[PATCH 0/3] IB/ehca: MR/MW fixes

linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed

* [PATCH 0/3] IB/ehca: MR/MW fixes
@ 2007-09-13 16:14 Joachim Fenkes
  2007-09-13 16:14 ` [PATCH 1/3] IB/ehca: Fix large page HW cap defines Joachim Fenkes
                   ` (2 more replies)
  0 siblings, 3 replies; 7+ messages in thread
From: Joachim Fenkes @ 2007-09-13 16:14 UTC (permalink / raw)
  To: LinuxPPC-Dev, LKML, OF-General, Roland Dreier, OF-EWG
  Cc: Stefan Roscher, Christoph Raisch

This patchset replaces Nam's previous MR/MW patch (posted by me). I split
the #define fixes into a separate patch and moved the "is the memory from
hugetlbfs?" code into ib_umem_get().

[1/3] fixes the page size HW cap defines
[2/3] adds the hugetlb test to ib_umem_get()
[3/3] finally uses the hugetlb flag in ehca_reg_user_mr()

The patches should apply cleanly, in order, on top of my previous 12-patch
set. Please review the changes and apply the patches for 2.6.24 if they are
okay.

Regards,
  Joachim

=2D-=20
Joachim Fenkes =A0-- =A0eHCA Linux Driver Developer and Hardware Tamer
IBM Deutschland Entwicklung GmbH =A0-- =A0Dept. 3627 (I/O Firmware Dev. 2)
Schoenaicher Strasse 220 =A0-- =A071032 Boeblingen =A0-- =A0Germany
eMail: fenkes@de.ibm.com

^ permalink raw reply	[flat|nested] 7+ messages in thread

* [PATCH 1/3] IB/ehca: Fix large page HW cap defines
  2007-09-13 16:14 [PATCH 0/3] IB/ehca: MR/MW fixes Joachim Fenkes
@ 2007-09-13 16:14 ` Joachim Fenkes
  2007-09-17 22:12   ` Roland Dreier
  2007-09-13 16:15 ` [PATCH 2/3] IB/umem: Add hugetlb flag to struct ib_umem Joachim Fenkes
  2007-09-13 16:16 ` [PATCH 3/3] IB/ehca: Make sure user pages are from hugetlb before using MR large pages Joachim Fenkes
  2 siblings, 1 reply; 7+ messages in thread
From: Joachim Fenkes @ 2007-09-13 16:14 UTC (permalink / raw)
  To: LinuxPPC-Dev, LKML, OF-General, Roland Dreier, OF-EWG
  Cc: Stefan Roscher, Christoph Raisch

From: Hoang-Nam Nguyen <hnguyen@de.ibm.com>

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
---
 drivers/infiniband/hw/ehca/ehca_classes.h |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/ehca/ehca_classes.h b/drivers/infiniband/hw/ehca/ehca_classes.h
index 206d4eb..c2edd4c 100644
--- a/drivers/infiniband/hw/ehca/ehca_classes.h
+++ b/drivers/infiniband/hw/ehca/ehca_classes.h
@@ -99,10 +99,10 @@ struct ehca_sport {
 	struct ehca_sma_attr saved_attr;
 };
 
-#define HCA_CAP_MR_PGSIZE_4K  1
-#define HCA_CAP_MR_PGSIZE_64K 2
-#define HCA_CAP_MR_PGSIZE_1M  4
-#define HCA_CAP_MR_PGSIZE_16M 8
+#define HCA_CAP_MR_PGSIZE_4K  0x80000000
+#define HCA_CAP_MR_PGSIZE_64K 0x40000000
+#define HCA_CAP_MR_PGSIZE_1M  0x20000000
+#define HCA_CAP_MR_PGSIZE_16M 0x10000000
 
 struct ehca_shca {
 	struct ib_device ib_device;
-- 
1.5.2

^ permalink raw reply related	[flat|nested] 7+ messages in thread

* Re: [PATCH 1/3] IB/ehca: Fix large page HW cap defines
  2007-09-13 16:14 ` [PATCH 1/3] IB/ehca: Fix large page HW cap defines Joachim Fenkes
@ 2007-09-17 22:12   ` Roland Dreier
  0 siblings, 0 replies; 7+ messages in thread
From: Roland Dreier @ 2007-09-17 22:12 UTC (permalink / raw)
  To: Joachim Fenkes
  Cc: LKML, OF-EWG, LinuxPPC-Dev, Christoph Raisch, OF-General,
	Stefan Roscher

obviously OK...applied.

