -1149,7 +1147,6 @@ struct mlx5_ib_dev {
 	/* Array with num_ports elements */
 	struct mlx5_ib_port	*port;
 	struct mlx5_sq_bfreg	bfreg;
-	struct mlx5_sq_bfreg	wc_bfreg;
 	struct mlx5_sq_bfreg	fp_bfreg;
 	struct mlx5_ib_delay_drop	delay_drop;
 	const struct mlx5_ib_profile	*profile;

index 8c16c9278ce4..1d09e515aec7 100644
@@ -1109,8 +1109,6 @@ static int _create_kernel_qp(struct mlx5_ib_dev *dev,
 
 	if (init_attr->qp_type == MLX5_IB_QPT_REG_UMR)
 		qp->bf.bfreg = &dev->fp_bfreg;
-	else if (qp->flags & MLX5_IB_QP_CREATE_WC_TEST)
-		qp->bf.bfreg = &dev->wc_bfreg;
 	else
 		qp->bf.bfreg = &dev->bfreg;
 

@@ -2961,14 +2959,6 @@ static void process_create_flag(struct mlx5_ib_dev *dev, int *flags, int flag,
 		return;
 	}
 
-	if (flag == MLX5_IB_QP_CREATE_WC_TEST) {
-		/*
-		 * Special case, if condition didn't meet, it won't be error,
-		 * just different in-kernel flow.
-		 */
-		*flags &= ~MLX5_IB_QP_CREATE_WC_TEST;
-		return;
-	}
 	mlx5_ib_dbg(dev, "Verbs create QP flag 0x%X is not supported\n", flag);
 }
 

@@ -3029,8 +3019,6 @@ static int process_create_flags(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 			    IB_QP_CREATE_PCI_WRITE_END_PADDING,
 			    MLX5_CAP_GEN(mdev, end_pad), qp);
 
-	process_create_flag(dev, &create_flags, MLX5_IB_QP_CREATE_WC_TEST,
-			    qp_type != MLX5_IB_QPT_REG_UMR, qp);
 	process_create_flag(dev, &create_flags, MLX5_IB_QP_CREATE_SQPN_QP1,
 			    true, qp);
 

@@ -4611,10 +4599,6 @@ static bool mlx5_ib_modify_qp_allowed(struct mlx5_ib_dev *dev,
 	if (qp->type == IB_QPT_RAW_PACKET || qp->type == MLX5_IB_QPT_REG_UMR)
 		return true;
 
-	/* Internal QP used for wc testing, with NOPs in wq */
-	if (qp->flags & MLX5_IB_QP_CREATE_WC_TEST)
-		return true;
-
 	return false;
 }
 

index 6574c145dc1e..91ab55289104 100644
@@ -1816,6 +1816,7 @@ int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
 	mutex_init(&dev->intf_state_mutex);
 	lockdep_set_class(&dev->intf_state_mutex, &dev->lock_key);
 	mutex_init(&dev->mlx5e_res.uplink_netdev_lock);
+	mutex_init(&dev->wc_state_lock);
 
 	mutex_init(&priv->bfregs.reg_head.lock);
 	mutex_init(&priv->bfregs.wc_head.lock);

@@ -1913,6 +1914,7 @@ void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
 	mutex_destroy(&priv->alloc_mutex);
 	mutex_destroy(&priv->bfregs.wc_head.lock);
 	mutex_destroy(&priv->bfregs.reg_head.lock);
+	mutex_destroy(&dev->wc_state_lock);
 	mutex_destroy(&dev->mlx5e_res.uplink_netdev_lock);
 	mutex_destroy(&dev->intf_state_mutex);
 	lockdep_unregister_key(&dev->lock_key);

434 lines not on this page · raw message

index ee50301a1a0e..145e2fb1b832 100644
@@ -766,6 +766,12 @@ struct mlx5_hca_cap {
 	u32 max[MLX5_UN_SZ_DW(hca_cap_union)];
 };
 
+enum mlx5_wc_state {
+	MLX5_WC_STATE_UNINITIALIZED,
+	MLX5_WC_STATE_UNSUPPORTED,
+	MLX5_WC_STATE_SUPPORTED,
+};
+
 struct mlx5_core_dev {
 	struct device *device;
 	enum mlx5_coredev_type coredev_type;

