Commit 11f552e2 authored by Michael Guralnik's avatar Michael Guralnik Committed by Jason Gunthorpe
Browse files

IB/mlx5: Test write combining support 

Linux can run in all sorts of physical machines and VMs where write
combining may or may not be supported. Currently there is no way to
reliably tell if the system supports WC, or not. The driver uses WC to
optimize posting work to the HCA, and getting this wrong in either
direction can cause a significant performance loss.

Add a test in mlx5_ib initialization process to test whether
write-combining is supported on the machine.  The test will run as part of
the enable_driver callback to ensure that the test runs after the device
is setup and can create and modify the QP needed, but runs before the
device is exposed to the users.

The test opens UD QP and posts NOP WQEs, the WQE written to the BlueFlame
is different from the WQE in memory, requesting CQE only on the BlueFlame
WQE. By checking whether we received a completion on one of these WQEs we
can know if BlueFlame succeeded and this write-combining must be
supported.

Change reporting of BlueFlame support to be dependent on write-combining
support instead of the FW's guess as to what the machine can do.

Link: https://lore.kernel.org/r/20191027062234.10993-1-leon@kernel.org


Signed-off-by: default avatarMichael Guralnik <michaelgur@mellanox.com>
Signed-off-by: default avatarLeon Romanovsky <leonro@mellanox.com>
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
parent 546d3009

drivers/infiniband/hw/mlx5/main.c

+14 −1
Original line number Diff line number Diff line
@@ -1796,7 +1796,7 @@ static int mlx5_ib_alloc_ucontext(struct ib_ucontext *uctx, 
		return -EINVAL;



	resp.qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp);

	if (mlx5_core_is_pf(dev->mdev) && MLX5_CAP_GEN(dev->mdev, bf))

	if (dev->wc_support)

		resp.bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev, log_bf_reg_size);

	resp.cache_line_size = cache_line_size();

	resp.max_sq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_sq);
@@ -6256,6 +6256,7 @@ static const struct ib_device_ops mlx5_ib_dev_ops = { 
	.disassociate_ucontext = mlx5_ib_disassociate_ucontext,

	.drain_rq = mlx5_ib_drain_rq,

	.drain_sq = mlx5_ib_drain_sq,

	.enable_driver = mlx5_ib_enable_driver,

	.fill_res_entry = mlx5_ib_fill_res_entry,

	.fill_stat_entry = mlx5_ib_fill_stat_entry,

	.get_dev_fw_str = get_dev_fw_str,
@@ -6699,6 +6700,18 @@ static void mlx5_ib_stage_devx_cleanup(struct mlx5_ib_dev *dev) 
	}

}



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;

}



void __mlx5_ib_remove(struct mlx5_ib_dev *dev,

		      const struct mlx5_ib_profile *profile,

		      int stage)

drivers/infiniband/hw/mlx5/mem.c

+199 −0
Original line number Diff line number Diff line
@@ -34,6 +34,7 @@ 
#include <rdma/ib_umem.h>

#include <rdma/ib_umem_odp.h>

#include "mlx5_ib.h"

#include <linux/jiffies.h>



/* @umem: umem object to scan

 * @addr: ib virtual address requested by the user
@@ -216,3 +217,201 @@ int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset) 
	*offset = buf_off >> ilog2(off_size);

	return 0;

}



#define WR_ID_BF 0xBF

#define WR_ID_END 0xBAD

#define TEST_WC_NUM_WQES 255

#define TEST_WC_POLLING_MAX_TIME_JIFFIES msecs_to_jiffies(100)

static int post_send_nop(struct mlx5_ib_dev *dev, struct ib_qp *ibqp, u64 wr_id,

			 bool signaled)

{

	struct mlx5_ib_qp *qp = to_mqp(ibqp);

	struct mlx5_wqe_ctrl_seg *ctrl;

	struct mlx5_bf *bf = &qp->bf;

