From 33eb12f2966ef9c529ce7138711a9c150a17931e Mon Sep 17 00:00:00 2001
From: Patrisious Haddad <phaddad@nvidia.com>
Date: Wed, 30 Dec 2020 15:02:40 +0200
Subject: RDMA/nldev: Return an error message on failure to turn auto mode

The bounded counter can't be reconfigured to be in auto mode, in attempt
to do it, the user will get an error, but without any hint why. Update
nldev interface to return an error message through extack mechanism.

Link: https://lore.kernel.org/r/20201230130240.180737-1-leon@kernel.org
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 include/rdma/rdma_counter.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

(limited to 'include')

diff --git a/include/rdma/rdma_counter.h b/include/rdma/rdma_counter.h
index eb99856e8b30..e75cf9742e04 100644
--- a/include/rdma/rdma_counter.h
+++ b/include/rdma/rdma_counter.h
@@ -46,7 +46,8 @@ struct rdma_counter {
 void rdma_counter_init(struct ib_device *dev);
 void rdma_counter_release(struct ib_device *dev);
 int rdma_counter_set_auto_mode(struct ib_device *dev, u8 port,
-			       bool on, enum rdma_nl_counter_mask mask);
+			       enum rdma_nl_counter_mask mask,
+			       struct netlink_ext_ack *extack);
 int rdma_counter_bind_qp_auto(struct ib_qp *qp, u8 port);
 int rdma_counter_unbind_qp(struct ib_qp *qp, bool force);
 
-- 
cgit v1.2.3


From 368c0159d492d7fbdb5791b40c9263ec4e97a10f Mon Sep 17 00:00:00 2001
From: Jianxin Xiong <jianxin.xiong@intel.com>
Date: Tue, 15 Dec 2020 13:27:13 -0800
Subject: RDMA/umem: Support importing dma-buf as user memory region

Dma-buf is a standard cross-driver buffer sharing mechanism that can be
used to support peer-to-peer access from RDMA devices.

Device memory exported via dma-buf is associated with a file descriptor.
This is passed to the user space as a property associated with the buffer
allocation. When the buffer is registered as a memory region, the file
descriptor is passed to the RDMA driver along with other parameters.

Implement the common code for importing dma-buf object and mapping dma-buf
pages.

Link: https://lore.kernel.org/r/1608067636-98073-2-git-send-email-jianxin.xiong@intel.com
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Acked-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Acked-by: Christian Koenig <christian.koenig@amd.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/Kconfig            |   1 +
 drivers/infiniband/core/Makefile      |   2 +-
 drivers/infiniband/core/umem.c        |   3 +
 drivers/infiniband/core/umem_dmabuf.c | 174 ++++++++++++++++++++++++++++++++++
 include/rdma/ib_umem.h                |  48 +++++++++-
 5 files changed, 224 insertions(+), 4 deletions(-)
 create mode 100644 drivers/infiniband/core/umem_dmabuf.c

(limited to 'include')

diff --git a/drivers/infiniband/Kconfig b/drivers/infiniband/Kconfig
index 9325e189a215..04a78d9f8fe3 100644
--- a/drivers/infiniband/Kconfig
+++ b/drivers/infiniband/Kconfig
@@ -41,6 +41,7 @@ config INFINIBAND_USER_MEM
 	bool
 	depends on INFINIBAND_USER_ACCESS != n
 	depends on MMU
+	select DMA_SHARED_BUFFER
 	default y
 
 config INFINIBAND_ON_DEMAND_PAGING
diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile
index ccf2670ef45e..8ab4eea5a0a5 100644
--- a/drivers/infiniband/core/Makefile
+++ b/drivers/infiniband/core/Makefile
@@ -40,5 +40,5 @@ ib_uverbs-y :=			uverbs_main.o uverbs_cmd.o uverbs_marshall.o \
 				uverbs_std_types_srq.o \
 				uverbs_std_types_wq.o \
 				uverbs_std_types_qp.o
-ib_uverbs-$(CONFIG_INFINIBAND_USER_MEM) += umem.o
+ib_uverbs-$(CONFIG_INFINIBAND_USER_MEM) += umem.o umem_dmabuf.o
 ib_uverbs-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += umem_odp.o
diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index 917338db7ac1..2dde99a9ba07 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -2,6 +2,7 @@
  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
  * Copyright (c) 2005 Cisco Systems.  All rights reserved.
  * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
+ * Copyright (c) 2020 Intel Corporation. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -278,6 +279,8 @@ void ib_umem_release(struct ib_umem *umem)
 {
 	if (!umem)
 		return;
+	if (umem->is_dmabuf)
+		return ib_umem_dmabuf_release(to_ib_umem_dmabuf(umem));
 	if (umem->is_odp)
 		return ib_umem_odp_release(to_ib_umem_odp(umem));
 
diff --git a/drivers/infiniband/core/umem_dmabuf.c b/drivers/infiniband/core/umem_dmabuf.c
new file mode 100644
index 000000000000..f9b5162d9260
--- /dev/null
+++ b/drivers/infiniband/core/umem_dmabuf.c
@@ -0,0 +1,174 @@
+// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
+/*
+ * Copyright (c) 2020 Intel Corporation. All rights reserved.
+ */
+
+#include <linux/dma-buf.h>
+#include <linux/dma-resv.h>
+#include <linux/dma-mapping.h>
+
+#include "uverbs.h"
+
+int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf)
+{
+	struct sg_table *sgt;
+	struct scatterlist *sg;
+	struct dma_fence *fence;
+	unsigned long start, end, cur = 0;
+	unsigned int nmap = 0;
+	int i;
+
+	dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
+
+	if (umem_dmabuf->sgt)
+		goto wait_fence;
+
+	sgt = dma_buf_map_attachment(umem_dmabuf->attach, DMA_BIDIRECTIONAL);
+	if (IS_ERR(sgt))
+		return PTR_ERR(sgt);
+
+	/* modify the sg list in-place to match umem address and length */
+
+	start = ALIGN_DOWN(umem_dmabuf->umem.address, PAGE_SIZE);
+	end = ALIGN(umem_dmabuf->umem.address + umem_dmabuf->umem.length,
+		    PAGE_SIZE);
+	for_each_sgtable_dma_sg(sgt, sg, i) {
+		if (start < cur + sg_dma_len(sg) && cur < end)
+			nmap++;
+		if (cur <= start && start < cur + sg_dma_len(sg)) {
+			unsigned long offset = start - cur;
+
+			umem_dmabuf->first_sg = sg;
+			umem_dmabuf->first_sg_offset = offset;
+			sg_dma_address(sg) += offset;
+			sg_dma_len(sg) -= offset;
+			cur += offset;
+		}
+		if (cur < end && end <= cur + sg_dma_len(sg)) {
+			unsigned long trim = cur + sg_dma_len(sg) - end;
+
+			umem_dmabuf->last_sg = sg;
+			umem_dmabuf->last_sg_trim = trim;
+			sg_dma_len(sg) -= trim;
+			break;
+		}
+		cur += sg_dma_len(sg);
+	}
+
+	umem_dmabuf->umem.sg_head.sgl = umem_dmabuf->first_sg;
+	umem_dmabuf->umem.sg_head.nents = nmap;
+	umem_dmabuf->umem.nmap = nmap;
+	umem_dmabuf->sgt = sgt;
+
+wait_fence:
+	/*
+	 * Although the sg list is valid now, the content of the pages
+	 * may be not up-to-date. Wait for the exporter to finish
+	 * the migration.
+	 */
+	fence = dma_resv_get_excl(umem_dmabuf->attach->dmabuf->resv);
+	if (fence)
+		return dma_fence_wait(fence, false);
+
+	return 0;
+}
+EXPORT_SYMBOL(ib_umem_dmabuf_map_pages);
+
+void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf)
+{
+	dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
+
+	if (!umem_dmabuf->sgt)
+		return;
+
+	/* retore the original sg list */
+	if (umem_dmabuf->first_sg) {
+		sg_dma_address(umem_dmabuf->first_sg) -=
+			umem_dmabuf->first_sg_offset;
+		sg_dma_len(umem_dmabuf->first_sg) +=
+			umem_dmabuf->first_sg_offset;
+		umem_dmabuf->first_sg = NULL;
+		umem_dmabuf->first_sg_offset = 0;
+	}
+	if (umem_dmabuf->last_sg) {
+		sg_dma_len(umem_dmabuf->last_sg) +=
+			umem_dmabuf->last_sg_trim;
+		umem_dmabuf->last_sg = NULL;
+		umem_dmabuf->last_sg_trim = 0;
+	}
+
+	dma_buf_unmap_attachment(umem_dmabuf->attach, umem_dmabuf->sgt,
+				 DMA_BIDIRECTIONAL);
+
+	umem_dmabuf->sgt = NULL;
+}
+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)
+{
+	struct dma_buf *dmabuf;
+	struct ib_umem_dmabuf *umem_dmabuf;
+	struct ib_umem *umem;
+	unsigned long end;
+	struct ib_umem_dmabuf *ret = ERR_PTR(-EINVAL);
+
+	if (check_add_overflow(offset, (unsigned long)size, &end))
+		return ret;
+
+	if (unlikely(!ops || !ops->move_notify))
+		return ret;
+
+	dmabuf = dma_buf_get(fd);
+	if (IS_ERR(dmabuf))
+		return ERR_CAST(dmabuf);
+
+	if (dmabuf->size < end)
+		goto out_release_dmabuf;
+
+	umem_dmabuf = kzalloc(sizeof(*umem_dmabuf), GFP_KERNEL);
+	if (!umem_dmabuf) {
+		ret = ERR_PTR(-ENOMEM);
+		goto out_release_dmabuf;
+	}
+
+	umem = &umem_dmabuf->umem;
+	umem->ibdev = device;
+	umem->length = size;
+	umem->address = offset;
+	umem->writable = ib_access_writable(access);
+	umem->is_dmabuf = 1;
+
+	if (!ib_umem_num_pages(umem))
+		goto out_free_umem;
+
+	umem_dmabuf->attach = dma_buf_dynamic_attach(
+					dmabuf,
+					device->dma_device,
+					ops,
+					umem_dmabuf);
+	if (IS_ERR(umem_dmabuf->attach)) {
+		ret = ERR_CAST(umem_dmabuf->attach);
+		goto out_free_umem;
+	}
+	return umem_dmabuf;
+
+out_free_umem:
+	kfree(umem_dmabuf);
+
+out_release_dmabuf:
+	dma_buf_put(dmabuf);
+	return ret;
+}
+EXPORT_SYMBOL(ib_umem_dmabuf_get);
+
+void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf)
+{
+	struct dma_buf *dmabuf = umem_dmabuf->attach->dmabuf;
+
+	dma_buf_detach(dmabuf, umem_dmabuf->attach);
+	dma_buf_put(dmabuf);
+	kfree(umem_dmabuf);
+}
diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h
index 7752211c9638..676c57f5ca80 100644
--- a/include/rdma/ib_umem.h
+++ b/include/rdma/ib_umem.h
@@ -1,6 +1,7 @@
 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
 /*
  * Copyright (c) 2007 Cisco Systems.  All rights reserved.
+ * Copyright (c) 2020 Intel Corporation.  All rights reserved.
  */
 
