Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

Pull rdma updates from Jason Gunthorpe:
 "Quite a small cycle this time, even with the rc8. I suppose everyone
  went to sleep over xmas.

   - Minor driver updates for hfi1, cxgb4, erdma, hns, irdma, mlx5, siw,
     mana

   - inline CQE support for hns

   - Have mlx5 display device error codes

   - Pinned DMABUF support for irdma

   - Continued rxe cleanups, particularly converting the MRs to use
     xarray

   - Improvements to what can be cached in the mlx5 mkey cache"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (61 commits)
  IB/mlx5: Extend debug control for CC parameters
  IB/hfi1: Fix sdma.h tx->num_descs off-by-one errors
  IB/hfi1: Fix math bugs in hfi1_can_pin_pages()
  RDMA/irdma: Add support for dmabuf pin memory regions
  RDMA/mlx5: Use query_special_contexts for mkeys
  net/mlx5e: Use query_special_contexts for mkeys
  net/mlx5: Change define name for 0x100 lkey value
  net/mlx5: Expose bits for querying special mkeys
  RDMA/rxe: Fix missing memory barriers in rxe_queue.h
  RDMA/mana_ib: Fix a bug when the PF indicates more entries for registering memory on first packet
  RDMA/rxe: Remove rxe_alloc()
  RDMA/cma: Distinguish between sockaddr_in and sockaddr_in6 by size
  Subject: RDMA/rxe: Handle zero length rdma
  iw_cxgb4: Fix potential NULL dereference in c4iw_fill_res_cm_id_entry()
  RDMA/mlx5: Use rdma_umem_for_each_dma_block()
  RDMA/umem: Remove unused 'work' member from struct ib_umem
  RDMA/irdma: Cap MSIX used to online CPUs + 1
  RDMA/mlx5: Check reg_create() create for errors
  RDMA/restrack: Correct spelling
  RDMA/cxgb4: Fix potential null-ptr-deref in pass_establish()
  ...

10 files changed, 600 insertions, 585 deletions

diff --git a/drivers/infiniband/sw/rxe/rxe.h b/drivers/infiniband/sw/rxe/rxe.h
index ab334900fcc3..2415f3704f57 100644
--- a/drivers/infiniband/sw/rxe/rxe.h
+++ b/drivers/infiniband/sw/rxe/rxe.h
@@ -57,6 +57,44 @@
 #define rxe_dbg_mw(mw, fmt, ...) ibdev_dbg((mw)->ibmw.device,		\
 		"mw#%d %s:  " fmt, (mw)->elem.index, __func__, ##__VA_ARGS__)
 
+/* responder states */
+enum resp_states {
+	RESPST_NONE,
+	RESPST_GET_REQ,
+	RESPST_CHK_PSN,
+	RESPST_CHK_OP_SEQ,
+	RESPST_CHK_OP_VALID,
+	RESPST_CHK_RESOURCE,
+	RESPST_CHK_LENGTH,
+	RESPST_CHK_RKEY,
+	RESPST_EXECUTE,
+	RESPST_READ_REPLY,
+	RESPST_ATOMIC_REPLY,
+	RESPST_ATOMIC_WRITE_REPLY,
+	RESPST_PROCESS_FLUSH,
+	RESPST_COMPLETE,
+	RESPST_ACKNOWLEDGE,
+	RESPST_CLEANUP,
+	RESPST_DUPLICATE_REQUEST,
+	RESPST_ERR_MALFORMED_WQE,
+	RESPST_ERR_UNSUPPORTED_OPCODE,
+	RESPST_ERR_MISALIGNED_ATOMIC,
+	RESPST_ERR_PSN_OUT_OF_SEQ,
+	RESPST_ERR_MISSING_OPCODE_FIRST,
+	RESPST_ERR_MISSING_OPCODE_LAST_C,
+	RESPST_ERR_MISSING_OPCODE_LAST_D1E,
+	RESPST_ERR_TOO_MANY_RDMA_ATM_REQ,
+	RESPST_ERR_RNR,
+	RESPST_ERR_RKEY_VIOLATION,
+	RESPST_ERR_INVALIDATE_RKEY,
+	RESPST_ERR_LENGTH,
+	RESPST_ERR_CQ_OVERFLOW,
+	RESPST_ERROR,
+	RESPST_RESET,
+	RESPST_DONE,
+	RESPST_EXIT,
+};
+
 void rxe_set_mtu(struct rxe_dev *rxe, unsigned int dev_mtu);
 
 int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name);
diff --git a/drivers/infiniband/sw/rxe/rxe_loc.h b/drivers/infiniband/sw/rxe/rxe_loc.h
index 948ce4902b10..1bb0cb479eb1 100644
--- a/drivers/infiniband/sw/rxe/rxe_loc.h
+++ b/drivers/infiniband/sw/rxe/rxe_loc.h
@@ -64,12 +64,16 @@ void rxe_mr_init_dma(int access, struct rxe_mr *mr);
 int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
 		     int access, struct rxe_mr *mr);
 int rxe_mr_init_fast(int max_pages, struct rxe_mr *mr);
-int rxe_flush_pmem_iova(struct rxe_mr *mr, u64 iova, int length);
-int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr, int length,
-		enum rxe_mr_copy_dir dir);
+int rxe_flush_pmem_iova(struct rxe_mr *mr, u64 iova, unsigned int length);
+int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr,
+		unsigned int length, enum rxe_mr_copy_dir dir);
 int copy_data(struct rxe_pd *pd, int access, struct rxe_dma_info *dma,
 	      void *addr, int length, enum rxe_mr_copy_dir dir);
-void *iova_to_vaddr(struct rxe_mr *mr, u64 iova, int length);
+int rxe_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
+		  int sg_nents, unsigned int *sg_offset);
+int rxe_mr_do_atomic_op(struct rxe_mr *mr, u64 iova, int opcode,
+			u64 compare, u64 swap_add, u64 *orig_val);
+int rxe_mr_do_atomic_write(struct rxe_mr *mr, u64 iova, u64 value);
 struct rxe_mr *lookup_mr(struct rxe_pd *pd, int access, u32 key,
 			 enum rxe_mr_lookup_type type);
 int mr_check_range(struct rxe_mr *mr, u64 iova, size_t length);
diff --git a/drivers/infiniband/sw/rxe/rxe_mr.c b/drivers/infiniband/sw/rxe/rxe_mr.c
index 072eac4b65d2..b10aa1580a64 100644
--- a/drivers/infiniband/sw/rxe/rxe_mr.c
+++ b/drivers/infiniband/sw/rxe/rxe_mr.c
@@ -26,22 +26,22 @@ u8 rxe_get_next_key(u32 last_key)
 
 int mr_check_range(struct rxe_mr *mr, u64 iova, size_t length)
 {
-
-
 	switch (mr->ibmr.type) {
 	case IB_MR_TYPE_DMA:
 		return 0;
 
 	case IB_MR_TYPE_USER:
 	case IB_MR_TYPE_MEM_REG:
-		if (iova < mr->ibmr.iova || length > mr->ibmr.length ||
-		    iova > mr->ibmr.iova + mr->ibmr.length - length)
-			return -EFAULT;
+		if (iova < mr->ibmr.iova ||
+		    iova + length > mr->ibmr.iova + mr->ibmr.length) {
+			rxe_dbg_mr(mr, "iova/length out of range");
+			return -EINVAL;
+		}
 		return 0;
 
 	default:
-		rxe_dbg_mr(mr, "type (%d) not supported\n", mr->ibmr.type);
-		return -EFAULT;
+		rxe_dbg_mr(mr, "mr type not supported\n");
+		return -EINVAL;
 	}
 }
 
@@ -62,57 +62,31 @@ static void rxe_mr_init(int access, struct rxe_mr *mr)
 	mr->lkey = mr->ibmr.lkey = lkey;
 	mr->rkey = mr->ibmr.rkey = rkey;
 
+	mr->access = access;
+	mr->ibmr.page_size = PAGE_SIZE;
+	mr->page_mask = PAGE_MASK;
+	mr->page_shift = PAGE_SHIFT;
 	mr->state = RXE_MR_STATE_INVALID;
 }
 
