diff mbox

nfs broken in net-next? -- now in mainline -- bisected: d9f5950f90292f7cc42834338dfd5f44dc4cc4ca

Message ID 86802c440912121825h20663affv9ff8d004974d0d31@mail.gmail.com
State Not Applicable, archived
Delegated to: David Miller
Headers show

Commit Message

Yinghai Lu Dec. 13, 2009, 2:25 a.m. UTC
On Sat, Dec 12, 2009 at 5:57 PM, Yinghai Lu <yinghai@kernel.org> wrote:
> On Sat, Dec 12, 2009 at 5:05 PM, Yinghai Lu <yinghai@kernel.org> wrote:
>>
>> [f86dcc5aa8c7908f2c287e7a211228df599e3e71] udp: dynamically
>> size hash tables at boot time
>
> commit f86dcc5aa8c7908f2c287e7a211228df599e3e71
> Author: Eric Dumazet <eric.dumazet@gmail.com>
> Date:   Wed Oct 7 00:37:59 2009 +0000
>
>    udp: dynamically size hash tables at boot time
>
>    UDP_HTABLE_SIZE was initialy defined to 128, which is a bit small for
>    several setups.
>
>    4000 active UDP sockets -> 32 sockets per chain in average. An
>    incoming frame has to lookup all sockets to find best match, so long
>    chains hurt latency.
>
>    Instead of a fixed size hash table that cant be perfect for every
>    needs, let UDP stack choose its table size at boot time like tcp/ip
>    route, using alloc_large_system_hash() helper
>
>    Add an optional boot parameter, uhash_entries=x so that an admin can
>    force a size between 256 and 65536 if needed, like thash_entries and
>    rhash_entries.
>
>    dmesg logs two new lines :
>    [    0.647039] UDP hash table entries: 512 (order: 0, 4096 bytes)
>    [    0.647099] UDP Lite hash table entries: 512 (order: 0, 4096 bytes)
>
>    Maximal size on 64bit arches would be 65536 slots, ie 1 MBytes for non
>    debugging spinlocks.
>
>    Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
>    Signed-off-by: David S. Miller <davem@davemloft.net>
>
> cause the problem: nfs mount fail.
>
> the setup is:
> 64bit kernel, have all needed drivers in kernel, and boot with
> ip=dhcp, root disk is 256M ramdisk.
> then try to mount nfs....
>

change entries default value from 65536 to 256,
nfs mount will work.

YH

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Yinghai Lu Dec. 13, 2009, 2:41 a.m. UTC | #1
On Sat, Dec 12, 2009 at 6:25 PM, Yinghai Lu <yinghai@kernel.org> wrote:
> On Sat, Dec 12, 2009 at 5:57 PM, Yinghai Lu <yinghai@kernel.org> wrote:
>> On Sat, Dec 12, 2009 at 5:05 PM, Yinghai Lu <yinghai@kernel.org> wrote:
>>>
>>> [f86dcc5aa8c7908f2c287e7a211228df599e3e71] udp: dynamically
>>> size hash tables at boot time
>>
>> commit f86dcc5aa8c7908f2c287e7a211228df599e3e71
>> Author: Eric Dumazet <eric.dumazet@gmail.com>
>> Date:   Wed Oct 7 00:37:59 2009 +0000
>>
>>    udp: dynamically size hash tables at boot time
>>
>>    UDP_HTABLE_SIZE was initialy defined to 128, which is a bit small for
>>    several setups.
>>
>>    4000 active UDP sockets -> 32 sockets per chain in average. An
>>    incoming frame has to lookup all sockets to find best match, so long
>>    chains hurt latency.
>>
>>    Instead of a fixed size hash table that cant be perfect for every
>>    needs, let UDP stack choose its table size at boot time like tcp/ip
>>    route, using alloc_large_system_hash() helper
>>
>>    Add an optional boot parameter, uhash_entries=x so that an admin can
>>    force a size between 256 and 65536 if needed, like thash_entries and
>>    rhash_entries.
>>
>>    dmesg logs two new lines :
>>    [    0.647039] UDP hash table entries: 512 (order: 0, 4096 bytes)
>>    [    0.647099] UDP Lite hash table entries: 512 (order: 0, 4096 bytes)
>>
>>    Maximal size on 64bit arches would be 65536 slots, ie 1 MBytes for non
>>    debugging spinlocks.
>>
>>    Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
>>    Signed-off-by: David S. Miller <davem@davemloft.net>
>>
>> cause the problem: nfs mount fail.
>>
>> the setup is:
>> 64bit kernel, have all needed drivers in kernel, and boot with
>> ip=dhcp, root disk is 256M ramdisk.
>> then try to mount nfs....
>>
>
> change entries default value from 65536 to 256,
> nfs mount will work.
>
> YH
>
> diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
> index 1f95348..57c13c3 100644
> --- a/net/ipv4/udp.c
> +++ b/net/ipv4/udp.c
> @@ -2054,7 +2054,7 @@ void udp4_proc_exit(void)
>  }
>  #endif /* CONFIG_PROC_FS */
>
> -static __initdata unsigned long uhash_entries;
> +static __initdata unsigned long uhash_entries = UDP_HTABLE_SIZE_MIN;
>  static int __init set_uhash_entries(char *str)
>  {
>        if (!str)
>

interesting:

m:~/dump # grep UDP dmesg.txt
[   28.996034] UDP hash table entries: 65536 (order: 11, 10485760 bytes)
[   29.032364] UDP-Lite hash table entries: 65536 (order: 11, 10485760 bytes)

will not work

but 32768 will work.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index 1f95348..57c13c3 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -2054,7 +2054,7 @@  void udp4_proc_exit(void)
 }
 #endif /* CONFIG_PROC_FS */

-static __initdata unsigned long uhash_entries;
+static __initdata unsigned long uhash_entries = UDP_HTABLE_SIZE_MIN;
 static int __init set_uhash_entries(char *str)
 {
        if (!str)