 #ifndef IB_UMEM_H
@@ -13,6 +14,7 @@
 
 struct ib_ucontext;
 struct ib_umem_odp;
+struct dma_buf_attach_ops;
 
 struct ib_umem {
 	struct ib_device       *ibdev;
@@ -22,12 +24,29 @@ struct ib_umem {
 	unsigned long		address;
 	u32 writable : 1;
 	u32 is_odp : 1;
+	u32 is_dmabuf : 1;
 	struct work_struct	work;
 	struct sg_table sg_head;
 	int             nmap;
 	unsigned int    sg_nents;
 };
 
+struct ib_umem_dmabuf {
+	struct ib_umem umem;
+	struct dma_buf_attachment *attach;
+	struct sg_table *sgt;
+	struct scatterlist *first_sg;
+	struct scatterlist *last_sg;
+	unsigned long first_sg_offset;
+	unsigned long last_sg_trim;
+	void *private;
+};
+
+static inline struct ib_umem_dmabuf *to_ib_umem_dmabuf(struct ib_umem *umem)
+{
+	return container_of(umem, struct ib_umem_dmabuf, umem);
+}
+
 /* Returns the offset of the umem start relative to the first page. */
 static inline int ib_umem_offset(struct ib_umem *umem)
 {
@@ -86,6 +105,7 @@ int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset,
 unsigned long ib_umem_find_best_pgsz(struct ib_umem *umem,
 				     unsigned long pgsz_bitmap,
 				     unsigned long virt);
+
 /**
  * ib_umem_find_best_pgoff - Find best HW page size
  *
@@ -116,6 +136,14 @@ static inline unsigned long ib_umem_find_best_pgoff(struct ib_umem *umem,
 				      dma_addr & pgoff_bitmask);
 }
 
+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);
+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);
+
 #else /* CONFIG_INFINIBAND_USER_MEM */
 
 #include <linux/err.h>
@@ -124,12 +152,12 @@ static inline struct ib_umem *ib_umem_get(struct ib_device *device,
 					  unsigned long addr, size_t size,
 					  int access)
 {
-	return ERR_PTR(-EINVAL);
+	return ERR_PTR(-EOPNOTSUPP);
 }
 static inline void ib_umem_release(struct ib_umem *umem) { }
 static inline int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset,
 		      		    size_t length) {
-	return -EINVAL;
+	return -EOPNOTSUPP;
 }
 static inline unsigned long ib_umem_find_best_pgsz(struct ib_umem *umem,
 						   unsigned long pgsz_bitmap,
@@ -143,7 +171,21 @@ static inline unsigned long ib_umem_find_best_pgoff(struct ib_umem *umem,
 {
 	return 0;
 }
+static inline
+struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device,
+					  unsigned long offset,
+					  size_t size, int fd,
+					  int access,
+					  struct dma_buf_attach_ops *ops)
+{
+	return ERR_PTR(-EOPNOTSUPP);
+}
+static inline int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf)
+{
+	return -EOPNOTSUPP;
+}
+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) { }
 
 #endif /* CONFIG_INFINIBAND_USER_MEM */
-
 #endif /* IB_UMEM_H */
-- 
cgit v1.2.3


From 3bc489e8827a93b14d27211ae0576b3c1de85000 Mon Sep 17 00:00:00 2001
From: Jianxin Xiong <jianxin.xiong@intel.com>
Date: Tue, 15 Dec 2020 13:27:14 -0800
Subject: RDMA/core: Add device method for registering dma-buf based memory
 region

Dma-buf based memory region requires one extra parameter and is processed
quite differently. Adding a separate method allows clean separation from
regular memory regions.

Link: https://lore.kernel.org/r/1608067636-98073-3-git-send-email-jianxin.xiong@intel.com
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Acked-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Acked-by: Christian Koenig <christian.koenig@amd.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/core/device.c | 1 +
 include/rdma/ib_verbs.h          | 6 +++++-
 2 files changed, 6 insertions(+), 1 deletion(-)

(limited to 'include')

diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index 3d08373c7797..c895d7bfa512 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -2681,6 +2681,7 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops)
 	SET_DEVICE_OP(dev_ops, read_counters);
 	SET_DEVICE_OP(dev_ops, reg_dm_mr);
 	SET_DEVICE_OP(dev_ops, reg_user_mr);
+	SET_DEVICE_OP(dev_ops, reg_user_mr_dmabuf);
 	SET_DEVICE_OP(dev_ops, req_ncomp_notif);
 	SET_DEVICE_OP(dev_ops, req_notify_cq);
 	SET_DEVICE_OP(dev_ops, rereg_user_mr);
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 9fed65bf9279..62e574c50555 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -2,7 +2,7 @@
 /*
  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
- * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004, 2020 Intel Corporation.  All rights reserved.
  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
@@ -2434,6 +2434,10 @@ struct ib_device_ops {
 	struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
 				     u64 virt_addr, int mr_access_flags,
 				     struct ib_udata *udata);
+	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 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,
-- 
cgit v1.2.3


From bfe0cc6eb2491aeae2bba02a305fcce13cd90624 Mon Sep 17 00:00:00 2001
From: Jianxin Xiong <jianxin.xiong@intel.com>
Date: Tue, 15 Dec 2020 13:27:15 -0800
Subject: RDMA/uverbs: Add uverbs command for dma-buf based MR registration

Implement a new uverbs ioctl method for memory registration with file
descriptor as an extra parameter.

Link: https://lore.kernel.org/r/1608067636-98073-4-git-send-email-jianxin.xiong@intel.com
Signed-off-by: Jianxin Xiong <jianxin.xiong@intel.com>
Reviewed-by: Sean Hefty <sean.hefty@intel.com>
Acked-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Acked-by: Christian Koenig <christian.koenig@amd.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/core/uverbs_std_types_mr.c | 117 +++++++++++++++++++++++++-
 include/uapi/rdma/ib_user_ioctl_cmds.h        |  14 +++
 2 files changed, 129 insertions(+), 2 deletions(-)

(limited to 'include')

diff --git a/drivers/infiniband/core/uverbs_std_types_mr.c b/drivers/infiniband/core/uverbs_std_types_mr.c
index dd4e76b26c74..f782d5e1aa25 100644
--- a/drivers/infiniband/core/uverbs_std_types_mr.c
+++ b/drivers/infiniband/core/uverbs_std_types_mr.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2018, Mellanox Technologies inc.  All rights reserved.
+ * Copyright (c) 2020, Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -182,6 +183,86 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QUERY_MR)(
 	return IS_UVERBS_COPY_ERR(ret) ? ret : 0;
 }
 
+static int UVERBS_HANDLER(UVERBS_METHOD_REG_DMABUF_MR)(
+	struct uverbs_attr_bundle *attrs)
+{
+	struct ib_uobject *uobj =
+		uverbs_attr_get_uobject(attrs, UVERBS_ATTR_REG_DMABUF_MR_HANDLE);
+	struct ib_pd *pd =
+		uverbs_attr_get_obj(attrs, UVERBS_ATTR_REG_DMABUF_MR_PD_HANDLE);
+	struct ib_device *ib_dev = pd->device;
+
+	u64 offset, length, iova;
+	u32 fd, access_flags;
+	struct ib_mr *mr;
+	int ret;
+
+	if (!ib_dev->ops.reg_user_mr_dmabuf)
+		return -EOPNOTSUPP;
+
+	ret = uverbs_copy_from(&offset, attrs,
+			       UVERBS_ATTR_REG_DMABUF_MR_OFFSET);
+	if (ret)
+		return ret;
+
+	ret = uverbs_copy_from(&length, attrs,
+			       UVERBS_ATTR_REG_DMABUF_MR_LENGTH);
+	if (ret)
+		return ret;
+
+	ret = uverbs_copy_from(&iova, attrs,
+			       UVERBS_ATTR_REG_DMABUF_MR_IOVA);
+	if (ret)
+		return ret;
+
+	if ((offset & ~PAGE_MASK) != (iova & ~PAGE_MASK))
+		return -EINVAL;
+
+	ret = uverbs_copy_from(&fd, attrs,
+			       UVERBS_ATTR_REG_DMABUF_MR_FD);
+	if (ret)
+		return ret;
+
+	ret = uverbs_get_flags32(&access_flags, attrs,
+				 UVERBS_ATTR_REG_DMABUF_MR_ACCESS_FLAGS,
+				 IB_ACCESS_LOCAL_WRITE |
+				 IB_ACCESS_REMOTE_READ |
+				 IB_ACCESS_REMOTE_WRITE |
+				 IB_ACCESS_REMOTE_ATOMIC |
+				 IB_ACCESS_RELAXED_ORDERING);
+	if (ret)
+		return ret;
+
+	ret = ib_check_mr_access(ib_dev, access_flags);
+	if (ret)
+		return ret;
+
+	mr = pd->device->ops.reg_user_mr_dmabuf(pd, offset, length, iova, fd,
+						access_flags,
+						&attrs->driver_udata);
+	if (IS_ERR(mr))
+		return PTR_ERR(mr);
+
+	mr->device = pd->device;
+	mr->pd = pd;
+	mr->type = IB_MR_TYPE_USER;
+	mr->uobject = uobj;
+	atomic_inc(&pd->usecnt);
+
+	uobj->object = mr;
+
+	uverbs_finalize_uobj_create(attrs, UVERBS_ATTR_REG_DMABUF_MR_HANDLE);
+
+	ret = uverbs_copy_to(attrs, UVERBS_ATTR_REG_DMABUF_MR_RESP_LKEY,
+			     &mr->lkey, sizeof(mr->lkey));
+	if (ret)
+		return ret;
+
+	ret = uverbs_copy_to(attrs, UVERBS_ATTR_REG_DMABUF_MR_RESP_RKEY,
+			     &mr->rkey, sizeof(mr->rkey));
+	return ret;
+}
+
 DECLARE_UVERBS_NAMED_METHOD(
 	UVERBS_METHOD_ADVISE_MR,
 	UVERBS_ATTR_IDR(UVERBS_ATTR_ADVISE_MR_PD_HANDLE,
@@ -247,6 +328,37 @@ DECLARE_UVERBS_NAMED_METHOD(
 			    UVERBS_ATTR_TYPE(u32),
 			    UA_MANDATORY));
 