-static int rxe_mr_alloc(struct rxe_mr *mr, int num_buf)
-{
-	int i;
-	int num_map;
-	struct rxe_map **map = mr->map;
-
-	num_map = (num_buf + RXE_BUF_PER_MAP - 1) / RXE_BUF_PER_MAP;
-
-	mr->map = kmalloc_array(num_map, sizeof(*map), GFP_KERNEL);
-	if (!mr->map)
-		goto err1;
-
-	for (i = 0; i < num_map; i++) {
-		mr->map[i] = kmalloc(sizeof(**map), GFP_KERNEL);
-		if (!mr->map[i])
-			goto err2;
-	}
-
-	BUILD_BUG_ON(!is_power_of_2(RXE_BUF_PER_MAP));
-
-	mr->map_shift = ilog2(RXE_BUF_PER_MAP);
-	mr->map_mask = RXE_BUF_PER_MAP - 1;
-
-	mr->num_buf = num_buf;
-	mr->num_map = num_map;
-	mr->max_buf = num_map * RXE_BUF_PER_MAP;
-
-	return 0;
-
-err2:
-	for (i--; i >= 0; i--)
-		kfree(mr->map[i]);
-
-	kfree(mr->map);
-	mr->map = NULL;
-err1:
-	return -ENOMEM;
-}
-
 void rxe_mr_init_dma(int access, struct rxe_mr *mr)
 {
 	rxe_mr_init(access, mr);
 
-	mr->access = access;
 	mr->state = RXE_MR_STATE_VALID;
 	mr->ibmr.type = IB_MR_TYPE_DMA;
 }
 
+static unsigned long rxe_mr_iova_to_index(struct rxe_mr *mr, u64 iova)
+{
+	return (iova >> mr->page_shift) - (mr->ibmr.iova >> mr->page_shift);
+}
+
+static unsigned long rxe_mr_iova_to_page_offset(struct rxe_mr *mr, u64 iova)
+{
+	return iova & (mr_page_size(mr) - 1);
+}
+
 static bool is_pmem_page(struct page *pg)
 {
 	unsigned long paddr = page_to_phys(pg);
@@ -122,86 +96,98 @@ static bool is_pmem_page(struct page *pg)
 				 IORES_DESC_PERSISTENT_MEMORY);
 }
 
+static int rxe_mr_fill_pages_from_sgt(struct rxe_mr *mr, struct sg_table *sgt)
+{
+	XA_STATE(xas, &mr->page_list, 0);
+	struct sg_page_iter sg_iter;
+	struct page *page;
+	bool persistent = !!(mr->access & IB_ACCESS_FLUSH_PERSISTENT);
+
+	__sg_page_iter_start(&sg_iter, sgt->sgl, sgt->orig_nents, 0);
+	if (!__sg_page_iter_next(&sg_iter))
+		return 0;
+
+	do {
+		xas_lock(&xas);
+		while (true) {
+			page = sg_page_iter_page(&sg_iter);
+
+			if (persistent && !is_pmem_page(page)) {
+				rxe_dbg_mr(mr, "Page can't be persistent\n");
+				xas_set_err(&xas, -EINVAL);
+				break;
+			}
+
+			xas_store(&xas, page);
+			if (xas_error(&xas))
+				break;
+			xas_next(&xas);
+			if (!__sg_page_iter_next(&sg_iter))
+				break;
+		}
+		xas_unlock(&xas);
+	} while (xas_nomem(&xas, GFP_KERNEL));
+
+	return xas_error(&xas);
+}
+
 int rxe_mr_init_user(struct rxe_dev *rxe, u64 start, u64 length, u64 iova,
 		     int access, struct rxe_mr *mr)
 {
-	struct rxe_map		**map;
-	struct rxe_phys_buf	*buf = NULL;
-	struct ib_umem		*umem;
-	struct sg_page_iter	sg_iter;
-	int			num_buf;
-	void			*vaddr;
+	struct ib_umem *umem;
 	int err;
 
+	rxe_mr_init(access, mr);
+
+	xa_init(&mr->page_list);
+
 	umem = ib_umem_get(&rxe->ib_dev, start, length, access);
 	if (IS_ERR(umem)) {
 		rxe_dbg_mr(mr, "Unable to pin memory region err = %d\n",
 			(int)PTR_ERR(umem));
-		err = PTR_ERR(umem);
-		goto err_out;
+		return PTR_ERR(umem);
 	}
 
-	num_buf = ib_umem_num_pages(umem);
-
-	rxe_mr_init(access, mr);
-
-	err = rxe_mr_alloc(mr, num_buf);
+	err = rxe_mr_fill_pages_from_sgt(mr, &umem->sgt_append.sgt);
 	if (err) {
-		rxe_dbg_mr(mr, "Unable to allocate memory for map\n");
-		goto err_release_umem;
+		ib_umem_release(umem);
+		return err;
 	}
 
-	mr->page_shift = PAGE_SHIFT;
-	mr->page_mask = PAGE_SIZE - 1;
-
-	num_buf			= 0;
-	map = mr->map;
-	if (length > 0) {
-		bool persistent_access = access & IB_ACCESS_FLUSH_PERSISTENT;
-
-		buf = map[0]->buf;
-		for_each_sgtable_page (&umem->sgt_append.sgt, &sg_iter, 0) {
-			struct page *pg = sg_page_iter_page(&sg_iter);
+	mr->umem = umem;
+	mr->ibmr.type = IB_MR_TYPE_USER;
+	mr->state = RXE_MR_STATE_VALID;
 
-			if (persistent_access && !is_pmem_page(pg)) {
-				rxe_dbg_mr(mr, "Unable to register persistent access to non-pmem device\n");
-				err = -EINVAL;
-				goto err_release_umem;
-			}
+	return 0;
+}
 
-			if (num_buf >= RXE_BUF_PER_MAP) {
-				map++;
-				buf = map[0]->buf;
-				num_buf = 0;
-			}
+static int rxe_mr_alloc(struct rxe_mr *mr, int num_buf)
+{
+	XA_STATE(xas, &mr->page_list, 0);
+	int i = 0;
+	int err;
 
-			vaddr = page_address(pg);
-			if (!vaddr) {
-				rxe_dbg_mr(mr, "Unable to get virtual address\n");
-				err = -ENOMEM;
-				goto err_release_umem;
-			}
-			buf->addr = (uintptr_t)vaddr;
-			buf->size = PAGE_SIZE;
-			num_buf++;
-			buf++;
+	xa_init(&mr->page_list);
 
+	do {
+		xas_lock(&xas);
+		while (i != num_buf) {
+			xas_store(&xas, XA_ZERO_ENTRY);
+			if (xas_error(&xas))
+				break;
+			xas_next(&xas);
+			i++;
 		}
-	}
+		xas_unlock(&xas);
+	} while (xas_nomem(&xas, GFP_KERNEL));
 
-	mr->umem = umem;
-	mr->access = access;
-	mr->offset = ib_umem_offset(umem);
-	mr->state = RXE_MR_STATE_VALID;
-	mr->ibmr.type = IB_MR_TYPE_USER;
-	mr->ibmr.page_size = PAGE_SIZE;
+	err = xas_error(&xas);
+	if (err)
+		return err;
 
-	return 0;
+	mr->num_buf = num_buf;
 
-err_release_umem:
-	ib_umem_release(umem);
-err_out:
-	return err;
+	return 0;
 }
 
 int rxe_mr_init_fast(int max_pages, struct rxe_mr *mr)
@@ -215,7 +201,6 @@ int rxe_mr_init_fast(int max_pages, struct rxe_mr *mr)
 	if (err)
 		goto err1;
 
-	mr->max_buf = max_pages;
 	mr->state = RXE_MR_STATE_FREE;
 	mr->ibmr.type = IB_MR_TYPE_MEM_REG;
 
@@ -225,187 +210,125 @@ err1:
 	return err;
 }
 
-static void lookup_iova(struct rxe_mr *mr, u64 iova, int *m_out, int *n_out,
-			size_t *offset_out)
+static int rxe_set_page(struct ib_mr *ibmr, u64 iova)
 {
-	size_t offset = iova - mr->ibmr.iova + mr->offset;
-	int			map_index;
-	int			buf_index;
-	u64			length;
-
-	if (likely(mr->page_shift)) {
-		*offset_out = offset & mr->page_mask;
-		offset >>= mr->page_shift;
-		*n_out = offset & mr->map_mask;
-		*m_out = offset >> mr->map_shift;
-	} else {
-		map_index = 0;
-		buf_index = 0;
+	struct rxe_mr *mr = to_rmr(ibmr);
+	struct page *page = virt_to_page(iova & mr->page_mask);
+	bool persistent = !!(mr->access & IB_ACCESS_FLUSH_PERSISTENT);
+	int err;
 
-		length = mr->map[map_index]->buf[buf_index].size;
+	if (persistent && !is_pmem_page(page)) {
+		rxe_dbg_mr(mr, "Page cannot be persistent\n");
+		return -EINVAL;
+	}
 
-		while (offset >= length) {
-			offset -= length;
-			buf_index++;
+	if (unlikely(mr->nbuf == mr->num_buf))
+		return -ENOMEM;
 
