netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
To: Alexander Lobakin <aleksander.lobakin@intel.com>
Cc: "David S. Miller" <davem@davemloft.net>,
	"Eric Dumazet" <edumazet@google.com>,
	"Jakub Kicinski" <kuba@kernel.org>,
	"Paolo Abeni" <pabeni@redhat.com>,
	"Toke Høiland-Jørgensen" <toke@redhat.com>,
	"Alexei Starovoitov" <ast@kernel.org>,
	"Daniel Borkmann" <daniel@iogearbox.net>,
	"John Fastabend" <john.fastabend@gmail.com>,
	"Andrii Nakryiko" <andrii@kernel.org>,
	"Stanislav Fomichev" <sdf@fomichev.me>,
	"Magnus Karlsson" <magnus.karlsson@intel.com>,
	nex.sw.ncis.osdt.itp.upstreaming@intel.com, bpf@vger.kernel.org,
	netdev@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH net-next v2 15/18] xsk: add generic XSk &xdp_buff -> skb conversion
Date: Fri, 18 Oct 2024 14:48:18 +0200	[thread overview]
Message-ID: <ZxJZEn43W4y8EwsD@boxer> (raw)
In-Reply-To: <20241015145350.4077765-16-aleksander.lobakin@intel.com>

On Tue, Oct 15, 2024 at 04:53:47PM +0200, Alexander Lobakin wrote:
> Same as with converting &xdp_buff to skb on Rx, the code which allocates
> a new skb and copies the XSk frame there is identical across the
> drivers, so make it generic. This includes copying all the frags if they
> are present in the original buff.
> System percpu Page Pools help here a lot: when available, allocate pages
> from there instead of the MM layer. This greatly improves XDP_PASS
> performance on XSk: instead of page_alloc() + page_free(), the net core
> recycles the same pages, so the only overhead left is memcpy()s.
> Note that the passed buff gets freed if the conversion is done w/o any
> error, assuming you don't need this buffer after you convert it to an
> skb.

AFAICT looks good.
Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>

I have to switch the context now so I'll finish reviewing the remainder of
this set on monday.