+DECLARE_UVERBS_NAMED_METHOD(
+	UVERBS_METHOD_REG_DMABUF_MR,
+	UVERBS_ATTR_IDR(UVERBS_ATTR_REG_DMABUF_MR_HANDLE,
+			UVERBS_OBJECT_MR,
+			UVERBS_ACCESS_NEW,
+			UA_MANDATORY),
+	UVERBS_ATTR_IDR(UVERBS_ATTR_REG_DMABUF_MR_PD_HANDLE,
+			UVERBS_OBJECT_PD,
+			UVERBS_ACCESS_READ,
+			UA_MANDATORY),
+	UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_OFFSET,
+			   UVERBS_ATTR_TYPE(u64),
+			   UA_MANDATORY),
+	UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_LENGTH,
+			   UVERBS_ATTR_TYPE(u64),
+			   UA_MANDATORY),
+	UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_IOVA,
+			   UVERBS_ATTR_TYPE(u64),
+			   UA_MANDATORY),
+	UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_FD,
+			   UVERBS_ATTR_TYPE(u32),
+			   UA_MANDATORY),
+	UVERBS_ATTR_FLAGS_IN(UVERBS_ATTR_REG_DMABUF_MR_ACCESS_FLAGS,
+			     enum ib_access_flags),
+	UVERBS_ATTR_PTR_OUT(UVERBS_ATTR_REG_DMABUF_MR_RESP_LKEY,
+			    UVERBS_ATTR_TYPE(u32),
+			    UA_MANDATORY),
+	UVERBS_ATTR_PTR_OUT(UVERBS_ATTR_REG_DMABUF_MR_RESP_RKEY,
+			    UVERBS_ATTR_TYPE(u32),
+			    UA_MANDATORY));
+
 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
 	UVERBS_METHOD_MR_DESTROY,
 	UVERBS_ATTR_IDR(UVERBS_ATTR_DESTROY_MR_HANDLE,
@@ -257,10 +369,11 @@ DECLARE_UVERBS_NAMED_METHOD_DESTROY(
 DECLARE_UVERBS_NAMED_OBJECT(
 	UVERBS_OBJECT_MR,
 	UVERBS_TYPE_ALLOC_IDR(uverbs_free_mr),
+	&UVERBS_METHOD(UVERBS_METHOD_ADVISE_MR),
 	&UVERBS_METHOD(UVERBS_METHOD_DM_MR_REG),
 	&UVERBS_METHOD(UVERBS_METHOD_MR_DESTROY),
-	&UVERBS_METHOD(UVERBS_METHOD_ADVISE_MR),
-	&UVERBS_METHOD(UVERBS_METHOD_QUERY_MR));
+	&UVERBS_METHOD(UVERBS_METHOD_QUERY_MR),
+	&UVERBS_METHOD(UVERBS_METHOD_REG_DMABUF_MR));
 
 const struct uapi_definition uverbs_def_obj_mr[] = {
 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(UVERBS_OBJECT_MR,
diff --git a/include/uapi/rdma/ib_user_ioctl_cmds.h b/include/uapi/rdma/ib_user_ioctl_cmds.h
index 7968a1845355..dafc7ebe545b 100644
--- a/include/uapi/rdma/ib_user_ioctl_cmds.h
+++ b/include/uapi/rdma/ib_user_ioctl_cmds.h
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2018, Mellanox Technologies inc.  All rights reserved.
+ * Copyright (c) 2020, Intel Corporation. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -251,6 +252,7 @@ enum uverbs_methods_mr {
 	UVERBS_METHOD_MR_DESTROY,
 	UVERBS_METHOD_ADVISE_MR,
 	UVERBS_METHOD_QUERY_MR,
+	UVERBS_METHOD_REG_DMABUF_MR,
 };
 
 enum uverbs_attrs_mr_destroy_ids {
@@ -272,6 +274,18 @@ enum uverbs_attrs_query_mr_cmd_attr_ids {
 	UVERBS_ATTR_QUERY_MR_RESP_IOVA,
 };
 
+enum uverbs_attrs_reg_dmabuf_mr_cmd_attr_ids {
+	UVERBS_ATTR_REG_DMABUF_MR_HANDLE,
+	UVERBS_ATTR_REG_DMABUF_MR_PD_HANDLE,
+	UVERBS_ATTR_REG_DMABUF_MR_OFFSET,
+	UVERBS_ATTR_REG_DMABUF_MR_LENGTH,
+	UVERBS_ATTR_REG_DMABUF_MR_IOVA,
+	UVERBS_ATTR_REG_DMABUF_MR_FD,
+	UVERBS_ATTR_REG_DMABUF_MR_ACCESS_FLAGS,
+	UVERBS_ATTR_REG_DMABUF_MR_RESP_LKEY,
+	UVERBS_ATTR_REG_DMABUF_MR_RESP_RKEY,
+};
+
 enum uverbs_attrs_create_counters_cmd_attr_ids {
 	UVERBS_ATTR_CREATE_COUNTERS_HANDLE,
 };
-- 
cgit v1.2.3


From 3ce60f443b143e649aa26cd3f668d645434647ac Mon Sep 17 00:00:00 2001
From: Parav Pandit <parav@nvidia.com>
Date: Wed, 3 Feb 2021 15:01:29 +0200
Subject: IB/mlx5: Move mlx5_port_caps from mlx5_core_dev to mlx5_ib_dev

mlx5_port_caps are RDMA specific capabilities. These are not used by the
mlx5_core_device at all. Move them to mlx5_ib_dev where it is used and
reduce the scope of it to multiple drivers.

Link: https://lore.kernel.org/r/20210203130133.4057329-2-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/hw/mlx5/mad.c     |  8 ++++----
 drivers/infiniband/hw/mlx5/main.c    | 14 ++++++--------
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  8 ++++++++
 drivers/infiniband/hw/mlx5/qp.c      |  6 +++---
 drivers/infiniband/hw/mlx5/wr.c      |  2 +-
 include/linux/mlx5/driver.h          |  8 --------
 6 files changed, 22 insertions(+), 24 deletions(-)

(limited to 'include')

diff --git a/drivers/infiniband/hw/mlx5/mad.c b/drivers/infiniband/hw/mlx5/mad.c
index 53dec6063245..e9d0a5269582 100644
--- a/drivers/infiniband/hw/mlx5/mad.c
+++ b/drivers/infiniband/hw/mlx5/mad.c
@@ -48,7 +48,7 @@ static bool can_do_mad_ifc(struct mlx5_ib_dev *dev, u8 port_num,
 	if (in_mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_SUBN_LID_ROUTED &&
 	    in_mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
 		return true;
-	return dev->mdev->port_caps[port_num - 1].has_smi;
+	return dev->port_caps[port_num - 1].has_smi;
 }
 
 static int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey,
@@ -299,7 +299,7 @@ int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port)
 
 	packet_error = be16_to_cpu(out_mad->status);
 
-	dev->mdev->port_caps[port - 1].ext_port_cap = (!err && !packet_error) ?
+	dev->port_caps[port - 1].ext_port_cap = (!err && !packet_error) ?
 		MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO : 0;
 
 out:
@@ -549,7 +549,7 @@ int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port,
 	props->port_cap_flags	= be32_to_cpup((__be32 *)(out_mad->data + 20));
 	props->gid_tbl_len	= out_mad->data[50];
 	props->max_msg_sz	= 1 << MLX5_CAP_GEN(mdev, log_max_msg);
-	props->pkey_tbl_len	= mdev->port_caps[port - 1].pkey_table_len;
+	props->pkey_tbl_len	= dev->port_caps[port - 1].pkey_table_len;
 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 46));
 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 48));
 	props->active_width	= out_mad->data[31] & 0xf;
@@ -589,7 +589,7 @@ int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port,
 
 	/* If reported active speed is QDR, check if is FDR-10 */
 	if (props->active_speed == 4) {
-		if (mdev->port_caps[port - 1].ext_port_cap &
+		if (dev->port_caps[port - 1].ext_port_cap &
 		    MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO) {
 			init_query_mad(in_mad);
 			in_mad->attr_id = MLX5_ATTR_EXTENDED_PORT_INFO;
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 40fb86db3376..d2f185ae2b9c 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -2946,8 +2946,8 @@ static int set_has_smi_cap(struct mlx5_ib_dev *dev)
 	int err;
 	int port;
 
-	for (port = 1; port <= ARRAY_SIZE(dev->mdev->port_caps); port++) {
-		dev->mdev->port_caps[port - 1].has_smi = false;
+	for (port = 1; port <= ARRAY_SIZE(dev->port_caps); port++) {
+		dev->port_caps[port - 1].has_smi = false;
 		if (MLX5_CAP_GEN(dev->mdev, port_type) ==
 		    MLX5_CAP_PORT_TYPE_IB) {
 			if (MLX5_CAP_GEN(dev->mdev, ib_virt)) {
@@ -2959,10 +2959,10 @@ static int set_has_smi_cap(struct mlx5_ib_dev *dev)
 						    port, err);
 					return err;
 				}
-				dev->mdev->port_caps[port - 1].has_smi =
+				dev->port_caps[port - 1].has_smi =
 					vport_ctx.has_smi;
 			} else {
-				dev->mdev->port_caps[port - 1].has_smi = true;
+				dev->port_caps[port - 1].has_smi = true;
 			}
 		}
 	}
@@ -3004,10 +3004,8 @@ static int __get_port_caps(struct mlx5_ib_dev *dev, u8 port)
 		goto out;
 	}
 
-	dev->mdev->port_caps[port - 1].pkey_table_len =
-					dprops->max_pkeys;
-	dev->mdev->port_caps[port - 1].gid_table_len =
-					pprops->gid_tbl_len;
+	dev->port_caps[port - 1].pkey_table_len = dprops->max_pkeys;
+	dev->port_caps[port - 1].gid_table_len = pprops->gid_tbl_len;
 	mlx5_ib_dbg(dev, "port %d: pkey_table_len %d, gid_table_len %d\n",
 		    port, dprops->max_pkeys, pprops->gid_tbl_len);
 
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 2fd2927abe45..c2f91c15b973 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -1036,6 +1036,13 @@ struct mlx5_var_table {
 	u64 num_var_hw_entries;
 };
 
+struct mlx5_port_caps {
+	int gid_table_len;
+	int pkey_table_len;
+	bool has_smi;
+	u8 ext_port_cap;
+};
+
 struct mlx5_ib_dev {
 	struct ib_device		ib_dev;
 	struct mlx5_core_dev		*mdev;
@@ -1096,6 +1103,7 @@ struct mlx5_ib_dev {
 	struct mlx5_var_table var_table;
 
 	struct xarray sig_mrs;
+	struct mlx5_port_caps port_caps[MLX5_MAX_PORTS];
 };
 
