[PATCH AUTOSEL 4.19 053/123] RDMA/mlx5: Fix fence type for IB_WR_LOCAL_INV WR

[PATCH AUTOSEL 4.19 053/123] RDMA/mlx5: Fix fence type for IB_WR_LOCAL_INV WR

[Sasha Levin]

From: Majd Dibbiny <majd@mellanox.com>

[ Upstream commit 074fca3a18e7e1e0d4d7dcc9d7badc43b90232f4 ]

Currently, for IB_WR_LOCAL_INV WR, when the next fence is None, the
current fence will be SMALL instead of Normal Fence.

Without this patch krping doesn't work on CX-5 devices and throws
following error:

The error messages are from CX5 driver are: (from server side)
[ 710.434014] mlx5_0:dump_cqe:278:(pid 2712): dump error cqe
[ 710.434016] 00000000 00000000 00000000 00000000
[ 710.434016] 00000000 00000000 00000000 00000000
[ 710.434017] 00000000 00000000 00000000 00000000
[ 710.434018] 00000000 93003204 100000b8 000524d2
[ 710.434019] krping: cq completion failed with wr_id 0 status 4 opcode 128 vender_err 32

Fixed the logic to set the correct fence type.

Fixes: 6e8484c5cf07 ("RDMA/mlx5: set UMR wqe fence according to HCA cap")
Signed-off-by: Majd Dibbiny <majd@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/hw/mlx5/qp.c | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index d53d954ac8af..183fe5c8ceb7 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -4413,17 +4413,18 @@ static int _mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 			goto out;
 		}
 
-		if (wr->opcode == IB_WR_LOCAL_INV ||
-		    wr->opcode == IB_WR_REG_MR) {
+		if (wr->opcode == IB_WR_REG_MR) {
 			fence = dev->umr_fence;
 			next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
-		} else if (wr->send_flags & IB_SEND_FENCE) {
-			if (qp->next_fence)
-				fence = MLX5_FENCE_MODE_SMALL_AND_FENCE;
-			else
-				fence = MLX5_FENCE_MODE_FENCE;
-		} else {
-			fence = qp->next_fence;
+		} else  {
+			if (wr->send_flags & IB_SEND_FENCE) {
+				if (qp->next_fence)
+					fence = MLX5_FENCE_MODE_SMALL_AND_FENCE;
+				else
+					fence = MLX5_FENCE_MODE_FENCE;
+			} else {
+				fence = qp->next_fence;
+			}
 		}
 
 		switch (ibqp->qp_type) {
-- 
2.17.1

[PATCH AUTOSEL 4.19 054/123] RDMA/core: Add GIDs while changing MAC addr only for registered ndev

[Sasha Levin]

From: Parav Pandit <parav@mellanox.com>

[ Upstream commit d52ef88a9f4be523425730da3239cf87bee936da ]

Currently when MAC address is changed, regardless of the netdev reg_state,
GID entries are removed and added to reflect the new MAC address and new
default GID entries.

When a bonding device is used and the underlying PCI device is removed
several netdevice events are generated. Two events of the interest are
CHANGEADDR and UNREGISTER event on lower(slave) netdevice of the bond
netdevice.

Sometimes CHANGEADDR event is generated when netdev state is
UNREGISTERING (after UNREGISTER event is generated). In this scenario, GID
entries for default GIDs are added and never deleted because GID entries
are deleted only when netdev state is < UNREGISTERED.

This leads to non zero reference count on the netdevice. Due to this, PCI
device unbind operation is getting stuck.

To avoid it, when changing mac address, add GID entries only if netdev is
in REGISTERED state.

Fixes: 03db3a2d81e6 ("IB/core: Add RoCE GID table management")
Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/core/roce_gid_mgmt.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/core/roce_gid_mgmt.c b/drivers/infiniband/core/roce_gid_mgmt.c
index ee366199b169..25d43c8f1c2a 100644
--- a/drivers/infiniband/core/roce_gid_mgmt.c
+++ b/drivers/infiniband/core/roce_gid_mgmt.c
@@ -767,8 +767,10 @@ static int netdevice_event(struct notifier_block *this, unsigned long event,
 
 	case NETDEV_CHANGEADDR:
 		cmds[0] = netdev_del_cmd;
-		cmds[1] = add_default_gid_cmd;
-		cmds[2] = add_cmd;
+		if (ndev->reg_state == NETREG_REGISTERED) {
+			cmds[1] = add_default_gid_cmd;
+			cmds[2] = add_cmd;
+		}
 		break;
 
 	case NETDEV_CHANGEUPPER:
-- 
2.17.1

[PATCH AUTOSEL 4.19 055/123] RDMA/bnxt_re: Fix system hang when registration with L2 driver fails

[Sasha Levin]

From: Selvin Xavier <selvin.xavier@broadcom.com>

[ Upstream commit 3c4b1419c33c2417836a63f8126834ee36968321 ]

Driver doesn't release rtnl lock if registration with
L2 driver (bnxt_re_register_netdev) fais and this causes
hang while requesting for the next lock.

