[PATCH] IB: Refactor umem to use linear SG table

[PATCH] IB: Refactor umem to use linear SG table

[Or Gerlitz]

From: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>

This patch does refactoring of the IB core umem code and vendor drivers to use 
linear SG table instead of chunk list. With this change the relevant code becomes 
more clear, no need for nested loops to build and use umem.

The patch includes: umem changes to work with linear SG table and Vendors driver 
changes to use SG instead of chunk list.

Cc: Faisal Latif <faisal.latif-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Cc: Mike Marciniszyn <infinipath-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Cc: Hoang-Nam Nguyen <hnguyen-tA70FqPdS9bQT0dZR+AlfA@public.gmane.org>
Cc: Steve Wise <swise-ut6Up61K2wZBDgjK7y7TUQ@public.gmane.org>
Signed-off-by: Shachar Raindel <raindel-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
Signed-off-by: Yishai Hadas <yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 drivers/infiniband/core/umem.c               |  120 ++++++-------
 drivers/infiniband/hw/amso1100/c2_provider.c |   23 +--
 drivers/infiniband/hw/cxgb3/iwch_provider.c  |   19 +--
 drivers/infiniband/hw/cxgb4/mem.c            |   39 ++---
 drivers/infiniband/hw/ehca/ehca_classes.h    |    2 +-
 drivers/infiniband/hw/ehca/ehca_mrmw.c       |  245 ++++++++++---------------
 drivers/infiniband/hw/ipath/ipath_mr.c       |   39 ++---
 drivers/infiniband/hw/mlx4/doorbell.c        |    4 +-
 drivers/infiniband/hw/mlx4/mr.c              |   39 ++--
 drivers/infiniband/hw/mlx5/doorbell.c        |    4 +-
 drivers/infiniband/hw/mlx5/mem.c             |   80 ++++----
 drivers/infiniband/hw/mthca/mthca_provider.c |   42 ++---
 drivers/infiniband/hw/nes/nes_verbs.c        |  253 +++++++++++++-------------
 drivers/infiniband/hw/ocrdma/ocrdma_verbs.c  |   66 ++++----
 drivers/infiniband/hw/qib/qib_mr.c           |   14 +-
 include/rdma/ib_umem.h                       |   11 +-
 16 files changed, 448 insertions(+), 552 deletions(-)

diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index a841123..a3a2e9c 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -42,29 +42,29 @@
 
 #include "uverbs.h"
 
-#define IB_UMEM_MAX_PAGE_CHUNK						\
-	((PAGE_SIZE - offsetof(struct ib_umem_chunk, page_list)) /	\
-	 ((void *) &((struct ib_umem_chunk *) 0)->page_list[1] -	\
-	  (void *) &((struct ib_umem_chunk *) 0)->page_list[0]))
 
 static void __ib_umem_release(struct ib_device *dev, struct ib_umem *umem, int dirty)
 {
-	struct ib_umem_chunk *chunk, *tmp;
+	struct scatterlist *sg;
+	struct page *page;
 	int i;
 
-	list_for_each_entry_safe(chunk, tmp, &umem->chunk_list, list) {
-		ib_dma_unmap_sg(dev, chunk->page_list,
-				chunk->nents, DMA_BIDIRECTIONAL);
-		for (i = 0; i < chunk->nents; ++i) {
-			struct page *page = sg_page(&chunk->page_list[i]);
+	if (umem->nmap > 0)
+		ib_dma_unmap_sg(dev, umem->sg_head.sgl,
+				umem->nmap,
+				DMA_BIDIRECTIONAL);
 
-			if (umem->writable && dirty)
-				set_page_dirty_lock(page);
-			put_page(page);
-		}
+	for_each_sg(umem->sg_head.sgl, sg, umem->npages, i) {
 
-		kfree(chunk);
+		page = sg_page(sg);
+		if (umem->writable && dirty)
+			set_page_dirty_lock(page);
+		put_page(page);
 	}
+
+	sg_free_table(&umem->sg_head);
+	return;
+
 }
 
 /**
@@ -81,15 +81,15 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 	struct ib_umem *umem;
 	struct page **page_list;
 	struct vm_area_struct **vma_list;
-	struct ib_umem_chunk *chunk;
 	unsigned long locked;
 	unsigned long lock_limit;
 	unsigned long cur_base;
 	unsigned long npages;
 	int ret;
-	int off;
 	int i;
 	DEFINE_DMA_ATTRS(attrs);
+	struct scatterlist *sg, *sg_list_start;
+	int need_release = 0;
 
 	if (dmasync)
 		dma_set_attr(DMA_ATTR_WRITE_BARRIER, &attrs);
@@ -97,7 +97,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 	if (!can_do_mlock())
 		return ERR_PTR(-EPERM);
 
-	umem = kmalloc(sizeof *umem, GFP_KERNEL);
+	umem = kzalloc(sizeof *umem, GFP_KERNEL);
 	if (!umem)
 		return ERR_PTR(-ENOMEM);
 
@@ -117,8 +117,6 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 	/* We assume the memory is from hugetlb until proved otherwise */
 	umem->hugetlb   = 1;
 
-	INIT_LIST_HEAD(&umem->chunk_list);
-
 	page_list = (struct page **) __get_free_page(GFP_KERNEL);
 	if (!page_list) {
 		kfree(umem);
@@ -147,7 +145,18 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 
 	cur_base = addr & PAGE_MASK;
 
-	ret = 0;
+	if (npages == 0) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = sg_alloc_table(&umem->sg_head, npages, GFP_KERNEL);
+	if (ret)
+		goto out;
+
+	need_release = 1;
+	sg_list_start = umem->sg_head.sgl;
+
 	while (npages) {
 		ret = get_user_pages(current, current->mm, cur_base,
 				     min_t(unsigned long, npages,
@@ -157,54 +166,38 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 		if (ret < 0)
 			goto out;
 
+		umem->npages += ret;
 		cur_base += ret * PAGE_SIZE;
 		npages   -= ret;
 
-		off = 0;
-
-		while (ret) {
-			chunk = kmalloc(sizeof *chunk + sizeof (struct scatterlist) *
-					min_t(int, ret, IB_UMEM_MAX_PAGE_CHUNK),
-					GFP_KERNEL);
-			if (!chunk) {
-				ret = -ENOMEM;
-				goto out;
-			}
-
-			chunk->nents = min_t(int, ret, IB_UMEM_MAX_PAGE_CHUNK);
-			sg_init_table(chunk->page_list, chunk->nents);
-			for (i = 0; i < chunk->nents; ++i) {
-				if (vma_list &&
-				    !is_vm_hugetlb_page(vma_list[i + off]))
-					umem->hugetlb = 0;
-				sg_set_page(&chunk->page_list[i], page_list[i + off], PAGE_SIZE, 0);
-			}
-
-			chunk->nmap = ib_dma_map_sg_attrs(context->device,
-							  &chunk->page_list[0],
-							  chunk->nents,
-							  DMA_BIDIRECTIONAL,
-							  &attrs);
-			if (chunk->nmap <= 0) {
-				for (i = 0; i < chunk->nents; ++i)
-					put_page(sg_page(&chunk->page_list[i]));
-				kfree(chunk);
-
-				ret = -ENOMEM;
-				goto out;
-			}
-
-			ret -= chunk->nents;
-			off += chunk->nents;
-			list_add_tail(&chunk->list, &umem->chunk_list);
+		for_each_sg(sg_list_start, sg, ret, i) {
+			if (vma_list && !is_vm_hugetlb_page(vma_list[i]))
+				umem->hugetlb = 0;
+
+			sg_set_page(sg, page_list[i], PAGE_SIZE, 0);
 		}
 
-		ret = 0;
+		/* preparing for next loop */
+		sg_list_start = sg;
 	}
 
+	umem->nmap = ib_dma_map_sg_attrs(context->device,
+				  umem->sg_head.sgl,
+				  umem->npages,
+				  DMA_BIDIRECTIONAL,
+				  &attrs);
+
+	if (umem->nmap <= 0) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	ret = 0;
+
 out:
 	if (ret < 0) {
-		__ib_umem_release(context->device, umem, 0);
+		if (need_release)
+			__ib_umem_release(context->device, umem, 0);
 		kfree(umem);
 	} else
 		current->mm->pinned_vm = locked;
@@ -278,17 +271,16 @@ EXPORT_SYMBOL(ib_umem_release);
 
 int ib_umem_page_count(struct ib_umem *umem)
 {
-	struct ib_umem_chunk *chunk;
 	int shift;
 	int i;
 	int n;
+	struct scatterlist *sg;
 
 	shift = ilog2(umem->page_size);
 
 	n = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		for (i = 0; i < chunk->nmap; ++i)
-			n += sg_dma_len(&chunk->page_list[i]) >> shift;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, i)
+		n += sg_dma_len(sg) >> shift;
 
 	return n;
 }
diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c b/drivers/infiniband/hw/amso1100/c2_provider.c
index 07eb3a8..8af33cf 100644
--- a/drivers/infiniband/hw/amso1100/c2_provider.c
+++ b/drivers/infiniband/hw/amso1100/c2_provider.c
@@ -431,9 +431,9 @@ static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	u64 *pages;
 	u64 kva = 0;
 	int shift, n, len;
-	int i, j, k;
+	int i, k, entry;
 	int err = 0;
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct c2_pd *c2pd = to_c2pd(pd);
 	struct c2_mr *c2mr;
 
