The patch titled
IB/ipath: use more appropriate gfp flags
has been removed from the -mm tree. Its filename is
ib-ipath-use-more-appropriate-gfp-flags.patch
This patch was dropped because it was merged into mainline or a subsystem tree
------------------------------------------------------
Subject: IB/ipath: use more appropriate gfp flags
From: "Bryan O'Sullivan" <bos@xxxxxxxxxxxxx>
This helps us to survive better when memory is fragmented.
Signed-off-by: Dave Olson <dave.olson@xxxxxxxxxx>
Signed-off-by: Bryan O'Sullivan <bryan.osullivan@xxxxxxxxxx>
Cc: "Michael S. Tsirkin" <mst@xxxxxxxxxxxxxx>
Cc: Roland Dreier <rolandd@xxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxx>
---
drivers/infiniband/hw/ipath/ipath_file_ops.c | 26 +++++++++--------
1 file changed, 14 insertions(+), 12 deletions(-)
diff -puN drivers/infiniband/hw/ipath/ipath_file_ops.c~ib-ipath-use-more-appropriate-gfp-flags drivers/infiniband/hw/ipath/ipath_file_ops.c
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c~ib-ipath-use-more-appropriate-gfp-flags
+++ a/drivers/infiniband/hw/ipath/ipath_file_ops.c
@@ -705,6 +705,15 @@ static int ipath_create_user_egr(struct
unsigned e, egrcnt, alloced, egrperchunk, chunk, egrsize, egroff;
size_t size;
int ret;
+ gfp_t gfp_flags;
+
+ /*
+ * GFP_USER, but without GFP_FS, so buffer cache can be
+ * coalesced (we hope); otherwise, even at order 4,
+ * heavy filesystem activity makes these fail, and we can
+ * use compound pages.
+ */
+ gfp_flags = __GFP_WAIT | __GFP_IO | __GFP_COMP;
egrcnt = dd->ipath_rcvegrcnt;
/* TID number offset for this port */
@@ -721,10 +730,8 @@ static int ipath_create_user_egr(struct
* memory pressure (creating large files and then copying them over
* NFS while doing lots of MPI jobs), we hit some allocation
* failures, even though we can sleep... (2.6.10) Still get
- * failures at 64K. 32K is the lowest we can go without waiting
- * more memory again. It seems likely that the coalescing in
- * free_pages, etc. still has issues (as it has had previously
- * during 2.6.x development).
+ * failures at 64K. 32K is the lowest we can go without wasting
+ * additional memory.
*/
size = 0x8000;
alloced = ALIGN(egrsize * egrcnt, size);
@@ -745,12 +752,6 @@ static int ipath_create_user_egr(struct
goto bail_rcvegrbuf;
}
for (e = 0; e < pd->port_rcvegrbuf_chunks; e++) {
- /*
- * GFP_USER, but without GFP_FS, so buffer cache can be
- * coalesced (we hope); otherwise, even at order 4,
- * heavy filesystem activity makes these fail
- */
- gfp_t gfp_flags = __GFP_WAIT | __GFP_IO | __GFP_COMP;
pd->port_rcvegrbuf[e] = dma_alloc_coherent(
&dd->pcidev->dev, size, &pd->port_rcvegrbuf_phys[e],
@@ -1167,9 +1168,10 @@ static int ipath_mmap(struct file *fp, s
ureg = dd->ipath_uregbase + dd->ipath_palign * pd->port_port;
- ipath_cdbg(MM, "ushare: pgaddr %llx vm_start=%lx, vmlen %lx\n",
+ ipath_cdbg(MM, "pgaddr %llx vm_start=%lx len %lx port %u:%u\n",
(unsigned long long) pgaddr, vma->vm_start,
- vma->vm_end - vma->vm_start);
+ vma->vm_end - vma->vm_start, dd->ipath_unit,
+ pd->port_port);
if (pgaddr == ureg)
ret = mmap_ureg(vma, dd, ureg);
_
Patches currently in -mm which might be from bos@xxxxxxxxxxxxx are
origin.patch
ib-ipath-fixes-a-bug-where-our-delay-for-eeprom-no.patch
-
To unsubscribe from this list: send the line "unsubscribe mm-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