-			if (buf_index == RXE_BUF_PER_MAP) {
-				map_index++;
-				buf_index = 0;
-			}
-			length = mr->map[map_index]->buf[buf_index].size;
-		}
+	err = xa_err(xa_store(&mr->page_list, mr->nbuf, page, GFP_KERNEL));
+	if (err)
+		return err;
 
-		*m_out = map_index;
-		*n_out = buf_index;
-		*offset_out = offset;
-	}
+	mr->nbuf++;
+	return 0;
 }
 
-void *iova_to_vaddr(struct rxe_mr *mr, u64 iova, int length)
+int rxe_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sgl,
+		  int sg_nents, unsigned int *sg_offset)
 {
-	size_t offset;
-	int m, n;
-	void *addr;
-
-	if (mr->state != RXE_MR_STATE_VALID) {
-		rxe_dbg_mr(mr, "Not in valid state\n");
-		addr = NULL;
-		goto out;
-	}
-
-	if (!mr->map) {
-		addr = (void *)(uintptr_t)iova;
-		goto out;
-	}
-
-	if (mr_check_range(mr, iova, length)) {
-		rxe_dbg_mr(mr, "Range violation\n");
-		addr = NULL;
-		goto out;
-	}
-
-	lookup_iova(mr, iova, &m, &n, &offset);
-
-	if (offset + length > mr->map[m]->buf[n].size) {
-		rxe_dbg_mr(mr, "Crosses page boundary\n");
-		addr = NULL;
-		goto out;
-	}
+	struct rxe_mr *mr = to_rmr(ibmr);
+	unsigned int page_size = mr_page_size(mr);
 
-	addr = (void *)(uintptr_t)mr->map[m]->buf[n].addr + offset;
+	mr->nbuf = 0;
+	mr->page_shift = ilog2(page_size);
+	mr->page_mask = ~((u64)page_size - 1);
+	mr->page_offset = mr->ibmr.iova & (page_size - 1);
 
-out:
-	return addr;
+	return ib_sg_to_pages(ibmr, sgl, sg_nents, sg_offset, rxe_set_page);
 }
 
-int rxe_flush_pmem_iova(struct rxe_mr *mr, u64 iova, int length)
+static int rxe_mr_copy_xarray(struct rxe_mr *mr, u64 iova, void *addr,
+			      unsigned int length, enum rxe_mr_copy_dir dir)
 {
-	size_t offset;
-
-	if (length == 0)
-		return 0;
-
-	if (mr->ibmr.type == IB_MR_TYPE_DMA)
-		return -EFAULT;
+	unsigned int page_offset = rxe_mr_iova_to_page_offset(mr, iova);
+	unsigned long index = rxe_mr_iova_to_index(mr, iova);
+	unsigned int bytes;
+	struct page *page;
+	void *va;
 
-	offset = (iova - mr->ibmr.iova + mr->offset) & mr->page_mask;
-	while (length > 0) {
-		u8 *va;
-		int bytes;
-
-		bytes = mr->ibmr.page_size - offset;
-		if (bytes > length)
-			bytes = length;
-
-		va = iova_to_vaddr(mr, iova, length);
-		if (!va)
+	while (length) {
+		page = xa_load(&mr->page_list, index);
+		if (!page)
 			return -EFAULT;
 
-		arch_wb_cache_pmem(va, bytes);
-
+		bytes = min_t(unsigned int, length,
+				mr_page_size(mr) - page_offset);
+		va = kmap_local_page(page);
+		if (dir == RXE_FROM_MR_OBJ)
+			memcpy(addr, va + page_offset, bytes);
+		else
+			memcpy(va + page_offset, addr, bytes);
+		kunmap_local(va);
+
+		page_offset = 0;
+		addr += bytes;
 		length -= bytes;
-		iova += bytes;
-		offset = 0;
+		index++;
 	}
 
 	return 0;
 }
 
-/* copy data from a range (vaddr, vaddr+length-1) to or from
- * a mr object starting at iova.
- */
-int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr, int length,
-		enum rxe_mr_copy_dir dir)
+static void rxe_mr_copy_dma(struct rxe_mr *mr, u64 iova, void *addr,
+			    unsigned int length, enum rxe_mr_copy_dir dir)
 {
-	int			err;
-	int			bytes;
-	u8			*va;
-	struct rxe_map		**map;
-	struct rxe_phys_buf	*buf;
-	int			m;
-	int			i;
-	size_t			offset;
+	unsigned int page_offset = iova & (PAGE_SIZE - 1);
+	unsigned int bytes;
+	struct page *page;
+	u8 *va;
 
-	if (length == 0)
-		return 0;
-
-	if (mr->ibmr.type == IB_MR_TYPE_DMA) {
-		u8 *src, *dest;
+	while (length) {
+		page = virt_to_page(iova & mr->page_mask);
+		bytes = min_t(unsigned int, length,
+				PAGE_SIZE - page_offset);
+		va = kmap_local_page(page);
+
+		if (dir == RXE_TO_MR_OBJ)
+			memcpy(va + page_offset, addr, bytes);
+		else
+			memcpy(addr, va + page_offset, bytes);
+
+		kunmap_local(va);
+		page_offset = 0;
+		iova += bytes;
+		addr += bytes;
+		length -= bytes;
+	}
+}
 
-		src = (dir == RXE_TO_MR_OBJ) ? addr : ((void *)(uintptr_t)iova);
+int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr,
+		unsigned int length, enum rxe_mr_copy_dir dir)
+{
+	int err;
 
-		dest = (dir == RXE_TO_MR_OBJ) ? ((void *)(uintptr_t)iova) : addr;
+	if (length == 0)
+		return 0;
 
-		memcpy(dest, src, length);
+	if (WARN_ON(!mr))
+		return -EINVAL;
 
+	if (mr->ibmr.type == IB_MR_TYPE_DMA) {
+		rxe_mr_copy_dma(mr, iova, addr, length, dir);
 		return 0;
 	}
 
-	WARN_ON_ONCE(!mr->map);
-
 	err = mr_check_range(mr, iova, length);
-	if (err) {
-		err = -EFAULT;
-		goto err1;
-	}
-
-	lookup_iova(mr, iova, &m, &i, &offset);
-
-	map = mr->map + m;
-	buf	= map[0]->buf + i;
-
-	while (length > 0) {
-		u8 *src, *dest;
-
-		va	= (u8 *)(uintptr_t)buf->addr + offset;
-		src = (dir == RXE_TO_MR_OBJ) ? addr : va;
-		dest = (dir == RXE_TO_MR_OBJ) ? va : addr;
-
-		bytes	= buf->size - offset;
-
-		if (bytes > length)
-			bytes = length;
-
-		memcpy(dest, src, bytes);
-
-		length	-= bytes;
-		addr	+= bytes;
-
-		offset	= 0;
-		buf++;
-		i++;
-
-		if (i == RXE_BUF_PER_MAP) {
-			i = 0;
-			map++;
-			buf = map[0]->buf;
-		}
+	if (unlikely(err)) {
+		rxe_dbg_mr(mr, "iova out of range");
+		return err;
 	}
 
-	return 0;
-
-err1:
-	return err;
+	return rxe_mr_copy_xarray(mr, iova, addr, length, dir);
 }
 
 /* copy data in or out of a wqe, i.e. sg list
@@ -477,7 +400,6 @@ int copy_data(
 
 		if (bytes > 0) {
 			iova = sge->addr + offset;
-
 			err = rxe_mr_copy(mr, iova, addr, bytes, dir);
 			if (err)
 				goto err2;
@@ -504,6 +426,165 @@ err1:
 	return err;
 }
 