 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 88be94215708..5274349dd998 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -3177,10 +3177,10 @@ static int mlx5_set_path(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 
 	if (ah_flags & IB_AH_GRH) {
 		if (grh->sgid_index >=
-		    dev->mdev->port_caps[port - 1].gid_table_len) {
+		    dev->port_caps[port - 1].gid_table_len) {
 			pr_err("sgid_index (%u) too large. max is %d\n",
 			       grh->sgid_index,
-			       dev->mdev->port_caps[port - 1].gid_table_len);
+			       dev->port_caps[port - 1].gid_table_len);
 			return -EINVAL;
 		}
 	}
@@ -4311,7 +4311,7 @@ int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	if (attr_mask & IB_QP_PKEY_INDEX) {
 		port = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
 		if (attr->pkey_index >=
-		    dev->mdev->port_caps[port - 1].pkey_table_len) {
+		    dev->port_caps[port - 1].pkey_table_len) {
 			mlx5_ib_dbg(dev, "invalid pkey index %d\n",
 				    attr->pkey_index);
 			goto out;
diff --git a/drivers/infiniband/hw/mlx5/wr.c b/drivers/infiniband/hw/mlx5/wr.c
index d6038fb6c50c..cf2852cba45c 100644
--- a/drivers/infiniband/hw/mlx5/wr.c
+++ b/drivers/infiniband/hw/mlx5/wr.c
@@ -1369,7 +1369,7 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 			handle_qpt_uc(wr, &seg, &size);
 			break;
 		case IB_QPT_SMI:
-			if (unlikely(!mdev->port_caps[qp->port - 1].has_smi)) {
+			if (unlikely(!dev->port_caps[qp->port - 1].has_smi)) {
 				mlx5_ib_warn(dev, "Send SMP MADs is not allowed\n");
 				err = -EPERM;
 				*bad_wr = wr;
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index f93bfe7473aa..11558c2e99f0 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -305,13 +305,6 @@ struct mlx5_cmd {
 	struct mlx5_cmd_stats *stats;
 };
 
-struct mlx5_port_caps {
-	int	gid_table_len;
-	int	pkey_table_len;
-	u8	ext_port_cap;
-	bool	has_smi;
-};
-
 struct mlx5_cmd_mailbox {
 	void	       *buf;
 	dma_addr_t	dma;
@@ -694,7 +687,6 @@ struct mlx5_core_dev {
 	u8			rev_id;
 	char			board_id[MLX5_BOARD_ID_LEN];
 	struct mlx5_cmd		cmd;
-	struct mlx5_port_caps	port_caps[MLX5_MAX_PORTS];
 	struct {
 		u32 hca_cur[MLX5_CAP_NUM][MLX5_UN_SZ_DW(hca_cap_union)];
 		u32 hca_max[MLX5_CAP_NUM][MLX5_UN_SZ_DW(hca_cap_union)];
-- 
cgit v1.2.3


From 7416790e22452bfa86de6b55638eacf7780c8f6f Mon Sep 17 00:00:00 2001
From: Parav Pandit <parav@nvidia.com>
Date: Wed, 3 Feb 2021 15:01:32 +0200
Subject: RDMA/core: Introduce and use API to read port immutable data

Currently mlx5 driver caches port GID table length for 2 ports.  It is
also cached by IB core as port immutable data.

When mlx5 representor ports are present, which are usually more than 2,
invalid access to port_caps array can happen while validating the GID
table length which is only for 2 ports.

To avoid this, take help of the IB cores port immutable data by exposing
an API to read the port immutable fields.

Remove mlx5 driver's internal cache, thereby reduce code and data.

Link: https://lore.kernel.org/r/20210203130133.4057329-5-leon@kernel.org
Signed-off-by: Parav Pandit <parav@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/core/device.c     | 14 +++++++++
 drivers/infiniband/hw/mlx5/main.c    | 55 +-----------------------------------
 drivers/infiniband/hw/mlx5/mlx5_ib.h |  1 -
 drivers/infiniband/hw/mlx5/qp.c      |  8 ++++--
 include/rdma/ib_verbs.h              |  3 ++
 5 files changed, 23 insertions(+), 58 deletions(-)

(limited to 'include')

diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index c895d7bfa512..051c018fb73c 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -848,6 +848,20 @@ static int setup_port_data(struct ib_device *device)
 	return 0;
 }
 
+/**
+ * ib_port_immutable_read() - Read rdma port's immutable data
+ * @dev - IB device
+ * @port - port number whose immutable data to read. It starts with index 1 and
+ *         valid upto including rdma_end_port().
+ */
+const struct ib_port_immutable*
+ib_port_immutable_read(struct ib_device *dev, unsigned int port)
+{
+	WARN_ON(!rdma_is_port_valid(dev, port));
+	return &dev->port_data[port].immutable;
+}
+EXPORT_SYMBOL(ib_port_immutable_read);
+
 void ib_get_device_fw_str(struct ib_device *dev, char *str)
 {
 	if (dev->ops.get_dev_fw_str)
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 176f2a866f12..1e1e3edcb1d5 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -2964,41 +2964,6 @@ static void get_ext_port_caps(struct mlx5_ib_dev *dev)
 		mlx5_query_ext_port_caps(dev, port);
 }
 
-static int __get_port_caps(struct mlx5_ib_dev *dev, u8 port)
-{
-	struct ib_port_attr *pprops = NULL;
-	int err = -ENOMEM;
-
-	pprops = kzalloc(sizeof(*pprops), GFP_KERNEL);
-	if (!pprops)
-		goto out;
-
-	err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
-	if (err) {
-		mlx5_ib_warn(dev, "query_port %d failed %d\n",
-			     port, err);
-		goto out;
-	}
-
-	dev->port_caps[port - 1].gid_table_len = pprops->gid_tbl_len;
-	mlx5_ib_dbg(dev, "port %d: pkey_table_len %d, gid_table_len %d\n",
-		    port, dev->pkey_table_len, pprops->gid_tbl_len);
-
-out:
-	kfree(pprops);
-	return err;
-}
-
-static int get_port_caps(struct mlx5_ib_dev *dev, u8 port)
-{
-	/* For representors use port 1, is this is the only native
-	 * port
-	 */
-	if (dev->is_rep)
-		return __get_port_caps(dev, 1);
-	return __get_port_caps(dev, port);
-}
-
 static u8 mlx5_get_umr_fence(u8 umr_fence_cap)
 {
 	switch (umr_fence_cap) {
@@ -3472,10 +3437,6 @@ static bool mlx5_ib_bind_slave_port(struct mlx5_ib_dev *ibdev,
 	if (err)
 		goto unbind;
 
-	err = get_port_caps(ibdev, mlx5_core_native_port_num(mpi->mdev));
-	if (err)
-		goto unbind;
-
 	err = mlx5_add_netdev_notifier(ibdev, port_num);
 	if (err) {
 		mlx5_ib_err(ibdev, "failed adding netdev notifier for port %u\n",
@@ -3553,11 +3514,9 @@ static int mlx5_ib_init_multiport_master(struct mlx5_ib_dev *dev)
 				break;
 			}
 		}
-		if (!bound) {
-			get_port_caps(dev, i + 1);
+		if (!bound)
 			mlx5_ib_dbg(dev, "no free port found for port %d\n",
 				    i + 1);
-		}
 	}
 
 	list_add_tail(&dev->ib_dev_list, &mlx5_ib_dev_list);
@@ -3940,18 +3899,6 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
 	if (err)
 		goto err_mp;
 
-	if (!mlx5_core_mp_enabled(mdev)) {
-		for (i = 1; i <= dev->num_ports; i++) {
-			err = get_port_caps(dev, i);
-			if (err)
-				break;
-		}
-	} else {
-		err = get_port_caps(dev, mlx5_core_native_port_num(mdev));
-	}
-	if (err)
-		goto err_mp;
-
 	err = mlx5_query_max_pkeys(&dev->ib_dev, &dev->pkey_table_len);
 	if (err)
 		goto err_mp;
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 36a92f3c29e3..0f567d570230 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -1037,7 +1037,6 @@ struct mlx5_var_table {
 };
 
 struct mlx5_port_caps {
-	int gid_table_len;
 	bool has_smi;
 	u8 ext_port_cap;
 };
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 475470237e0b..358c44c5a8fc 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -3176,11 +3176,13 @@ static int mlx5_set_path(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 			 alt ? attr->alt_pkey_index : attr->pkey_index);
 
 	if (ah_flags & IB_AH_GRH) {
-		if (grh->sgid_index >=
-		    dev->port_caps[port - 1].gid_table_len) {
+		const struct ib_port_immutable *immutable;
+
+		immutable = ib_port_immutable_read(&dev->ib_dev, port);
+		if (grh->sgid_index >= immutable->gid_tbl_len) {
 			pr_err("sgid_index (%u) too large. max is %d\n",
 			       grh->sgid_index,
-			       dev->port_caps[port - 1].gid_table_len);
+			       immutable->gid_tbl_len);
 			return -EINVAL;
 		}
 	}
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 62e574c50555..ca28fca5736b 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -4674,4 +4674,7 @@ static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
 
 	return (u32)(v & IB_GRH_FLOWLABEL_MASK);
 }
+
+const struct ib_port_immutable*
+ib_port_immutable_read(struct ib_device *dev, unsigned int port);
 #endif /* IB_VERBS_H */
-- 
cgit v1.2.3


From db72438c9319cfd37e3c237a7754ca862ae12d63 Mon Sep 17 00:00:00 2001
From: Yishai Hadas <yishaih@nvidia.com>
Date: Tue, 2 Feb 2021 09:13:09 +0200
Subject: RDMA/mlx5: Cleanup the synchronize_srcu() from the ODP flow

Cleanup the synchronize_srcu() from the ODP flow as it was found to be a
very heavy time consumer as part of dereg_mr.

For example de-registration of 10000 ODP MRs each with size of 2M hugepage
took 19.6 sec comparing de-registration of same number of non ODP MRs that
took 172 ms.

The new locking scheme uses the wait_event() mechanism which follows the
use count of the MR instead of using synchronize_srcu().

By that change, the time required for the above test took 95 ms which is
even better than the non ODP flow.

Once fully dropped the srcu usage, had to come with a lock to protect the
XA access.

As part of using the above mechanism we could also clean the
num_deferred_work stuff and follow the use count instead.

