Re: [RFC net-next] net: veth: reduce page_pool memory footprint using half page per-buffer

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



From: Lorenzo Bianconi <lorenzo@xxxxxxxxxx>
Date: Fri, 12 May 2023 15:08:13 +0200

> In order to reduce page_pool memory footprint, rely on
> page_pool_dev_alloc_frag routine and reduce buffer size
> (VETH_PAGE_POOL_FRAG_SIZE) to PAGE_SIZE / 2 in order to consume one page
> for two 1500B frames. Reduce VETH_XDP_PACKET_HEADROOM to 192 from 256
> (XDP_PACKET_HEADROOM) to fit max_head_size in VETH_PAGE_POOL_FRAG_SIZE.
> Please note, using default values (CONFIG_MAX_SKB_FRAGS=17), maximum
> supported MTU is now reduced to 36350B.

I thought we're stepping away from page splitting bit by bit O_o
Primarily for the reasons you mentioned / worked around here: it creates
several significant limitations and at least on 64-bit systems it
doesn't scale anymore. 192 bytes of headroom is less than what XDP
expects (isn't it? Isn't 256 standard-standard, so that skb XDP path
reallocates heads only to have 256+ there?), 384 bytes of shinfo can
change anytime and even now page split simply blocks you from increasing
MAX_SKB_FRAGS even by one. Not speaking of MTU limitations etc.
BTW Intel drivers suffer from the very same things due solely to page
split (and I'm almost done with converting at least some of them to Page
Pool and 1 page per buffer model), I don't recommend deliberately
falling into that pit =\ :D

> 
> Signed-off-by: Lorenzo Bianconi <lorenzo@xxxxxxxxxx>
> ---
>  drivers/net/veth.c | 39 +++++++++++++++++++++++++--------------
>  1 file changed, 25 insertions(+), 14 deletions(-)
[...]

Thanks,
Olek




[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux