Re: [2.6.24 regression][BUGFIX] numactl --interleave=all doesn't works on memoryless node.

From: KOSAKI Motohiro
Date: Thu Feb 07 2008 - 03:32:26 EST


Hi Lee-san

Unfortunately, 2.6.24-mm1 can't boot on fujitsu machine.
(hmm, origin.patch cause regression to pci initialization ;-)

instead, I tested 2.6.24 + your patch.
it seem work good :)

Tested-by: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx>

and, I have a bit comment.


> /* Do sanity checking on a policy */
> -static int mpol_check_policy(int mode, nodemask_t *nodes)
> +static int mpol_check_policy(int mode, nodemask_t *nodes, int was_empty)

was_empty argument is a bit ugly.
Could we unify mpol_check_policy and contextualize_policy?
mpol_check_policy only called from contextualize_policy.

> - return nodes_subset(*nodes, node_states[N_HIGH_MEMORY]) ? 0 : -EINVAL;
> + return 0;

Could we N_POSSIBLE check?

I attached the patch for my idea explain.
on my test environment, your patch and mine works good both.

- kosaki


---
mm/mempolicy.c | 47 +++++++++++++++++++++--------------------------
1 file changed, 21 insertions(+), 26 deletions(-)

Index: b/mm/mempolicy.c
===================================================================
--- a/mm/mempolicy.c 2008-02-07 17:19:09.000000000 +0900
+++ b/mm/mempolicy.c 2008-02-07 17:24:28.000000000 +0900
@@ -114,9 +114,25 @@ static void mpol_rebind_policy(struct me
const nodemask_t *newmask);

/* Do sanity checking on a policy */
-static int mpol_check_policy(int mode, nodemask_t *nodes, int was_empty)
+static int mpol_check_policy(int mode, nodemask_t *nodes)
{
- int is_empty = nodes_empty(*nodes);
+ int was_empty;
+ int is_empty;
+
+ if (!nodes)
+ return 0;
+
+ /*
+ * Remember whether in coming nodemask was empty, If not,
+ * restrict the nodes to the allowed nodes in the cpuset.
+ * This is guaranteed to be a subset of nodes with memory.
+ */
+ cpuset_update_task_memory_state();
+ was_empty = nodes_empty(*nodes);
+ if (!was_empty)
+ nodes_and(*nodes, *nodes, cpuset_current_mems_allowed);
+
+ is_empty = nodes_empty(*nodes);

switch (mode) {
case MPOL_DEFAULT:
@@ -144,7 +160,7 @@ static int mpol_check_policy(int mode, n
return -EINVAL;
break;
}
- return 0;
+ return nodes_subset(*nodes, node_states[N_POSSIBLE]) ? 0 : -EINVAL;
}

/* Generate a custom zonelist for the BIND policy. */
@@ -432,27 +448,6 @@ static int mbind_range(struct vm_area_st
return err;
}

-static int contextualize_policy(int mode, nodemask_t *nodes)
-{
- int was_empty;
-
- if (!nodes)
- return 0;
-
- /*
- * Remember whether in coming nodemask was empty, If not,
- * restrict the nodes to the allowed nodes in the cpuset.
- * This is guaranteed to be a subset of nodes with memory.
- */
- cpuset_update_task_memory_state();
- was_empty = nodes_empty(*nodes);
- if (!was_empty)
- nodes_and(*nodes, *nodes, cpuset_current_mems_allowed);
-
- return mpol_check_policy(mode, nodes, was_empty);
-}
-
-
/*
* Update task->flags PF_MEMPOLICY bit: set iff non-default
* mempolicy. Allows more rapid checking of this (combined perhaps
@@ -488,7 +483,7 @@ static long do_set_mempolicy(int mode, n
{
struct mempolicy *new;

- if (contextualize_policy(mode, nodes))
+ if (mpol_check_policy(mode, nodes))
return -EINVAL;
new = mpol_new(mode, nodes);
if (IS_ERR(new))
@@ -817,7 +812,7 @@ static long do_mbind(unsigned long start
if (end == start)
return 0;

- if (contextualize_policy(mode, nmask))
+ if (mpol_check_policy(mode, nmask))
return -EINVAL;

new = mpol_new(mode, nmask);


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