[PATCH v4 0/2] nvme/pci: PRP list DMA pool partitioning

[Caleb Sander Mateos]

On Tue, Apr 22, 2025 at 10:48 AM Keith Busch <kbusch at kernel.org> wrote:
>
> On Tue, Apr 22, 2025 at 10:19:57AM -0600, Caleb Sander Mateos wrote:
> > This reduces the _raw_spin_lock_irqsave overhead by about half, to
> > 1.2%.
>
> Could you try this atop your series? I hope to see if we can squeeze a
> little more out by keeping the spinlock and list links local to the node
> using them.

That does help save some CPU time in dma_pool_alloc too. It's down
from 0.87% of CPU time to 0.52% while doing 5% more IOPS. I'll include
your patch (with a few tweaks) in a v5 of this series. Hopefully
you're okay if I include your Signed-off-by.

Thanks,
Caleb