@@ -452,10 +452,7 @@ static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	shift = ffs(c2mr->umem->page_size) - 1;
-
-	n = 0;
-	list_for_each_entry(chunk, &c2mr->umem->chunk_list, list)
-		n += chunk->nents;
+	n = c2mr->umem->nmap;
 
 	pages = kmalloc(n * sizeof(u64), GFP_KERNEL);
 	if (!pages) {
@@ -464,14 +461,12 @@ static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	i = 0;
-	list_for_each_entry(chunk, &c2mr->umem->chunk_list, list) {
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> shift;
-			for (k = 0; k < len; ++k) {
-				pages[i++] =
-					sg_dma_address(&chunk->page_list[j]) +
-					(c2mr->umem->page_size * k);
-			}
+	for_each_sg(c2mr->umem->sg_head.sgl, sg, c2mr->umem->nmap, entry) {
+		len = sg_dma_len(sg) >> shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] =
+				sg_dma_address(sg) +
+				(c2mr->umem->page_size * k);
 		}
 	}
 
diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c
index d228383..811b24a 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_provider.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c
@@ -618,14 +618,13 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	__be64 *pages;
 	int shift, n, len;
-	int i, j, k;
+	int i, k, entry;
 	int err = 0;
-	struct ib_umem_chunk *chunk;
 	struct iwch_dev *rhp;
 	struct iwch_pd *php;
 	struct iwch_mr *mhp;
 	struct iwch_reg_user_mr_resp uresp;
-
+	struct scatterlist *sg;
 	PDBG("%s ib_pd %p\n", __func__, pd);
 
 	php = to_iwch_pd(pd);
@@ -645,9 +644,7 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	shift = ffs(mhp->umem->page_size) - 1;
 
-	n = 0;
-	list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
-		n += chunk->nents;
+	n = mhp->umem->nmap;
 
 	err = iwch_alloc_pbl(mhp, n);
 	if (err)
@@ -661,12 +658,10 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	i = n = 0;
 
-	list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> shift;
+	for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
+			len = sg_dma_len(sg) >> shift;
 			for (k = 0; k < len; ++k) {
-				pages[i++] = cpu_to_be64(sg_dma_address(
-					&chunk->page_list[j]) +
+				pages[i++] = cpu_to_be64(sg_dma_address(sg) +
 					mhp->umem->page_size * k);
 				if (i == PAGE_SIZE / sizeof *pages) {
 					err = iwch_write_pbl(mhp, pages, i, n);
@@ -676,7 +671,7 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 					i = 0;
 				}
 			}
-		}
+	}
 
 	if (i)
 		err = iwch_write_pbl(mhp, pages, i, n);
diff --git a/drivers/infiniband/hw/cxgb4/mem.c b/drivers/infiniband/hw/cxgb4/mem.c
index 84e4500..aa9d3f6 100644
--- a/drivers/infiniband/hw/cxgb4/mem.c
+++ b/drivers/infiniband/hw/cxgb4/mem.c
@@ -678,9 +678,9 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	__be64 *pages;
 	int shift, n, len;
-	int i, j, k;
+	int i, k, entry;
 	int err = 0;
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct c4iw_dev *rhp;
 	struct c4iw_pd *php;
 	struct c4iw_mr *mhp;
@@ -710,10 +710,7 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	shift = ffs(mhp->umem->page_size) - 1;
 
-	n = 0;
-	list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
-		n += chunk->nents;
-
+	n = mhp->umem->nmap;
 	err = alloc_pbl(mhp, n);
 	if (err)
 		goto err;
@@ -726,24 +723,22 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	i = n = 0;
 
-	list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> shift;
-			for (k = 0; k < len; ++k) {
-				pages[i++] = cpu_to_be64(sg_dma_address(
-					&chunk->page_list[j]) +
-					mhp->umem->page_size * k);
-				if (i == PAGE_SIZE / sizeof *pages) {
-					err = write_pbl(&mhp->rhp->rdev,
-					      pages,
-					      mhp->attr.pbl_addr + (n << 3), i);
-					if (err)
-						goto pbl_done;
-					n += i;
-					i = 0;
-				}
+	for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
+		len = sg_dma_len(sg) >> shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] = cpu_to_be64(sg_dma_address(sg) +
+				mhp->umem->page_size * k);
+			if (i == PAGE_SIZE / sizeof *pages) {
+				err = write_pbl(&mhp->rhp->rdev,
+				      pages,
+				      mhp->attr.pbl_addr + (n << 3), i);
+				if (err)
+					goto pbl_done;
+				n += i;
+				i = 0;
 			}
 		}
+	}
 
 	if (i)
 		err = write_pbl(&mhp->rhp->rdev, pages,
diff --git a/drivers/infiniband/hw/ehca/ehca_classes.h b/drivers/infiniband/hw/ehca/ehca_classes.h
index f08f6ea..bd45e0f 100644
--- a/drivers/infiniband/hw/ehca/ehca_classes.h
+++ b/drivers/infiniband/hw/ehca/ehca_classes.h
@@ -322,7 +322,7 @@ struct ehca_mr_pginfo {
 		} phy;
 		struct { /* type EHCA_MR_PGI_USER section */
 			struct ib_umem *region;
-			struct ib_umem_chunk *next_chunk;
+			struct scatterlist *next_sg;
 			u64 next_nmap;
 		} usr;
 		struct { /* type EHCA_MR_PGI_FMR section */
diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.c b/drivers/infiniband/hw/ehca/ehca_mrmw.c
index bcfb0c1..7168f59 100644
--- a/drivers/infiniband/hw/ehca/ehca_mrmw.c
+++ b/drivers/infiniband/hw/ehca/ehca_mrmw.c
@@ -400,10 +400,7 @@ reg_user_mr_fallback:
 	pginfo.num_hwpages = num_hwpages;
 	pginfo.u.usr.region = e_mr->umem;
 	pginfo.next_hwpage = e_mr->umem->offset / hwpage_size;
-	pginfo.u.usr.next_chunk = list_prepare_entry(pginfo.u.usr.next_chunk,
-						     (&e_mr->umem->chunk_list),
-						     list);
-
+	pginfo.u.usr.next_sg = pginfo.u.usr.region->sg_head.sgl;
 	ret = ehca_reg_mr(shca, e_mr, (u64 *)virt, length, mr_access_flags,
 			  e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
 			  &e_mr->ib.ib_mr.rkey, EHCA_REG_MR);
@@ -1858,61 +1855,39 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
 				  u64 *kpage)
 {
 	int ret = 0;
-	struct ib_umem_chunk *prev_chunk;
-	struct ib_umem_chunk *chunk;
 	u64 pgaddr;
-	u32 i = 0;
 	u32 j = 0;
 	int hwpages_per_kpage = PAGE_SIZE / pginfo->hwpage_size;
-
-	/* loop over desired chunk entries */
-	chunk      = pginfo->u.usr.next_chunk;
-	prev_chunk = pginfo->u.usr.next_chunk;
-	list_for_each_entry_continue(
-		chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
-		for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
-			pgaddr = page_to_pfn(sg_page(&chunk->page_list[i]))
-				<< PAGE_SHIFT ;
-			*kpage = pgaddr + (pginfo->next_hwpage *
-					   pginfo->hwpage_size);
-			if ( !(*kpage) ) {
-				ehca_gen_err("pgaddr=%llx "
-					     "chunk->page_list[i]=%llx "
-					     "i=%x next_hwpage=%llx",
-					     pgaddr, (u64)sg_dma_address(
-						     &chunk->page_list[i]),
-					     i, pginfo->next_hwpage);
-				return -EFAULT;
-			}
-			(pginfo->hwpage_cnt)++;
-			(pginfo->next_hwpage)++;
-			kpage++;
-			if (pginfo->next_hwpage % hwpages_per_kpage == 0) {
-				(pginfo->kpage_cnt)++;
-				(pginfo->u.usr.next_nmap)++;
-				pginfo->next_hwpage = 0;
-				i++;
-			}
-			j++;
-			if (j >= number) break;
+	struct scatterlist **sg = &pginfo->u.usr.next_sg;
+
+	while (*sg != NULL) {
+		pgaddr = page_to_pfn(sg_page(*sg))
+			<< PAGE_SHIFT;
+		*kpage = pgaddr + (pginfo->next_hwpage *
+				   pginfo->hwpage_size);
+		if (!(*kpage)) {
+			ehca_gen_err("pgaddr=%llx "
+				     "sg_dma_address=%llx "
+				     "entry=%llx next_hwpage=%llx",
+				     pgaddr, (u64)sg_dma_address(*sg),
+				     pginfo->u.usr.next_nmap,
+				     pginfo->next_hwpage);
+			return -EFAULT;
 		}
-		if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
-		    (j >= number)) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-			break;
-		} else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-		} else if (j >= number)
+		(pginfo->hwpage_cnt)++;
+		(pginfo->next_hwpage)++;
+		kpage++;
+		if (pginfo->next_hwpage % hwpages_per_kpage == 0) {
+			(pginfo->kpage_cnt)++;
+			(pginfo->u.usr.next_nmap)++;
+			pginfo->next_hwpage = 0;
+			*sg = sg_next(*sg);
+		}
+		j++;
+		if (j >= number)
 			break;
-		else
-			prev_chunk = chunk;
 	}
-	pginfo->u.usr.next_chunk =
-		list_prepare_entry(prev_chunk,
-				   (&(pginfo->u.usr.region->chunk_list)),
-				   list);
+
 	return ret;
 }
 