^ permalink raw reply	[flat|nested] 7+ messages in thread

* [PATCH 2/3] IB/umem: Add hugetlb flag to struct ib_umem
  2007-09-13 16:14 [PATCH 0/3] IB/ehca: MR/MW fixes Joachim Fenkes
  2007-09-13 16:14 ` [PATCH 1/3] IB/ehca: Fix large page HW cap defines Joachim Fenkes
@ 2007-09-13 16:15 ` Joachim Fenkes
  2007-09-19 17:16   ` Roland Dreier
  2007-09-13 16:16 ` [PATCH 3/3] IB/ehca: Make sure user pages are from hugetlb before using MR large pages Joachim Fenkes
  2 siblings, 1 reply; 7+ messages in thread
From: Joachim Fenkes @ 2007-09-13 16:15 UTC (permalink / raw)
  To: LinuxPPC-Dev, LKML, OF-General, Roland Dreier, OF-EWG
  Cc: Stefan Roscher, Christoph Raisch

During ib_umem_get(), determine whether all pages from the memory region are
hugetlb pages and report this in the "hugetlb" field. Low-level driver can
use this information if they need it.

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
---
 drivers/infiniband/core/umem.c |   20 +++++++++++++++++++-
 include/rdma/ib_umem.h         |    1 +
 2 files changed, 20 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index 664d2fa..2f54e29 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -37,6 +37,7 @@
 #include <linux/mm.h>
 #include <linux/dma-mapping.h>
 #include <linux/sched.h>
+#include <linux/hugetlb.h>
 
 #include "uverbs.h"
 
@@ -75,6 +76,7 @@ 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;
@@ -104,6 +106,9 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 	 */
 	umem->writable  = !!(access & ~IB_ACCESS_REMOTE_READ);
 
+	/* 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);
@@ -112,6 +117,14 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 		return ERR_PTR(-ENOMEM);
 	}
 
+	/*
+	 * if we can't alloc the vma_list, it's not so bad;
+	 * just assume the memory is not hugetlb memory
+	 */
+	vma_list = (struct vm_area_struct **) __get_free_page(GFP_KERNEL);
+	if (!vma_list)
+		umem->hugetlb = 0;
+
 	npages = PAGE_ALIGN(size + umem->offset) >> PAGE_SHIFT;
 
 	down_write(&current->mm->mmap_sem);
@@ -131,7 +144,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 		ret = get_user_pages(current, current->mm, cur_base,
 				     min_t(int, npages,
 					   PAGE_SIZE / sizeof (struct page *)),
-				     1, !umem->writable, page_list, NULL);
+				     1, !umem->writable, page_list, vma_list);
 
 		if (ret < 0)
 			goto out;
@@ -152,6 +165,9 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 
 			chunk->nents = min_t(int, ret, IB_UMEM_MAX_PAGE_CHUNK);
 			for (i = 0; i < chunk->nents; ++i) {
+				if (vma_list &&
+				    !is_vm_hugetlb_page(vma_list[i + off]))
+					umem->hugetlb = 0;
 				chunk->page_list[i].page   = page_list[i + off];
 				chunk->page_list[i].offset = 0;
 				chunk->page_list[i].length = PAGE_SIZE;
@@ -186,6 +202,8 @@ out:
 		current->mm->locked_vm = locked;
 
 	up_write(&current->mm->mmap_sem);
+	if (vma_list)
+		free_page((unsigned long) vma_list);
 	free_page((unsigned long) page_list);
 
 	return ret < 0 ? ERR_PTR(ret) : umem;
diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h
index c533d6c..2229842 100644
--- a/include/rdma/ib_umem.h
+++ b/include/rdma/ib_umem.h
@@ -45,6 +45,7 @@ struct ib_umem {
 	int			offset;
 	int			page_size;
 	int                     writable;
+	int                     hugetlb;
 	struct list_head	chunk_list;
 	struct work_struct	work;
 	struct mm_struct       *mm;
-- 
1.5.2

^ permalink raw reply related	[flat|nested] 7+ messages in thread

* Re: [PATCH 2/3] IB/umem: Add hugetlb flag to struct ib_umem
  2007-09-13 16:15 ` [PATCH 2/3] IB/umem: Add hugetlb flag to struct ib_umem Joachim Fenkes
@ 2007-09-19 17:16   ` Roland Dreier
  0 siblings, 0 replies; 7+ messages in thread
From: Roland Dreier @ 2007-09-19 17:16 UTC (permalink / raw)
  To: Joachim Fenkes
  Cc: LKML, OF-EWG, LinuxPPC-Dev, Christoph Raisch, OF-General,
	Stefan Roscher

This looks realy nice to me... a very clean patch.

I'll add this to 2.6.24 unless someone objects soon...

^ permalink raw reply	[flat|nested] 7+ messages in thread

* [PATCH 3/3] IB/ehca: Make sure user pages are from hugetlb before using MR large pages
  2007-09-13 16:14 [PATCH 0/3] IB/ehca: MR/MW fixes Joachim Fenkes
  2007-09-13 16:14 ` [PATCH 1/3] IB/ehca: Fix large page HW cap defines Joachim Fenkes
  2007-09-13 16:15 ` [PATCH 2/3] IB/umem: Add hugetlb flag to struct ib_umem Joachim Fenkes
@ 2007-09-13 16:16 ` Joachim Fenkes
  2007-09-20 16:29   ` Roland Dreier
  2 siblings, 1 reply; 7+ messages in thread
From: Joachim Fenkes @ 2007-09-13 16:16 UTC (permalink / raw)
  To: LinuxPPC-Dev, LKML, OF-General, Roland Dreier, OF-EWG
  Cc: Stefan Roscher, Christoph Raisch

...because, on virtualized hardware like System p, we can't be sure that the
physical pages behind them are contiguous.

Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
---
 drivers/infiniband/hw/ehca/ehca_mrmw.c |   25 +++++++++++++++----------
 1 files changed, 15 insertions(+), 10 deletions(-)

diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.c b/drivers/infiniband/hw/ehca/ehca_mrmw.c
index 4c8f3b3..4ba8b7c 100644
--- a/drivers/infiniband/hw/ehca/ehca_mrmw.c
+++ b/drivers/infiniband/hw/ehca/ehca_mrmw.c
@@ -51,6 +51,7 @@
 
 #define NUM_CHUNKS(length, chunk_size) \
 	(((length) + (chunk_size - 1)) / (chunk_size))
+
 /* max number of rpages (per hcall register_rpages) */
 #define MAX_RPAGES 512
 
@@ -64,6 +65,11 @@ enum ehca_mr_pgsize {
 	EHCA_MR_PGSIZE16M = 0x1000000L
 };
 
+#define EHCA_MR_PGSHIFT4K  12
+#define EHCA_MR_PGSHIFT64K 16
+#define EHCA_MR_PGSHIFT1M  20
+#define EHCA_MR_PGSHIFT16M 24
+
 static u32 ehca_encode_hwpage_size(u32 pgsize)
 {
 	u32 idx = 0;
@@ -347,17 +353,16 @@ struct ib_mr *ehca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	/* select proper hw_pgsize */
 	if (ehca_mr_largepage &&
 	    (shca->hca_cap_mr_pgsize & HCA_CAP_MR_PGSIZE_16M)) {
-		if (length <= EHCA_MR_PGSIZE4K
-		    && PAGE_SIZE == EHCA_MR_PGSIZE4K)
-			hwpage_size = EHCA_MR_PGSIZE4K;
-		else if (length <= EHCA_MR_PGSIZE64K)
-			hwpage_size = EHCA_MR_PGSIZE64K;
-		else if (length <= EHCA_MR_PGSIZE1M)
-			hwpage_size = EHCA_MR_PGSIZE1M;
-		else
-			hwpage_size = EHCA_MR_PGSIZE16M;
+		int page_shift = PAGE_SHIFT;
+		if (e_mr->umem->hugetlb) {
+			/* determine page_shift, clamp between 4K and 16M */
+			page_shift = (fls64(length - 1) + 3) & ~3;
+			page_shift = min(max(page_shift, EHCA_MR_PGSHIFT4K),
+					 EHCA_MR_PGSHIFT16M);
+		}
+		hwpage_size = 1UL << page_shift;
 	} else
-		hwpage_size = EHCA_MR_PGSIZE4K;
+		hwpage_size = EHCA_MR_PGSIZE4K; /* ehca1 only supports 4k */
 	ehca_dbg(pd->device, "hwpage_size=%lx", hwpage_size);
 
 reg_user_mr_fallback:
-- 
1.5.2

^ permalink raw reply related	[flat|nested] 7+ messages in thread

* Re: [PATCH 3/3] IB/ehca: Make sure user pages are from hugetlb before using MR large pages
  2007-09-13 16:16 ` [PATCH 3/3] IB/ehca: Make sure user pages are from hugetlb before using MR large pages Joachim Fenkes
@ 2007-09-20 16:29   ` Roland Dreier
  0 siblings, 0 replies; 7+ messages in thread
From: Roland Dreier @ 2007-09-20 16:29 UTC (permalink / raw)
  To: Joachim Fenkes
  Cc: LKML, OF-EWG, LinuxPPC-Dev, Christoph Raisch, OF-General,
	Stefan Roscher

thanks, applied this and the umem patch...

^ permalink raw reply	[flat|nested] 7+ messages in thread

end of thread, other threads:[~2007-09-20 16:29 UTC | newest]

Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-09-13 16:14 [PATCH 0/3] IB/ehca: MR/MW fixes Joachim Fenkes
2007-09-13 16:14 ` [PATCH 1/3] IB/ehca: Fix large page HW cap defines Joachim Fenkes
2007-09-17 22:12   ` Roland Dreier
2007-09-13 16:15 ` [PATCH 2/3] IB/umem: Add hugetlb flag to struct ib_umem Joachim Fenkes
2007-09-19 17:16   ` Roland Dreier
2007-09-13 16:16 ` [PATCH 3/3] IB/ehca: Make sure user pages are from hugetlb before using MR large pages Joachim Fenkes
2007-09-20 16:29   ` Roland Dreier

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).