netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Sasha Levin <sashal@kernel.org>
To: linux-kernel@vger.kernel.org, stable@vger.kernel.org
Cc: Eric Dumazet <edumazet@google.com>,
	Jason Xing <kerneljasonxing@gmail.com>,
	Kuniyuki Iwashima <kuniyu@amazon.com>,
	Jakub Kicinski <kuba@kernel.org>, Sasha Levin <sashal@kernel.org>,
	ncardwell@google.com, davem@davemloft.net, dsahern@kernel.org,
	pabeni@redhat.com, netdev@vger.kernel.org
Subject: [PATCH AUTOSEL 5.10 039/114] tcp: bring back NUMA dispersion in inet_ehash_locks_alloc()
Date: Mon,  5 May 2025 19:17:02 -0400	[thread overview]
Message-ID: <20250505231817.2697367-39-sashal@kernel.org> (raw)
In-Reply-To: <20250505231817.2697367-1-sashal@kernel.org>

From: Eric Dumazet <edumazet@google.com>

[ Upstream commit f8ece40786c9342249aa0a1b55e148ee23b2a746 ]

We have platforms with 6 NUMA nodes and 480 cpus.

inet_ehash_locks_alloc() currently allocates a single 64KB page
to hold all ehash spinlocks. This adds more pressure on a single node.

Change inet_ehash_locks_alloc() to use vmalloc() to spread
the spinlocks on all online nodes, driven by NUMA policies.

At boot time, NUMA policy is interleave=all, meaning that
tcp_hashinfo.ehash_locks gets hash dispersion on all nodes.

Tested:

lack5:~# grep inet_ehash_locks_alloc /proc/vmallocinfo
0x00000000d9aec4d1-0x00000000a828b652   69632 inet_ehash_locks_alloc+0x90/0x100 pages=16 vmalloc N0=2 N1=3 N2=3 N3=3 N4=3 N5=2

lack5:~# echo 8192 >/proc/sys/net/ipv4/tcp_child_ehash_entries
lack5:~# numactl --interleave=all unshare -n bash -c "grep inet_ehash_locks_alloc /proc/vmallocinfo"
0x000000004e99d30c-0x00000000763f3279   36864 inet_ehash_locks_alloc+0x90/0x100 pages=8 vmalloc N0=1 N1=2 N2=2 N3=1 N4=1 N5=1
0x00000000d9aec4d1-0x00000000a828b652   69632 inet_ehash_locks_alloc+0x90/0x100 pages=16 vmalloc N0=2 N1=3 N2=3 N3=3 N4=3 N5=2

lack5:~# numactl --interleave=0,5 unshare -n bash -c "grep inet_ehash_locks_alloc /proc/vmallocinfo"
0x00000000fd73a33e-0x0000000004b9a177   36864 inet_ehash_locks_alloc+0x90/0x100 pages=8 vmalloc N0=4 N5=4
0x00000000d9aec4d1-0x00000000a828b652   69632 inet_ehash_locks_alloc+0x90/0x100 pages=16 vmalloc N0=2 N1=3 N2=3 N3=3 N4=3 N5=2

lack5:~# echo 1024 >/proc/sys/net/ipv4/tcp_child_ehash_entries
lack5:~# numactl --interleave=all unshare -n bash -c "grep inet_ehash_locks_alloc /proc/vmallocinfo"
0x00000000db07d7a2-0x00000000ad697d29    8192 inet_ehash_locks_alloc+0x90/0x100 pages=1 vmalloc N2=1
0x00000000d9aec4d1-0x00000000a828b652   69632 inet_ehash_locks_alloc+0x90/0x100 pages=16 vmalloc N0=2 N1=3 N2=3 N3=3 N4=3 N5=2

Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Jason Xing <kerneljasonxing@gmail.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://patch.msgid.link/20250305130550.1865988-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 net/ipv4/inet_hashtables.c | 37 ++++++++++++++++++++++++++-----------
 1 file changed, 26 insertions(+), 11 deletions(-)

diff --git a/net/ipv4/inet_hashtables.c b/net/ipv4/inet_hashtables.c
index 0fb5d758264fe..fea74ab2a4bec 100644
--- a/net/ipv4/inet_hashtables.c
+++ b/net/ipv4/inet_hashtables.c
@@ -934,22 +934,37 @@ int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
 {
 	unsigned int locksz = sizeof(spinlock_t);
 	unsigned int i, nblocks = 1;
+	spinlock_t *ptr = NULL;
 
-	if (locksz != 0) {
-		/* allocate 2 cache lines or at least one spinlock per cpu */
-		nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
-		nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
+	if (locksz == 0)
+		goto set_mask;
 