@@ -1920,20 +1895,19 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
  * check given pages for contiguous layout
  * last page addr is returned in prev_pgaddr for further check
  */
-static int ehca_check_kpages_per_ate(struct scatterlist *page_list,
-				     int start_idx, int end_idx,
+static int ehca_check_kpages_per_ate(struct scatterlist **sg,
+				     int num_pages,
 				     u64 *prev_pgaddr)
 {
-	int t;
-	for (t = start_idx; t <= end_idx; t++) {
-		u64 pgaddr = page_to_pfn(sg_page(&page_list[t])) << PAGE_SHIFT;
+	for (; *sg && num_pages > 0; *sg = sg_next(*sg), num_pages--) {
+		u64 pgaddr = page_to_pfn(sg_page(*sg)) << PAGE_SHIFT;
 		if (ehca_debug_level >= 3)
 			ehca_gen_dbg("chunk_page=%llx value=%016llx", pgaddr,
 				     *(u64 *)__va(pgaddr));
 		if (pgaddr - PAGE_SIZE != *prev_pgaddr) {
 			ehca_gen_err("uncontiguous page found pgaddr=%llx "
-				     "prev_pgaddr=%llx page_list_i=%x",
-				     pgaddr, *prev_pgaddr, t);
+				     "prev_pgaddr=%llx entries_left_in_hwpage=%x",
+				     pgaddr, *prev_pgaddr, num_pages);
 			return -EINVAL;
 		}
 		*prev_pgaddr = pgaddr;
@@ -1947,111 +1921,80 @@ static int ehca_set_pagebuf_user2(struct ehca_mr_pginfo *pginfo,
 				  u64 *kpage)
 {
 	int ret = 0;
-	struct ib_umem_chunk *prev_chunk;
-	struct ib_umem_chunk *chunk;
 	u64 pgaddr, prev_pgaddr;
-	u32 i = 0;
 	u32 j = 0;
 	int kpages_per_hwpage = pginfo->hwpage_size / PAGE_SIZE;
 	int nr_kpages = kpages_per_hwpage;
+	struct scatterlist **sg = &pginfo->u.usr.next_sg;
+
+	while (*sg != NULL) {
 
-	/* loop over desired chunk entries */
-	chunk      = pginfo->u.usr.next_chunk;
-	prev_chunk = pginfo->u.usr.next_chunk;
-	list_for_each_entry_continue(
-		chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
-		for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
-			if (nr_kpages == kpages_per_hwpage) {
-				pgaddr = ( page_to_pfn(sg_page(&chunk->page_list[i]))
-					   << PAGE_SHIFT );
-				*kpage = pgaddr;
-				if ( !(*kpage) ) {
-					ehca_gen_err("pgaddr=%llx i=%x",
-						     pgaddr, i);
+		if (nr_kpages == kpages_per_hwpage) {
+			pgaddr = (page_to_pfn(sg_page(*sg))
+				   << PAGE_SHIFT);
+			*kpage = pgaddr;
+			if (!(*kpage)) {
+				ehca_gen_err("pgaddr=%llx entry=%llx",
+					     pgaddr, pginfo->u.usr.next_nmap);
+				ret = -EFAULT;
+				return ret;
+			}
+			/*
+			 * The first page in a hwpage must be aligned;
+			 * the first MR page is exempt from this rule.
+			 */
+			if (pgaddr & (pginfo->hwpage_size - 1)) {
+				if (pginfo->hwpage_cnt) {
+					ehca_gen_err(
+						"invalid alignment "
+						"pgaddr=%llx entry=%llx "
+						"mr_pgsize=%llx",
+						pgaddr, pginfo->u.usr.next_nmap,
+						pginfo->hwpage_size);
 					ret = -EFAULT;
 					return ret;
 				}
-				/*
-				 * The first page in a hwpage must be aligned;
-				 * the first MR page is exempt from this rule.
-				 */
-				if (pgaddr & (pginfo->hwpage_size - 1)) {
-					if (pginfo->hwpage_cnt) {
-						ehca_gen_err(
-							"invalid alignment "
-							"pgaddr=%llx i=%x "
-							"mr_pgsize=%llx",
-							pgaddr, i,
-							pginfo->hwpage_size);
-						ret = -EFAULT;
-						return ret;
-					}
-					/* first MR page */
-					pginfo->kpage_cnt =
-						(pgaddr &
-						 (pginfo->hwpage_size - 1)) >>
-						PAGE_SHIFT;
-					nr_kpages -= pginfo->kpage_cnt;
-					*kpage = pgaddr &
-						 ~(pginfo->hwpage_size - 1);
-				}
-				if (ehca_debug_level >= 3) {
-					u64 val = *(u64 *)__va(pgaddr);
-					ehca_gen_dbg("kpage=%llx chunk_page=%llx "
-						     "value=%016llx",
-						     *kpage, pgaddr, val);
-				}
-				prev_pgaddr = pgaddr;
-				i++;
-				pginfo->kpage_cnt++;
-				pginfo->u.usr.next_nmap++;
-				nr_kpages--;
-				if (!nr_kpages)
-					goto next_kpage;
-				continue;
+				/* first MR page */
+				pginfo->kpage_cnt =
+					(pgaddr &
+					 (pginfo->hwpage_size - 1)) >>
+					PAGE_SHIFT;
+				nr_kpages -= pginfo->kpage_cnt;
+				*kpage = pgaddr &
+					 ~(pginfo->hwpage_size - 1);
 			}
-			if (i + nr_kpages > chunk->nmap) {
-				ret = ehca_check_kpages_per_ate(
-					chunk->page_list, i,
-					chunk->nmap - 1, &prev_pgaddr);
-				if (ret) return ret;
-				pginfo->kpage_cnt += chunk->nmap - i;
-				pginfo->u.usr.next_nmap += chunk->nmap - i;
-				nr_kpages -= chunk->nmap - i;
-				break;
+			if (ehca_debug_level >= 3) {
+				u64 val = *(u64 *)__va(pgaddr);
+				ehca_gen_dbg("kpage=%llx page=%llx "
+					     "value=%016llx",
+					     *kpage, pgaddr, val);
 			}
+			prev_pgaddr = pgaddr;
+			*sg = sg_next(*sg);
+			pginfo->kpage_cnt++;
+			pginfo->u.usr.next_nmap++;
+			nr_kpages--;
+			if (!nr_kpages)
+				goto next_kpage;
+			continue;
+		}
+
+		ret = ehca_check_kpages_per_ate(sg, nr_kpages,
+						&prev_pgaddr);
+		if (ret)
+			return ret;
+		pginfo->kpage_cnt += nr_kpages;
+		pginfo->u.usr.next_nmap += nr_kpages;
 
