| Message ID | 20170821222817.17376-13-hauke@hauke-m.de (mailing list archive) |
|---|---|
| State | Accepted |
| Headers | show |
On Tue, 2017-08-22 at 00:28 +0200, Hauke Mehrtens wrote: > This reverts commit 12e8fd6fd3802 ("lib/rhashtable.c: use kvzalloc() > in > bucket_table_alloc() when possible") from upstream kernel. > Kernel versions older that 4.12 do not have kvzalloc() try to avoid > the > usage of this. Can't we backport these? There was a patch from AceLan Kao to do so in a pretty simple fashion, but that should still be better than a patch? johannes -- To unsubscribe from this list: send the line "unsubscribe backports" in
On 09/06/2017 05:05 PM, Johannes Berg wrote: > On Tue, 2017-08-22 at 00:28 +0200, Hauke Mehrtens wrote: >> This reverts commit 12e8fd6fd3802 ("lib/rhashtable.c: use kvzalloc() >> in >> bucket_table_alloc() when possible") from upstream kernel. >> Kernel versions older that 4.12 do not have kvzalloc() try to avoid >> the >> usage of this. > > Can't we backport these? There was a patch from AceLan Kao to do so in > a pretty simple fashion, but that should still be better than a patch? > > johannes Hi, I think AceLan Kao's patch misses the implementation of kvmalloc_node(). We can implement this for node == NUMA_NO_NODE by using __vmalloc() instead of __vmalloc_node_flags_caller() which should be fine. Hauke -- To unsubscribe from this list: send the line "unsubscribe backports" in
On Thu, 2017-09-07 at 00:12 +0200, Hauke Mehrtens wrote: > On 09/06/2017 05:05 PM, Johannes Berg wrote: > > On Tue, 2017-08-22 at 00:28 +0200, Hauke Mehrtens wrote: > > > This reverts commit 12e8fd6fd3802 ("lib/rhashtable.c: use > > > kvzalloc() > > > in > > > bucket_table_alloc() when possible") from upstream kernel. > > > Kernel versions older that 4.12 do not have kvzalloc() try to > > > avoid > > > the > > > usage of this. > > > > Can't we backport these? There was a patch from AceLan Kao to do so > > in > > a pretty simple fashion, but that should still be better than a > > patch? > > > > johannes > > Hi, > > I think AceLan Kao's patch misses the implementation of > kvmalloc_node(). > We can implement this for node == NUMA_NO_NODE by using __vmalloc() > instead of __vmalloc_node_flags_caller() which should be fine. Sounds reasonable to me. johannes -- To unsubscribe from this list: send the line "unsubscribe backports" in
diff --git a/patches/lib-rhashtable.patch b/patches/lib-rhashtable.patch index 9c262b02..35424efb 100644 --- a/patches/lib-rhashtable.patch +++ b/patches/lib-rhashtable.patch @@ -27,6 +27,19 @@ if (!tbl->locks) return -ENOMEM; for (i = 0; i < size; i++) --- -1.9.1 +@@ -226,10 +226,11 @@ static struct bucket_table *bucket_table_alloc(struct rhashtable *ht, + int i; + + size = sizeof(*tbl) + nbuckets * sizeof(tbl->buckets[0]); +- if (gfp != GFP_KERNEL) ++ if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER) || ++ gfp != GFP_KERNEL) + tbl = kzalloc(size, gfp | __GFP_NOWARN | __GFP_NORETRY); +- else +- tbl = kvzalloc(size, gfp); ++ if (tbl == NULL && gfp == GFP_KERNEL) ++ tbl = vzalloc(size); + + size = nbuckets; +
This reverts commit 12e8fd6fd3802 ("lib/rhashtable.c: use kvzalloc() in bucket_table_alloc() when possible") from upstream kernel. Kernel versions older that 4.12 do not have kvzalloc() try to avoid the usage of this. Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de> --- patches/lib-rhashtable.patch | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)