netdev - Re: [RFC PATCH v2 03/14] xsk: add umem fill queue support and mmap

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]

Date:   Thu, 12 Apr 2018 05:15:42 +0300
From:   "Michael S. Tsirkin" <mst@...hat.com>
To:     Björn Töpel <bjorn.topel@...il.com>
Cc:     magnus.karlsson@...el.com, alexander.h.duyck@...el.com,
        alexander.duyck@...il.com, john.fastabend@...il.com, ast@...com,
        brouer@...hat.com, willemdebruijn.kernel@...il.com,
        daniel@...earbox.net, netdev@...r.kernel.org,
        michael.lundkvist@...csson.com, jesse.brandeburg@...el.com,
        anjali.singhai@...el.com, qi.z.zhang@...el.com,
        ravineet.singh@...csson.com
Subject: Re: [RFC PATCH v2 03/14] xsk: add umem fill queue support and mmap

On Tue, Mar 27, 2018 at 06:59:08PM +0200, Björn Töpel wrote:
> @@ -30,4 +31,18 @@ struct xdp_umem_reg {
>  	__u32 frame_headroom; /* Frame head room */
>  };
>  
> +/* Pgoff for mmaping the rings */
> +#define XDP_UMEM_PGOFF_FILL_QUEUE	0x100000000
> +
> +struct xdp_queue {
> +	__u32 head_idx __attribute__((aligned(64)));
> +	__u32 tail_idx __attribute__((aligned(64)));
> +};
> +
> +/* Used for the fill and completion queues for buffers */
> +struct xdp_umem_queue {
> +	struct xdp_queue ptrs;
> +	__u32 desc[0] __attribute__((aligned(64)));
> +};
> +
>  #endif /* _LINUX_IF_XDP_H */

So IIUC it's a head/tail ring of 32 bit descriptors.

In my experience (from implementing ptr_ring) this
implies that head/tail cache lines bounce a lot between
CPUs. Caching will help some. You are also forced to
use barriers to check validity which is slow on
some architectures.

If instead you can use a special descriptor value (e.g. 0) as
a valid signal, things work much better:

- you read descriptor atomically, if it's not 0 it's fine
- same with write - write 0 to pass it to the other side
- there is a data dependency so no need for barriers (except on dec alpha)
- no need for power of 2 limitations, you can make it any size you like
- easy to resize too

architecture (if not implementation) would be shared with ptr_ring
so some of the optimization ideas like batched updates could
be lifted from there.

When I was building ptr_ring, any head/tail design underperformed
storing valid flag with data itself. YMMV.

-- 
MST