[  371.635416] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  371.635417] kworker/u48:1   D    0   634      2 0x80000000
[  371.635423] Workqueue: bnxt_re bnxt_re_task [bnxt_re]
[  371.635424] Call Trace:
[  371.635426]  ? __schedule+0x36b/0xbd0
[  371.635429]  schedule+0x39/0x90
[  371.635430]  schedule_preempt_disabled+0x11/0x20
[  371.635431]  __mutex_lock+0x45b/0x9c0
[  371.635433]  ? __mutex_lock+0x16d/0x9c0
[  371.635435]  ? bnxt_re_ib_reg+0x2b/0xb30 [bnxt_re]
[  371.635438]  ? wake_up_klogd+0x37/0x40
[  371.635442]  bnxt_re_ib_reg+0x2b/0xb30 [bnxt_re]
[  371.635447]  bnxt_re_task+0xfd/0x180 [bnxt_re]
[  371.635449]  process_one_work+0x216/0x5b0
[  371.635450]  ? process_one_work+0x189/0x5b0
[  371.635453]  worker_thread+0x4e/0x3d0
[  371.635455]  kthread+0x10e/0x140
[  371.635456]  ? process_one_work+0x5b0/0x5b0
[  371.635458]  ? kthread_stop+0x220/0x220
[  371.635460]  ret_from_fork+0x3a/0x50
[  371.635477] INFO: task NetworkManager:1228 blocked for more than 120 seconds.
[  371.635478]       Tainted: G    B      OE     4.20.0-rc1+ #42
[  371.635479] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.

Release the rtnl_lock correctly in the failure path.

Fixes: de5c95d0f518 ("RDMA/bnxt_re: Fix system crash during RDMA resource initialization")
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/hw/bnxt_re/main.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index 85cd1a3593d6..bb3f16273938 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -1252,6 +1252,7 @@ static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
 	/* Registered a new RoCE device instance to netdev */
 	rc = bnxt_re_register_netdev(rdev);
 	if (rc) {
+		rtnl_unlock();
 		pr_err("Failed to register with netedev: %#x\n", rc);
 		return -EINVAL;
 	}
-- 
2.17.1

[PATCH AUTOSEL 4.19 056/123] RDMA/bnxt_re: Avoid accessing the device structure after it is freed

[Sasha Levin]

From: Selvin Xavier <selvin.xavier@broadcom.com>

[ Upstream commit a6c66d6a08b88cc10aca9d3f65cfae31e7652a99 ]

When bnxt_re_ib_reg returns failure, the device structure gets
freed. Driver tries to access the device pointer
after it is freed.

[ 4871.034744] Failed to register with netedev: 0xffffffa1
[ 4871.034765] infiniband (null): Failed to register with IB: 0xffffffea
[ 4871.046430] ==================================================================
[ 4871.046437] BUG: KASAN: use-after-free in bnxt_re_task+0x63/0x180 [bnxt_re]
[ 4871.046439] Write of size 4 at addr ffff880fa8406f48 by task kworker/u48:2/17813

[ 4871.046443] CPU: 20 PID: 17813 Comm: kworker/u48:2 Kdump: loaded Tainted: G B OE  4.20.0-rc1+ #42
[ 4871.046444] Hardware name: Dell Inc. PowerEdge R730/0599V5, BIOS 1.0.4 08/28/2014
[ 4871.046447] Workqueue: bnxt_re bnxt_re_task [bnxt_re]
[ 4871.046449] Call Trace:
[ 4871.046454]  dump_stack+0x91/0xeb
[ 4871.046458]  print_address_description+0x6a/0x2a0
[ 4871.046461]  kasan_report+0x176/0x2d0
[ 4871.046463]  ? bnxt_re_task+0x63/0x180 [bnxt_re]
[ 4871.046466]  bnxt_re_task+0x63/0x180 [bnxt_re]
[ 4871.046470]  process_one_work+0x216/0x5b0
[ 4871.046471]  ? process_one_work+0x189/0x5b0
[ 4871.046475]  worker_thread+0x4e/0x3d0
[ 4871.046479]  kthread+0x10e/0x140
[ 4871.046480]  ? process_one_work+0x5b0/0x5b0
[ 4871.046482]  ? kthread_stop+0x220/0x220
[ 4871.046486]  ret_from_fork+0x3a/0x50

[ 4871.046492] The buggy address belongs to the page:
[ 4871.046494] page:ffffea003ea10180 count:0 mapcount:0 mapping:0000000000000000 index:0x0
[ 4871.046495] flags: 0x57ffffc0000000()
[ 4871.046498] raw: 0057ffffc0000000 0000000000000000 ffffea003ea10188 0000000000000000
[ 4871.046500] raw: 0000000000000000 0000000000000000 00000000ffffffff 0000000000000000
[ 4871.046501] page dumped because: kasan: bad access detected

Avoid accessing the device structure once it is freed.

Fixes: 497158aa5f52 ("RDMA/bnxt_re: Fix the ib_reg failure cleanup")
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/hw/bnxt_re/main.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/infiniband/hw/bnxt_re/main.c b/drivers/infiniband/hw/bnxt_re/main.c
index bb3f16273938..22bd9784fa2e 100644
--- a/drivers/infiniband/hw/bnxt_re/main.c
+++ b/drivers/infiniband/hw/bnxt_re/main.c
@@ -1462,6 +1462,7 @@ static void bnxt_re_task(struct work_struct *work)
 				"Failed to register with IB: %#x", rc);
 			bnxt_re_remove_one(rdev);
 			bnxt_re_dev_unreg(rdev);
+			goto exit;
 		}
 		break;
 	case NETDEV_UP:
@@ -1485,6 +1486,7 @@ static void bnxt_re_task(struct work_struct *work)
 	}
 	smp_mb__before_atomic();
 	atomic_dec(&rdev->sched_count);
+exit:
 	kfree(re_work);
 }
 
-- 
2.17.1

[PATCH AUTOSEL 4.19 057/123] RDMA/rdmavt: Fix rvt_create_ah function signature

[Sasha Levin]

From: Kamal Heib <kamalheib1@gmail.com>

[ Upstream commit 4f32fb921b153ae9ea280e02a3e91509fffc03d3 ]

rdmavt uses a crazy system that looses the type checking when assinging
functions to struct ib_device function pointers. Because of this the
signature to this function was not changed when the below commit revised
things.

Fix the signature so we are not calling a function pointer with a
mismatched signature.

Fixes: 477864c8fcd9 ("IB/core: Let create_ah return extended response to user")
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/sw/rdmavt/ah.c | 4 +++-
 drivers/infiniband/sw/rdmavt/ah.h | 3 ++-
 2 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/sw/rdmavt/ah.c b/drivers/infiniband/sw/rdmavt/ah.c
index 89ec0f64abfc..084bb4baebb5 100644
--- a/drivers/infiniband/sw/rdmavt/ah.c
+++ b/drivers/infiniband/sw/rdmavt/ah.c
@@ -91,13 +91,15 @@ EXPORT_SYMBOL(rvt_check_ah);
  * rvt_create_ah - create an address handle
  * @pd: the protection domain
  * @ah_attr: the attributes of the AH
+ * @udata: pointer to user's input output buffer information.
  *
  * This may be called from interrupt context.
  *
  * Return: newly allocated ah
  */
 struct ib_ah *rvt_create_ah(struct ib_pd *pd,
-			    struct rdma_ah_attr *ah_attr)
+			    struct rdma_ah_attr *ah_attr,
+			    struct ib_udata *udata)
 {
 	struct rvt_ah *ah;
 	struct rvt_dev_info *dev = ib_to_rvt(pd->device);
diff --git a/drivers/infiniband/sw/rdmavt/ah.h b/drivers/infiniband/sw/rdmavt/ah.h
index 16105af99189..25271b48a683 100644
--- a/drivers/infiniband/sw/rdmavt/ah.h
+++ b/drivers/infiniband/sw/rdmavt/ah.h
@@ -51,7 +51,8 @@
 #include <rdma/rdma_vt.h>
 
 struct ib_ah *rvt_create_ah(struct ib_pd *pd,
-			    struct rdma_ah_attr *ah_attr);
+			    struct rdma_ah_attr *ah_attr,
+			    struct ib_udata *udata);
 int rvt_destroy_ah(struct ib_ah *ibah);
 int rvt_modify_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
 int rvt_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
-- 
2.17.1