@@ -824,6 +830,9 @@ struct mlx5_core_dev {
 #endif
 	u64 num_ipsec_offloads;
 	struct mlx5_sd          *sd;
+	enum mlx5_wc_state wc_state;
+	/* sync write combining state */
+	struct mutex wc_state_lock;
 };
 
 struct mlx5_db {

@@ -1377,4 +1386,6 @@ static inline bool mlx5_is_macsec_roce_supported(struct mlx5_core_dev *mdev)
 enum {
 	MLX5_OCTWORD = 16,
 };
+
+bool mlx5_wc_support_get(struct mlx5_core_dev *mdev);
 #endif /* MLX5_DRIVER_H */

index ec4031e851e6..d7a21998d4ca 100644
@@ -4117,6 +4117,7 @@ static void mlx5_ib_stage_pre_ib_reg_umr_cleanup(struct mlx5_ib_dev *dev)
 {
 	mlx5_mkey_cache_cleanup(dev);
 	mlx5r_umr_resource_cleanup(dev);
+	mlx5r_umr_cleanup(dev);
 }
 
 static void mlx5_ib_stage_ib_reg_cleanup(struct mlx5_ib_dev *dev)

@@ -4128,7 +4129,7 @@ static int mlx5_ib_stage_post_ib_reg_umr_init(struct mlx5_ib_dev *dev)
 {
 	int ret;
 
-	ret = mlx5r_umr_resource_init(dev);
+	ret = mlx5r_umr_init(dev);
 	if (ret)
 		return ret;
 

index a74e6f5798f6..34e34808958e 100644
@@ -751,6 +751,8 @@ struct umr_common {
 	 */
 	struct mutex lock;
 	unsigned int state;
+	/* Protects from repeat UMR QP creation */
+	struct mutex init_lock;
 };
 
 #define NUM_MKEYS_PER_PAGE \

index ccb741dc1b37..ba8754e7d362 100644
@@ -1480,6 +1480,7 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 	struct ib_umem *umem;
+	int err;
 
 	if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM))
 		return ERR_PTR(-EOPNOTSUPP);

@@ -1487,6 +1488,10 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	mlx5_ib_dbg(dev, "start 0x%llx, iova 0x%llx, length 0x%llx, access_flags 0x%x\n",
 		    start, iova, length, access_flags);
 
+	err = mlx5r_umr_resource_init(dev);
+	if (err)
+		return ERR_PTR(err);
+
 	if (access_flags & IB_ACCESS_ON_DEMAND)
 		return create_user_odp_mr(pd, start, length, iova, access_flags,
 					  udata);

@@ -1534,6 +1539,10 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset,
 		    "offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x\n",
 		    offset, virt_addr, length, fd, access_flags);
 
+	err = mlx5r_umr_resource_init(dev);
+	if (err)
+		return ERR_PTR(err);
+
 	/* dmabuf requires xlt update via umr to work. */
 	if (!mlx5r_umr_can_load_pas(dev, length))
 		return ERR_PTR(-EINVAL);

101 lines not on this page · raw message

index 3799bb758e49..5f734dc72bef 100644
@@ -16,6 +16,9 @@
 int mlx5r_umr_resource_init(struct mlx5_ib_dev *dev);
 void mlx5r_umr_resource_cleanup(struct mlx5_ib_dev *dev);
 
+int mlx5r_umr_init(struct mlx5_ib_dev *dev);
+void mlx5r_umr_cleanup(struct mlx5_ib_dev *dev);
+
 static inline bool mlx5r_umr_can_load_pas(struct mlx5_ib_dev *dev,
 					  size_t length)
 {

index d7a21998d4ca..a7003316d438 100644
@@ -2854,37 +2854,72 @@ static u8 mlx5_get_umr_fence(u8 umr_fence_cap)
 	}
 }
 
-static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev)
+int mlx5_ib_dev_res_cq_init(struct mlx5_ib_dev *dev)
 {
 	struct mlx5_ib_resources *devr = &dev->devr;
-	struct ib_srq_init_attr attr;
-	struct ib_device *ibdev;
 	struct ib_cq_init_attr cq_attr = {.cqe = 1};
-	int port;
+	struct ib_device *ibdev;
+	struct ib_pd *pd;
+	struct ib_cq *cq;
 	int ret = 0;
 
