Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1228810 > unrolled thread

[RFC PATCH] iommu: enable the last bit in iommu_area_alloc()

Started byWei Yang <weiyang@linux.vnet.ibm.com>
First post2015-09-20 04:20 +0200
Last post2015-09-20 04:20 +0200
Articles 1 — 1 participant

Back to article view | Back to linux.kernel


Contents

  [RFC PATCH] iommu: enable the last bit in iommu_area_alloc() Wei Yang <weiyang@linux.vnet.ibm.com> - 2015-09-20 04:20 +0200

#1228810 — [RFC PATCH] iommu: enable the last bit in iommu_area_alloc()

FromWei Yang <weiyang@linux.vnet.ibm.com>
Date2015-09-20 04:20 +0200
Subject[RFC PATCH] iommu: enable the last bit in iommu_area_alloc()
Message-ID<qaCop-39s-1@gated-at.bofh.it>
In 'commit a66022c45775 ("iommu-helper: use bitmap library")',
iommu_area_alloc() uses bitmap_find_next_zero_area() to lookup available
iommu space.

When given "start, size, nr", bitmap_find_next_zero_area() is looking for a
range with nr zero bit in [start, size) instead of [start, size]. This
means the last bit is already excluded. By decrease size at the beginning,
the last iommu page will not be allocated.

This patch removes the decrease on size.

Signed-off-by: Wei Yang <weiyang@linux.vnet.ibm.com>

---

I may missed something, while the code makes me a little confused.

I found two users of iommu_area_alloc(), one in powernv platform and one in
lib/iommu-common.c. The "limit", which is passed to
bitmap_find_next_zero_area() as the "size" are both set to pool->end in these
two cases. While the pool->end in these two cases are calculated differently.

On powernv platform, iommu_init_table() sets 
	p->end = p->start + tbl->poolsize;
While in iommu_tbl_pool_init(), 
        pools[i].end = pools[i].start + iommu->poolsize -1;

In both case, it will not do harm to system, except we have one more less
iommu page in the second case. 

And then in current code, iommu_area_alloc() will decrease the size by one
again.

I may missed something, currently I think the implementation in powernv
platform is correct while iommu_area_alloc() eats the last iommu page.

Tests:

I have apply this on top of v4.2 and enforce to use dma_iommu_ops for each
device. Transfered on guest image with 4GB by scp and check the checksum are
the same.

---
 lib/iommu-helper.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/lib/iommu-helper.c b/lib/iommu-helper.c
index c27e269..2866004 100644
--- a/lib/iommu-helper.c
+++ b/lib/iommu-helper.c
@@ -23,8 +23,6 @@ unsigned long iommu_area_alloc(unsigned long *map, unsigned long size,
 {
 	unsigned long index;
 
-	/* We don't want the last of the limit */
-	size -= 1;
 again:
 	index = bitmap_find_next_zero_area(map, size, start, nr, align_mask);
 	if (index < size) {
-- 
2.5.0

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[toc] | [standalone]


Back to top | Article view | linux.kernel


csiph-web