[PATCH AUTOSEL 4.19 064/123] RDMA/hns: Bugfix pbl configuration for rereg mr

[Sasha Levin]

From: Yixian Liu <liuyixian@huawei.com>

[ Upstream commit ca088320a02537f36c243ac21794525d8eabb3bd ]

Current hns driver assigned the first two PBL page addresses from previous
registered MR to the hardware when reregister MR changing the memory
locations occurred. This will lead to PBL addressing error as the PBL has
already been released. This patch fixes this wrong assignment by using the
page address from new allocated PBL.

Fixes: a2c80b7b4119 ("RDMA/hns: Add rereg mr support for hip08")
Signed-off-by: Yixian Liu <liuyixian@huawei.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 128 ++++++++++-----------
 1 file changed, 60 insertions(+), 68 deletions(-)

diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
index 0218c0f8c2a7..a442b29e7611 100644
--- a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
+++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
@@ -1661,10 +1661,9 @@ static int hns_roce_v2_set_mac(struct hns_roce_dev *hr_dev, u8 phy_port,
 	return hns_roce_cmq_send(hr_dev, &desc, 1);
 }
 
-static int hns_roce_v2_write_mtpt(void *mb_buf, struct hns_roce_mr *mr,
-				  unsigned long mtpt_idx)
+static int set_mtpt_pbl(struct hns_roce_v2_mpt_entry *mpt_entry,
+			struct hns_roce_mr *mr)
 {
-	struct hns_roce_v2_mpt_entry *mpt_entry;
 	struct scatterlist *sg;
 	u64 page_addr;
 	u64 *pages;
@@ -1672,6 +1671,53 @@ static int hns_roce_v2_write_mtpt(void *mb_buf, struct hns_roce_mr *mr,
 	int len;
 	int entry;
 
+	mpt_entry->pbl_size = cpu_to_le32(mr->pbl_size);
+	mpt_entry->pbl_ba_l = cpu_to_le32(lower_32_bits(mr->pbl_ba >> 3));
+	roce_set_field(mpt_entry->byte_48_mode_ba,
+		       V2_MPT_BYTE_48_PBL_BA_H_M, V2_MPT_BYTE_48_PBL_BA_H_S,
+		       upper_32_bits(mr->pbl_ba >> 3));
+
+	pages = (u64 *)__get_free_page(GFP_KERNEL);
+	if (!pages)
+		return -ENOMEM;
+
+	i = 0;
+	for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
+		len = sg_dma_len(sg) >> PAGE_SHIFT;
+		for (j = 0; j < len; ++j) {
+			page_addr = sg_dma_address(sg) +
+				(j << mr->umem->page_shift);
+			pages[i] = page_addr >> 6;
+			/* Record the first 2 entry directly to MTPT table */
+			if (i >= HNS_ROCE_V2_MAX_INNER_MTPT_NUM - 1)
+				goto found;
+			i++;
+		}
+	}
+found:
+	mpt_entry->pa0_l = cpu_to_le32(lower_32_bits(pages[0]));
+	roce_set_field(mpt_entry->byte_56_pa0_h, V2_MPT_BYTE_56_PA0_H_M,
+		       V2_MPT_BYTE_56_PA0_H_S, upper_32_bits(pages[0]));
+
+	mpt_entry->pa1_l = cpu_to_le32(lower_32_bits(pages[1]));
+	roce_set_field(mpt_entry->byte_64_buf_pa1, V2_MPT_BYTE_64_PA1_H_M,
+		       V2_MPT_BYTE_64_PA1_H_S, upper_32_bits(pages[1]));
+	roce_set_field(mpt_entry->byte_64_buf_pa1,
+		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_M,
+		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_S,
+		       mr->pbl_buf_pg_sz + PG_SHIFT_OFFSET);
+
+	free_page((unsigned long)pages);
+
+	return 0;
+}
+
+static int hns_roce_v2_write_mtpt(void *mb_buf, struct hns_roce_mr *mr,
+				  unsigned long mtpt_idx)
+{
+	struct hns_roce_v2_mpt_entry *mpt_entry;
+	int ret;
+
 	mpt_entry = mb_buf;
 	memset(mpt_entry, 0, sizeof(*mpt_entry));
 
