On 4/1/25 01:42, Joanne Koong wrote: > On Mon, Mar 31, 2025 at 1:57 PM Joanne Koong <joannelkoong@xxxxxxxxx> wrote: >> >> There is a 1:1 mapping between cpus and queues. Allocate the queue on >> the numa node associated with the cpu to help reduce memory access >> latencies. >> >> Signed-off-by: Joanne Koong <joannelkoong@xxxxxxxxx> >> --- >> fs/fuse/dev_uring.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/fs/fuse/dev_uring.c b/fs/fuse/dev_uring.c >> index accdce2977c5..0762d6229ac6 100644 >> --- a/fs/fuse/dev_uring.c >> +++ b/fs/fuse/dev_uring.c >> @@ -256,7 +256,7 @@ static struct fuse_ring_queue *fuse_uring_create_queue(struct fuse_ring *ring, >> struct fuse_ring_queue *queue; >> struct list_head *pq; >> >> - queue = kzalloc(sizeof(*queue), GFP_KERNEL_ACCOUNT); >> + queue = kzalloc_node(sizeof(*queue), GFP_KERNEL_ACCOUNT, cpu_to_node(qid)); >> if (!queue) >> return NULL; >> pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL); > > On the same note I guess we should also allocate pq on the > corresponding numa node too. So this is supposed to be called from a thread that already runs on this numa node and then kmalloc will allocate anyway on the right node, afaik. Do you have a use case where this is called from another node? If you do, all allocations in this file should be changed. Thanks, Bernd