+int rxe_flush_pmem_iova(struct rxe_mr *mr, u64 iova, unsigned int length)
+{
+	unsigned int page_offset;
+	unsigned long index;
+	struct page *page;
+	unsigned int bytes;
+	int err;
+	u8 *va;
+
+	/* mr must be valid even if length is zero */
+	if (WARN_ON(!mr))
+		return -EINVAL;
+
+	if (length == 0)
+		return 0;
+
+	if (mr->ibmr.type == IB_MR_TYPE_DMA)
+		return -EFAULT;
+
+	err = mr_check_range(mr, iova, length);
+	if (err)
+		return err;
+
+	while (length > 0) {
+		index = rxe_mr_iova_to_index(mr, iova);
+		page = xa_load(&mr->page_list, index);
+		page_offset = rxe_mr_iova_to_page_offset(mr, iova);
+		if (!page)
+			return -EFAULT;
+		bytes = min_t(unsigned int, length,
+				mr_page_size(mr) - page_offset);
+
+		va = kmap_local_page(page);
+		arch_wb_cache_pmem(va + page_offset, bytes);
+		kunmap_local(va);
+
+		length -= bytes;
+		iova += bytes;
+		page_offset = 0;
+	}
+
+	return 0;
+}
+
+/* Guarantee atomicity of atomic operations at the machine level. */
+static DEFINE_SPINLOCK(atomic_ops_lock);
+
+int rxe_mr_do_atomic_op(struct rxe_mr *mr, u64 iova, int opcode,
+			u64 compare, u64 swap_add, u64 *orig_val)
+{
+	unsigned int page_offset;
+	struct page *page;
+	u64 value;
+	u64 *va;
+
+	if (unlikely(mr->state != RXE_MR_STATE_VALID)) {
+		rxe_dbg_mr(mr, "mr not in valid state");
+		return RESPST_ERR_RKEY_VIOLATION;
+	}
+
+	if (mr->ibmr.type == IB_MR_TYPE_DMA) {
+		page_offset = iova & (PAGE_SIZE - 1);
+		page = virt_to_page(iova & PAGE_MASK);
+	} else {
+		unsigned long index;
+		int err;
+
+		err = mr_check_range(mr, iova, sizeof(value));
+		if (err) {
+			rxe_dbg_mr(mr, "iova out of range");
+			return RESPST_ERR_RKEY_VIOLATION;
+		}
+		page_offset = rxe_mr_iova_to_page_offset(mr, iova);
+		index = rxe_mr_iova_to_index(mr, iova);
+		page = xa_load(&mr->page_list, index);
+		if (!page)
+			return RESPST_ERR_RKEY_VIOLATION;
+	}
+
+	if (unlikely(page_offset & 0x7)) {
+		rxe_dbg_mr(mr, "iova not aligned");
+		return RESPST_ERR_MISALIGNED_ATOMIC;
+	}
+
+	va = kmap_local_page(page);
+
+	spin_lock_bh(&atomic_ops_lock);
+	value = *orig_val = va[page_offset >> 3];
+
+	if (opcode == IB_OPCODE_RC_COMPARE_SWAP) {
+		if (value == compare)
+			va[page_offset >> 3] = swap_add;
+	} else {
+		value += swap_add;
+		va[page_offset >> 3] = value;
+	}
+	spin_unlock_bh(&atomic_ops_lock);
+
+	kunmap_local(va);
+
+	return 0;
+}
+
+#if defined CONFIG_64BIT
+/* only implemented or called for 64 bit architectures */
+int rxe_mr_do_atomic_write(struct rxe_mr *mr, u64 iova, u64 value)
+{
+	unsigned int page_offset;
+	struct page *page;
+	u64 *va;
+
+	/* See IBA oA19-28 */
+	if (unlikely(mr->state != RXE_MR_STATE_VALID)) {
+		rxe_dbg_mr(mr, "mr not in valid state");
+		return RESPST_ERR_RKEY_VIOLATION;
+	}
+
+	if (mr->ibmr.type == IB_MR_TYPE_DMA) {
+		page_offset = iova & (PAGE_SIZE - 1);
+		page = virt_to_page(iova & PAGE_MASK);
+	} else {
+		unsigned long index;
+		int err;
+
+		/* See IBA oA19-28 */
+		err = mr_check_range(mr, iova, sizeof(value));
+		if (unlikely(err)) {
+			rxe_dbg_mr(mr, "iova out of range");
+			return RESPST_ERR_RKEY_VIOLATION;
+		}
+		page_offset = rxe_mr_iova_to_page_offset(mr, iova);
+		index = rxe_mr_iova_to_index(mr, iova);
+		page = xa_load(&mr->page_list, index);
+		if (!page)
+			return RESPST_ERR_RKEY_VIOLATION;
+	}
+
+	/* See IBA A19.4.2 */
+	if (unlikely(page_offset & 0x7)) {
+		rxe_dbg_mr(mr, "misaligned address");
+		return RESPST_ERR_MISALIGNED_ATOMIC;
+	}
+
+	va = kmap_local_page(page);
+
+	/* Do atomic write after all prior operations have completed */
+	smp_store_release(&va[page_offset >> 3], value);
+
+	kunmap_local(va);
+
+	return 0;
+}
+#else
+int rxe_mr_do_atomic_write(struct rxe_mr *mr, u64 iova, u64 value)
+{
+	return RESPST_ERR_UNSUPPORTED_OPCODE;
+}
+#endif
+
 int advance_dma_data(struct rxe_dma_info *dma, unsigned int length)
 {
 	struct rxe_sge		*sge	= &dma->sge[dma->cur_sge];
@@ -537,12 +618,6 @@ int advance_dma_data(struct rxe_dma_info *dma, unsigned int length)
 	return 0;
 }
 
-/* (1) find the mr corresponding to lkey/rkey
- *     depending on lookup_type
- * (2) verify that the (qp) pd matches the mr pd
- * (3) verify that the mr can support the requested access
- * (4) verify that mr state is valid
- */
 struct rxe_mr *lookup_mr(struct rxe_pd *pd, int access, u32 key,
 			 enum rxe_mr_lookup_type type)
 {
@@ -656,22 +731,17 @@ int rxe_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata)
 		return -EINVAL;
 
 	rxe_cleanup(mr);
-
+	kfree_rcu(mr);
 	return 0;
 }
 
 void rxe_mr_cleanup(struct rxe_pool_elem *elem)
 {
 	struct rxe_mr *mr = container_of(elem, typeof(*mr), elem);
-	int i;
 
 	rxe_put(mr_pd(mr));
 	ib_umem_release(mr->umem);
 
-	if (mr->map) {
-		for (i = 0; i < mr->num_map; i++)
-			kfree(mr->map[i]);
-
-		kfree(mr->map);
-	}
+	if (mr->ibmr.type != IB_MR_TYPE_DMA)
+		xa_destroy(&mr->page_list);
 }
diff --git a/drivers/infiniband/sw/rxe/rxe_pool.c b/drivers/infiniband/sw/rxe/rxe_pool.c
index 1151c0b5ccea..6215c6de3a84 100644
--- a/drivers/infiniband/sw/rxe/rxe_pool.c
+++ b/drivers/infiniband/sw/rxe/rxe_pool.c
@@ -116,55 +116,12 @@ void rxe_pool_cleanup(struct rxe_pool *pool)
 	WARN_ON(!xa_empty(&pool->xa));
 }
 
-void *rxe_alloc(struct rxe_pool *pool)
-{
-	struct rxe_pool_elem *elem;
-	void *obj;
-	int err;
-
-	if (WARN_ON(!(pool->type == RXE_TYPE_MR)))
-		return NULL;
-
-	if (atomic_inc_return(&pool->num_elem) > pool->max_elem)
-		goto err_cnt;
-
-	obj = kzalloc(pool->elem_size, GFP_KERNEL);
-	if (!obj)
-		goto err_cnt;
-
-	elem = (struct rxe_pool_elem *)((u8 *)obj + pool->elem_offset);
-
-	elem->pool = pool;
-	elem->obj = obj;
-	kref_init(&elem->ref_cnt);
-	init_completion(&elem->complete);
-
-	/* allocate index in array but leave pointer as NULL so it
-	 * can't be looked up until rxe_finalize() is called
-	 */
-	err = xa_alloc_cyclic(&pool->xa, &elem->index, NULL, pool->limit,
-			      &pool->next, GFP_KERNEL);
-	if (err < 0)
-		goto err_free;
-
-	return obj;
-
-err_free:
-	kfree(obj);
-err_cnt:
-	atomic_dec(&pool->num_elem);
-	return NULL;
-}
-
 int __rxe_add_to_pool(struct rxe_pool *pool, struct rxe_pool_elem *elem,
 				bool sleepable)
 {
 	int err;
 	gfp_t gfp_flags;
 
-	if (WARN_ON(pool->type == RXE_TYPE_MR))
-		return -EINVAL;
-
 	if (atomic_inc_return(&pool->num_elem) > pool->max_elem)
 		goto err_cnt;
 
@@ -275,9 +232,6 @@ int __rxe_cleanup(struct rxe_pool_elem *elem, bool sleepable)
 	if (pool->cleanup)
 		pool->cleanup(elem);
 
-	if (pool->type == RXE_TYPE_MR)
-		kfree_rcu(elem->obj);
-
 	atomic_dec(&pool->num_elem);
 
 	return err;
diff --git a/drivers/infiniband/sw/rxe/rxe_pool.h b/drivers/infiniband/sw/rxe/rxe_pool.h
index 9d83cb32092f..b42e26427a70 100644
--- a/drivers/infiniband/sw/rxe/rxe_pool.h
+++ b/drivers/infiniband/sw/rxe/rxe_pool.h
@@ -54,9 +54,6 @@ void rxe_pool_init(struct rxe_dev *rxe, struct rxe_pool *pool,
 /* free resources from object pool */
 void rxe_pool_cleanup(struct rxe_pool *pool);
 