 
> Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
> ---
>  include/net/xdp.h |   1 +
>  net/core/xdp.c    | 138 ++++++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 139 insertions(+)
> 
> diff --git a/include/net/xdp.h b/include/net/xdp.h
> index 83e3f4648caa..69728b2d75d5 100644
> --- a/include/net/xdp.h
> +++ b/include/net/xdp.h
> @@ -331,6 +331,7 @@ void xdp_warn(const char *msg, const char *func, const int line);
>  #define XDP_WARN(msg) xdp_warn(msg, __func__, __LINE__)
>  
>  struct sk_buff *xdp_build_skb_from_buff(const struct xdp_buff *xdp);
> +struct sk_buff *xdp_build_skb_from_zc(struct xdp_buff *xdp);
>  struct xdp_frame *xdp_convert_zc_to_xdp_frame(struct xdp_buff *xdp);
>  struct sk_buff *__xdp_build_skb_from_frame(struct xdp_frame *xdpf,
>  					   struct sk_buff *skb,
> diff --git a/net/core/xdp.c b/net/core/xdp.c
> index 371c26c203b2..116153b88d26 100644
> --- a/net/core/xdp.c
> +++ b/net/core/xdp.c
> @@ -22,6 +22,8 @@
>  #include <trace/events/xdp.h>
>  #include <net/xdp_sock_drv.h>
>  
> +#include "dev.h"
> +
>  #define REG_STATE_NEW		0x0
>  #define REG_STATE_REGISTERED	0x1
>  #define REG_STATE_UNREGISTERED	0x2
> @@ -682,6 +684,142 @@ struct sk_buff *xdp_build_skb_from_buff(const struct xdp_buff *xdp)
>  }
>  EXPORT_SYMBOL_GPL(xdp_build_skb_from_buff);
>  
> +/**
> + * xdp_copy_frags_from_zc - copy the frags from an XSk buff to an skb
> + * @skb: skb to copy frags to
> + * @xdp: XSk &xdp_buff from which the frags will be copied
> + * @pp: &page_pool backing page allocation, if available
> + *
> + * Copy all frags from an XSk &xdp_buff to an skb to pass it up the stack.
> + * Allocate a new page / page frag for each frag, copy it and attach to
> + * the skb.
> + *
> + * Return: true on success, false on page allocation fail.
> + */
> +static noinline bool xdp_copy_frags_from_zc(struct sk_buff *skb,
> +					    const struct xdp_buff *xdp,
> +					    struct page_pool *pp)
> +{
> +	const struct skb_shared_info *xinfo;
> +	struct skb_shared_info *sinfo;
> +	u32 nr_frags, ts;
> +
> +	xinfo = xdp_get_shared_info_from_buff(xdp);
> +	nr_frags = xinfo->nr_frags;
> +	sinfo = skb_shinfo(skb);
> +
> +#if IS_ENABLED(CONFIG_PAGE_POOL)
> +	ts = 0;
> +#else
> +	ts = xinfo->xdp_frags_truesize ? : nr_frags * xdp->frame_sz;
> +#endif
> +
> +	for (u32 i = 0; i < nr_frags; i++) {
> +		u32 len = skb_frag_size(&xinfo->frags[i]);
> +		void *data;
> +#if IS_ENABLED(CONFIG_PAGE_POOL)
> +		u32 truesize = len;
> +
> +		data = page_pool_dev_alloc_va(pp, &truesize);
> +		ts += truesize;
> +#else
> +		data = napi_alloc_frag(len);
> +#endif
> +		if (unlikely(!data))
> +			return false;
> +
> +		memcpy(data, skb_frag_address(&xinfo->frags[i]),
> +		       LARGEST_ALIGN(len));
> +		__skb_fill_page_desc(skb, sinfo->nr_frags++,
> +				     virt_to_page(data),
> +				     offset_in_page(data), len);
> +	}
> +
> +	xdp_update_skb_shared_info(skb, nr_frags, xinfo->xdp_frags_size,
> +				   ts, false);
> +
> +	return true;
> +}
> +
> +/**
> + * xdp_build_skb_from_zc - create an skb from an XSk &xdp_buff
> + * @xdp: source XSk buff
> + *
> + * Similar to xdp_build_skb_from_buff(), but for XSk frames. Allocate an skb
> + * head, new page for the head, copy the data and initialize the skb fields.
> + * If there are frags, allocate new pages for them and copy.
> + * If Page Pool is available, the function allocates memory from the system
> + * percpu pools to try recycling the pages, otherwise it uses the NAPI page
> + * frag caches.
> + * If new skb was built successfully, @xdp is returned to XSk pool's freelist.
> + * On error, it remains untouched and the caller must take care of this.
> + *
> + * Return: new &sk_buff on success, %NULL on error.
> + */
> +struct sk_buff *xdp_build_skb_from_zc(struct xdp_buff *xdp)
> +{
> +	const struct xdp_rxq_info *rxq = xdp->rxq;
> +	u32 len = xdp->data_end - xdp->data_meta;
> +	struct page_pool *pp;
> +	struct sk_buff *skb;
> +	int metalen;
> +#if IS_ENABLED(CONFIG_PAGE_POOL)
> +	u32 truesize;
> +	void *data;
> +
> +	pp = this_cpu_read(system_page_pool);
> +	truesize = xdp->frame_sz;
> +
> +	data = page_pool_dev_alloc_va(pp, &truesize);
> +	if (unlikely(!data))
> +		return NULL;
> +
> +	skb = napi_build_skb(data, truesize);
> +	if (unlikely(!skb)) {
> +		page_pool_free_va(pp, data, true);
> +		return NULL;
> +	}
> +
> +	skb_mark_for_recycle(skb);
> +	skb_reserve(skb, xdp->data_meta - xdp->data_hard_start);
> +#else /* !CONFIG_PAGE_POOL */
> +	struct napi_struct *napi;
> +
> +	pp = NULL;
> +	napi = napi_by_id(rxq->napi_id);
> +	if (likely(napi))
> +		skb = napi_alloc_skb(napi, len);
> +	else
> +		skb = __netdev_alloc_skb_ip_align(rxq->dev, len,
> +						  GFP_ATOMIC | __GFP_NOWARN);
> +	if (unlikely(!skb))
> +		return NULL;
> +#endif /* !CONFIG_PAGE_POOL */
> +
> +	memcpy(__skb_put(skb, len), xdp->data_meta, LARGEST_ALIGN(len));
> +
> +	metalen = xdp->data - xdp->data_meta;
> +	if (metalen > 0) {
> +		skb_metadata_set(skb, metalen);
> +		__skb_pull(skb, metalen);
> +	}
> +
> +	skb_record_rx_queue(skb, rxq->queue_index);
> +
> +	if (unlikely(xdp_buff_has_frags(xdp)) &&
> +	    unlikely(!xdp_copy_frags_from_zc(skb, xdp, pp))) {
> +		napi_consume_skb(skb, true);
> +		return NULL;
> +	}
> +
> +	xsk_buff_free(xdp);
> +
> +	skb->protocol = eth_type_trans(skb, rxq->dev);
> +
> +	return skb;
> +}
> +EXPORT_SYMBOL_GPL(xdp_build_skb_from_zc);
> +
>  struct sk_buff *__xdp_build_skb_from_frame(struct xdp_frame *xdpf,
>  					   struct sk_buff *skb,
>  					   struct net_device *dev)
> -- 
> 2.46.2
> 

  reply	other threads:[~2024-10-18 12:48 UTC|newest]