-			ret = ehca_check_kpages_per_ate(chunk->page_list, i,
-							i + nr_kpages - 1,
-							&prev_pgaddr);
-			if (ret) return ret;
-			i += nr_kpages;
-			pginfo->kpage_cnt += nr_kpages;
-			pginfo->u.usr.next_nmap += nr_kpages;
 next_kpage:
-			nr_kpages = kpages_per_hwpage;
-			(pginfo->hwpage_cnt)++;
-			kpage++;
-			j++;
-			if (j >= number) break;
-		}
-		if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
-		    (j >= number)) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-			break;
-		} else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-		} else if (j >= number)
+		nr_kpages = kpages_per_hwpage;
+		(pginfo->hwpage_cnt)++;
+		kpage++;
+		j++;
+		if (j >= number)
 			break;
-		else
-			prev_chunk = chunk;
 	}
-	pginfo->u.usr.next_chunk =
-		list_prepare_entry(prev_chunk,
-				   (&(pginfo->u.usr.region->chunk_list)),
-				   list);
+
 	return ret;
 }
 
diff --git a/drivers/infiniband/hw/ipath/ipath_mr.c b/drivers/infiniband/hw/ipath/ipath_mr.c
index e346d38..5e61e9b 100644
--- a/drivers/infiniband/hw/ipath/ipath_mr.c
+++ b/drivers/infiniband/hw/ipath/ipath_mr.c
@@ -188,8 +188,8 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	struct ipath_mr *mr;
 	struct ib_umem *umem;
-	struct ib_umem_chunk *chunk;
-	int n, m, i;
+	int n, m, entry;
+	struct scatterlist *sg;
 	struct ib_mr *ret;
 
 	if (length == 0) {
@@ -202,10 +202,7 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	if (IS_ERR(umem))
 		return (void *) umem;
 
-	n = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		n += chunk->nents;
-
+	n = umem->nmap;
 	mr = alloc_mr(n, &to_idev(pd->device)->lk_table);
 	if (!mr) {
 		ret = ERR_PTR(-ENOMEM);
@@ -224,22 +221,20 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	m = 0;
 	n = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list) {
-		for (i = 0; i < chunk->nents; i++) {
-			void *vaddr;
-
-			vaddr = page_address(sg_page(&chunk->page_list[i]));
-			if (!vaddr) {
-				ret = ERR_PTR(-EINVAL);
-				goto bail;
-			}
-			mr->mr.map[m]->segs[n].vaddr = vaddr;
-			mr->mr.map[m]->segs[n].length = umem->page_size;
-			n++;
-			if (n == IPATH_SEGSZ) {
-				m++;
-				n = 0;
-			}
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		void *vaddr;
+
+		vaddr = page_address(sg_page(sg));
+		if (!vaddr) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+		mr->mr.map[m]->segs[n].vaddr = vaddr;
+		mr->mr.map[m]->segs[n].length = umem->page_size;
+		n++;
+		if (n == IPATH_SEGSZ) {
+			m++;
+			n = 0;
 		}
 	}
 	ret = &mr->ibmr;
diff --git a/drivers/infiniband/hw/mlx4/doorbell.c b/drivers/infiniband/hw/mlx4/doorbell.c
index 8aee423..c517409 100644
--- a/drivers/infiniband/hw/mlx4/doorbell.c
+++ b/drivers/infiniband/hw/mlx4/doorbell.c
@@ -45,7 +45,6 @@ int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt,
 			struct mlx4_db *db)
 {
 	struct mlx4_ib_user_db_page *page;
-	struct ib_umem_chunk *chunk;
 	int err = 0;
 
 	mutex_lock(&context->db_page_mutex);
@@ -73,8 +72,7 @@ int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt,
 	list_add(&page->list, &context->db_page_list);
 
 found:
-	chunk = list_entry(page->umem->chunk_list.next, struct ib_umem_chunk, list);
-	db->dma		= sg_dma_address(chunk->page_list) + (virt & ~PAGE_MASK);
+	db->dma = sg_dma_address(page->umem->sg_head.sgl) + (virt & ~PAGE_MASK);
 	db->u.user_page = page;
 	++page->refcnt;
 
diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
index e471f08..cb2a872 100644
--- a/drivers/infiniband/hw/mlx4/mr.c
+++ b/drivers/infiniband/hw/mlx4/mr.c
@@ -90,11 +90,11 @@ int mlx4_ib_umem_write_mtt(struct mlx4_ib_dev *dev, struct mlx4_mtt *mtt,
 			   struct ib_umem *umem)
 {
 	u64 *pages;
-	struct ib_umem_chunk *chunk;
-	int i, j, k;
+	int i, k, entry;
 	int n;
 	int len;
 	int err = 0;
+	struct scatterlist *sg;
 
 	pages = (u64 *) __get_free_page(GFP_KERNEL);
 	if (!pages)
@@ -102,26 +102,25 @@ int mlx4_ib_umem_write_mtt(struct mlx4_ib_dev *dev, struct mlx4_mtt *mtt,
 
 	i = n = 0;
 
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> mtt->page_shift;
-			for (k = 0; k < len; ++k) {
-				pages[i++] = sg_dma_address(&chunk->page_list[j]) +
-					umem->page_size * k;
-				/*
-				 * Be friendly to mlx4_write_mtt() and
-				 * pass it chunks of appropriate size.
-				 */
-				if (i == PAGE_SIZE / sizeof (u64)) {
-					err = mlx4_write_mtt(dev->dev, mtt, n,
-							     i, pages);
-					if (err)
-						goto out;
-					n += i;
-					i = 0;
-				}
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> mtt->page_shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] = sg_dma_address(sg) +
+				umem->page_size * k;
+			/*
+			 * Be friendly to mlx4_write_mtt() and
+			 * pass it chunks of appropriate size.
+			 */
+			if (i == PAGE_SIZE / sizeof (u64)) {
+				err = mlx4_write_mtt(dev->dev, mtt, n,
+						     i, pages);
+				if (err)
+					goto out;
+				n += i;
+				i = 0;
 			}
 		}
+	}
 
 	if (i)
 		err = mlx4_write_mtt(dev->dev, mtt, n, i, pages);
diff --git a/drivers/infiniband/hw/mlx5/doorbell.c b/drivers/infiniband/hw/mlx5/doorbell.c
index 256a233..ece028f 100644
--- a/drivers/infiniband/hw/mlx5/doorbell.c
+++ b/drivers/infiniband/hw/mlx5/doorbell.c
@@ -47,7 +47,6 @@ int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
 			struct mlx5_db *db)
 {
 	struct mlx5_ib_user_db_page *page;
-	struct ib_umem_chunk *chunk;
 	int err = 0;
 
 	mutex_lock(&context->db_page_mutex);
@@ -75,8 +74,7 @@ int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
 	list_add(&page->list, &context->db_page_list);
 
 found:
-	chunk = list_entry(page->umem->chunk_list.next, struct ib_umem_chunk, list);
-	db->dma		= sg_dma_address(chunk->page_list) + (virt & ~PAGE_MASK);
+	db->dma = sg_dma_address(page->umem->sg_head.sgl) + (virt & ~PAGE_MASK);
 	db->u.user_page = page;
 	++page->refcnt;
 
diff --git a/drivers/infiniband/hw/mlx5/mem.c b/drivers/infiniband/hw/mlx5/mem.c
index 3a53228..8499aec 100644
--- a/drivers/infiniband/hw/mlx5/mem.c
+++ b/drivers/infiniband/hw/mlx5/mem.c
@@ -44,16 +44,17 @@
 void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift,
 			int *ncont, int *order)
 {
-	struct ib_umem_chunk *chunk;
 	unsigned long tmp;
 	unsigned long m;
-	int i, j, k;
+	int i, k;
 	u64 base = 0;
 	int p = 0;
 	int skip;
 	int mask;
 	u64 len;
 	u64 pfn;
+	struct scatterlist *sg;
+	int entry;
 
 	addr = addr >> PAGE_SHIFT;
 	tmp = (unsigned long)addr;
@@ -61,32 +62,31 @@ void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift,
 	skip = 1 << m;
 	mask = skip - 1;
 	i = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; j++) {
-			len = sg_dma_len(&chunk->page_list[j]) >> PAGE_SHIFT;
-			pfn = sg_dma_address(&chunk->page_list[j]) >> PAGE_SHIFT;
-			for (k = 0; k < len; k++) {
-				if (!(i & mask)) {
-					tmp = (unsigned long)pfn;
-					m = min(m, find_first_bit(&tmp, sizeof(tmp)));
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> PAGE_SHIFT;
+		pfn = sg_dma_address(sg) >> PAGE_SHIFT;
+		for (k = 0; k < len; k++) {
+			if (!(i & mask)) {
+				tmp = (unsigned long)pfn;
+				m = min(m, find_first_bit(&tmp, sizeof(tmp)));
+				skip = 1 << m;
+				mask = skip - 1;
+				base = pfn;
+				p = 0;
+			} else {
+				if (base + p != pfn) {
+					tmp = (unsigned long)p;
+					m = find_first_bit(&tmp, sizeof(tmp));
 					skip = 1 << m;
 					mask = skip - 1;
 					base = pfn;
 					p = 0;
-				} else {
-					if (base + p != pfn) {
-						tmp = (unsigned long)p;
-						m = find_first_bit(&tmp, sizeof(tmp));
-						skip = 1 << m;
-						mask = skip - 1;
-						base = pfn;
-						p = 0;
-					}
 				}
-				p++;
-				i++;
 			}
+			p++;
+			i++;
 		}
+	}
 
 	if (i) {
 		m = min_t(unsigned long, ilog2(roundup_pow_of_two(i)), m);
@@ -112,32 +112,32 @@ void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
 {
 	int shift = page_shift - PAGE_SHIFT;
 	int mask = (1 << shift) - 1;
-	struct ib_umem_chunk *chunk;
-	int i, j, k;
+	int i, k;
 	u64 cur = 0;
 	u64 base;
 	int len;
+	struct scatterlist *sg;
+	int entry;
 
 	i = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; j++) {
-			len = sg_dma_len(&chunk->page_list[j]) >> PAGE_SHIFT;
-			base = sg_dma_address(&chunk->page_list[j]);
-			for (k = 0; k < len; k++) {
-				if (!(i & mask)) {
-					cur = base + (k << PAGE_SHIFT);
-					if (umr)
-						cur |= 3;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> PAGE_SHIFT;
+		base = sg_dma_address(sg);
+		for (k = 0; k < len; k++) {
+			if (!(i & mask)) {
+				cur = base + (k << PAGE_SHIFT);
+				if (umr)
+					cur |= 3;
 
-					pas[i >> shift] = cpu_to_be64(cur);
-					mlx5_ib_dbg(dev, "pas[%d] 0x%llx\n",
-						    i >> shift, be64_to_cpu(pas[i >> shift]));
-				}  else
-					mlx5_ib_dbg(dev, "=====> 0x%llx\n",
-						    base + (k << PAGE_SHIFT));
-				i++;
-			}
+				pas[i >> shift] = cpu_to_be64(cur);
+				mlx5_ib_dbg(dev, "pas[%d] 0x%llx\n",
+					    i >> shift, be64_to_cpu(pas[i >> shift]));
+			}  else
+				mlx5_ib_dbg(dev, "=====> 0x%llx\n",
+					    base + (k << PAGE_SHIFT));
+			i++;
 		}
+	}
 }
 
 int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset)
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 5b71d43..6440800 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -976,12 +976,12 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				       u64 virt, int acc, struct ib_udata *udata)
 {
 	struct mthca_dev *dev = to_mdev(pd->device);
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct mthca_mr *mr;
 	struct mthca_reg_mr ucmd;
 	u64 *pages;
 	int shift, n, len;
-	int i, j, k;
+	int i, k, entry;
 	int err = 0;
 	int write_mtt_size;
 
@@ -1009,10 +1009,7 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	shift = ffs(mr->umem->page_size) - 1;
-
-	n = 0;
-	list_for_each_entry(chunk, &mr->umem->chunk_list, list)
-		n += chunk->nents;
+	n = mr->umem->nmap;
 
 	mr->mtt = mthca_alloc_mtt(dev, n);
 	if (IS_ERR(mr->mtt)) {
@@ -1030,25 +1027,24 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
 
-	list_for_each_entry(chunk, &mr->umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> shift;
-			for (k = 0; k < len; ++k) {
-				pages[i++] = sg_dma_address(&chunk->page_list[j]) +
-					mr->umem->page_size * k;
-				/*
-				 * Be friendly to write_mtt and pass it chunks
-				 * of appropriate size.
-				 */
-				if (i == write_mtt_size) {
-					err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
-					if (err)
-						goto mtt_done;
-					n += i;
-					i = 0;
-				}
+	for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
+		len = sg_dma_len(sg) >> shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] = sg_dma_address(sg) +
+				mr->umem->page_size * k;
+			/*
+			 * Be friendly to write_mtt and pass it chunks
+			 * of appropriate size.
+			 */
+			if (i == write_mtt_size) {
+				err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
+				if (err)
+					goto mtt_done;
+				n += i;
+				i = 0;
 			}
 		}
+	}
 
 	if (i)
 		err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c
index 8308e36..32d3682 100644
--- a/drivers/infiniband/hw/nes/nes_verbs.c
+++ b/drivers/infiniband/hw/nes/nes_verbs.c
@@ -2307,7 +2307,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	struct ib_mr *ibmr = ERR_PTR(-EINVAL);
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct nes_ucontext *nes_ucontext;
 	struct nes_pbl *nespbl;
 	struct nes_mr *nesmr;
@@ -2315,7 +2315,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	struct nes_mem_reg_req req;
 	struct nes_vpbl vpbl;
 	struct nes_root_vpbl root_vpbl;
-	int nmap_index, page_index;
+	int entry, page_index;
 	int page_count = 0;
 	int err, pbl_depth = 0;
 	int chunk_pages;
@@ -2330,6 +2330,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	u16 pbl_count;
 	u8 single_page = 1;
 	u8 stag_key;
+	int first_page = 1;
 
 	region = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 	if (IS_ERR(region)) {
@@ -2380,128 +2381,125 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			}
 			nesmr->region = region;
 
-			list_for_each_entry(chunk, &region->chunk_list, list) {
-				nes_debug(NES_DBG_MR, "Chunk: nents = %u, nmap = %u .\n",
-						chunk->nents, chunk->nmap);
-				for (nmap_index = 0; nmap_index < chunk->nmap; ++nmap_index) {
-					if (sg_dma_address(&chunk->page_list[nmap_index]) & ~PAGE_MASK) {
-						ib_umem_release(region);
-						nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
-						nes_debug(NES_DBG_MR, "Unaligned Memory Buffer: 0x%x\n",
-								(unsigned int) sg_dma_address(&chunk->page_list[nmap_index]));
-						ibmr = ERR_PTR(-EINVAL);
-						kfree(nesmr);
-						goto reg_user_mr_err;
-					}
+			for_each_sg(region->sg_head.sgl, sg, region->nmap, entry) {
+				if (sg_dma_address(sg) & ~PAGE_MASK) {
+					ib_umem_release(region);
+					nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
+					nes_debug(NES_DBG_MR, "Unaligned Memory Buffer: 0x%x\n",
+						  (unsigned int) sg_dma_address(sg));
+					ibmr = ERR_PTR(-EINVAL);
+					kfree(nesmr);
+					goto reg_user_mr_err;
+				}
 
-					if (!sg_dma_len(&chunk->page_list[nmap_index])) {
-						ib_umem_release(region);
-						nes_free_resource(nesadapter, nesadapter->allocated_mrs,
-								stag_index);
-						nes_debug(NES_DBG_MR, "Invalid Buffer Size\n");
-						ibmr = ERR_PTR(-EINVAL);
-						kfree(nesmr);
-						goto reg_user_mr_err;
-					}
+				if (!sg_dma_len(sg)) {
+					ib_umem_release(region);
+					nes_free_resource(nesadapter, nesadapter->allocated_mrs,
+							  stag_index);
+					nes_debug(NES_DBG_MR, "Invalid Buffer Size\n");
+					ibmr = ERR_PTR(-EINVAL);
+					kfree(nesmr);
+					goto reg_user_mr_err;
+				}
 
-					region_length += sg_dma_len(&chunk->page_list[nmap_index]);
-					chunk_pages = sg_dma_len(&chunk->page_list[nmap_index]) >> 12;
-					region_length -= skip_pages << 12;
-					for (page_index=skip_pages; page_index < chunk_pages; page_index++) {
-						skip_pages = 0;
-						if ((page_count!=0)&&(page_count<<12)-(region->offset&(4096-1))>=region->length)
-							goto enough_pages;
-						if ((page_count&0x01FF) == 0) {
-							if (page_count >= 1024 * 512) {
+				region_length += sg_dma_len(sg);
+				chunk_pages = sg_dma_len(sg) >> 12;
+				region_length -= skip_pages << 12;
+				for (page_index = skip_pages; page_index < chunk_pages; page_index++) {
+					skip_pages = 0;
+					if ((page_count != 0) && (page_count<<12)-(region->offset&(4096-1)) >= region->length)
+						goto enough_pages;
+					if ((page_count&0x01FF) == 0) {
+						if (page_count >= 1024 * 512) {
+							ib_umem_release(region);
+							nes_free_resource(nesadapter,
+									  nesadapter->allocated_mrs, stag_index);
+							kfree(nesmr);
+							ibmr = ERR_PTR(-E2BIG);
+							goto reg_user_mr_err;
+						}
+						if (root_pbl_index == 1) {
+							root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev,
+									8192, &root_vpbl.pbl_pbase);
+							nes_debug(NES_DBG_MR, "Allocating root PBL, va = %p, pa = 0x%08X\n",
+								  root_vpbl.pbl_vbase, (unsigned int)root_vpbl.pbl_pbase);
+							if (!root_vpbl.pbl_vbase) {
 								ib_umem_release(region);
-								nes_free_resource(nesadapter,
-										nesadapter->allocated_mrs, stag_index);
+								pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
+										    vpbl.pbl_pbase);
+								nes_free_resource(nesadapter, nesadapter->allocated_mrs,
+										  stag_index);
 								kfree(nesmr);
-								ibmr = ERR_PTR(-E2BIG);
+								ibmr = ERR_PTR(-ENOMEM);
 								goto reg_user_mr_err;
 							}
-							if (root_pbl_index == 1) {
-								root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev,
-										8192, &root_vpbl.pbl_pbase);
-								nes_debug(NES_DBG_MR, "Allocating root PBL, va = %p, pa = 0x%08X\n",
-										root_vpbl.pbl_vbase, (unsigned int)root_vpbl.pbl_pbase);
-								if (!root_vpbl.pbl_vbase) {
-									ib_umem_release(region);
-									pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
-											vpbl.pbl_pbase);
-									nes_free_resource(nesadapter, nesadapter->allocated_mrs,
-											stag_index);
-									kfree(nesmr);
-									ibmr = ERR_PTR(-ENOMEM);
-									goto reg_user_mr_err;
-								}
-								root_vpbl.leaf_vpbl = kzalloc(sizeof(*root_vpbl.leaf_vpbl)*1024,
-										GFP_KERNEL);
-								if (!root_vpbl.leaf_vpbl) {
-									ib_umem_release(region);
-									pci_free_consistent(nesdev->pcidev, 8192, root_vpbl.pbl_vbase,
-											root_vpbl.pbl_pbase);
-									pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
-											vpbl.pbl_pbase);
-									nes_free_resource(nesadapter, nesadapter->allocated_mrs,
-											stag_index);
-									kfree(nesmr);
-									ibmr = ERR_PTR(-ENOMEM);
-									goto reg_user_mr_err;
-								}
-								root_vpbl.pbl_vbase[0].pa_low =
-										cpu_to_le32((u32)vpbl.pbl_pbase);
-								root_vpbl.pbl_vbase[0].pa_high =
-										cpu_to_le32((u32)((((u64)vpbl.pbl_pbase) >> 32)));
-								root_vpbl.leaf_vpbl[0] = vpbl;
-							}
-							vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096,
-									&vpbl.pbl_pbase);
-							nes_debug(NES_DBG_MR, "Allocating leaf PBL, va = %p, pa = 0x%08X\n",
-									vpbl.pbl_vbase, (unsigned int)vpbl.pbl_pbase);
-							if (!vpbl.pbl_vbase) {
+							root_vpbl.leaf_vpbl = kzalloc(sizeof(*root_vpbl.leaf_vpbl)*1024,
+									GFP_KERNEL);
+							if (!root_vpbl.leaf_vpbl) {
 								ib_umem_release(region);
-								nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
-								ibmr = ERR_PTR(-ENOMEM);
+								pci_free_consistent(nesdev->pcidev, 8192, root_vpbl.pbl_vbase,
+										    root_vpbl.pbl_pbase);
+								pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
+										    vpbl.pbl_pbase);
+								nes_free_resource(nesadapter, nesadapter->allocated_mrs,
+										  stag_index);
 								kfree(nesmr);
+								ibmr = ERR_PTR(-ENOMEM);
 								goto reg_user_mr_err;
 							}
-							if (1 <= root_pbl_index) {
-								root_vpbl.pbl_vbase[root_pbl_index].pa_low =
-										cpu_to_le32((u32)vpbl.pbl_pbase);
-								root_vpbl.pbl_vbase[root_pbl_index].pa_high =
-										cpu_to_le32((u32)((((u64)vpbl.pbl_pbase)>>32)));
-								root_vpbl.leaf_vpbl[root_pbl_index] = vpbl;
-							}
-							root_pbl_index++;
-							cur_pbl_index = 0;
+							root_vpbl.pbl_vbase[0].pa_low =
+									cpu_to_le32((u32)vpbl.pbl_pbase);
+							root_vpbl.pbl_vbase[0].pa_high =
+									cpu_to_le32((u32)((((u64)vpbl.pbl_pbase) >> 32)));
+							root_vpbl.leaf_vpbl[0] = vpbl;
 						}
-						if (single_page) {
-							if (page_count != 0) {
-								if ((last_dma_addr+4096) !=
-										(sg_dma_address(&chunk->page_list[nmap_index])+
-										(page_index*4096)))
-									single_page = 0;
-								last_dma_addr = sg_dma_address(&chunk->page_list[nmap_index])+
-										(page_index*4096);
-							} else {
-								first_dma_addr = sg_dma_address(&chunk->page_list[nmap_index])+
-										(page_index*4096);
-								last_dma_addr = first_dma_addr;
-							}
+						vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096,
+								&vpbl.pbl_pbase);
+						nes_debug(NES_DBG_MR, "Allocating leaf PBL, va = %p, pa = 0x%08X\n",
+							  vpbl.pbl_vbase, (unsigned int)vpbl.pbl_pbase);
+						if (!vpbl.pbl_vbase) {
+							ib_umem_release(region);
+							nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
+							ibmr = ERR_PTR(-ENOMEM);
+							kfree(nesmr);
+							goto reg_user_mr_err;
+						}
+						if (1 <= root_pbl_index) {
+							root_vpbl.pbl_vbase[root_pbl_index].pa_low =
+									cpu_to_le32((u32)vpbl.pbl_pbase);
+							root_vpbl.pbl_vbase[root_pbl_index].pa_high =
+									cpu_to_le32((u32)((((u64)vpbl.pbl_pbase)>>32)));
+							root_vpbl.leaf_vpbl[root_pbl_index] = vpbl;
+						}
+						root_pbl_index++;
+						cur_pbl_index = 0;
+					}
+					if (single_page) {
+						if (page_count != 0) {
+							if ((last_dma_addr+4096) !=
+									(sg_dma_address(sg)+
+									(page_index*4096)))
+								single_page = 0;
+							last_dma_addr = sg_dma_address(sg)+
+									(page_index*4096);
+						} else {
+							first_dma_addr = sg_dma_address(sg)+
+									(page_index*4096);
+							last_dma_addr = first_dma_addr;
 						}
-
-						vpbl.pbl_vbase[cur_pbl_index].pa_low =
-								cpu_to_le32((u32)(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096)));
-						vpbl.pbl_vbase[cur_pbl_index].pa_high =
-								cpu_to_le32((u32)((((u64)(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096))) >> 32)));
-						cur_pbl_index++;
-						page_count++;
 					}
+
+					vpbl.pbl_vbase[cur_pbl_index].pa_low =
+							cpu_to_le32((u32)(sg_dma_address(sg)+
+							(page_index*4096)));
+					vpbl.pbl_vbase[cur_pbl_index].pa_high =
+							cpu_to_le32((u32)((((u64)(sg_dma_address(sg)+
+							(page_index*4096))) >> 32)));
+					cur_pbl_index++;
+					page_count++;
 				}
 			}
+
 			enough_pages:
 			nes_debug(NES_DBG_MR, "calculating stag, stag_index=0x%08x, driver_key=0x%08x,"
 					" stag_key=0x%08x\n",
@@ -2613,25 +2611,28 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				  nespbl->pbl_size, (unsigned long) nespbl->pbl_pbase,
 				  (void *) nespbl->pbl_vbase, nespbl->user_base);
 
-			list_for_each_entry(chunk, &region->chunk_list, list) {
-				for (nmap_index = 0; nmap_index < chunk->nmap; ++nmap_index) {
-					chunk_pages = sg_dma_len(&chunk->page_list[nmap_index]) >> 12;
-					chunk_pages += (sg_dma_len(&chunk->page_list[nmap_index]) & (4096-1)) ? 1 : 0;
-					nespbl->page = sg_page(&chunk->page_list[0]);
-					for (page_index=0; page_index<chunk_pages; page_index++) {
-						((__le32 *)pbl)[0] = cpu_to_le32((u32)
-								(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096)));
-						((__le32 *)pbl)[1] = cpu_to_le32(((u64)
-								(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096)))>>32);
-						nes_debug(NES_DBG_MR, "pbl=%p, *pbl=0x%016llx, 0x%08x%08x\n", pbl,
-								(unsigned long long)*pbl,
-								le32_to_cpu(((__le32 *)pbl)[1]), le32_to_cpu(((__le32 *)pbl)[0]));
-						pbl++;
-					}
+			for_each_sg(region->sg_head.sgl, sg, region->nmap, entry) {
+				chunk_pages = sg_dma_len(sg) >> 12;
+				chunk_pages += (sg_dma_len(sg) & (4096-1)) ? 1 : 0;
+				if (first_page) {
+					nespbl->page = sg_page(sg);
+					first_page = 0;
+				}
+
+				for (page_index = 0; page_index < chunk_pages; page_index++) {
+					((__le32 *)pbl)[0] = cpu_to_le32((u32)
+							(sg_dma_address(sg)+
+							(page_index*4096)));
+					((__le32 *)pbl)[1] = cpu_to_le32(((u64)
+							(sg_dma_address(sg)+
+							(page_index*4096)))>>32);
+					nes_debug(NES_DBG_MR, "pbl=%p, *pbl=0x%016llx, 0x%08x%08x\n", pbl,
+						  (unsigned long long)*pbl,
+						  le32_to_cpu(((__le32 *)pbl)[1]), le32_to_cpu(((__le32 *)pbl)[0]));
+					pbl++;
 				}
 			}
+
 			if (req.reg_type == IWNES_MEMREG_TYPE_QP) {
 				list_add_tail(&nespbl->list, &nes_ucontext->qp_reg_mem_list);
 			} else {
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
index 7686dce..50e42d3 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -726,10 +726,10 @@ static void build_user_pbes(struct ocrdma_dev *dev, struct ocrdma_mr *mr,
 			    u32 num_pbes)
 {
 	struct ocrdma_pbe *pbe;
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct ocrdma_pbl *pbl_tbl = mr->hwmr.pbl_table;
 	struct ib_umem *umem = mr->umem;
-	int i, shift, pg_cnt, pages, pbe_cnt, total_num_pbes = 0;
+	int shift, pg_cnt, pages, pbe_cnt, entry, total_num_pbes = 0;
 
 	if (!mr->hwmr.num_pbes)
 		return;
@@ -739,39 +739,37 @@ static void build_user_pbes(struct ocrdma_dev *dev, struct ocrdma_mr *mr,
 
 	shift = ilog2(umem->page_size);
 
-	list_for_each_entry(chunk, &umem->chunk_list, list) {
-		/* get all the dma regions from the chunk. */
-		for (i = 0; i < chunk->nmap; i++) {
-			pages = sg_dma_len(&chunk->page_list[i]) >> shift;
-			for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) {
-				/* store the page address in pbe */
-				pbe->pa_lo =
-				    cpu_to_le32(sg_dma_address
-						(&chunk->page_list[i]) +
-						(umem->page_size * pg_cnt));
-				pbe->pa_hi =
-				    cpu_to_le32(upper_32_bits
-						((sg_dma_address
-						  (&chunk->page_list[i]) +
-						  umem->page_size * pg_cnt)));
-				pbe_cnt += 1;
-				total_num_pbes += 1;
-				pbe++;
-
-				/* if done building pbes, issue the mbx cmd. */
-				if (total_num_pbes == num_pbes)
-					return;
-
-				/* if the given pbl is full storing the pbes,
-				 * move to next pbl.
-				 */
-				if (pbe_cnt ==
-					(mr->hwmr.pbl_size / sizeof(u64))) {
-					pbl_tbl++;
-					pbe = (struct ocrdma_pbe *)pbl_tbl->va;
-					pbe_cnt = 0;
-				}
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		pages = sg_dma_len(sg) >> shift;
+		for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) {
+			/* store the page address in pbe */
+			pbe->pa_lo =
+			    cpu_to_le32(sg_dma_address
+					(sg) +
+					(umem->page_size * pg_cnt));
+			pbe->pa_hi =
+			    cpu_to_le32(upper_32_bits
+					((sg_dma_address
+					  (sg) +
+					  umem->page_size * pg_cnt)));
+			pbe_cnt += 1;
+			total_num_pbes += 1;
+			pbe++;
+
+			/* if done building pbes, issue the mbx cmd. */
+			if (total_num_pbes == num_pbes)
+				return;
+
+			/* if the given pbl is full storing the pbes,
+			 * move to next pbl.
+			 */
+			if (pbe_cnt ==
+				(mr->hwmr.pbl_size / sizeof(u64))) {
+				pbl_tbl++;
+				pbe = (struct ocrdma_pbe *)pbl_tbl->va;
+				pbe_cnt = 0;
 			}
+
 		}
 	}
 }
diff --git a/drivers/infiniband/hw/qib/qib_mr.c b/drivers/infiniband/hw/qib/qib_mr.c
index e6687de..9bbb553 100644
--- a/drivers/infiniband/hw/qib/qib_mr.c
+++ b/drivers/infiniband/hw/qib/qib_mr.c
@@ -232,8 +232,8 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	struct qib_mr *mr;
 	struct ib_umem *umem;
-	struct ib_umem_chunk *chunk;
-	int n, m, i;
+	struct scatterlist *sg;
+	int n, m, entry;
 	struct ib_mr *ret;
 
 	if (length == 0) {
@@ -246,9 +246,7 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	if (IS_ERR(umem))
 		return (void *) umem;
 
-	n = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		n += chunk->nents;
+	n = umem->nmap;
 
 	mr = alloc_mr(n, pd);
 	if (IS_ERR(mr)) {
@@ -268,11 +266,10 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 		mr->mr.page_shift = ilog2(umem->page_size);
 	m = 0;
 	n = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list) {
-		for (i = 0; i < chunk->nents; i++) {
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
 			void *vaddr;
 
-			vaddr = page_address(sg_page(&chunk->page_list[i]));
+			vaddr = page_address(sg_page(sg));
 			if (!vaddr) {
 				ret = ERR_PTR(-EINVAL);
 				goto bail;
@@ -284,7 +281,6 @@ struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				m++;
 				n = 0;
 			}
-		}
 	}
 	ret = &mr->ibmr;
 
diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h
index 9ee0d2e..1ea0b65 100644
--- a/include/rdma/ib_umem.h
+++ b/include/rdma/ib_umem.h
@@ -46,17 +46,12 @@ struct ib_umem {
 	int			page_size;
 	int                     writable;
 	int                     hugetlb;
-	struct list_head	chunk_list;
 	struct work_struct	work;
 	struct mm_struct       *mm;
 	unsigned long		diff;
-};
-
-struct ib_umem_chunk {
-	struct list_head	list;
-	int                     nents;
-	int                     nmap;
-	struct scatterlist      page_list[0];
+	struct sg_table sg_head;
+	int             nmap;
+	int             npages;
 };
 
 #ifdef CONFIG_INFINIBAND_USER_MEM
-- 
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] IB: Refactor umem to use linear SG table

[Or Gerlitz]

On 28/01/2014 13:40, Or Gerlitz wrote:
> From: Yishai Hadas<yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
>
> This patch does refactoring of the IB core umem code and vendor drivers to use
> linear SG table instead of chunk list. With this change the relevant code becomes
> more clear, no need for nested loops to build and use umem.
>
> The patch includes: umem changes to work with linear SG table and Vendors driver
> changes to use SG instead of chunk list.
>
> Cc: Faisal Latif<faisal.latif-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Cc: Mike Marciniszyn<infinipath-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> Cc: Hoang-Nam Nguyen<hnguyen-tA70FqPdS9bQT0dZR+AlfA@public.gmane.org>
> Cc: Steve Wise<swise-ut6Up61K2wZBDgjK7y7TUQ@public.gmane.org>
> Signed-off-by: Shachar Raindel<raindel-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> Signed-off-by: Yishai Hadas<yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> ---
>   drivers/infiniband/core/umem.c               |  120 ++++++-------
>   drivers/infiniband/hw/amso1100/c2_provider.c |   23 +--
>   drivers/infiniband/hw/cxgb3/iwch_provider.c  |   19 +--
>   drivers/infiniband/hw/cxgb4/mem.c            |   39 ++---
>   drivers/infiniband/hw/ehca/ehca_classes.h    |    2 +-
>   drivers/infiniband/hw/ehca/ehca_mrmw.c       |  245 ++++++++++---------------
>   drivers/infiniband/hw/ipath/ipath_mr.c       |   39 ++---
>   drivers/infiniband/hw/mlx4/doorbell.c        |    4 +-
>   drivers/infiniband/hw/mlx4/mr.c              |   39 ++--
>   drivers/infiniband/hw/mlx5/doorbell.c        |    4 +-
>   drivers/infiniband/hw/mlx5/mem.c             |   80 ++++----
>   drivers/infiniband/hw/mthca/mthca_provider.c |   42 ++---
>   drivers/infiniband/hw/nes/nes_verbs.c        |  253 +++++++++++++-------------
>   drivers/infiniband/hw/ocrdma/ocrdma_verbs.c  |   66 ++++----
>   drivers/infiniband/hw/qib/qib_mr.c           |   14 +-
>   include/rdma/ib_umem.h                       |   11 +-
>   16 files changed, 448 insertions(+), 552 deletions(-)

While looking on this patch I suddenly noticed that unlike the rest of 
the hw drivers, the usnic driver doesn't call ib_umem_get but rather 
implements it -- while duplicating code/functionality which exists in 
the IB core, agree? why?