-/* allocate an object from pool */
-void *rxe_alloc(struct rxe_pool *pool);
-
 /* connect already allocated object to pool */
 int __rxe_add_to_pool(struct rxe_pool *pool, struct rxe_pool_elem *elem,
 				bool sleepable);
diff --git a/drivers/infiniband/sw/rxe/rxe_queue.h b/drivers/infiniband/sw/rxe/rxe_queue.h
index ed44042782fa..c711cb98b949 100644
--- a/drivers/infiniband/sw/rxe/rxe_queue.h
+++ b/drivers/infiniband/sw/rxe/rxe_queue.h
@@ -35,19 +35,26 @@
 /**
  * enum queue_type - type of queue
  * @QUEUE_TYPE_TO_CLIENT:	Queue is written by rxe driver and
- *				read by client. Used by rxe driver only.
+ *				read by client which may be a user space
+ *				application or a kernel ulp.
+ *				Used by rxe internals only.
  * @QUEUE_TYPE_FROM_CLIENT:	Queue is written by client and
- *				read by rxe driver. Used by rxe driver only.
- * @QUEUE_TYPE_TO_DRIVER:	Queue is written by client and
- *				read by rxe driver. Used by kernel client only.
- * @QUEUE_TYPE_FROM_DRIVER:	Queue is written by rxe driver and
- *				read by client. Used by kernel client only.
+ *				read by rxe driver.
+ *				Used by rxe internals only.
+ * @QUEUE_TYPE_FROM_ULP:	Queue is written by kernel ulp and
+ *				read by rxe driver.
+ *				Used by kernel verbs APIs only on
+ *				behalf of ulps.
+ * @QUEUE_TYPE_TO_ULP:		Queue is written by rxe driver and
+ *				read by kernel ulp.
+ *				Used by kernel verbs APIs only on
+ *				behalf of ulps.
  */
 enum queue_type {
 	QUEUE_TYPE_TO_CLIENT,
 	QUEUE_TYPE_FROM_CLIENT,
-	QUEUE_TYPE_TO_DRIVER,
-	QUEUE_TYPE_FROM_DRIVER,
+	QUEUE_TYPE_FROM_ULP,
+	QUEUE_TYPE_TO_ULP,
 };
 
 struct rxe_queue_buf;
@@ -62,9 +69,9 @@ struct rxe_queue {
 	u32			index_mask;
 	enum queue_type		type;
 	/* private copy of index for shared queues between
-	 * kernel space and user space. Kernel reads and writes
+	 * driver and clients. Driver reads and writes
 	 * this copy and then replicates to rxe_queue_buf
-	 * for read access by user space.
+	 * for read access by clients.
 	 */
 	u32			index;
 };
@@ -97,19 +104,21 @@ static inline u32 queue_get_producer(const struct rxe_queue *q,
 
 	switch (type) {
 	case QUEUE_TYPE_FROM_CLIENT:
-		/* protect user index */
+		/* used by rxe, client owns the index */
 		prod = smp_load_acquire(&q->buf->producer_index);
 		break;
 	case QUEUE_TYPE_TO_CLIENT:
+		/* used by rxe which owns the index */
 		prod = q->index;
 		break;
-	case QUEUE_TYPE_FROM_DRIVER:
-		/* protect driver index */
-		prod = smp_load_acquire(&q->buf->producer_index);
-		break;
-	case QUEUE_TYPE_TO_DRIVER:
+	case QUEUE_TYPE_FROM_ULP:
+		/* used by ulp which owns the index */
 		prod = q->buf->producer_index;
 		break;
+	case QUEUE_TYPE_TO_ULP:
+		/* used by ulp, rxe owns the index */
+		prod = smp_load_acquire(&q->buf->producer_index);
+		break;
 	}
 
 	return prod;
@@ -122,19 +131,21 @@ static inline u32 queue_get_consumer(const struct rxe_queue *q,
 
 	switch (type) {
 	case QUEUE_TYPE_FROM_CLIENT:
+		/* used by rxe which owns the index */
 		cons = q->index;
 		break;
 	case QUEUE_TYPE_TO_CLIENT:
-		/* protect user index */
+		/* used by rxe, client owns the index */
 		cons = smp_load_acquire(&q->buf->consumer_index);
 		break;
-	case QUEUE_TYPE_FROM_DRIVER:
-		cons = q->buf->consumer_index;
-		break;
-	case QUEUE_TYPE_TO_DRIVER:
-		/* protect driver index */
+	case QUEUE_TYPE_FROM_ULP:
+		/* used by ulp, rxe owns the index */
 		cons = smp_load_acquire(&q->buf->consumer_index);
 		break;
+	case QUEUE_TYPE_TO_ULP:
+		/* used by ulp which owns the index */
+		cons = q->buf->consumer_index;
+		break;
 	}
 
 	return cons;
@@ -172,24 +183,31 @@ static inline void queue_advance_producer(struct rxe_queue *q,
 
 	switch (type) {
 	case QUEUE_TYPE_FROM_CLIENT:
-		pr_warn("%s: attempt to advance client index\n",
-			__func__);
+		/* used by rxe, client owns the index */
+		if (WARN_ON(1))
+			pr_warn("%s: attempt to advance client index\n",
+				__func__);
 		break;
 	case QUEUE_TYPE_TO_CLIENT:
+		/* used by rxe which owns the index */
 		prod = q->index;
 		prod = (prod + 1) & q->index_mask;
 		q->index = prod;
-		/* protect user index */
+		/* release so client can read it safely */
 		smp_store_release(&q->buf->producer_index, prod);
 		break;
-	case QUEUE_TYPE_FROM_DRIVER:
-		pr_warn("%s: attempt to advance driver index\n",
-			__func__);
-		break;
-	case QUEUE_TYPE_TO_DRIVER:
+	case QUEUE_TYPE_FROM_ULP:
+		/* used by ulp which owns the index */
 		prod = q->buf->producer_index;
 		prod = (prod + 1) & q->index_mask;
-		q->buf->producer_index = prod;
+		/* release so rxe can read it safely */
+		smp_store_release(&q->buf->producer_index, prod);
+		break;
+	case QUEUE_TYPE_TO_ULP:
+		/* used by ulp, rxe owns the index */
+		if (WARN_ON(1))
+			pr_warn("%s: attempt to advance driver index\n",
+				__func__);
 		break;
 	}
 }
@@ -201,24 +219,30 @@ static inline void queue_advance_consumer(struct rxe_queue *q,
 
 	switch (type) {
 	case QUEUE_TYPE_FROM_CLIENT:
-		cons = q->index;
-		cons = (cons + 1) & q->index_mask;
+		/* used by rxe which owns the index */
+		cons = (q->index + 1) & q->index_mask;
 		q->index = cons;
-		/* protect user index */
+		/* release so client can read it safely */
 		smp_store_release(&q->buf->consumer_index, cons);
 		break;
 	case QUEUE_TYPE_TO_CLIENT:
-		pr_warn("%s: attempt to advance client index\n",
-			__func__);
+		/* used by rxe, client owns the index */
+		if (WARN_ON(1))
+			pr_warn("%s: attempt to advance client index\n",
+				__func__);
+		break;
+	case QUEUE_TYPE_FROM_ULP:
+		/* used by ulp, rxe owns the index */
+		if (WARN_ON(1))
+			pr_warn("%s: attempt to advance driver index\n",
+				__func__);
 		break;
-	case QUEUE_TYPE_FROM_DRIVER:
+	case QUEUE_TYPE_TO_ULP:
+		/* used by ulp which owns the index */
 		cons = q->buf->consumer_index;
 		cons = (cons + 1) & q->index_mask;
-		q->buf->consumer_index = cons;
-		break;
-	case QUEUE_TYPE_TO_DRIVER:
-		pr_warn("%s: attempt to advance driver index\n",
-			__func__);
+		/* release so rxe can read it safely */
+		smp_store_release(&q->buf->consumer_index, cons);
 		break;
 	}
 }
diff --git a/drivers/infiniband/sw/rxe/rxe_resp.c b/drivers/infiniband/sw/rxe/rxe_resp.c
index c74972244f08..0cc1ba91d48c 100644
--- a/drivers/infiniband/sw/rxe/rxe_resp.c
+++ b/drivers/infiniband/sw/rxe/rxe_resp.c
@@ -10,43 +10,6 @@
 #include "rxe_loc.h"
 #include "rxe_queue.h"
 