Link: https://lore.kernel.org/r/20210202071309.2057998-1-leon@kernel.org
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/hw/mlx5/devx.c            |  13 +-
 drivers/infiniband/hw/mlx5/main.c            |   5 -
 drivers/infiniband/hw/mlx5/mlx5_ib.h         |  31 +++-
 drivers/infiniband/hw/mlx5/mr.c              |  26 ++--
 drivers/infiniband/hw/mlx5/odp.c             | 224 ++++++++++-----------------
 drivers/net/ethernet/mellanox/mlx5/core/mr.c |   1 +
 include/linux/mlx5/driver.h                  |   2 +
 7 files changed, 127 insertions(+), 175 deletions(-)

(limited to 'include')

diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index e6d2709eb09e..e39661db0d2f 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -1310,9 +1310,9 @@ static int devx_handle_mkey_indirect(struct devx_obj *obj,
 	mkey->size = MLX5_GET64(mkc, mkc, len);
 	mkey->pd = MLX5_GET(mkc, mkc, pd);
 	devx_mr->ndescs = MLX5_GET(mkc, mkc, translations_octword_size);
+	init_waitqueue_head(&mkey->wait);
 
-	return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mkey->key), mkey,
-			       GFP_KERNEL));
+	return mlx5r_store_odp_mkey(dev, mkey);
 }
 
 static int devx_handle_mkey_create(struct mlx5_ib_dev *dev,
@@ -1385,16 +1385,15 @@ static int devx_obj_cleanup(struct ib_uobject *uobject,
 	int ret;
 
 	dev = mlx5_udata_to_mdev(&attrs->driver_udata);
-	if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) {
+	if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY &&
+	    xa_erase(&obj->ib_dev->odp_mkeys,
+		     mlx5_base_mkey(obj->devx_mr.mmkey.key)))
 		/*
 		 * The pagefault_single_data_segment() does commands against
 		 * the mmkey, we must wait for that to stop before freeing the
 		 * mkey, as another allocation could get the same mkey #.
 		 */
-		xa_erase(&obj->ib_dev->odp_mkeys,
-			 mlx5_base_mkey(obj->devx_mr.mmkey.key));
-		synchronize_srcu(&dev->odp_srcu);
-	}
+		mlx5r_deref_wait_odp_mkey(&obj->devx_mr.mmkey);
 
 	if (obj->flags & DEVX_OBJ_FLAGS_DCT)
 		ret = mlx5_core_destroy_dct(obj->ib_dev, &obj->core_dct);
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index ce4a9eba53f9..93dd568a5397 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -3869,7 +3869,6 @@ static void mlx5_ib_stage_init_cleanup(struct mlx5_ib_dev *dev)
 {
 	mlx5_ib_cleanup_multiport_master(dev);
 	WARN_ON(!xa_empty(&dev->odp_mkeys));
-	cleanup_srcu_struct(&dev->odp_srcu);
 	mutex_destroy(&dev->cap_mask_mutex);
 	WARN_ON(!xa_empty(&dev->sig_mrs));
 	WARN_ON(!bitmap_empty(dev->dm.memic_alloc_pages, MLX5_MAX_MEMIC_PAGES));
@@ -3914,10 +3913,6 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev)
 
 	dev->ib_dev.num_comp_vectors    = mlx5_comp_vectors_count(mdev);
 
-	err = init_srcu_struct(&dev->odp_srcu);
-	if (err)
-		goto err_mp;
-
 	mutex_init(&dev->cap_mask_mutex);
 	INIT_LIST_HEAD(&dev->qp_list);
 	spin_lock_init(&dev->reset_flow_resource_lock);
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index ab52083634e6..88cc26e008fc 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -684,11 +684,8 @@ struct mlx5_ib_mr {
 	u64			pi_iova;
 
 	/* For ODP and implicit */
-	atomic_t		num_deferred_work;
-	wait_queue_head_t       q_deferred_work;
 	struct xarray		implicit_children;
 	union {
-		struct rcu_head rcu;
 		struct list_head elm;
 		struct work_struct work;
 	} odp_destroy;
@@ -1068,11 +1065,6 @@ struct mlx5_ib_dev {
 	u64			odp_max_size;
 	struct mlx5_ib_pf_eq	odp_pf_eq;
 
-	/*
-	 * Sleepable RCU that prevents destruction of MRs while they are still
-	 * being used by a page fault handler.
-	 */
-	struct srcu_struct      odp_srcu;
 	struct xarray		odp_mkeys;
 
 	u32			null_mkey;
@@ -1599,6 +1591,29 @@ static inline bool mlx5_ib_can_reconfig_with_umr(struct mlx5_ib_dev *dev,
 	return true;
 }
 
+static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev,
+				       struct mlx5_core_mkey *mmkey)
+{
+	refcount_set(&mmkey->usecount, 1);
+
+	return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key),
+			       mmkey, GFP_KERNEL));
+}
+
+/* deref an mkey that can participate in ODP flow */
+static inline void mlx5r_deref_odp_mkey(struct mlx5_core_mkey *mmkey)
+{
+	if (refcount_dec_and_test(&mmkey->usecount))
+		wake_up(&mmkey->wait);
+}
+
+/* deref an mkey that can participate in ODP flow and wait for relese */
+static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_core_mkey *mmkey)
+{
+	mlx5r_deref_odp_mkey(mmkey);
+	wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0);
+}
+
 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev);
 
 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
index a63ef7c66e38..db05b0e0a8d7 100644
--- a/drivers/infiniband/hw/mlx5/mr.c
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -158,6 +158,7 @@ static void create_mkey_callback(int status, struct mlx5_async_work *context)
 	mr->mmkey.type = MLX5_MKEY_MR;
 	mr->mmkey.key |= mlx5_idx_to_mkey(
 		MLX5_GET(create_mkey_out, mr->out, mkey_index));
+	init_waitqueue_head(&mr->mmkey.wait);
 
 	WRITE_ONCE(dev->cache.last_add, jiffies);
 
@@ -1551,10 +1552,7 @@ static struct ib_mr *create_user_odp_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	odp->private = mr;
-	init_waitqueue_head(&mr->q_deferred_work);
-	atomic_set(&mr->num_deferred_work, 0);
-	err = xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mr->mmkey.key),
-			      &mr->mmkey, GFP_KERNEL));
+	err = mlx5r_store_odp_mkey(dev, &mr->mmkey);
 	if (err)
 		goto err_dereg_mr;
 
@@ -1651,10 +1649,7 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset,
 
 	atomic_add(ib_umem_num_pages(mr->umem), &dev->mdev->priv.reg_pages);
 	umem_dmabuf->private = mr;
-	init_waitqueue_head(&mr->q_deferred_work);
-	atomic_set(&mr->num_deferred_work, 0);
-	err = xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mr->mmkey.key),
-			      &mr->mmkey, GFP_KERNEL));
+	err = mlx5r_store_odp_mkey(dev, &mr->mmkey);
 	if (err)
 		goto err_dereg_mr;
 
@@ -2330,9 +2325,7 @@ int mlx5_ib_alloc_mw(struct ib_mw *ibmw, struct ib_udata *udata)
 	}
 
 	if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) {
-		err = xa_err(xa_store(&dev->odp_mkeys,
-				      mlx5_base_mkey(mw->mmkey.key), &mw->mmkey,
-				      GFP_KERNEL));
+		err = mlx5r_store_odp_mkey(dev, &mw->mmkey);
 		if (err)
 			goto free_mkey;
 	}
@@ -2352,14 +2345,13 @@ int mlx5_ib_dealloc_mw(struct ib_mw *mw)
 	struct mlx5_ib_dev *dev = to_mdev(mw->device);
 	struct mlx5_ib_mw *mmw = to_mmw(mw);
 
-	if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) {
-		xa_erase(&dev->odp_mkeys, mlx5_base_mkey(mmw->mmkey.key));
+	if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) &&
+	    xa_erase(&dev->odp_mkeys, mlx5_base_mkey(mmw->mmkey.key)))
 		/*
-		 * pagefault_single_data_segment() may be accessing mmw under
-		 * SRCU if the user bound an ODP MR to this MW.
+		 * pagefault_single_data_segment() may be accessing mmw
+		 * if the user bound an ODP MR to this MW.
 		 */
-		synchronize_srcu(&dev->odp_srcu);
-	}
+		mlx5r_deref_wait_odp_mkey(&mmw->mmkey);
 
 	return mlx5_core_destroy_mkey(dev->mdev, &mmw->mmkey);
 }
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index e77b1db73893..374698186662 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -115,7 +115,6 @@ static void populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries,
 	 * xarray would be protected by the umem_mutex, however that is not
 	 * possible. Instead this uses a weaker update-then-lock pattern:
 	 *
-	 *  srcu_read_lock()
 	 *    xa_store()
 	 *    mutex_lock(umem_mutex)
 	 *     mlx5_ib_update_xlt()
@@ -126,12 +125,9 @@ static void populate_klm(struct mlx5_klm *pklm, size_t idx, size_t nentries,
 	 * before destroying.
 	 *
 	 * The umem_mutex provides the acquire/release semantic needed to make
-	 * the xa_store() visible to a racing thread. While SRCU is not
-	 * technically required, using it gives consistent use of the SRCU
-	 * locking around the xarray.
+	 * the xa_store() visible to a racing thread.
 	 */
 	lockdep_assert_held(&to_ib_umem_odp(imr->umem)->umem_mutex);
-	lockdep_assert_held(&mr_to_mdev(imr)->odp_srcu);
 
 	for (; pklm != end; pklm++, idx++) {
 		struct mlx5_ib_mr *mtt = xa_load(&imr->implicit_children, idx);
@@ -207,8 +203,8 @@ static void dma_fence_odp_mr(struct mlx5_ib_mr *mr)
 }
 
 /*
- * This must be called after the mr has been removed from implicit_children
- * and the SRCU synchronized.  NOTE: The MR does not necessarily have to be
+ * This must be called after the mr has been removed from implicit_children.
+ * NOTE: The MR does not necessarily have to be
  * empty here, parallel page faults could have raced with the free process and
  * added pages to it.
  */
@@ -218,19 +214,15 @@ static void free_implicit_child_mr(struct mlx5_ib_mr *mr, bool need_imr_xlt)
 	struct ib_umem_odp *odp_imr = to_ib_umem_odp(imr->umem);
 	struct ib_umem_odp *odp = to_ib_umem_odp(mr->umem);
 	unsigned long idx = ib_umem_start(odp) >> MLX5_IMR_MTT_SHIFT;
-	int srcu_key;
 
