linux-kernel - Re: [PATCH v4 2/2] nvme/pci: make PRP list DMA pools per-NUMA-node

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <aAfE_eofwrOIQ3Sw@kbusch-mbp.dhcp.thefacebook.com>
Date: Tue, 22 Apr 2025 10:34:05 -0600
From: Keith Busch <kbusch@...nel.org>
To: Caleb Sander Mateos <csander@...estorage.com>
Cc: Jens Axboe <axboe@...nel.dk>, Christoph Hellwig <hch@....de>,
	Sagi Grimberg <sagi@...mberg.me>,
	Kanchan Joshi <joshi.k@...sung.com>, linux-nvme@...ts.infradead.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH v4 2/2] nvme/pci: make PRP list DMA pools per-NUMA-node

On Tue, Apr 22, 2025 at 10:19:59AM -0600, Caleb Sander Mateos wrote:
> NVMe commands with more than 4 KB of data allocate PRP list pages from
> the per-nvme_device dma_pool prp_page_pool or prp_small_pool. Each call
> to dma_pool_alloc() and dma_pool_free() takes the per-dma_pool spinlock.
> These device-global spinlocks are a significant source of contention
> when many CPUs are submitting to the same NVMe devices. On a workload
> issuing 32 KB reads from 16 CPUs (8 hypertwin pairs) across 2 NUMA nodes
> to 23 NVMe devices, we observed 2.4% of CPU time spent in
> _raw_spin_lock_irqsave called from dma_pool_alloc and dma_pool_free.
> 
> Ideally, the dma_pools would be per-hctx to minimize
> contention. But that could impose considerable resource costs in a
> system with many NVMe devices and CPUs.
> 
> As a compromise, allocate per-NUMA-node PRP list DMA pools. Map each
> nvme_queue to the set of DMA pools corresponding to its device and its
> hctx's NUMA node. This reduces the _raw_spin_lock_irqsave overhead by
> about half, to 1.2%. Preventing the sharing of PRP list pages across
> NUMA nodes also makes them cheaper to initialize.

I was hoping for an even greater improvement, but still good. Maybe we
can do slightly better if we also pass the numa_node to
dma_pool_create() and allocate the 'struct dma_pool' on the same node
that's going to use it.

Looks good,

Reviewed-by: Keith Busch <kbusch@...nel.org>