[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <10257218-3b58-fd2e-c38f-ce320ca62de5@iogearbox.net>
Date: Mon, 31 Aug 2020 22:20:59 +0200
From: Daniel Borkmann <daniel@...earbox.net>
To: Magnus Karlsson <magnus.karlsson@...el.com>, bjorn.topel@...el.com,
ast@...nel.org, netdev@...r.kernel.org, jonathan.lemon@...il.com,
maximmi@...lanox.com
Cc: bpf@...r.kernel.org, jeffrey.t.kirsher@...el.com,
anthony.l.nguyen@...el.com, maciej.fijalkowski@...el.com,
maciejromanfijalkowski@...il.com, cristian.dumitrescu@...el.com
Subject: Re: [PATCH bpf-next v5 00/15] xsk: support shared umems between
devices and queues
On 8/28/20 10:26 AM, Magnus Karlsson wrote:
[...]
> v4 -> v5:
>
> * Fixed performance problem with sharing a umem between different
> queues on the same netdev. Sharing the dma_pages array between
> buffer pool instances was a bad idea. It led to many cross-core
> snoop traffic messages that degraded performance. The solution: only
> map the dma mappings once as before, but copy the dma_addr_t to a per
> buffer pool array so that this sharing dissappears.
> * Added patch 10 that improves performance with 3% for l2fwd with a
> simple fix that is now possible, as we pass the buffer pool to the driver.
> * xp_dma_unmap() did not honor the refcount. Fixed. [Maxim]
> * Fixed bisectabilty problem in patch 5 [Maxim]
Applied, thanks!
Powered by blists - more mailing lists