Hello mm devs,
The patch 13808910713a: "Memoryless nodes: Generic management of
nodemasks for various purposes" from Oct 16, 2007, leads to the
following Smatch static checker warning:
./include/linux/nodemask.h:416 node_state()
warn: passing negative bit value '(-1)-1023' to 'test_bit()'
./include/linux/nodemask.h
406 /*
407 * The following particular system nodemasks and operations
408 * on them manage all possible and online nodes.
409 */
410
411 extern nodemask_t node_states[NR_NODE_STATES];
412
413 #if MAX_NUMNODES > 1
414 static inline int node_state(int node, enum node_states state)
415 {
--> 416 return node_isset(node, node_states[state]);
417 }
Smatch thinks there are a bunch of callers which seem to pass
NUMA_NO_NODE (-1) as the "node" value. Passing -1 as a bit number to
test_bit() is a bug.
Looking through the callers it feels like Smatch is probably correct.
drivers/iommu/intel/dmar.c | dmar_parse_one_rhsa | node_state | PARAM_VALUE | 0 | node | (-1)-64
drivers/acpi/numa/hmat.c | alloc_memory_initiator | node_state | PARAM_VALUE | 0 | node | (-1)-64
drivers/acpi/numa/hmat.c | hmat_parse_locality | node_state | PARAM_VALUE | 0 | node | (-1)-64
drivers/acpi/numa/hmat.c | hmat_register_target_initiators | node_state | PARAM_VALUE | 0 | node | (-1)-64
drivers/base/node.c | node_set_perf_attrs | node_state | PARAM_VALUE | 0 | node | (-1)-64
drivers/base/node.c | node_add_cache | node_state | PARAM_VALUE | 0 | node | (-1)-64
drivers/base/node.c | register_memory_node_under_compute_node | node_state | PARAM_VALUE | 0 | node | (-1)-64
drivers/base/node.c | register_memory_node_under_compute_node | node_state | PARAM_VALUE | 0 | node | (-1)-64
kernel/sched/fair.c | task_numa_placement | node_state | PARAM_VALUE | 0 | node | (-1)-64
regards,
dan carpenter