-enum resp_states {
-	RESPST_NONE,
-	RESPST_GET_REQ,
-	RESPST_CHK_PSN,
-	RESPST_CHK_OP_SEQ,
-	RESPST_CHK_OP_VALID,
-	RESPST_CHK_RESOURCE,
-	RESPST_CHK_LENGTH,
-	RESPST_CHK_RKEY,
-	RESPST_EXECUTE,
-	RESPST_READ_REPLY,
-	RESPST_ATOMIC_REPLY,
-	RESPST_ATOMIC_WRITE_REPLY,
-	RESPST_PROCESS_FLUSH,
-	RESPST_COMPLETE,
-	RESPST_ACKNOWLEDGE,
-	RESPST_CLEANUP,
-	RESPST_DUPLICATE_REQUEST,
-	RESPST_ERR_MALFORMED_WQE,
-	RESPST_ERR_UNSUPPORTED_OPCODE,
-	RESPST_ERR_MISALIGNED_ATOMIC,
-	RESPST_ERR_PSN_OUT_OF_SEQ,
-	RESPST_ERR_MISSING_OPCODE_FIRST,
-	RESPST_ERR_MISSING_OPCODE_LAST_C,
-	RESPST_ERR_MISSING_OPCODE_LAST_D1E,
-	RESPST_ERR_TOO_MANY_RDMA_ATM_REQ,
-	RESPST_ERR_RNR,
-	RESPST_ERR_RKEY_VIOLATION,
-	RESPST_ERR_INVALIDATE_RKEY,
-	RESPST_ERR_LENGTH,
-	RESPST_ERR_CQ_OVERFLOW,
-	RESPST_ERROR,
-	RESPST_RESET,
-	RESPST_DONE,
-	RESPST_EXIT,
-};
-
 static char *resp_state_name[] = {
 	[RESPST_NONE]				= "NONE",
 	[RESPST_GET_REQ]			= "GET_REQ",
@@ -457,13 +420,23 @@ static enum resp_states rxe_resp_check_length(struct rxe_qp *qp,
 	return RESPST_CHK_RKEY;
 }
 
+/* if the reth length field is zero we can assume nothing
+ * about the rkey value and should not validate or use it.
+ * Instead set qp->resp.rkey to 0 which is an invalid rkey
+ * value since the minimum index part is 1.
+ */
 static void qp_resp_from_reth(struct rxe_qp *qp, struct rxe_pkt_info *pkt)
 {
+	unsigned int length = reth_len(pkt);
+
 	qp->resp.va = reth_va(pkt);
 	qp->resp.offset = 0;
-	qp->resp.rkey = reth_rkey(pkt);
-	qp->resp.resid = reth_len(pkt);
-	qp->resp.length = reth_len(pkt);
+	qp->resp.resid = length;
+	qp->resp.length = length;
+	if (pkt->mask & RXE_READ_OR_WRITE_MASK && length == 0)
+		qp->resp.rkey = 0;
+	else
+		qp->resp.rkey = reth_rkey(pkt);
 }
 
 static void qp_resp_from_atmeth(struct rxe_qp *qp, struct rxe_pkt_info *pkt)
@@ -474,6 +447,10 @@ static void qp_resp_from_atmeth(struct rxe_qp *qp, struct rxe_pkt_info *pkt)
 	qp->resp.resid = sizeof(u64);
 }
 
+/* resolve the packet rkey to qp->resp.mr or set qp->resp.mr to NULL
+ * if an invalid rkey is received or the rdma length is zero. For middle
+ * or last packets use the stored value of mr.
+ */
 static enum resp_states check_rkey(struct rxe_qp *qp,
 				   struct rxe_pkt_info *pkt)
 {
@@ -510,10 +487,12 @@ static enum resp_states check_rkey(struct rxe_qp *qp,
 		return RESPST_EXECUTE;
 	}
 
-	/* A zero-byte op is not required to set an addr or rkey. See C9-88 */
+	/* A zero-byte read or write op is not required to
+	 * set an addr or rkey. See C9-88
+	 */
 	if ((pkt->mask & RXE_READ_OR_WRITE_MASK) &&
-	    (pkt->mask & RXE_RETH_MASK) &&
-	    reth_len(pkt) == 0) {
+	    (pkt->mask & RXE_RETH_MASK) && reth_len(pkt) == 0) {
+		qp->resp.mr = NULL;
 		return RESPST_EXECUTE;
 	}
 
@@ -592,6 +571,7 @@ skip_check_range:
 	return RESPST_EXECUTE;
 
 err:
+	qp->resp.mr = NULL;
 	if (mr)
 		rxe_put(mr);
 	if (mw)
@@ -725,17 +705,12 @@ static enum resp_states process_flush(struct rxe_qp *qp,
 	return RESPST_ACKNOWLEDGE;
 }
 
-/* Guarantee atomicity of atomic operations at the machine level. */
-static DEFINE_SPINLOCK(atomic_ops_lock);
-
 static enum resp_states atomic_reply(struct rxe_qp *qp,
-					 struct rxe_pkt_info *pkt)
+				     struct rxe_pkt_info *pkt)
 {
-	u64 *vaddr;
-	enum resp_states ret;
 	struct rxe_mr *mr = qp->resp.mr;
 	struct resp_res *res = qp->resp.res;
-	u64 value;
+	int err;
 
 	if (!res) {
 		res = rxe_prepare_res(qp, pkt, RXE_ATOMIC_MASK);
@@ -743,32 +718,14 @@ static enum resp_states atomic_reply(struct rxe_qp *qp,
 	}
 
 	if (!res->replay) {
-		if (mr->state != RXE_MR_STATE_VALID) {
-			ret = RESPST_ERR_RKEY_VIOLATION;
-			goto out;
-		}
-
-		vaddr = iova_to_vaddr(mr, qp->resp.va + qp->resp.offset,
-					sizeof(u64));
-
-		/* check vaddr is 8 bytes aligned. */
-		if (!vaddr || (uintptr_t)vaddr & 7) {
-			ret = RESPST_ERR_MISALIGNED_ATOMIC;
-			goto out;
-		}
+		u64 iova = qp->resp.va + qp->resp.offset;
 
-		spin_lock_bh(&atomic_ops_lock);
-		res->atomic.orig_val = value = *vaddr;
-
-		if (pkt->opcode == IB_OPCODE_RC_COMPARE_SWAP) {
-			if (value == atmeth_comp(pkt))
-				value = atmeth_swap_add(pkt);
-		} else {
-			value += atmeth_swap_add(pkt);
-		}
-
-		*vaddr = value;
-		spin_unlock_bh(&atomic_ops_lock);
+		err = rxe_mr_do_atomic_op(mr, iova, pkt->opcode,
+					  atmeth_comp(pkt),
+					  atmeth_swap_add(pkt),
+					  &res->atomic.orig_val);
+		if (err)
+			return err;
 
 		qp->resp.msn++;
 
@@ -780,35 +737,35 @@ static enum resp_states atomic_reply(struct rxe_qp *qp,
 		qp->resp.status = IB_WC_SUCCESS;
 	}
 
-	ret = RESPST_ACKNOWLEDGE;
-out:
-	return ret;
+	return RESPST_ACKNOWLEDGE;
 }
 
-#ifdef CONFIG_64BIT
-static enum resp_states do_atomic_write(struct rxe_qp *qp,
-					struct rxe_pkt_info *pkt)
+static enum resp_states atomic_write_reply(struct rxe_qp *qp,
+					   struct rxe_pkt_info *pkt)
 {
-	struct rxe_mr *mr = qp->resp.mr;
-	int payload = payload_size(pkt);
-	u64 src, *dst;
-
-	if (mr->state != RXE_MR_STATE_VALID)
-		return RESPST_ERR_RKEY_VIOLATION;
+	struct resp_res *res = qp->resp.res;
+	struct rxe_mr *mr;
+	u64 value;
+	u64 iova;
+	int err;
 
-	memcpy(&src, payload_addr(pkt), payload);
+	if (!res) {
+		res = rxe_prepare_res(qp, pkt, RXE_ATOMIC_WRITE_MASK);
+		qp->resp.res = res;
+	}
 
-	dst = iova_to_vaddr(mr, qp->resp.va + qp->resp.offset, payload);
-	/* check vaddr is 8 bytes aligned. */
-	if (!dst || (uintptr_t)dst & 7)
-		return RESPST_ERR_MISALIGNED_ATOMIC;
+	if (res->replay)
+		return RESPST_ACKNOWLEDGE;
 
-	/* Do atomic write after all prior operations have completed */
-	smp_store_release(dst, src);
+	mr = qp->resp.mr;
+	value = *(u64 *)payload_addr(pkt);
+	iova = qp->resp.va + qp->resp.offset;
 
-	/* decrease resp.resid to zero */
-	qp->resp.resid -= sizeof(payload);
+	err = rxe_mr_do_atomic_write(mr, iova, value);
+	if (err)
+		return err;
 
+	qp->resp.resid = 0;
 	qp->resp.msn++;
 