@@ -1686,7 +1732,6 @@ static int hns_roce_v2_write_mtpt(void *mb_buf, struct hns_roce_mr *mr,
 		       mr->pbl_ba_pg_sz + PG_SHIFT_OFFSET);
 	roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PD_M,
 		       V2_MPT_BYTE_4_PD_S, mr->pd);
-	mpt_entry->byte_4_pd_hop_st = cpu_to_le32(mpt_entry->byte_4_pd_hop_st);
 
 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RA_EN_S, 0);
 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_R_INV_EN_S, 1);
@@ -1700,13 +1745,11 @@ static int hns_roce_v2_write_mtpt(void *mb_buf, struct hns_roce_mr *mr,
 		     (mr->access & IB_ACCESS_REMOTE_WRITE ? 1 : 0));
 	roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_LW_EN_S,
 		     (mr->access & IB_ACCESS_LOCAL_WRITE ? 1 : 0));
-	mpt_entry->byte_8_mw_cnt_en = cpu_to_le32(mpt_entry->byte_8_mw_cnt_en);
 
 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_PA_S,
 		     mr->type == MR_TYPE_MR ? 0 : 1);
 	roce_set_bit(mpt_entry->byte_12_mw_pa, V2_MPT_BYTE_12_INNER_PA_VLD_S,
 		     1);
-	mpt_entry->byte_12_mw_pa = cpu_to_le32(mpt_entry->byte_12_mw_pa);
 
 	mpt_entry->len_l = cpu_to_le32(lower_32_bits(mr->size));
 	mpt_entry->len_h = cpu_to_le32(upper_32_bits(mr->size));
@@ -1717,53 +1760,9 @@ static int hns_roce_v2_write_mtpt(void *mb_buf, struct hns_roce_mr *mr,
 	if (mr->type == MR_TYPE_DMA)
 		return 0;
 
-	mpt_entry->pbl_size = cpu_to_le32(mr->pbl_size);
-
-	mpt_entry->pbl_ba_l = cpu_to_le32(lower_32_bits(mr->pbl_ba >> 3));
-	roce_set_field(mpt_entry->byte_48_mode_ba, V2_MPT_BYTE_48_PBL_BA_H_M,
-		       V2_MPT_BYTE_48_PBL_BA_H_S,
-		       upper_32_bits(mr->pbl_ba >> 3));
-	mpt_entry->byte_48_mode_ba = cpu_to_le32(mpt_entry->byte_48_mode_ba);
-
-	pages = (u64 *)__get_free_page(GFP_KERNEL);
-	if (!pages)
-		return -ENOMEM;
-
-	i = 0;
-	for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
-		len = sg_dma_len(sg) >> PAGE_SHIFT;
-		for (j = 0; j < len; ++j) {
-			page_addr = sg_dma_address(sg) +
-				    (j << mr->umem->page_shift);
-			pages[i] = page_addr >> 6;
-
-			/* Record the first 2 entry directly to MTPT table */
-			if (i >= HNS_ROCE_V2_MAX_INNER_MTPT_NUM - 1)
-				goto found;
-			i++;
-		}
-	}
+	ret = set_mtpt_pbl(mpt_entry, mr);
 
-found:
-	mpt_entry->pa0_l = cpu_to_le32(lower_32_bits(pages[0]));
-	roce_set_field(mpt_entry->byte_56_pa0_h, V2_MPT_BYTE_56_PA0_H_M,
-		       V2_MPT_BYTE_56_PA0_H_S,
-		       upper_32_bits(pages[0]));
-	mpt_entry->byte_56_pa0_h = cpu_to_le32(mpt_entry->byte_56_pa0_h);
-
-	mpt_entry->pa1_l = cpu_to_le32(lower_32_bits(pages[1]));
-	roce_set_field(mpt_entry->byte_64_buf_pa1, V2_MPT_BYTE_64_PA1_H_M,
-		       V2_MPT_BYTE_64_PA1_H_S, upper_32_bits(pages[1]));
-
-	free_page((unsigned long)pages);
-
-	roce_set_field(mpt_entry->byte_64_buf_pa1,
-		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_M,
-		       V2_MPT_BYTE_64_PBL_BUF_PG_SZ_S,
-		       mr->pbl_buf_pg_sz + PG_SHIFT_OFFSET);
-	mpt_entry->byte_64_buf_pa1 = cpu_to_le32(mpt_entry->byte_64_buf_pa1);
-
-	return 0;
+	return ret;
 }
 
