LU-13294 libcfs: incorrect rotor behaviour

Signed int cpt rotor is set to -1 on initialization.
cfs_cpt_spread_node() improperly handles this value
via "if (!rotor--)" check. The condition is never true
with negative rotor values, so for_each_node_mask()
only exits with node = MAX_NUMNODES.

kmalloc_node() attempts to determine the zonelist based
on the passed node id and maps MAX_NUMNODES to some
random pointer. Crash.

BUG: unable to handle kernel paging request at 0000000100002007
IP: [<ffffffff847c0da7>] __alloc_pages_nodemask+0x97/0x420

Change-Id: I4df74e394bdfc2a918d66aa12e6852ff0f6738ab
Signed-off-by: Andrew Perepechko <c17827@cray.com>
Cray-bug-id: LUS-8492
Reviewed-on: https://review.whamcloud.com/37709
Reviewed-by: Alexander Boyko <c17825@cray.com>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Reviewed-by: Alexander Zarochentsev <alexander.zarochentsev@hpe.com>
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/libcfs/include/libcfs/libcfs_cpu.h b/libcfs/include/libcfs/libcfs_cpu.h
index 34db1f6..404b07a 100644 (file)
--- a/libcfs/include/libcfs/libcfs_cpu.h
+++ b/libcfs/include/libcfs/libcfs_cpu.h
@@ -91,7 +91,7 @@ struct cfs_cpu_partition {
        /* NUMA distance between CPTs */
        unsigned int                    *cpt_distance;
        /* spread rotor for NUMA allocator */
-       int                              cpt_spread_rotor;
+       unsigned int                     cpt_spread_rotor;
        /* NUMA node if cpt_nodemask is empty */
        int                              cpt_node;
 };
@@ -101,7 +101,7 @@ struct cfs_cpu_partition {
 struct cfs_cpt_table {
 #ifdef CONFIG_SMP
        /* spread rotor for NUMA allocator */
-       int                              ctb_spread_rotor;
+       unsigned int                     ctb_spread_rotor;
        /* maximum NUMA distance between all nodes in table */
        unsigned int                     ctb_distance;
        /* partitions tables */

diff --git a/libcfs/libcfs/libcfs_cpu.c b/libcfs/libcfs/libcfs_cpu.c
index 808169a..68046a7 100644 (file)
--- a/libcfs/libcfs/libcfs_cpu.c
+++ b/libcfs/libcfs/libcfs_cpu.c
@@ -627,7 +627,7 @@ int cfs_cpt_spread_node(struct cfs_cpt_table *cptab, int cpt)
 {
        nodemask_t *mask;
        int weight;
-       int rotor;
+       unsigned int rotor;
        int node = 0;
 
        /* convert CPU partition ID to HW node id */