-	/* implicit_child_mr's are not allowed to have deferred work */
-	WARN_ON(atomic_read(&mr->num_deferred_work));
+	mlx5r_deref_wait_odp_mkey(&mr->mmkey);
 
 	if (need_imr_xlt) {
-		srcu_key = srcu_read_lock(&mr_to_mdev(mr)->odp_srcu);
 		mutex_lock(&odp_imr->umem_mutex);
 		mlx5_ib_update_xlt(mr->parent, idx, 1, 0,
 				   MLX5_IB_UPD_XLT_INDIRECT |
 				   MLX5_IB_UPD_XLT_ATOMIC);
 		mutex_unlock(&odp_imr->umem_mutex);
-		srcu_read_unlock(&mr_to_mdev(mr)->odp_srcu, srcu_key);
 	}
 
 	dma_fence_odp_mr(mr);
@@ -238,26 +230,16 @@ static void free_implicit_child_mr(struct mlx5_ib_mr *mr, bool need_imr_xlt)
 	mr->parent = NULL;
 	mlx5_mr_cache_free(mr_to_mdev(mr), mr);
 	ib_umem_odp_release(odp);
-	if (atomic_dec_and_test(&imr->num_deferred_work))
-		wake_up(&imr->q_deferred_work);
 }
 
 static void free_implicit_child_mr_work(struct work_struct *work)
 {
 	struct mlx5_ib_mr *mr =
 		container_of(work, struct mlx5_ib_mr, odp_destroy.work);
+	struct mlx5_ib_mr *imr = mr->parent;
 
 	free_implicit_child_mr(mr, true);
-}
-
-static void free_implicit_child_mr_rcu(struct rcu_head *head)
-{
-	struct mlx5_ib_mr *mr =
-		container_of(head, struct mlx5_ib_mr, odp_destroy.rcu);
-
-	/* Freeing a MR is a sleeping operation, so bounce to a work queue */
-	INIT_WORK(&mr->odp_destroy.work, free_implicit_child_mr_work);
-	queue_work(system_unbound_wq, &mr->odp_destroy.work);
+	mlx5r_deref_odp_mkey(&imr->mmkey);
 }
 
 static void destroy_unused_implicit_child_mr(struct mlx5_ib_mr *mr)
@@ -266,21 +248,14 @@ static void destroy_unused_implicit_child_mr(struct mlx5_ib_mr *mr)
 	unsigned long idx = ib_umem_start(odp) >> MLX5_IMR_MTT_SHIFT;
 	struct mlx5_ib_mr *imr = mr->parent;
 
-	xa_lock(&imr->implicit_children);
-	/*
-	 * This can race with mlx5_ib_free_implicit_mr(), the first one to
-	 * reach the xa lock wins the race and destroys the MR.
-	 */
-	if (__xa_cmpxchg(&imr->implicit_children, idx, mr, NULL, GFP_ATOMIC) !=
-	    mr)
-		goto out_unlock;
+	if (!refcount_inc_not_zero(&imr->mmkey.usecount))
+		return;
 
-	atomic_inc(&imr->num_deferred_work);
-	call_srcu(&mr_to_mdev(mr)->odp_srcu, &mr->odp_destroy.rcu,
-		  free_implicit_child_mr_rcu);
+	xa_erase(&imr->implicit_children, idx);
 
-out_unlock:
-	xa_unlock(&imr->implicit_children);
+	/* Freeing a MR is a sleeping operation, so bounce to a work queue */
+	INIT_WORK(&mr->odp_destroy.work, free_implicit_child_mr_work);
+	queue_work(system_unbound_wq, &mr->odp_destroy.work);
 }
 
 static bool mlx5_ib_invalidate_range(struct mmu_interval_notifier *mni,
@@ -492,6 +467,12 @@ static struct mlx5_ib_mr *implicit_get_child_mr(struct mlx5_ib_mr *imr,
 	mr->parent = imr;
 	odp->private = mr;
 
+	/*
+	 * First refcount is owned by the xarray and second refconut
+	 * is returned to the caller.
+	 */
+	refcount_set(&mr->mmkey.usecount, 2);
+
 	err = mlx5_ib_update_xlt(mr, 0,
 				 MLX5_IMR_MTT_ENTRIES,
 				 PAGE_SHIFT,
@@ -502,27 +483,28 @@ static struct mlx5_ib_mr *implicit_get_child_mr(struct mlx5_ib_mr *imr,
 		goto out_mr;
 	}
 
-	/*
-	 * Once the store to either xarray completes any error unwind has to
-	 * use synchronize_srcu(). Avoid this with xa_reserve()
-	 */
-	ret = xa_cmpxchg(&imr->implicit_children, idx, NULL, mr,
-			 GFP_KERNEL);
+	xa_lock(&imr->implicit_children);
+	ret = __xa_cmpxchg(&imr->implicit_children, idx, NULL, mr,
+			   GFP_KERNEL);
 	if (unlikely(ret)) {
 		if (xa_is_err(ret)) {
 			ret = ERR_PTR(xa_err(ret));
-			goto out_mr;
+			goto out_lock;
 		}
 		/*
 		 * Another thread beat us to creating the child mr, use
 		 * theirs.
 		 */
-		goto out_mr;
+		refcount_inc(&ret->mmkey.usecount);
+		goto out_lock;
 	}
+	xa_unlock(&imr->implicit_children);
 
 	mlx5_ib_dbg(mr_to_mdev(imr), "key %x mr %p\n", mr->mmkey.key, mr);
 	return mr;
 
+out_lock:
+	xa_unlock(&imr->implicit_children);
 out_mr:
 	mlx5_mr_cache_free(mr_to_mdev(imr), mr);
 out_umem:
@@ -561,8 +543,6 @@ struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
 	imr->ibmr.device = &dev->ib_dev;
 	imr->umem = &umem_odp->umem;
 	imr->is_odp_implicit = true;
-	atomic_set(&imr->num_deferred_work, 0);
-	init_waitqueue_head(&imr->q_deferred_work);
 	xa_init(&imr->implicit_children);
 
 	err = mlx5_ib_update_xlt(imr, 0,
@@ -574,8 +554,7 @@ struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
 	if (err)
 		goto out_mr;
 
-	err = xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(imr->mmkey.key),
-			      &imr->mmkey, GFP_KERNEL));
+	err = mlx5r_store_odp_mkey(dev, &imr->mmkey);
 	if (err)
 		goto out_mr;
 
@@ -593,51 +572,24 @@ void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *imr)
 {
 	struct ib_umem_odp *odp_imr = to_ib_umem_odp(imr->umem);
 	struct mlx5_ib_dev *dev = mr_to_mdev(imr);
-	struct list_head destroy_list;
 	struct mlx5_ib_mr *mtt;
-	struct mlx5_ib_mr *tmp;
 	unsigned long idx;
 
-	INIT_LIST_HEAD(&destroy_list);
-
 	xa_erase(&dev->odp_mkeys, mlx5_base_mkey(imr->mmkey.key));
-	/*
-	 * This stops the SRCU protected page fault path from touching either
-	 * the imr or any children. The page fault path can only reach the
-	 * children xarray via the imr.
-	 */
-	synchronize_srcu(&dev->odp_srcu);
-
 	/*
 	 * All work on the prefetch list must be completed, xa_erase() prevented
 	 * new work from being created.
 	 */
-	wait_event(imr->q_deferred_work, !atomic_read(&imr->num_deferred_work));
-
+	mlx5r_deref_wait_odp_mkey(&imr->mmkey);
 	/*
 	 * At this point it is forbidden for any other thread to enter
 	 * pagefault_mr() on this imr. It is already forbidden to call
 	 * pagefault_mr() on an implicit child. Due to this additions to
 	 * implicit_children are prevented.
+	 * In addition, any new call to destroy_unused_implicit_child_mr()
+	 * may return immediately.
 	 */
 
-	/*
-	 * Block destroy_unused_implicit_child_mr() from incrementing
-	 * num_deferred_work.
-	 */
-	xa_lock(&imr->implicit_children);
-	xa_for_each (&imr->implicit_children, idx, mtt) {
-		__xa_erase(&imr->implicit_children, idx);
-		list_add(&mtt->odp_destroy.elm, &destroy_list);
-	}
-	xa_unlock(&imr->implicit_children);
-
-	/*
-	 * Wait for any concurrent destroy_unused_implicit_child_mr() to
-	 * complete.
-	 */
-	wait_event(imr->q_deferred_work, !atomic_read(&imr->num_deferred_work));
-
 	/*
 	 * Fence the imr before we destroy the children. This allows us to
 	 * skip updating the XLT of the imr during destroy of the child mkey
@@ -645,8 +597,10 @@ void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *imr)
 	 */
 	mlx5_mr_cache_invalidate(imr);
 
-	list_for_each_entry_safe (mtt, tmp, &destroy_list, odp_destroy.elm)
+	xa_for_each(&imr->implicit_children, idx, mtt) {
+		xa_erase(&imr->implicit_children, idx);
 		free_implicit_child_mr(mtt, false);
+	}
 
 	mlx5_mr_cache_free(dev, imr);
 	ib_umem_odp_release(odp_imr);
@@ -665,9 +619,7 @@ void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr)
 	xa_erase(&mr_to_mdev(mr)->odp_mkeys, mlx5_base_mkey(mr->mmkey.key));
 
 	/* Wait for all running page-fault handlers to finish. */
-	synchronize_srcu(&mr_to_mdev(mr)->odp_srcu);
-
-	wait_event(mr->q_deferred_work, !atomic_read(&mr->num_deferred_work));
+	mlx5r_deref_wait_odp_mkey(&mr->mmkey);
 
 	dma_fence_odp_mr(mr);
 }
@@ -686,10 +638,7 @@ void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr)
 	/* Prevent new page faults and prefetch requests from succeeding */
 	xa_erase(&mr_to_mdev(mr)->odp_mkeys, mlx5_base_mkey(mr->mmkey.key));
 
-	/* Wait for all running page-fault handlers to finish. */
-	synchronize_srcu(&mr_to_mdev(mr)->odp_srcu);
-
-	wait_event(mr->q_deferred_work, !atomic_read(&mr->num_deferred_work));
+	mlx5r_deref_wait_odp_mkey(&mr->mmkey);
 
 	dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
 	mlx5_mr_cache_invalidate(mr);
@@ -780,8 +729,10 @@ static int pagefault_implicit_mr(struct mlx5_ib_mr *imr,
 		struct mlx5_ib_mr *mtt;
 		u64 len;
 
+		xa_lock(&imr->implicit_children);
 		mtt = xa_load(&imr->implicit_children, idx);
 		if (unlikely(!mtt)) {
+			xa_unlock(&imr->implicit_children);
 			mtt = implicit_get_child_mr(imr, idx);
 			if (IS_ERR(mtt)) {
 				ret = PTR_ERR(mtt);
@@ -789,6 +740,9 @@ static int pagefault_implicit_mr(struct mlx5_ib_mr *imr,
 			}
 			upd_start_idx = min(upd_start_idx, idx);
 			upd_len = idx - upd_start_idx + 1;
+		} else {
+			refcount_inc(&mtt->mmkey.usecount);
+			xa_unlock(&imr->implicit_children);
 		}
 
 		umem_odp = to_ib_umem_odp(mtt->umem);
@@ -797,6 +751,9 @@ static int pagefault_implicit_mr(struct mlx5_ib_mr *imr,
 
 		ret = pagefault_real_mr(mtt, umem_odp, user_va, len,
 					bytes_mapped, flags);
+
+		mlx5r_deref_odp_mkey(&mtt->mmkey);
+
 		if (ret < 0)
 			goto out;
 		user_va += len;
@@ -888,7 +845,6 @@ static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt,
 {
 	struct ib_umem_odp *odp = to_ib_umem_odp(mr->umem);
 