Thread overview: 42+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-10-15 14:53 [PATCH net-next v2 00/18] idpf: XDP chapter III: core XDP changes (+libeth_xdp) Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 01/18] jump_label: export static_key_slow_{inc,dec}_cpuslocked() Alexander Lobakin
2024-10-17 11:06   ` Maciej Fijalkowski
2024-10-21 13:53     ` Alexander Lobakin
2024-10-22 12:52       ` Maciej Fijalkowski
2024-10-15 14:53 ` [PATCH net-next v2 02/18] skbuff: allow 2-4-argument skb_frag_dma_map() Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 03/18] unroll: add generic loop unroll helpers Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 04/18] bpf, xdp: constify some bpf_prog * function arguments Alexander Lobakin
2024-10-17 11:12   ` Maciej Fijalkowski
2024-10-21 13:56     ` Alexander Lobakin
2024-10-22 12:55       ` Maciej Fijalkowski
2024-10-15 14:53 ` [PATCH net-next v2 05/18] xdp, xsk: constify read-only arguments of some static inline helpers Alexander Lobakin
2024-10-17 11:14   ` Maciej Fijalkowski
2024-10-21 13:57     ` Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 06/18] xdp: allow attaching already registered memory model to xdp_rxq_info Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 07/18] net: Register system page pool as an XDP memory model Alexander Lobakin
2024-10-17 11:32   ` Maciej Fijalkowski
2024-10-21 14:00     ` Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 08/18] page_pool: make page_pool_put_page_bulk() actually handle array of pages Alexander Lobakin
2024-10-17 11:33   ` Maciej Fijalkowski
2024-10-21 14:03     ` Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 09/18] page_pool: allow mixing PPs within one bulk Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 10/18] xdp: get rid of xdp_frame::mem.id Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 11/18] xdp: add generic xdp_buff_add_frag() Alexander Lobakin
2024-10-17 12:26   ` Maciej Fijalkowski
2024-10-21 14:10     ` Alexander Lobakin
2024-10-22 13:00       ` Maciej Fijalkowski
2024-10-15 14:53 ` [PATCH net-next v2 12/18] xdp: add generic xdp_build_skb_from_buff() Alexander Lobakin
2024-10-17 12:34   ` Maciej Fijalkowski
2024-10-21 14:20     ` Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 13/18] xsk: allow attaching XSk pool via xdp_rxq_info_reg_mem_model() Alexander Lobakin
2024-10-17 12:49   ` Maciej Fijalkowski
2024-10-21 14:23     ` Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 14/18] xsk: make xsk_buff_add_frag really add a frag via __xdp_buff_add_frag() Alexander Lobakin
2024-10-17 13:04   ` Maciej Fijalkowski
2024-10-15 14:53 ` [PATCH net-next v2 15/18] xsk: add generic XSk &xdp_buff -> skb conversion Alexander Lobakin
2024-10-18 12:48   ` Maciej Fijalkowski [this message]
2024-10-15 14:53 ` [PATCH net-next v2 16/18] xsk: add helper to get &xdp_desc's DMA and meta pointer in one go Alexander Lobakin
2024-10-22 15:42   ` Maciej Fijalkowski
2024-10-23 14:50     ` Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 17/18] libeth: support native XDP and register memory model Alexander Lobakin
2024-10-15 14:53 ` [PATCH net-next v2 18/18] libeth: add a couple of XDP helpers (libeth_xdp) Alexander Lobakin

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=ZxJZEn43W4y8EwsD@boxer \
    --to=maciej.fijalkowski@intel.com \
    --cc=aleksander.lobakin@intel.com \
    --cc=andrii@kernel.org \
    --cc=ast@kernel.org \
    --cc=bpf@vger.kernel.org \
    --cc=daniel@iogearbox.net \
    --cc=davem@davemloft.net \
    --cc=edumazet@google.com \
    --cc=john.fastabend@gmail.com \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=magnus.karlsson@intel.com \
    --cc=netdev@vger.kernel.org \
    --cc=nex.sw.ncis.osdt.itp.upstreaming@intel.com \
    --cc=pabeni@redhat.com \
    --cc=sdf@fomichev.me \
    --cc=toke@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).