-	ibdev = &dev->ib_dev;
 
-	if (!MLX5_CAP_GEN(dev->mdev, xrc))
-		return -EOPNOTSUPP;
+	/*
+	 * devr->c0 is set once, never changed until device unload.
+	 * Avoid taking the mutex if initialization is already done.
+	 */
+	if (devr->c0)
+		return 0;
 
-	devr->p0 = ib_alloc_pd(ibdev, 0);
-	if (IS_ERR(devr->p0))
-		return PTR_ERR(devr->p0);
+	mutex_lock(&devr->cq_lock);
+	if (devr->c0)
+		goto unlock;
 
-	devr->c0 = ib_create_cq(ibdev, NULL, NULL, NULL, &cq_attr);
-	if (IS_ERR(devr->c0)) {
-		ret = PTR_ERR(devr->c0);
-		goto error1;
+	ibdev = &dev->ib_dev;
+	pd = ib_alloc_pd(ibdev, 0);
+	if (IS_ERR(pd)) {
+		ret = PTR_ERR(pd);
+		mlx5_ib_err(dev, "Couldn't allocate PD for res init, err=%d\n", ret);
+		goto unlock;
 	}
 
-	ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn0, 0);
-	if (ret)
-		goto error2;
+	cq = ib_create_cq(ibdev, NULL, NULL, NULL, &cq_attr);
+	if (IS_ERR(cq)) {
+		ret = PTR_ERR(cq);
+		mlx5_ib_err(dev, "Couldn't create CQ for res init, err=%d\n", ret);
+		ib_dealloc_pd(pd);
+		goto unlock;
+	}
 
-	ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn1, 0);
+	devr->p0 = pd;
+	devr->c0 = cq;
+
+unlock:
+	mutex_unlock(&devr->cq_lock);
+	return ret;
+}
+
+int mlx5_ib_dev_res_srq_init(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_ib_resources *devr = &dev->devr;
+	struct ib_srq_init_attr attr;
+	struct ib_srq *s0, *s1;
+	int ret = 0;
+
+	/*
+	 * devr->s1 is set once, never changed until device unload.
+	 * Avoid taking the mutex if initialization is already done.
+	 */
+	if (devr->s1)
+		return 0;
+
+	mutex_lock(&devr->srq_lock);
+	if (devr->s1)
+		goto unlock;
+
+	ret = mlx5_ib_dev_res_cq_init(dev);
 	if (ret)
-		goto error3;
+		goto unlock;
 
 	memset(&attr, 0, sizeof(attr));
 	attr.attr.max_sge = 1;

@@ -2892,10 +2927,11 @@ static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev)
 	attr.srq_type = IB_SRQT_XRC;
 	attr.ext.cq = devr->c0;
 
-	devr->s0 = ib_create_srq(devr->p0, &attr);
-	if (IS_ERR(devr->s0)) {
-		ret = PTR_ERR(devr->s0);
-		goto err_create;
+	s0 = ib_create_srq(devr->p0, &attr);
+	if (IS_ERR(s0)) {
+		ret = PTR_ERR(s0);
+		mlx5_ib_err(dev, "Couldn't create SRQ 0 for res init, err=%d\n", ret);
+		goto unlock;
 	}
 
 	memset(&attr, 0, sizeof(attr));

@@ -2903,29 +2939,48 @@ static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev)
 	attr.attr.max_wr = 1;
 	attr.srq_type = IB_SRQT_BASIC;
 
-	devr->s1 = ib_create_srq(devr->p0, &attr);
-	if (IS_ERR(devr->s1)) {
-		ret = PTR_ERR(devr->s1);
-		goto error6;
+	s1 = ib_create_srq(devr->p0, &attr);
+	if (IS_ERR(s1)) {
+		ret = PTR_ERR(s1);
+		mlx5_ib_err(dev, "Couldn't create SRQ 1 for res init, err=%d\n", ret);
+		ib_destroy_srq(s0);
+	}
+
+	devr->s0 = s0;
+	devr->s1 = s1;
+
+unlock:
+	mutex_unlock(&devr->srq_lock);
+	return ret;
+}
+
+static int mlx5_ib_dev_res_init(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_ib_resources *devr = &dev->devr;
+	int port;
+	int ret;
+
+	if (!MLX5_CAP_GEN(dev->mdev, xrc))
+		return -EOPNOTSUPP;
+
+	ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn0, 0);
+	if (ret)
+		return ret;
+
+	ret = mlx5_cmd_xrcd_alloc(dev->mdev, &devr->xrcdn1, 0);
+	if (ret) {
+		mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0);
+		return ret;
 	}
 