-		/* no more locks than number of hash buckets */
-		nblocks = min(nblocks, hashinfo->ehash_mask + 1);
+	/* Allocate 2 cache lines or at least one spinlock per cpu. */
+	nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U) * num_possible_cpus();
 
-		hashinfo->ehash_locks = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
-		if (!hashinfo->ehash_locks)
-			return -ENOMEM;
+	/* At least one page per NUMA node. */
+	nblocks = max(nblocks, num_online_nodes() * PAGE_SIZE / locksz);
+
+	nblocks = roundup_pow_of_two(nblocks);
+
+	/* No more locks than number of hash buckets. */
+	nblocks = min(nblocks, hashinfo->ehash_mask + 1);
 
-		for (i = 0; i < nblocks; i++)
-			spin_lock_init(&hashinfo->ehash_locks[i]);
+	if (num_online_nodes() > 1) {
+		/* Use vmalloc() to allow NUMA policy to spread pages
+		 * on all available nodes if desired.
+		 */
+		ptr = vmalloc_array(nblocks, locksz);
+	}
+	if (!ptr) {
+		ptr = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
+		if (!ptr)
+			return -ENOMEM;
 	}
+	for (i = 0; i < nblocks; i++)
+		spin_lock_init(&ptr[i]);
+	hashinfo->ehash_locks = ptr;
+set_mask:
 	hashinfo->ehash_locks_mask = nblocks - 1;
 	return 0;
 }
-- 
2.39.5


  parent reply	other threads:[~2025-05-05 23:19 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <20250505231817.2697367-1-sashal@kernel.org>
2025-05-05 23:16 ` [PATCH AUTOSEL 5.10 008/114] SUNRPC: rpc_clnt_set_transport() must not change the autobind setting Sasha Levin
2025-05-05 23:16 ` [PATCH AUTOSEL 5.10 009/114] SUNRPC: rpcbind should never reset the port to the value '0' Sasha Levin
2025-05-05 23:16 ` [PATCH AUTOSEL 5.10 027/114] tcp: reorganize tcp_in_ack_event() and tcp_count_delivered() Sasha Levin
2025-05-05 23:16 ` [PATCH AUTOSEL 5.10 033/114] netfilter: conntrack: Bound nf_conntrack sysctl writes Sasha Levin
2025-05-05 23:16 ` [PATCH AUTOSEL 5.10 036/114] ipv6: save dontfrag in cork Sasha Levin
2025-05-05 23:17 ` Sasha Levin [this message]
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 041/114] ieee802154: ca8210: Use proper setters and getters for bitwise types Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 049/114] net: ethernet: ti: cpsw_new: populate netdev of_node Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 050/114] net: pktgen: fix mpls maximum labels list parsing Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 051/114] ipv4: fib: Move fib_valid_key_len() to rtm_to_fib_config() Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 059/114] net/mlx5: Avoid report two health errors on same syndrome Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 061/114] net: xgene-v2: remove incorrect ACPI_PTR annotation Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 062/114] bonding: report duplicate MAC address in all situations Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 075/114] net: pktgen: fix access outside of user given buffer in pktgen_thread_write() Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 077/114] bpf: Prevent unsafe access to the sock fields in the BPF timestamping callback Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 079/114] eth: mlx4: don't try to complete XDP frames in netpoll Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 082/114] net/mlx5: Modify LSB bitmask in temperature event to include only the first bit Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 083/114] net/mlx5: Apply rate-limiting to high temperature warning Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 090/114] net/mlx4_core: Avoid impossible mlx4_db_alloc() order value Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 093/114] net/mlx5: Extend Ethtool loopback selftest to support non-linear SKB Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 094/114] net/mlx5e: set the tx_queue_len for pfifo_fast Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 095/114] net/mlx5e: reduce rep rxq depth to 256 for ECPF Sasha Levin
2025-05-05 23:17 ` [PATCH AUTOSEL 5.10 096/114] ip: fib_rules: Fetch net from fib_rule in fib[46]_rule_configure() Sasha Levin
2025-05-05 23:18 ` [PATCH AUTOSEL 5.10 100/114] vxlan: Annotate FDB data races Sasha Levin
2025-05-05 23:18 ` [PATCH AUTOSEL 5.10 101/114] net-sysfs: prevent uncleared queues from being re-added Sasha Levin

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=20250505231817.2697367-39-sashal@kernel.org \
    --to=sashal@kernel.org \
    --cc=davem@davemloft.net \
    --cc=dsahern@kernel.org \
    --cc=edumazet@google.com \
    --cc=kerneljasonxing@gmail.com \
    --cc=kuba@kernel.org \
    --cc=kuniyu@amazon.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=ncardwell@google.com \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=stable@vger.kernel.org \
    /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).