 	/* next expected psn, read handles this separately */
@@ -817,29 +774,8 @@ static enum resp_states do_atomic_write(struct rxe_qp *qp,
 
 	qp->resp.opcode = pkt->opcode;
 	qp->resp.status = IB_WC_SUCCESS;
-	return RESPST_ACKNOWLEDGE;
-}
-#else
-static enum resp_states do_atomic_write(struct rxe_qp *qp,
-					struct rxe_pkt_info *pkt)
-{
-	return RESPST_ERR_UNSUPPORTED_OPCODE;
-}
-#endif /* CONFIG_64BIT */
 
-static enum resp_states atomic_write_reply(struct rxe_qp *qp,
-					   struct rxe_pkt_info *pkt)
-{
-	struct resp_res *res = qp->resp.res;
-
-	if (!res) {
-		res = rxe_prepare_res(qp, pkt, RXE_ATOMIC_WRITE_MASK);
-		qp->resp.res = res;
-	}
-
-	if (res->replay)
-		return RESPST_ACKNOWLEDGE;
-	return do_atomic_write(qp, pkt);
+	return RESPST_ACKNOWLEDGE;
 }
 
 static struct sk_buff *prepare_ack_packet(struct rxe_qp *qp,
@@ -966,7 +902,11 @@ static enum resp_states read_reply(struct rxe_qp *qp,
 	}
 
 	if (res->state == rdatm_res_state_new) {
-		if (!res->replay) {
+		if (!res->replay || qp->resp.length == 0) {
+			/* if length == 0 mr will be NULL (is ok)
+			 * otherwise qp->resp.mr holds a ref on mr
+			 * which we transfer to mr and drop below.
+			 */
 			mr = qp->resp.mr;
 			qp->resp.mr = NULL;
 		} else {
@@ -980,6 +920,10 @@ static enum resp_states read_reply(struct rxe_qp *qp,
 		else
 			opcode = IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST;
 	} else {
+		/* re-lookup mr from rkey on all later packets.
+		 * length will be non-zero. This can fail if someone
+		 * modifies or destroys the mr since the first packet.
+		 */
 		mr = rxe_recheck_mr(qp, res->read.rkey);
 		if (!mr)
 			return RESPST_ERR_RKEY_VIOLATION;
@@ -997,18 +941,16 @@ static enum resp_states read_reply(struct rxe_qp *qp,
 	skb = prepare_ack_packet(qp, &ack_pkt, opcode, payload,
 				 res->cur_psn, AETH_ACK_UNLIMITED);
 	if (!skb) {
-		if (mr)
-			rxe_put(mr);
-		return RESPST_ERR_RNR;
+		state = RESPST_ERR_RNR;
+		goto err_out;
 	}
 
 	err = rxe_mr_copy(mr, res->read.va, payload_addr(&ack_pkt),
 			  payload, RXE_FROM_MR_OBJ);
-	if (mr)
-		rxe_put(mr);
 	if (err) {
 		kfree_skb(skb);
-		return RESPST_ERR_RKEY_VIOLATION;
+		state = RESPST_ERR_RKEY_VIOLATION;
+		goto err_out;
 	}
 
 	if (bth_pad(&ack_pkt)) {
@@ -1017,9 +959,12 @@ static enum resp_states read_reply(struct rxe_qp *qp,
 		memset(pad, 0, bth_pad(&ack_pkt));
 	}
 
+	/* rxe_xmit_packet always consumes the skb */
 	err = rxe_xmit_packet(qp, &ack_pkt, skb);
-	if (err)
-		return RESPST_ERR_RNR;
+	if (err) {
+		state = RESPST_ERR_RNR;
+		goto err_out;
+	}
 
 	res->read.va += payload;
 	res->read.resid -= payload;
@@ -1036,6 +981,9 @@ static enum resp_states read_reply(struct rxe_qp *qp,
 		state = RESPST_CLEANUP;
 	}
 
+err_out:
+	if (mr)
+		rxe_put(mr);
 	return state;
 }
 
diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.c b/drivers/infiniband/sw/rxe/rxe_verbs.c
index 025b35bf014e..e14050a69276 100644
--- a/drivers/infiniband/sw/rxe/rxe_verbs.c
+++ b/drivers/infiniband/sw/rxe/rxe_verbs.c
@@ -245,7 +245,7 @@ static int post_one_recv(struct rxe_rq *rq, const struct ib_recv_wr *ibwr)
 	int num_sge = ibwr->num_sge;
 	int full;
 
-	full = queue_full(rq->queue, QUEUE_TYPE_TO_DRIVER);
+	full = queue_full(rq->queue, QUEUE_TYPE_FROM_ULP);
 	if (unlikely(full))
 		return -ENOMEM;
 
@@ -256,7 +256,7 @@ static int post_one_recv(struct rxe_rq *rq, const struct ib_recv_wr *ibwr)
 	for (i = 0; i < num_sge; i++)
 		length += ibwr->sg_list[i].length;
 
-	recv_wqe = queue_producer_addr(rq->queue, QUEUE_TYPE_TO_DRIVER);
+	recv_wqe = queue_producer_addr(rq->queue, QUEUE_TYPE_FROM_ULP);
 	recv_wqe->wr_id = ibwr->wr_id;
 
 	memcpy(recv_wqe->dma.sge, ibwr->sg_list,
@@ -268,7 +268,7 @@ static int post_one_recv(struct rxe_rq *rq, const struct ib_recv_wr *ibwr)
 	recv_wqe->dma.cur_sge		= 0;
 	recv_wqe->dma.sge_offset	= 0;
 
-	queue_advance_producer(rq->queue, QUEUE_TYPE_TO_DRIVER);
+	queue_advance_producer(rq->queue, QUEUE_TYPE_FROM_ULP);
 
 	return 0;
 }
@@ -623,17 +623,17 @@ static int post_one_send(struct rxe_qp *qp, const struct ib_send_wr *ibwr,
 
 	spin_lock_irqsave(&qp->sq.sq_lock, flags);
 
-	full = queue_full(sq->queue, QUEUE_TYPE_TO_DRIVER);
+	full = queue_full(sq->queue, QUEUE_TYPE_FROM_ULP);
 
 	if (unlikely(full)) {
 		spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
 		return -ENOMEM;
 	}
 
-	send_wqe = queue_producer_addr(sq->queue, QUEUE_TYPE_TO_DRIVER);
+	send_wqe = queue_producer_addr(sq->queue, QUEUE_TYPE_FROM_ULP);
 	init_send_wqe(qp, ibwr, mask, length, send_wqe);
 
-	queue_advance_producer(sq->queue, QUEUE_TYPE_TO_DRIVER);
+	queue_advance_producer(sq->queue, QUEUE_TYPE_FROM_ULP);
 
 	spin_unlock_irqrestore(&qp->sq.sq_lock, flags);
 
@@ -821,12 +821,12 @@ static int rxe_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
 
 	spin_lock_irqsave(&cq->cq_lock, flags);
 	for (i = 0; i < num_entries; i++) {
-		cqe = queue_head(cq->queue, QUEUE_TYPE_FROM_DRIVER);
+		cqe = queue_head(cq->queue, QUEUE_TYPE_TO_ULP);
 		if (!cqe)
 			break;
 
 		memcpy(wc++, &cqe->ibwc, sizeof(*wc));
-		queue_advance_consumer(cq->queue, QUEUE_TYPE_FROM_DRIVER);
+		queue_advance_consumer(cq->queue, QUEUE_TYPE_TO_ULP);
 	}
 	spin_unlock_irqrestore(&cq->cq_lock, flags);
 
@@ -838,7 +838,7 @@ static int rxe_peek_cq(struct ib_cq *ibcq, int wc_cnt)
 	struct rxe_cq *cq = to_rcq(ibcq);
 	int count;
 
-	count = queue_count(cq->queue, QUEUE_TYPE_FROM_DRIVER);
+	count = queue_count(cq->queue, QUEUE_TYPE_TO_ULP);
 
 	return (count > wc_cnt) ? wc_cnt : count;
 }
@@ -854,7 +854,7 @@ static int rxe_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
 	if (cq->notify != IB_CQ_NEXT_COMP)
 		cq->notify = flags & IB_CQ_SOLICITED_MASK;
 
-	empty = queue_empty(cq->queue, QUEUE_TYPE_FROM_DRIVER);
+	empty = queue_empty(cq->queue, QUEUE_TYPE_TO_ULP);
 
 	if ((flags & IB_CQ_REPORT_MISSED_EVENTS) && !empty)
 		ret = 1;
@@ -869,10 +869,17 @@ static struct ib_mr *rxe_get_dma_mr(struct ib_pd *ibpd, int access)
 	struct rxe_dev *rxe = to_rdev(ibpd->device);
 	struct rxe_pd *pd = to_rpd(ibpd);
 	struct rxe_mr *mr;
+	int err;
 