-	lockdep_assert_held(&mr_to_mdev(mr)->odp_srcu);
 	if (unlikely(io_virt < mr->mmkey.iova))
 		return -EFAULT;
 
@@ -980,7 +936,7 @@ static int pagefault_single_data_segment(struct mlx5_ib_dev *dev,
 					 u32 *bytes_committed,
 					 u32 *bytes_mapped)
 {
-	int npages = 0, srcu_key, ret, i, outlen, cur_outlen = 0, depth = 0;
+	int npages = 0, ret, i, outlen, cur_outlen = 0, depth = 0;
 	struct pf_frame *head = NULL, *frame;
 	struct mlx5_core_mkey *mmkey;
 	struct mlx5_ib_mr *mr;
@@ -989,14 +945,14 @@ static int pagefault_single_data_segment(struct mlx5_ib_dev *dev,
 	size_t offset;
 	int ndescs;
 
-	srcu_key = srcu_read_lock(&dev->odp_srcu);
-
 	io_virt += *bytes_committed;
 	bcnt -= *bytes_committed;
 
 next_mr:
+	xa_lock(&dev->odp_mkeys);
 	mmkey = xa_load(&dev->odp_mkeys, mlx5_base_mkey(key));
 	if (!mmkey) {
+		xa_unlock(&dev->odp_mkeys);
 		mlx5_ib_dbg(
 			dev,
 			"skipping non ODP MR (lkey=0x%06x) in page fault handler.\n",
@@ -1009,12 +965,15 @@ next_mr:
 		 * faulted.
 		 */
 		ret = 0;
-		goto srcu_unlock;
+		goto end;
 	}
+	refcount_inc(&mmkey->usecount);
+	xa_unlock(&dev->odp_mkeys);
+
 	if (!mkey_is_eq(mmkey, key)) {
 		mlx5_ib_dbg(dev, "failed to find mkey %x\n", key);
 		ret = -EFAULT;
-		goto srcu_unlock;
+		goto end;
 	}
 
 	switch (mmkey->type) {
@@ -1023,7 +982,7 @@ next_mr:
 
 		ret = pagefault_mr(mr, io_virt, bcnt, bytes_mapped, 0);
 		if (ret < 0)
-			goto srcu_unlock;
+			goto end;
 
 		mlx5_update_odp_stats(mr, faults, ret);
 
@@ -1038,7 +997,7 @@ next_mr:
 		if (depth >= MLX5_CAP_GEN(dev->mdev, max_indirection)) {
 			mlx5_ib_dbg(dev, "indirection level exceeded\n");
 			ret = -EFAULT;
-			goto srcu_unlock;
+			goto end;
 		}
 
 		outlen = MLX5_ST_SZ_BYTES(query_mkey_out) +
@@ -1049,7 +1008,7 @@ next_mr:
 			out = kzalloc(outlen, GFP_KERNEL);
 			if (!out) {
 				ret = -ENOMEM;
-				goto srcu_unlock;
+				goto end;
 			}
 			cur_outlen = outlen;
 		}
@@ -1059,7 +1018,7 @@ next_mr:
 
 		ret = mlx5_core_query_mkey(dev->mdev, mmkey, out, outlen);
 		if (ret)
-			goto srcu_unlock;
+			goto end;
 
 		offset = io_virt - MLX5_GET64(query_mkey_out, out,
 					      memory_key_mkey_entry.start_addr);
@@ -1073,7 +1032,7 @@ next_mr:
 			frame = kzalloc(sizeof(*frame), GFP_KERNEL);
 			if (!frame) {
 				ret = -ENOMEM;
-				goto srcu_unlock;
+				goto end;
 			}
 
 			frame->key = be32_to_cpu(pklm->key);
@@ -1092,7 +1051,7 @@ next_mr:
 	default:
 		mlx5_ib_dbg(dev, "wrong mkey type %d\n", mmkey->type);
 		ret = -EFAULT;
-		goto srcu_unlock;
+		goto end;
 	}
 
 	if (head) {
@@ -1105,10 +1064,13 @@ next_mr:
 		depth = frame->depth;
 		kfree(frame);
 
+		mlx5r_deref_odp_mkey(mmkey);
 		goto next_mr;
 	}
 
-srcu_unlock:
+end:
+	if (mmkey)
+		mlx5r_deref_odp_mkey(mmkey);
 	while (head) {
 		frame = head;
 		head = frame->next;
@@ -1116,7 +1078,6 @@ srcu_unlock:
 	}
 	kfree(out);
 
-	srcu_read_unlock(&dev->odp_srcu, srcu_key);
 	*bytes_committed = 0;
 	return ret ? ret : npages;
 }
@@ -1824,8 +1785,8 @@ static void destroy_prefetch_work(struct prefetch_mr_work *work)
 	u32 i;
 
 	for (i = 0; i < work->num_sge; ++i)
-		if (atomic_dec_and_test(&work->frags[i].mr->num_deferred_work))
-			wake_up(&work->frags[i].mr->q_deferred_work);
+		mlx5r_deref_odp_mkey(&work->frags[i].mr->mmkey);
+
 	kvfree(work);
 }
 
@@ -1835,24 +1796,30 @@ get_prefetchable_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 	struct mlx5_core_mkey *mmkey;
-	struct mlx5_ib_mr *mr;
-
-	lockdep_assert_held(&dev->odp_srcu);
+	struct mlx5_ib_mr *mr = NULL;
 
+	xa_lock(&dev->odp_mkeys);
 	mmkey = xa_load(&dev->odp_mkeys, mlx5_base_mkey(lkey));
 	if (!mmkey || mmkey->key != lkey || mmkey->type != MLX5_MKEY_MR)
-		return NULL;
+		goto end;
 
 	mr = container_of(mmkey, struct mlx5_ib_mr, mmkey);
 
-	if (mr->ibmr.pd != pd)
-		return NULL;
+	if (mr->ibmr.pd != pd) {
+		mr = NULL;
+		goto end;
+	}
 
 	/* prefetch with write-access must be supported by the MR */
 	if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE &&
-	    !mr->umem->writable)
-		return NULL;
+	    !mr->umem->writable) {
+		mr = NULL;
+		goto end;
+	}
 
+	refcount_inc(&mmkey->usecount);
+end:
+	xa_unlock(&dev->odp_mkeys);
 	return mr;
 }
 
@@ -1860,17 +1827,12 @@ static void mlx5_ib_prefetch_mr_work(struct work_struct *w)
 {
 	struct prefetch_mr_work *work =
 		container_of(w, struct prefetch_mr_work, work);
-	struct mlx5_ib_dev *dev;
 	u32 bytes_mapped = 0;
-	int srcu_key;
 	int ret;
 	u32 i;
 
 	/* We rely on IB/core that work is executed if we have num_sge != 0 only. */
 	WARN_ON(!work->num_sge);
-	dev = mr_to_mdev(work->frags[0].mr);
-	/* SRCU should be held when calling to mlx5_odp_populate_xlt() */
-	srcu_key = srcu_read_lock(&dev->odp_srcu);
 	for (i = 0; i < work->num_sge; ++i) {
 		ret = pagefault_mr(work->frags[i].mr, work->frags[i].io_virt,
 				   work->frags[i].length, &bytes_mapped,
@@ -1879,7 +1841,6 @@ static void mlx5_ib_prefetch_mr_work(struct work_struct *w)
 			continue;
 		mlx5_update_odp_stats(work->frags[i].mr, prefetch, ret);
 	}
-	srcu_read_unlock(&dev->odp_srcu, srcu_key);
 
 	destroy_prefetch_work(work);
 }
@@ -1903,9 +1864,6 @@ static bool init_prefetch_work(struct ib_pd *pd,
 			work->num_sge = i;
 			return false;
 		}
-
-		/* Keep the MR pointer will valid outside the SRCU */
-		atomic_inc(&work->frags[i].mr->num_deferred_work);
 	}
 	work->num_sge = num_sge;
 	return true;
@@ -1916,42 +1874,35 @@ static int mlx5_ib_prefetch_sg_list(struct ib_pd *pd,
 				    u32 pf_flags, struct ib_sge *sg_list,
 				    u32 num_sge)
 {
-	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 	u32 bytes_mapped = 0;
-	int srcu_key;
 	int ret = 0;
 	u32 i;
 
-	srcu_key = srcu_read_lock(&dev->odp_srcu);
 	for (i = 0; i < num_sge; ++i) {
 		struct mlx5_ib_mr *mr;
 
 		mr = get_prefetchable_mr(pd, advice, sg_list[i].lkey);
-		if (!mr) {
-			ret = -ENOENT;
-			goto out;
-		}
+		if (!mr)
+			return -ENOENT;
 		ret = pagefault_mr(mr, sg_list[i].addr, sg_list[i].length,
 				   &bytes_mapped, pf_flags);
-		if (ret < 0)
-			goto out;
+		if (ret < 0) {
+			mlx5r_deref_odp_mkey(&mr->mmkey);
+			return ret;
+		}
 		mlx5_update_odp_stats(mr, prefetch, ret);
+		mlx5r_deref_odp_mkey(&mr->mmkey);
 	}
-	ret = 0;
 
-out:
-	srcu_read_unlock(&dev->odp_srcu, srcu_key);
-	return ret;
+	return 0;
 }
 
 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
 			       enum ib_uverbs_advise_mr_advice advice,
 			       u32 flags, struct ib_sge *sg_list, u32 num_sge)
 {
-	struct mlx5_ib_dev *dev = to_mdev(pd->device);
 	u32 pf_flags = 0;
 	struct prefetch_mr_work *work;
-	int srcu_key;
 
 	if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH)
 		pf_flags |= MLX5_PF_FLAGS_DOWNGRADE;
@@ -1967,13 +1918,10 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
 	if (!work)
 		return -ENOMEM;
 
-	srcu_key = srcu_read_lock(&dev->odp_srcu);
 	if (!init_prefetch_work(pd, advice, pf_flags, work, sg_list, num_sge)) {
-		srcu_read_unlock(&dev->odp_srcu, srcu_key);
 		destroy_prefetch_work(work);
 		return -EINVAL;
 	}
 	queue_work(system_unbound_wq, &work->work);
-	srcu_read_unlock(&dev->odp_srcu, srcu_key);
 	return 0;
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/mr.c b/drivers/net/ethernet/mellanox/mlx5/core/mr.c
index 9eb51f06d3ae..50af84e76fb6 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/mr.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/mr.c
@@ -56,6 +56,7 @@ int mlx5_core_create_mkey(struct mlx5_core_dev *dev,
 	mkey->size = MLX5_GET64(mkc, mkc, len);
 	mkey->key |= mlx5_idx_to_mkey(mkey_index);
 	mkey->pd = MLX5_GET(mkc, mkc, pd);
+	init_waitqueue_head(&mkey->wait);
 
 	mlx5_core_dbg(dev, "out 0x%x, mkey 0x%x\n", mkey_index, mkey->key);
 	return 0;
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 11558c2e99f0..b0a59a18a708 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -366,6 +366,8 @@ struct mlx5_core_mkey {
 	u32			key;
 	u32			pd;
 	u32			type;
+	struct wait_queue_head wait;
+	refcount_t usecount;
 };
 
 #define MLX5_24BIT_MASK		((1 << 24) - 1)
-- 
cgit v1.2.3


From 633d61021298f690f823ff51bcdab906e3644fe1 Mon Sep 17 00:00:00 2001
From: Christoph Lameter <cl@linux.com>
Date: Thu, 28 Jan 2021 18:46:47 +0000
Subject: RDMA/ipoib: Remove racy Subnet Manager sendonly join checks

When a system receives a REREG event from the SM, then the SM information
in the kernel is marked as invalid and a request is sent to the SM to
update the information. The SM information is invalid in that time period.

