[PATCH v2 2/2] nvme/pci: make PRP list DMA pools per-NUMA-node

[Caleb Sander Mateos]

On Mon, Apr 21, 2025 at 9:34 AM Keith Busch <kbusch at kernel.org> wrote:
>
> On Mon, Apr 21, 2025 at 10:17:25AM -0600, Caleb Sander Mateos wrote:
> > +static void nvme_release_prp_pools(struct nvme_dev *dev)
> > +{
> > +     struct nvme_prp_dma_pools *pools_end = dev->prp_pools + nr_node_ids;
> > +     struct nvme_prp_dma_pools *prp_pools;
> > +
> > +     for (prp_pools = dev->prp_pools; prp_pools < pools_end; prp_pools++) {
> > +             if (!prp_pools->small)
> > +                     continue;
> > +
> > +             dma_pool_destroy(prp_pools->large);
> > +             dma_pool_destroy(prp_pools->small);
> > +     }
> > +}
>
> A minor difference in style, I think indexing looks cleaner than
> incrementing pointers:

Sure, can do that instead.

>
> static void nvme_release_prp_pools(struct nvme_dev *dev)
> {
>         int i;
>
>         for (i = 0; i < nr_node_ids; i++) {
>                 dma_pool_destroy(dev->prp_pools[i].small);
>                 dma_pool_destroy(dev->prp_pools[i].large);
>         }
> }
>
> Note, dma_pool_destroy() already checks for NULL, so no need to check
> before calling it.

In that case, nvme_setup_prp_pools() would need to clear the large
dma_pool pointer in case of partial initialization (the large pool is
initialized successfully but the small pool fails). I can make that
switch if you'd prefer.

Best,
Caleb