The patch titled
mempolicy-create-mempolicy_operations-structure fix
has been added to the -mm tree. Its filename is
mempolicy-create-mempolicy_operations-structure-fix.patch
Before you just go and hit "reply", please:
a) Consider who else should be cc'ed
b) Prefer to cc a suitable mailing list as well
c) Ideally: find the original patch on the mailing list and do a
reply-to-all to that, adding suitable additional cc's
*** Remember to use Documentation/SubmitChecklist when testing your code ***
See http://www.zip.com.au/~akpm/linux/patches/stuff/added-to-mm.txt to find
out what to do about this
The current -mm tree may be found at http://userweb.kernel.org/~akpm/mmotm/
------------------------------------------------------
Subject: mempolicy-create-mempolicy_operations-structure fix
From: Lee Schermerhorn <Lee.Schermerhorn@xxxxxx>
The subject patch causes a regression in the numactl package mempolicy
regression tests.
I'm using the numactl-1.0.2 package with the patches available at:
http://free.linux.hp.com/~lts/Patches/Numactl/numactl-1.0.2-patches-080226.tar.gz
The numastat and regress patches in that tarball are necessary for
recent kernels, else the tests will report false failures.
The following patch fixes the regression.
Specifically this patch fixes problems introduced by the rework of mpol_new()
in patch 2/4 in the second series. As a result, we're no longer accepting
NULL/empty nodemask with MPOL_PREFERRED as "local" allocation. This breaks
the numactl regression tests.
"numactl --localalloc" <some command>" fails with invalid argument.
This patch fixes the regression by again treating NULL/empty nodemask with
MPOL_PREFERRED as "local allocation", and special casing this condition, as
needed, in mpol_new(), mpol_new_preferred() and mpol_rebind_preferred().
It also appears that the patch series listed above required a non-empty
nodemask with MPOL_DEFAULT. However, I didn't test that. With this
patch, MPOL_DEFAULT effectively ignores the nodemask--empty or not.
This is a change in behavior that I have argued against, but the
regression tests don't test this, so I'm not going to attempt to address
it with this patch.
Note: I have no tests to test the 'STATIC_NODES and 'RELATIVE_NODES
behavior to ensure that this patch doesn't regress those.
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@xxxxxx>
Cc: Paul Jackson <pj@xxxxxxx>
Cc: Christoph Lameter <clameter@xxxxxxx>
Cc: Lee Schermerhorn <Lee.Schermerhorn@xxxxxx>
Cc: Andi Kleen <ak@xxxxxxx>
Cc: Eric Whitney <eric.whitney@xxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
mm/mempolicy.c | 54 +++++++++++++++++++++++++++++++----------------
1 file changed, 36 insertions(+), 18 deletions(-)
diff -puN mm/mempolicy.c~mempolicy-create-mempolicy_operations-structure-fix mm/mempolicy.c
--- a/mm/mempolicy.c~mempolicy-create-mempolicy_operations-structure-fix
+++ a/mm/mempolicy.c
@@ -158,9 +158,12 @@ static int mpol_new_interleave(struct me
static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
{
- if (nodes_empty(*nodes))
- return -EINVAL;
- pol->v.preferred_node = first_node(*nodes);
+ if (!nodes)
+ pol->v.preferred_node = -1; /* local allocation */
+ else if (nodes_empty(*nodes))
+ return -EINVAL; /* no allowed nodes */
+ else
+ pol->v.preferred_node = first_node(*nodes);
return 0;
}
