numademo: fix wrong node input #32
Conversation
| @@ -436,7 +436,7 @@ void test(enum test type) | |||
| numa_set_localalloc(); | |||
| memtest("local allocation", numa_alloc(msize)); | |||
|
|
|||
| numa_set_preferred((node_to_use[i]+1) % numnodes ); | |||
There was a problem hiding this comment.
This code preceded memtest("setting wrong preferred node", numa_alloc(msize));, but after your change it behaves just the same as memtest("setting wrong preferred node", numa_alloc(msize));...
In a way, this is trying to find the "next" node to create a mismatch.
Perhaps a possible approach would be earlier on when listing valid nodes create a map that points to the next valid node from the current one?
There was a problem hiding this comment.
Perhaps this might work?
numa_set_preferred(node_to_use[(i + 1) % numnodes]);That assumes that above in get_node_list(), got_nodes in the end matches numnodes, perhaps adding an assertion for that would be good there...
There was a problem hiding this comment.
@filbranden The actual trick I missed earlier and this fix is the precedence with which i was used. As you pointed out, (node_to_use[(i + 1) % numnodes]) rightly provides "wrongly preferred node". And yes, got_nodes should match the numnodes, i have added a check for it.
In few sparse node systems, test fails with invalid argument at set_mempolicy through numa_preferred_node. Patch fixes it by providing a valid wrong node. Signed-off-by: Harish <harish@linux.vnet.ibm.com>
harish-24
force-pushed
the
fix_wrong_node
branch
from
5f0e6a8 to
f9880e5
Compare
In few sparse node systems, test fails with invalid argument at set_mempolicy through numa_preferred_node. Patch fixes it by providing a valid wrong node.
Tested on both sparse and contiguous types of nodes.
Signed-off-by: Harish harish@linux.vnet.ibm.com