[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <d16327f3-33c0-61c1-3adf-78f7edcbec6a@gmail.com>
Date: Fri, 26 Feb 2021 15:38:06 +0100
From: Eric Dumazet <eric.dumazet@...il.com>
To: Yejune Deng <yejune.deng@...il.com>, davem@...emloft.net,
yoshfuji@...ux-ipv6.org, dsahern@...nel.org, kuba@...nel.org
Cc: netdev@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] inetpeer: use else if instead of if to reduce judgment
On 2/26/21 11:57 AM, Yejune Deng wrote:
> In inet_initpeers(), if si.totalram <= (8192*1024)/PAGE_SIZE, it will
> be judged three times. Use else if instead of if, it only needs to be
> judged once.
>
> Signed-off-by: Yejune Deng <yejune.deng@...il.com>
> ---
> net/ipv4/inetpeer.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/net/ipv4/inetpeer.c b/net/ipv4/inetpeer.c
> index ff327a62c9ce..07cd1f8204b3 100644
> --- a/net/ipv4/inetpeer.c
> +++ b/net/ipv4/inetpeer.c
> @@ -81,12 +81,12 @@ void __init inet_initpeers(void)
> * <kuznet@....inr.ac.ru>. I don't have any opinion about the values
> * myself. --SAW
> */
> - if (si.totalram <= (32768*1024)/PAGE_SIZE)
> + if (si.totalram <= (8192 * 1024) / PAGE_SIZE)
> + inet_peer_threshold >>= 4; /* about 128KB */
> + else if (si.totalram <= (16384 * 1024) / PAGE_SIZE)
> + inet_peer_threshold >>= 2; /* about 512KB */
> + else if (si.totalram <= (32768 * 1024) / PAGE_SIZE)
> inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
If you really want to change this stuff, I would suggest updating comments,
because nowadays, struct inet_peer on IA32 uses 128 bytes.
So 32768 entries would consume 4 MB,
16384 entries would consume 2 MB
and 4096 entries would consume 512KB
Another idea would be to get rid of the cascade and use something that
will not need to be adjusted in the future.
diff --git a/net/ipv4/inetpeer.c b/net/ipv4/inetpeer.c
index ff327a62c9ce9b1794104c3c924f5f2b9820ac8b..d5f486bd8c35234f99b22842e756a10531e070d6 100644
--- a/net/ipv4/inetpeer.c
+++ b/net/ipv4/inetpeer.c
@@ -65,7 +65,7 @@ EXPORT_SYMBOL_GPL(inet_peer_base_init);
#define PEER_MAX_GC 32
/* Exported for sysctl_net_ipv4. */
-int inet_peer_threshold __read_mostly = 65536 + 128; /* start to throw entries more
+int inet_peer_threshold __read_mostly; /* start to throw entries more
* aggressively at this stage */
int inet_peer_minttl __read_mostly = 120 * HZ; /* TTL under high load: 120 sec */
int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min */
@@ -73,20 +73,13 @@ int inet_peer_maxttl __read_mostly = 10 * 60 * HZ; /* usual time to live: 10 min
/* Called from ip_output.c:ip_init */
void __init inet_initpeers(void)
{
- struct sysinfo si;
+ u64 nr_entries;
- /* Use the straight interface to information about memory. */
- si_meminfo(&si);
- /* The values below were suggested by Alexey Kuznetsov
- * <kuznet@....inr.ac.ru>. I don't have any opinion about the values
- * myself. --SAW
- */
- if (si.totalram <= (32768*1024)/PAGE_SIZE)
- inet_peer_threshold >>= 1; /* max pool size about 1MB on IA32 */
- if (si.totalram <= (16384*1024)/PAGE_SIZE)
- inet_peer_threshold >>= 1; /* about 512KB */
- if (si.totalram <= (8192*1024)/PAGE_SIZE)
- inet_peer_threshold >>= 2; /* about 128KB */
+ /* 1% of physical memory */
+ nr_entries = div64_ul((u64)totalram_pages() << PAGE_SHIFT,
+ 100 * L1_CACHE_ALIGN(sizeof(struct inet_peer)));
+
+ inet_peer_threshold = clamp_val(nr_entries, 4096, 65536 + 128);
peer_cachep = kmem_cache_create("inet_peer_cache",
sizeof(struct inet_peer),
Powered by blists - more mailing lists