@@ -178,34 +181,43 @@ static struct mempolicy *mpol_new(unsign
{
struct mempolicy *policy;
nodemask_t cpuset_context_nmask;
+ int localalloc = 0;
int ret;
pr_debug("setting mode %d flags %d nodes[0] %lx\n",
mode, flags, nodes ? nodes_addr(*nodes)[0] : -1);
- if (nodes && nodes_empty(*nodes) && mode != MPOL_PREFERRED)
- return ERR_PTR(-EINVAL);
if (mode == MPOL_DEFAULT)
return NULL;
+ if (!nodes || nodes_empty(*nodes)) {
+ if (mode != MPOL_PREFERRED)
+ return ERR_PTR(-EINVAL);
+ localalloc = 1; /* special case: no mode flags */
+ }
policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
if (!policy)
return ERR_PTR(-ENOMEM);
atomic_set(&policy->refcnt, 1);
- cpuset_update_task_memory_state();
- if (flags & MPOL_F_RELATIVE_NODES)
- mpol_relative_nodemask(&cpuset_context_nmask, nodes,
- &cpuset_current_mems_allowed);
- else
- nodes_and(cpuset_context_nmask, *nodes,
- cpuset_current_mems_allowed);
policy->policy = mode;
- policy->flags = flags;
- if (mpol_store_user_nodemask(policy))
- policy->w.user_nodemask = *nodes;
- else
- policy->w.cpuset_mems_allowed = cpuset_mems_allowed(current);
- ret = mpol_ops[mode].create(policy, &cpuset_context_nmask);
+ if (!localalloc) {
+ policy->flags = flags;
+ cpuset_update_task_memory_state();
+ if (flags & MPOL_F_RELATIVE_NODES)
+ mpol_relative_nodemask(&cpuset_context_nmask, nodes,
+ &cpuset_current_mems_allowed);
+ else
+ nodes_and(cpuset_context_nmask, *nodes,
+ cpuset_current_mems_allowed);
+ if (mpol_store_user_nodemask(policy))
+ policy->w.user_nodemask = *nodes;
+ else
+ policy->w.cpuset_mems_allowed =
+ cpuset_mems_allowed(current);
+ }
+
+ ret = mpol_ops[mode].create(policy,
+ localalloc ? NULL : &cpuset_context_nmask);
if (ret < 0) {
kmem_cache_free(policy_cache, policy);
return ERR_PTR(ret);
@@ -247,6 +259,10 @@ static void mpol_rebind_preferred(struct
{
nodemask_t tmp;
+ /*
+ * check 'STATIC_NODES first, as preferred_node == -1 may be
+ * a temporary, "fallback" state for this policy.
+ */
if (pol->flags & MPOL_F_STATIC_NODES) {
int node = first_node(pol->w.user_nodemask);
@@ -254,6 +270,8 @@ static void mpol_rebind_preferred(struct
pol->v.preferred_node = node;
else
pol->v.preferred_node = -1;
+ } else if (pol->v.preferred_node == -1) {
+ return; /* no remap required for explicit local alloc */
} else if (pol->flags & MPOL_F_RELATIVE_NODES) {
mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
pol->v.preferred_node = first_node(tmp);
_
Patches currently in -mm which might be from Lee.Schermerhorn@xxxxxx are
fix-mempolicy-reference-counting-bugs.patch
mm-filter-based-on-a-nodemask-as-well-as-a-gfp_mask-doc-fixes.patch
mm-filter-based-on-a-nodemask-as-well-as-a-gfp_mask-make-dequeue_huge_page_vma-obey-mpol_bind-nodemask.patch
mm-filter-based-on-a-nodemask-as-well-as-a-gfp_mask-make-dequeue_huge_page_vma-obey-mpol_bind-nodemask-rework.patch
mempolicy-convert-mpol-constants-to-enum.patch
mempolicy-support-optional-mode-flags.patch
mempolicy-add-mpol_f_static_nodes-flag.patch
mempolicy-add-mpol_f_relative_nodes-flag.patch
mempolicy-update-numa-memory-policy-documentation.patch
mempolicy-move-rebind-functions.patch
mempolicy-create-mempolicy_operations-structure.patch
mempolicy-create-mempolicy_operations-structure-fix.patch
mempolicy-small-header-file-cleanup.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