Or.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] IB: Refactor umem to use linear SG table

[Upinder Malhi (umalhi)]

Or,
	umem_get internally calls dma_map_sg, which is punted to Intel's IOMMU driver's dma_map_sg. The Intel's IOMMU driver will program the IOMMU page tables, to monitor DMA's from the addresses in it's page tables to prevent cache coherency issues between DMA devices and CPU's.  usNIC also programs the Intel IOMMU via the API in iommu.h, for translating virtual address to physical address.  As part of programming the translation, the Intel IOMMU automatically will take care of cache coherency issues on these translation entries.  However, if a page is programmed in the IOMMU via the dma_map_sg API and the iommu API, the Intel IOMMU gets confused (confusion is that the cache coherency protocol ceases to work).  Therefore, usNIC cannot call umem_get.

Any particular reason that the code in usNIC that walks the chunk sg's isn't ported to the linear SG table?

Upinder

On Jan 28, 2014, at 5:01 AM, Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> wrote:

> On 28/01/2014 13:40, Or Gerlitz wrote:
>> From: Yishai Hadas<yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
>> 
>> This patch does refactoring of the IB core umem code and vendor drivers to use
>> linear SG table instead of chunk list. With this change the relevant code becomes
>> more clear, no need for nested loops to build and use umem.
>> 
>> The patch includes: umem changes to work with linear SG table and Vendors driver
>> changes to use SG instead of chunk list.
>> 
>> Cc: Faisal Latif<faisal.latif-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
>> Cc: Mike Marciniszyn<infinipath-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
>> Cc: Hoang-Nam Nguyen<hnguyen-tA70FqPdS9bQT0dZR+AlfA@public.gmane.org>
>> Cc: Steve Wise<swise-ut6Up61K2wZBDgjK7y7TUQ@public.gmane.org>
>> Signed-off-by: Shachar Raindel<raindel-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
>> Signed-off-by: Yishai Hadas<yishaih-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
>> ---
>>  drivers/infiniband/core/umem.c               |  120 ++++++-------
>>  drivers/infiniband/hw/amso1100/c2_provider.c |   23 +--
>>  drivers/infiniband/hw/cxgb3/iwch_provider.c  |   19 +--
>>  drivers/infiniband/hw/cxgb4/mem.c            |   39 ++---
>>  drivers/infiniband/hw/ehca/ehca_classes.h    |    2 +-
>>  drivers/infiniband/hw/ehca/ehca_mrmw.c       |  245 ++++++++++---------------
>>  drivers/infiniband/hw/ipath/ipath_mr.c       |   39 ++---
>>  drivers/infiniband/hw/mlx4/doorbell.c        |    4 +-
>>  drivers/infiniband/hw/mlx4/mr.c              |   39 ++--
>>  drivers/infiniband/hw/mlx5/doorbell.c        |    4 +-
>>  drivers/infiniband/hw/mlx5/mem.c             |   80 ++++----
>>  drivers/infiniband/hw/mthca/mthca_provider.c |   42 ++---
>>  drivers/infiniband/hw/nes/nes_verbs.c        |  253 +++++++++++++-------------
>>  drivers/infiniband/hw/ocrdma/ocrdma_verbs.c  |   66 ++++----
>>  drivers/infiniband/hw/qib/qib_mr.c           |   14 +-
>>  include/rdma/ib_umem.h                       |   11 +-
>>  16 files changed, 448 insertions(+), 552 deletions(-)
> 
> While looking on this patch I suddenly noticed that unlike the rest of the hw drivers, the usnic driver doesn't call ib_umem_get but rather implements it -- while duplicating code/functionality which exists in the IB core, agree? why?
> 
> Or.
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH] IB: Refactor umem to use linear SG table

[Shachar Raindel]

Hi Upinder,

> -----Original Message-----
> From: Upinder Malhi (umalhi) [mailto:umalhi-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org]
> Sent: Saturday, February 01, 2014 3:46 AM
> To: Or Gerlitz
> Cc: Roland Dreier; <linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>; Yishai Hadas; Faisal Latif;
> Mike Marciniszyn; Hoang-Nam Nguyen; Steve Wise; Shachar Raindel
> Subject: Re: [PATCH] IB: Refactor umem to use linear SG table
> 
> Or,
> 	umem_get internally calls dma_map_sg, which is punted to Intel's
> IOMMU driver's dma_map_sg. The Intel's IOMMU driver will program the
> IOMMU page tables, to monitor DMA's from the addresses in it's page tables
> to prevent cache coherency issues between DMA devices and CPU's.  usNIC also
> programs the Intel IOMMU via the API in iommu.h, for translating virtual
> address to physical address.  As part of programming the translation, the Intel
> IOMMU automatically will take care of cache coherency issues on these
> translation entries.  However, if a page is programmed in the IOMMU via the
> dma_map_sg API and the iommu API, the Intel IOMMU gets confused
> (confusion is that the cache coherency protocol ceases to work).  Therefore,
> usNIC cannot call umem_get.

Can you provide some additional detail regarding the confusion - what causes it?
Is the issue with the same page being mapped more than once in general?
Does the double mapping need to address the same device?

You can probably offer a patch to ib_umem_get, passing a flag to it identifying if
it should also do the DMA mapping, and use the same code base as the rest of the
IB drivers.

A better solution would be to fix Intel's IOMMU code such that it works properly if
the page is mapped using both dma_map_sg and the IOMMU API.


> 
> Any particular reason that the code in usNIC that walks the chunk sg's isn't
> ported to the linear SG table?

As your code didn't use the API which we changed, we preferred avoiding changing
it, to avoid introducing issues into the code, especially when we don't have the
needed hardware to test your code on.

You are more than welcome to implement a similar change in your code, or port
your code to use the ib_umem_get API.

Thanks,
--Shachar
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] IB: Refactor umem to use linear SG table

[Or Gerlitz]

On Mon, Feb 3, 2014, Shachar Raindel <raindel-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> wrote:
>> From: Upinder Malhi (umalhi) [mailto:umalhi-FYB4Gu1CFyUAvxtiuMwx3w@public.gmane.org]

>> Any particular reason that the code in usNIC that walks the chunk sg's isn't
>> ported to the linear SG table?

> As your code didn't use the API which we changed, we preferred avoiding changing
> it, to avoid introducing issues into the code, especially when we don't have the
> needed hardware to test your code on.

Indeed, thanks Shachar for stressing this point

> You are more than welcome to implement a similar change in your code, or port
> your code to use the ib_umem_get API.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] IB: Refactor umem to use linear SG table

[Upinder Malhi (umalhi)]

On Feb 2, 2014, at 11:30 PM, Shachar Raindel <raindel-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> wrote:
Hi Scachar,

> Can you provide some additional detail regarding the confusion - what causes it?
> Is the issue with the same page being mapped more than once in general?
> Does the double mapping need to address the same device?
> 
> You can probably offer a patch to ib_umem_get, passing a flag to it identifying if
> it should also do the DMA mapping, and use the same code base as the rest of the
> IB drivers.
> 
> A better solution would be to fix Intel's IOMMU code such that it works properly if
> the page is mapped using both dma_map_sg and the IOMMU API.
There is no reason, whatsoever, to call dma_map_sg if iommu_map has already been called on a page.  The IOMMU API is the superset of the dma_map_sg API's.  I didn't explore why cache coherency stopped working after we behaved badly and called both APIs.
I did think about passing a flag to ib_umem_get, but  the call to dma_map_sg was deeply nested within two loops, hence it wasn't clean.  But with the port to the sg table, those loops have disappeared and that solution is feasible now.
> 
> 
>> 
>> Any particular reason that the code in usNIC that walks the chunk sg's isn't
>> ported to the linear SG table?
> 
> As your code didn't use the API which we changed, we preferred avoiding changing
> it, to avoid introducing issues into the code, especially when we don't have the
> needed hardware to test your code on.
> 
> You are more than welcome to implement a similar change in your code, or port
> your code to use the ib_umem_get API.
Ack.   --
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] IB: Refactor umem to use linear SG table

[Upinder Malhi (umalhi)]

On Feb 2, 2014, at 11:30 PM, Shachar Raindel <raindel-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> wrote:

Follow up.
> You can probably offer a patch to ib_umem_get, passing a flag to it identifying if
> it should also do the DMA mapping, and use the same code base as the rest of the
> IB drivers.
The usnic memory management stuff is entirely freed of ib objects.  If we were to move to using ib_umem_get, the code would get corrupted with ib objects.  There are also other subtle minor differences.  The overlap with ib_umem_get is minor, so I am not going to port this code over to use ib_umem_get.
> 
> A better solution would be to fix Intel's IOMMU code such that it works properly if
> the page is mapped using both dma_map_sg and the IOMMU API.
> 
> 
>> 
>> Any particular reason that the code in usNIC that walks the chunk sg's isn't
>> ported to the linear SG table?
> 
> As your code didn't use the API which we changed, we preferred avoiding changing
> it, to avoid introducing issues into the code, especially when we don't have the
> needed hardware to test your code on.
> 
> You are more than welcome to implement a similar change in your code, or port
> your code to use the ib_umem_get API.
I will submit a patch for this.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html