-	mr = rxe_alloc(&rxe->mr_pool);
-	if (!mr)
-		return ERR_PTR(-ENOMEM);
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr) {
+		err = -ENOMEM;
+		goto err_out;
+	}
+
+	err = rxe_add_to_pool(&rxe->mr_pool, mr);
+	if (err)
+		goto err_free;
 
 	rxe_get(pd);
 	mr->ibmr.pd = ibpd;
@@ -880,8 +887,12 @@ static struct ib_mr *rxe_get_dma_mr(struct ib_pd *ibpd, int access)
 
 	rxe_mr_init_dma(access, mr);
 	rxe_finalize(mr);
-
 	return &mr->ibmr;
+
+err_free:
+	kfree(mr);
+err_out:
+	return ERR_PTR(err);
 }
 
 static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd,
@@ -895,9 +906,15 @@ static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd,
 	struct rxe_pd *pd = to_rpd(ibpd);
 	struct rxe_mr *mr;
 
-	mr = rxe_alloc(&rxe->mr_pool);
-	if (!mr)
-		return ERR_PTR(-ENOMEM);
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr) {
+		err = -ENOMEM;
+		goto err_out;
+	}
+
+	err = rxe_add_to_pool(&rxe->mr_pool, mr);
+	if (err)
+		goto err_free;
 
 	rxe_get(pd);
 	mr->ibmr.pd = ibpd;
@@ -905,14 +922,16 @@ static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd,
 
 	err = rxe_mr_init_user(rxe, start, length, iova, access, mr);
 	if (err)
-		goto err1;
+		goto err_cleanup;
 
 	rxe_finalize(mr);
-
 	return &mr->ibmr;
 
-err1:
+err_cleanup:
 	rxe_cleanup(mr);
+err_free:
+	kfree(mr);
+err_out:
 	return ERR_PTR(err);
 }
 
@@ -927,9 +946,15 @@ static struct ib_mr *rxe_alloc_mr(struct ib_pd *ibpd, enum ib_mr_type mr_type,
 	if (mr_type != IB_MR_TYPE_MEM_REG)
 		return ERR_PTR(-EINVAL);
 
-	mr = rxe_alloc(&rxe->mr_pool);
-	if (!mr)
-		return ERR_PTR(-ENOMEM);
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr) {
+		err = -ENOMEM;
+		goto err_out;
+	}
+
+	err = rxe_add_to_pool(&rxe->mr_pool, mr);
+	if (err)
+		goto err_free;
 
 	rxe_get(pd);
 	mr->ibmr.pd = ibpd;
@@ -937,53 +962,19 @@ static struct ib_mr *rxe_alloc_mr(struct ib_pd *ibpd, enum ib_mr_type mr_type,
 
 	err = rxe_mr_init_fast(max_num_sg, mr);
 	if (err)
-		goto err1;
+		goto err_cleanup;
 
 	rxe_finalize(mr);
-
 	return &mr->ibmr;
 
-err1:
+err_cleanup:
 	rxe_cleanup(mr);
+err_free:
+	kfree(mr);
+err_out:
 	return ERR_PTR(err);
 }
 
-static int rxe_set_page(struct ib_mr *ibmr, u64 addr)
-{
-	struct rxe_mr *mr = to_rmr(ibmr);
-	struct rxe_map *map;
-	struct rxe_phys_buf *buf;
-
-	if (unlikely(mr->nbuf == mr->num_buf))
-		return -ENOMEM;
-
-	map = mr->map[mr->nbuf / RXE_BUF_PER_MAP];
-	buf = &map->buf[mr->nbuf % RXE_BUF_PER_MAP];
-
-	buf->addr = addr;
-	buf->size = ibmr->page_size;
-	mr->nbuf++;
-
-	return 0;
-}
-
-static int rxe_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
-			 int sg_nents, unsigned int *sg_offset)
-{
-	struct rxe_mr *mr = to_rmr(ibmr);
-	int n;
-
-	mr->nbuf = 0;
-
-	n = ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, rxe_set_page);
-
-	mr->page_shift = ilog2(ibmr->page_size);
-	mr->page_mask = ibmr->page_size - 1;
-	mr->offset = ibmr->iova & mr->page_mask;
-
-	return n;
-}
-
 static ssize_t parent_show(struct device *device,
 			   struct device_attribute *attr, char *buf)
 {
diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.h b/drivers/infiniband/sw/rxe/rxe_verbs.h
index 19ddfa890480..c269ae2a3224 100644
--- a/drivers/infiniband/sw/rxe/rxe_verbs.h
+++ b/drivers/infiniband/sw/rxe/rxe_verbs.h
@@ -283,17 +283,6 @@ enum rxe_mr_lookup_type {
 	RXE_LOOKUP_REMOTE,
 };
 
-#define RXE_BUF_PER_MAP		(PAGE_SIZE / sizeof(struct rxe_phys_buf))
-
-struct rxe_phys_buf {
-	u64      addr;
-	u64      size;
-};
-
-struct rxe_map {
-	struct rxe_phys_buf	buf[RXE_BUF_PER_MAP];
-};
-
 static inline int rkey_is_mw(u32 rkey)
 {
 	u32 index = rkey >> 8;
@@ -310,25 +299,24 @@ struct rxe_mr {
 	u32			lkey;
 	u32			rkey;
 	enum rxe_mr_state	state;
-	u32			offset;
 	int			access;
+	atomic_t		num_mw;
 
-	int			page_shift;
-	int			page_mask;
-	int			map_shift;
-	int			map_mask;
+	unsigned int		page_offset;
+	unsigned int		page_shift;
+	u64			page_mask;
 
 	u32			num_buf;
 	u32			nbuf;
 
-	u32			max_buf;
-	u32			num_map;
-
-	atomic_t		num_mw;
-
-	struct rxe_map		**map;
+	struct xarray		page_list;
 };
 
+static inline unsigned int mr_page_size(struct rxe_mr *mr)
+{
+	return mr ? mr->ibmr.page_size : PAGE_SIZE;
+}
+
 enum rxe_mw_state {
 	RXE_MW_STATE_INVALID	= RXE_MR_STATE_INVALID,
 	RXE_MW_STATE_FREE	= RXE_MR_STATE_FREE,
diff --git a/drivers/infiniband/sw/siw/siw_mem.c b/drivers/infiniband/sw/siw/siw_mem.c
index b2b33dd3b4fa..f51ab2ccf151 100644
--- a/drivers/infiniband/sw/siw/siw_mem.c
+++ b/drivers/infiniband/sw/siw/siw_mem.c
@@ -398,7 +398,7 @@ struct siw_umem *siw_umem_get(u64 start, u64 len, bool writable)
 
 	mlock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 
-	if (num_pages + atomic64_read(&mm_s->pinned_vm) > mlock_limit) {
+	if (atomic64_add_return(num_pages, &mm_s->pinned_vm) > mlock_limit) {
 		rv = -ENOMEM;
 		goto out_sem_up;
 	}
@@ -411,30 +411,27 @@ struct siw_umem *siw_umem_get(u64 start, u64 len, bool writable)
 		goto out_sem_up;
 	}
 	for (i = 0; num_pages; i++) {
-		int got, nents = min_t(int, num_pages, PAGES_PER_CHUNK);
-
-		umem->page_chunk[i].plist =
+		int nents = min_t(int, num_pages, PAGES_PER_CHUNK);
+		struct page **plist =
 			kcalloc(nents, sizeof(struct page *), GFP_KERNEL);
-		if (!umem->page_chunk[i].plist) {
+
+		if (!plist) {
 			rv = -ENOMEM;
 			goto out_sem_up;
 		}
-		got = 0;
+		umem->page_chunk[i].plist = plist;
 		while (nents) {
-			struct page **plist = &umem->page_chunk[i].plist[got];
-
 			rv = pin_user_pages(first_page_va, nents, foll_flags,
 					    plist, NULL);
 			if (rv < 0)
 				goto out_sem_up;
 
 			umem->num_pages += rv;
-			atomic64_add(rv, &mm_s->pinned_vm);
 			first_page_va += rv * PAGE_SIZE;
+			plist += rv;
 			nents -= rv;
-			got += rv;
+			num_pages -= rv;
 		}
-		num_pages -= got;
 	}
 out_sem_up:
 	mmap_read_unlock(mm_s);
@@ -442,6 +439,10 @@ out_sem_up:
 	if (rv > 0)
 		return umem;
 
+	/* Adjust accounting for pages not pinned */
+	if (num_pages)
+		atomic64_sub(num_pages, &mm_s->pinned_vm);
+
 	siw_umem_release(umem, false);
 
 	return ERR_PTR(rv);