git
/
linux
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

netfilter: nft_set_pipapo: use avx2 algorithm for insertions too 

Always prefer the avx2 implementation if its available.
This greatly improves insertion performance (each insertion
checks if the new element would overlap with an existing one):

time nft -f - <<EOF
table ip pipapo {
	set s {
		typeof ip saddr . tcp dport
		flags interval
		size 800000
		elements = { 10.1.1.1 - 10.1.1.4 . 3996,
[.. 800k entries elided .. ]

before:
real    1m55.993s
user    0m2.505s
sys     1m53.296s

after:
real    0m42.586s
user    0m2.554s
sys     0m39.811s

Fold patch from Sebastian:

kernel_fpu_begin_mask()/ _end() remains in pipapo_get_avx2() where it is
required.

A followup patch will add local_lock_t to struct nft_pipapo_scratch in
order to protect the map pointer. The lock can not be acquired in
preemption disabled context which is what kernel_fpu_begin*() does.

Link: https://lore.kernel.org/netfilter-devel/20250818110213.1319982-2-bigeasy@linutronix.de/
Co-developed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
This commit is contained in:
Florian Westphal 2025-08-15 16:36:58 +02:00
parent 416e53e395
commit 84c1da7b38
3 changed files with 48 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
net/netfilter/nft_set_pipapo.c
View File

@ -397,7 +397,7 @@ int pipapo_refill(unsigned long *map, unsigned int len, unsigned int rules,
}
/**
* pipapo_get() - Get matching element reference given key data
* pipapo_get_slow() - Get matching element reference given key data
* @m: storage containing the set elements
* @data: Key data to be matched against existing elements
* @genmask: If set, check that element is active in given genmask
@ -414,9 +414,9 @@ int pipapo_refill(unsigned long *map, unsigned int len, unsigned int rules,
*
* Return: pointer to &struct nft_pipapo_elem on match, NULL otherwise.
*/
static struct nft_pipapo_elem *pipapo_get(const struct nft_pipapo_match *m,
const u8 *data, u8 genmask,
u64 tstamp)
static struct nft_pipapo_elem *pipapo_get_slow(const struct nft_pipapo_match *m,
const u8 *data, u8 genmask,
u64 tstamp)
{
struct nft_pipapo_scratch *scratch;
unsigned long *res_map, *fill_map;
@ -502,6 +502,41 @@ out:
return NULL;
}
/**
* pipapo_get() - Get matching element reference given key data
* @m: Storage containing the set elements
* @data: Key data to be matched against existing elements
* @genmask: If set, check that element is active in given genmask
* @tstamp: Timestamp to check for expired elements
*
* This is a dispatcher function, either calling out the generic C
* implementation or, if available, the AVX2 one.
* This helper is only called from the control plane, with either RCU
* read lock or transaction mutex held.
*
* Return: pointer to &struct nft_pipapo_elem on match, NULL otherwise.
*/
static struct nft_pipapo_elem *pipapo_get(const struct nft_pipapo_match *m,
const u8 *data, u8 genmask,
u64 tstamp)
{
struct nft_pipapo_elem *e;
local_bh_disable();
#if defined(CONFIG_X86_64) && !defined(CONFIG_UML)
if (boot_cpu_has(X86_FEATURE_AVX2) && boot_cpu_has(X86_FEATURE_AVX) &&
irq_fpu_usable()) {
e = pipapo_get_avx2(m, data, genmask, tstamp);
local_bh_enable();
return e;
}
#endif
e = pipapo_get_slow(m, data, genmask, tstamp);
local_bh_enable();
return e;
}
/**
* nft_pipapo_lookup() - Dataplane fronted for main lookup function
* @net: Network namespace
@ -523,7 +558,7 @@ nft_pipapo_lookup(const struct net *net, const struct nft_set *set,
const struct nft_pipapo_elem *e;
m = rcu_dereference(priv->match);
e = pipapo_get(m, (const u8 *)key, genmask, get_jiffies_64());
e = pipapo_get_slow(m, (const u8 *)key, genmask, get_jiffies_64());
return e ? &e->ext : NULL;
}

8
net/netfilter/nft_set_pipapo_avx2.c
View File

@ -1149,9 +1149,9 @@ static inline void pipapo_resmap_init_avx2(const struct nft_pipapo_match *m, uns
*
* Return: pointer to &struct nft_pipapo_elem on match, NULL otherwise.
*/
static struct nft_pipapo_elem *pipapo_get_avx2(const struct nft_pipapo_match *m,
const u8 *data, u8 genmask,
u64 tstamp)
struct nft_pipapo_elem *pipapo_get_avx2(const struct nft_pipapo_match *m,
const u8 *data, u8 genmask,
u64 tstamp)
{
struct nft_pipapo_scratch *scratch;
const struct nft_pipapo_field *f;
@ -1261,7 +1261,7 @@ next_match:
*
* This function is called from the data path. It will search for
* an element matching the given key in the current active copy using
* the AVX2 routines if the fpu is usable or fall back to the generic
* the AVX2 routines if the FPU is usable or fall back to the generic
* implementation of the algorithm otherwise.
*
* Return: nftables API extension pointer or NULL if no match.

4
net/netfilter/nft_set_pipapo_avx2.h
View File

@ -5,8 +5,12 @@
#include <asm/fpu/xstate.h>
#define NFT_PIPAPO_ALIGN (XSAVE_YMM_SIZE / BITS_PER_BYTE)
struct nft_pipapo_match;
bool nft_pipapo_avx2_estimate(const struct nft_set_desc *desc, u32 features,
struct nft_set_estimate *est);
struct nft_pipapo_elem *pipapo_get_avx2(const struct nft_pipapo_match *m,
const u8 *data, u8 genmask,
u64 tstamp);
#endif /* defined(CONFIG_X86_64) && !defined(CONFIG_UML) */
#endif /* _NFT_SET_PIPAPO_AVX2_H */