	__be32 mmio_wqe[16] = {};

	unsigned long flags;

	unsigned int idx;

	int i;



	if (unlikely(dev->mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR))

		return -EIO;



	spin_lock_irqsave(&qp->sq.lock, flags);



	idx = qp->sq.cur_post & (qp->sq.wqe_cnt - 1);

	ctrl = mlx5_frag_buf_get_wqe(&qp->sq.fbc, idx);



	memset(ctrl, 0, sizeof(struct mlx5_wqe_ctrl_seg));

	ctrl->fm_ce_se = signaled ? MLX5_WQE_CTRL_CQ_UPDATE : 0;

	ctrl->opmod_idx_opcode =

		cpu_to_be32(((u32)(qp->sq.cur_post) << 8) | MLX5_OPCODE_NOP);

	ctrl->qpn_ds = cpu_to_be32((sizeof(struct mlx5_wqe_ctrl_seg) / 16) |

				   (qp->trans_qp.base.mqp.qpn << 8));



	qp->sq.wrid[idx] = wr_id;

	qp->sq.w_list[idx].opcode = MLX5_OPCODE_NOP;

	qp->sq.wqe_head[idx] = qp->sq.head + 1;

	qp->sq.cur_post += DIV_ROUND_UP(sizeof(struct mlx5_wqe_ctrl_seg),

					MLX5_SEND_WQE_BB);

	qp->sq.w_list[idx].next = qp->sq.cur_post;

	qp->sq.head++;



	memcpy(mmio_wqe, ctrl, sizeof(*ctrl));

	((struct mlx5_wqe_ctrl_seg *)&mmio_wqe)->fm_ce_se |=

		MLX5_WQE_CTRL_CQ_UPDATE;



	/* Make sure that descriptors are written before

	 * updating doorbell record and ringing the doorbell

	 */

	wmb();



	qp->db.db[MLX5_SND_DBR] = cpu_to_be32(qp->sq.cur_post);



	/* Make sure doorbell record is visible to the HCA before

	 * we hit doorbell

	 */

	wmb();

	for (i = 0; i < 8; i++)

		mlx5_write64(&mmio_wqe[i * 2],

			     bf->bfreg->map + bf->offset + i * 8);



	bf->offset ^= bf->buf_size;



	spin_unlock_irqrestore(&qp->sq.lock, flags);



	return 0;

}



static int test_wc_poll_cq_result(struct mlx5_ib_dev *dev, struct ib_cq *cq)

{

	int ret;

	struct ib_wc wc = {};

	unsigned long end = jiffies + TEST_WC_POLLING_MAX_TIME_JIFFIES;



	do {

		ret = ib_poll_cq(cq, 1, &wc);

		if (ret < 0 || wc.status)

			return ret < 0 ? ret : -EINVAL;

		if (ret)

			break;

	} while (!time_after(jiffies, end));



	if (!ret)

		return -ETIMEDOUT;



	if (wc.wr_id != WR_ID_BF)

		ret = 0;



	return ret;

}



static int test_wc_do_send(struct mlx5_ib_dev *dev, struct ib_qp *qp)

{

	int err, i;



	for (i = 0; i < TEST_WC_NUM_WQES; i++) {

		err = post_send_nop(dev, qp, WR_ID_BF, false);

		if (err)

			return err;

	}



	return post_send_nop(dev, qp, WR_ID_END, true);

}



int mlx5_ib_test_wc(struct mlx5_ib_dev *dev)

{

	struct ib_cq_init_attr cq_attr = { .cqe = TEST_WC_NUM_WQES + 1 };

	int port_type_cap = MLX5_CAP_GEN(dev->mdev, port_type);

	struct ib_qp_init_attr qp_init_attr = {

		.cap = { .max_send_wr = TEST_WC_NUM_WQES },

		.qp_type = IB_QPT_UD,

		.sq_sig_type = IB_SIGNAL_REQ_WR,

		.create_flags = MLX5_IB_QP_CREATE_WC_TEST,

	};

	struct ib_qp_attr qp_attr = { .port_num = 1 };

	struct ib_device *ibdev = &dev->ib_dev;

	struct ib_qp *qp;

	struct ib_cq *cq;

	struct ib_pd *pd;

	int ret;



	if (!MLX5_CAP_GEN(dev->mdev, bf))

		return 0;



	if (!dev->mdev->roce.roce_en &&

	    port_type_cap == MLX5_CAP_PORT_TYPE_ETH) {

		if (mlx5_core_is_pf(dev->mdev))

			dev->wc_support = true;

		return 0;

	}



	ret = mlx5_alloc_bfreg(dev->mdev, &dev->wc_bfreg, true, false);

	if (ret)

		goto print_err;



	if (!dev->wc_bfreg.wc)

		goto out1;



	pd = ib_alloc_pd(ibdev, 0);

	if (IS_ERR(pd)) {

		ret = PTR_ERR(pd);

		goto out1;

	}



	cq = ib_create_cq(ibdev, NULL, NULL, NULL, &cq_attr);

	if (IS_ERR(cq)) {

		ret = PTR_ERR(cq);

		goto out2;

	}



	qp_init_attr.recv_cq = cq;

	qp_init_attr.send_cq = cq;

	qp = ib_create_qp(pd, &qp_init_attr);

	if (IS_ERR(qp)) {

		ret = PTR_ERR(qp);

		goto out3;

	}



	qp_attr.qp_state = IB_QPS_INIT;

	ret = ib_modify_qp(qp, &qp_attr,

			   IB_QP_STATE | IB_QP_PORT | IB_QP_PKEY_INDEX |

				   IB_QP_QKEY);

	if (ret)

		goto out4;



	qp_attr.qp_state = IB_QPS_RTR;

	ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);

	if (ret)

		goto out4;



	qp_attr.qp_state = IB_QPS_RTS;

	ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);

	if (ret)

		goto out4;



	ret = test_wc_do_send(dev, qp);

	if (ret < 0)

		goto out4;



	ret = test_wc_poll_cq_result(dev, cq);

	if (ret > 0) {

		dev->wc_support = true;

		ret = 0;

	}



out4:

	ib_destroy_qp(qp);

out3:

	ib_destroy_cq(cq);

out2:

	ib_dealloc_pd(pd);

out1:

	mlx5_free_bfreg(dev->mdev, &dev->wc_bfreg);

print_err:

	if (ret)

		mlx5_ib_err(

			dev,

			"Error %d while trying to test write-combining support\n",

			ret);

	return ret;

}

drivers/infiniband/hw/mlx5/mlx5_ib.h

+6 −0
Original line number Diff line number Diff line
@@ -248,6 +248,7 @@ 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;
@@ -966,6 +967,7 @@ struct mlx5_ib_dev { 
	u8				fill_delay:1;

	u8				is_rep:1;

	u8				lag_active:1;

	u8				wc_support:1;

	struct umr_common		umrc;

	/* sync used page count stats

	 */
@@ -993,6 +995,7 @@ 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;
@@ -1506,4 +1509,7 @@ static inline bool mlx5_ib_can_use_umr(struct mlx5_ib_dev *dev, 


	return true;

}



int mlx5_ib_enable_driver(struct ib_device *dev);

int mlx5_ib_test_wc(struct mlx5_ib_dev *dev);

#endif /* MLX5_IB_H */

drivers/infiniband/hw/mlx5/qp.c

+4 −2
Original line number Diff line number Diff line
@@ -1041,11 +1041,14 @@ static int create_kernel_qp(struct mlx5_ib_dev *dev, 
					IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK |

					IB_QP_CREATE_IPOIB_UD_LSO |

					IB_QP_CREATE_NETIF_QP |

					MLX5_IB_QP_CREATE_SQPN_QP1))

					MLX5_IB_QP_CREATE_SQPN_QP1 |

					MLX5_IB_QP_CREATE_WC_TEST))

		return -EINVAL;



	if (init_attr->qp_type == MLX5_IB_QPT_REG_UMR)

		qp->bf.bfreg = &dev->fp_bfreg;

	else if (init_attr->create_flags & MLX5_IB_QP_CREATE_WC_TEST)

		qp->bf.bfreg = &dev->wc_bfreg;

	else

		qp->bf.bfreg = &dev->bfreg;
@@ -5328,7 +5331,6 @@ out: 
		 * we hit doorbell */

		wmb();



		/* currently we support only regular doorbells */

		mlx5_write64((__be32 *)ctrl, bf->bfreg->map + bf->offset);

		/* Make sure doorbells don't leak out of SQ spinlock

		 * and reach the HCA out of order.

CRA Git | Maintained and supported by SUSTech CRA and CCSE