 static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev,
@@ -1772,6 +1771,7 @@ static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev,
 					u64 size, void *mb_buf)
 {
 	struct hns_roce_v2_mpt_entry *mpt_entry = mb_buf;
+	int ret = 0;
 
 	if (flags & IB_MR_REREG_PD) {
 		roce_set_field(mpt_entry->byte_4_pd_hop_st, V2_MPT_BYTE_4_PD_M,
@@ -1784,14 +1784,14 @@ static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev,
 			     V2_MPT_BYTE_8_BIND_EN_S,
 			     (mr_access_flags & IB_ACCESS_MW_BIND ? 1 : 0));
 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en,
-			   V2_MPT_BYTE_8_ATOMIC_EN_S,
-			   (mr_access_flags & IB_ACCESS_REMOTE_ATOMIC ? 1 : 0));
+			     V2_MPT_BYTE_8_ATOMIC_EN_S,
+			     mr_access_flags & IB_ACCESS_REMOTE_ATOMIC ? 1 : 0);
 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RR_EN_S,
-			     (mr_access_flags & IB_ACCESS_REMOTE_READ ? 1 : 0));
+			     mr_access_flags & IB_ACCESS_REMOTE_READ ? 1 : 0);
 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_RW_EN_S,
-			    (mr_access_flags & IB_ACCESS_REMOTE_WRITE ? 1 : 0));
+			     mr_access_flags & IB_ACCESS_REMOTE_WRITE ? 1 : 0);
 		roce_set_bit(mpt_entry->byte_8_mw_cnt_en, V2_MPT_BYTE_8_LW_EN_S,
-			     (mr_access_flags & IB_ACCESS_LOCAL_WRITE ? 1 : 0));
+			     mr_access_flags & IB_ACCESS_LOCAL_WRITE ? 1 : 0);
 	}
 
 	if (flags & IB_MR_REREG_TRANS) {
@@ -1800,21 +1800,13 @@ static int hns_roce_v2_rereg_write_mtpt(struct hns_roce_dev *hr_dev,
 		mpt_entry->len_l = cpu_to_le32(lower_32_bits(size));
 		mpt_entry->len_h = cpu_to_le32(upper_32_bits(size));
 
-		mpt_entry->pbl_size = cpu_to_le32(mr->pbl_size);
-		mpt_entry->pbl_ba_l =
-				cpu_to_le32(lower_32_bits(mr->pbl_ba >> 3));
-		roce_set_field(mpt_entry->byte_48_mode_ba,
-			       V2_MPT_BYTE_48_PBL_BA_H_M,
-			       V2_MPT_BYTE_48_PBL_BA_H_S,
-			       upper_32_bits(mr->pbl_ba >> 3));
-		mpt_entry->byte_48_mode_ba =
-				cpu_to_le32(mpt_entry->byte_48_mode_ba);
-
 		mr->iova = iova;
 		mr->size = size;
+
+		ret = set_mtpt_pbl(mpt_entry, mr);
 	}
 
-	return 0;
+	return ret;
 }
 
 static void *get_cqe_v2(struct hns_roce_cq *hr_cq, int n)
-- 
2.17.1

[PATCH AUTOSEL 4.19 076/123] IB/mlx5: Fix page fault handling for MW

[Sasha Levin]

From: Artemy Kovalyov <artemyko@mellanox.com>

[ Upstream commit 75b7b86bdb0df37e08e44b6c1f99010967f81944 ]

Memory windows are implemented with an indirect MKey, when a page fault
event comes for a MW Mkey we need to find the MR at the end of the list of
the indirect MKeys by iterating on all items from the first to the last.

The offset calculated during this process has to be zeroed after the first
iteration or the next iteration will start from a wrong address, resulting
incorrect ODP faulting behavior.

Fixes: db570d7deafb ("IB/mlx5: Add ODP support to MW")
Signed-off-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/infiniband/hw/mlx5/odp.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index d216e0d2921d..9e1cac8cb260 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -724,6 +724,7 @@ static int pagefault_single_data_segment(struct mlx5_ib_dev *dev,
 			head = frame;
 
 			bcnt -= frame->bcnt;
+			offset = 0;
 		}
 		break;
 
-- 
2.17.1