 	for (port = 0; port < ARRAY_SIZE(devr->ports); ++port)
 		INIT_WORK(&devr->ports[port].pkey_change_work,
 			  pkey_change_handler);
 
-	return 0;
+	mutex_init(&devr->cq_lock);
+	mutex_init(&devr->srq_lock);
 
-error6:
-	ib_destroy_srq(devr->s0);
-err_create:
-	mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn1, 0);
-error3:
-	mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0);
-error2:
-	ib_destroy_cq(devr->c0);
-error1:
-	ib_dealloc_pd(devr->p0);
-	return ret;
+	return 0;
 }
 
 static void mlx5_ib_dev_res_cleanup(struct mlx5_ib_dev *dev)

@@ -2942,12 +2997,20 @@ static void mlx5_ib_dev_res_cleanup(struct mlx5_ib_dev *dev)
 	for (port = 0; port < ARRAY_SIZE(devr->ports); ++port)
 		cancel_work_sync(&devr->ports[port].pkey_change_work);
 
-	ib_destroy_srq(devr->s1);
-	ib_destroy_srq(devr->s0);
+	/* After s0/s1 init, they are not unset during the device lifetime. */
+	if (devr->s1) {
+		ib_destroy_srq(devr->s1);
+		ib_destroy_srq(devr->s0);
+	}
 	mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn1, 0);
 	mlx5_cmd_xrcd_dealloc(dev->mdev, devr->xrcdn0, 0);
-	ib_destroy_cq(devr->c0);
-	ib_dealloc_pd(devr->p0);
+	/* After p0/c0 init, they are not unset during the device lifetime. */
+	if (devr->c0) {
+		ib_destroy_cq(devr->c0);
+		ib_dealloc_pd(devr->p0);
+	}
+	mutex_destroy(&devr->cq_lock);
+	mutex_destroy(&devr->srq_lock);
 }
 
 static u32 get_core_cap_flags(struct ib_device *ibdev,

index 34e34808958e..b63623336d83 100644
@@ -824,11 +824,13 @@ struct mlx5_ib_port_resources {
 
 struct mlx5_ib_resources {
 	struct ib_cq	*c0;
+	struct mutex cq_lock;
 	u32 xrcdn0;
 	u32 xrcdn1;
 	struct ib_pd	*p0;
 	struct ib_srq	*s0;
 	struct ib_srq	*s1;
+	struct mutex srq_lock;
 	struct mlx5_ib_port_resources ports[2];
 };
 

@@ -1272,6 +1274,8 @@ to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
 		struct mlx5_user_mmap_entry, rdma_entry);
 }
 
+int mlx5_ib_dev_res_cq_init(struct mlx5_ib_dev *dev);
+int mlx5_ib_dev_res_srq_init(struct mlx5_ib_dev *dev);
 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
 			struct mlx5_db *db);
 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);

index 1d09e515aec7..be288cc7a3c0 100644
@@ -3236,6 +3236,10 @@ int mlx5_ib_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *attr,
 	enum ib_qp_type type;
 	int err;
 
+	err = mlx5_ib_dev_res_srq_init(dev);
+	if (err)
+		return err;
+
 	err = check_qp_type(dev, attr, &type);
 	if (err)
 		return err;

index 32c6643d0f7a..ca62bd0ddea8 100644
@@ -213,6 +213,10 @@ int mlx5_ib_create_srq(struct ib_srq *ib_srq,
 		return -EINVAL;
 	}
 
+	err = mlx5_ib_dev_res_cq_init(dev);
+	if (err)
+		return err;
+
 	mutex_init(&srq->mutex);
 	spin_lock_init(&srq->lock);
 	srq->msrq.max    = roundup_pow_of_two(init_attr->attr.max_wr + 1);