However, receiving a REREG also occurs simultaneously in user space
applications that are now trying to rejoin the multicast groups. Some of
those may be sendonly multicast groups which are then failing.

If the SM information is invalid then ib_sa_sendonly_fullmem_support()
returns false. That is wrong because it just means that we do not know yet
if the potentially new SM supports sendonly joins.

Sendonly join was introduced in 2015 and all the Subnet managers have
supported it ever since. So there is no point in checking if a subnet
manager supports it.

Should an old opensm get a request for a sendonly join then the request
will fail. The code that is removed here accomodated that situation and
fell back to a full join.

Falling back to a full join is problematic in itself. The reason to use
the sendonly join was to reduce the traffic on the Infiniband fabric
otherwise one could have just stayed with the regular join.  So this patch
may cause users of very old opensms to discover that lots of traffic
needlessly crosses their IB fabrics.

Link: https://lore.kernel.org/r/alpine.DEB.2.22.394.2101281845160.13303@www.lameter.com
Signed-off-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/core/cma.c                  | 11 -----------
 drivers/infiniband/core/sa_query.c             | 24 ------------------------
 drivers/infiniband/ulp/ipoib/ipoib.h           |  1 -
 drivers/infiniband/ulp/ipoib/ipoib_main.c      |  2 --
 drivers/infiniband/ulp/ipoib/ipoib_multicast.c | 13 +------------
 include/rdma/ib_sa.h                           |  4 ----
 6 files changed, 1 insertion(+), 54 deletions(-)

(limited to 'include')

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index c51b84b2d2f3..58ee7004c8d8 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -4542,17 +4542,6 @@ static int cma_join_ib_multicast(struct rdma_id_private *id_priv,
 	rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr));
 	rec.join_state = mc->join_state;
 
-	if ((rec.join_state == BIT(SENDONLY_FULLMEMBER_JOIN)) &&
-	    (!ib_sa_sendonly_fullmem_support(&sa_client,
-					     id_priv->id.device,
-					     id_priv->id.port_num))) {
-		dev_warn(
-			&id_priv->id.device->dev,
-			"RDMA CM: port %u Unable to multicast join: SM doesn't support Send Only Full Member option\n",
-			id_priv->id.port_num);
-		return -EOPNOTSUPP;
-	}
-
 	comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID |
 		    IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE |
 		    IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL |
diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
index 5bd047042e68..9ef1a355131b 100644
--- a/drivers/infiniband/core/sa_query.c
+++ b/drivers/infiniband/core/sa_query.c
@@ -1951,30 +1951,6 @@ err1:
 }
 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
 
-bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
-				    struct ib_device *device,
-				    u8 port_num)
-{
-	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
-	struct ib_sa_port *port;
-	bool ret = false;
-	unsigned long flags;
-
-	if (!sa_dev)
-		return ret;
-
-	port  = &sa_dev->port[port_num - sa_dev->start_port];
-
-	spin_lock_irqsave(&port->classport_lock, flags);
-	if ((port->classport_info.valid) &&
-	    (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
-		ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
-			& IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
-	spin_unlock_irqrestore(&port->classport_lock, flags);
-	return ret;
-}
-EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
-
 struct ib_classport_info_context {
 	struct completion	done;
 	struct ib_sa_query	*sa_query;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 3440dc48d02c..179ff1d068e5 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -413,7 +413,6 @@ struct ipoib_dev_priv {
 	u64	hca_caps;
 	struct ipoib_ethtool_st ethtool;
 	unsigned int max_send_sge;
-	bool sm_fullmember_sendonly_support;
 	const struct net_device_ops	*rn_ops;
 };
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index a6f413491321..e16b40c09f82 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -141,8 +141,6 @@ int ipoib_open(struct net_device *dev)
 
 	set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
 
-	priv->sm_fullmember_sendonly_support = false;
-
 	if (ipoib_ib_dev_open(dev)) {
 		if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
 			return 0;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
index e3e4447c0f51..5b3154503bf4 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
@@ -333,15 +333,6 @@ void ipoib_mcast_carrier_on_task(struct work_struct *work)
 		ipoib_dbg(priv, "Keeping carrier off until IB port is active\n");
 		return;
 	}
-	/*
-	 * Check if can send sendonly MCG's with sendonly-fullmember join state.
-	 * It done here after the successfully join to the broadcast group,
-	 * because the broadcast group must always be joined first and is always
-	 * re-joined if the SM changes substantially.
-	 */
-	priv->sm_fullmember_sendonly_support =
-		ib_sa_sendonly_fullmem_support(&ipoib_sa_client,
-					       priv->ca, priv->port);
 	/*
 	 * Take rtnl_lock to avoid racing with ipoib_stop() and
 	 * turning the carrier back on while a device is being
@@ -537,9 +528,7 @@ static int ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast)
 		 * most closely emulates the behavior, from a user space
 		 * application perspective, of Ethernet multicast operation.
 		 */
-		if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) &&
-		    priv->sm_fullmember_sendonly_support)
-			/* SM supports sendonly-fullmember, otherwise fallback to full-member */
+		if (test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
 			rec.join_state = SENDONLY_FULLMEMBER_JOIN;
 	}
 	spin_unlock_irq(&priv->lock);
diff --git a/include/rdma/ib_sa.h b/include/rdma/ib_sa.h
index 693285e76f13..4c52c2fd22a1 100644
--- a/include/rdma/ib_sa.h
+++ b/include/rdma/ib_sa.h
@@ -547,10 +547,6 @@ int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
 					       void *context),
 			      void *context, struct ib_sa_query **sa_query);
 
-bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
-				    struct ib_device *device,
-				    u8 port_num);
-
 static inline bool sa_path_is_roce(struct sa_path_rec *rec)
 {
 	return ((rec->rec_type == SA_PATH_REC_TYPE_ROCE_V1) ||
-- 
cgit v1.2.3


From 7232c132d13aafd178ba18c1099b2cb98d104b8c Mon Sep 17 00:00:00 2001
From: Tal Gilboa <talgi@nvidia.com>
Date: Thu, 11 Feb 2021 10:55:49 +0200
Subject: RDMA/mlx5: Allow CQ creation without attached EQs

The traditional DevX CQ creation flow goes through mlx5_core_create_cq()
which checks that the given EQN corresponds to an existing EQ and attaches
a devx handler to the EQN for the CQ.

In some cases the EQ will not be a kernel EQ, but will be controlled by
modify CQ, don't block creating these just because the EQN can't be found
in the kernel.

Link: https://lore.kernel.org/r/20210211085549.1277674-1-leon@kernel.org
Signed-off-by: Tal Gilboa <talgi@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/infiniband/hw/mlx5/devx.c | 13 ++++++++++++-
 include/linux/mlx5/mlx5_ifc.h     |  5 +++--
 2 files changed, 15 insertions(+), 3 deletions(-)

(limited to 'include')

diff --git a/drivers/infiniband/hw/mlx5/devx.c b/drivers/infiniband/hw/mlx5/devx.c
index e39661db0d2f..ebc2a4355fa5 100644
--- a/drivers/infiniband/hw/mlx5/devx.c
+++ b/drivers/infiniband/hw/mlx5/devx.c
@@ -1439,6 +1439,16 @@ out:
 	rcu_read_unlock();
 }
 
+static bool is_apu_thread_cq(struct mlx5_ib_dev *dev, const void *in)
+{
+	if (!MLX5_CAP_GEN(dev->mdev, apu) ||
+	    !MLX5_GET(cqc, MLX5_ADDR_OF(create_cq_in, in, cq_context),
+		      apu_thread_cq))
+		return false;
+
+	return true;
+}
+
 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
 	struct uverbs_attr_bundle *attrs)
 {
@@ -1492,7 +1502,8 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
 		obj->flags |= DEVX_OBJ_FLAGS_DCT;
 		err = mlx5_core_create_dct(dev, &obj->core_dct, cmd_in,
 					   cmd_in_len, cmd_out, cmd_out_len);
-	} else if (opcode == MLX5_CMD_OP_CREATE_CQ) {
+	} else if (opcode == MLX5_CMD_OP_CREATE_CQ &&
+		   !is_apu_thread_cq(dev, cmd_in)) {
 		obj->flags |= DEVX_OBJ_FLAGS_CQ;
 		obj->core_cq.comp = devx_cq_comp;
 		err = mlx5_core_create_cq(dev->mdev, &obj->core_cq,
diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h
index cf692fc17f41..768e097c633b 100644
--- a/include/linux/mlx5/mlx5_ifc.h
+++ b/include/linux/mlx5/mlx5_ifc.h
@@ -1634,7 +1634,8 @@ struct mlx5_ifc_cmd_hca_cap_bits {
 	u8         sf_set_partition[0x1];
 	u8         reserved_at_682[0x1];
 	u8         log_max_sf[0x5];
-	u8         reserved_at_688[0x8];
+	u8         apu[0x1];
+	u8         reserved_at_689[0x7];
 	u8         log_min_sf_size[0x8];
 	u8         max_num_sf_partitions[0x8];
 
@@ -3816,7 +3817,7 @@ struct mlx5_ifc_cqc_bits {
 	u8         status[0x4];
 	u8         reserved_at_4[0x2];
 	u8         dbr_umem_valid[0x1];
-	u8         reserved_at_7[0x1];
+	u8         apu_thread_cq[0x1];
 	u8         cqe_sz[0x3];
 	u8         cc[0x1];
 	u8         reserved_at_c[0x1];
-- 
cgit v1.2.3