Re: l2fwd mmap memory failed

[jerry <jerry.lilijun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org>]

Hi,

On 2014/9/14 19:12, Zhang, Jerry wrote:
> Hi,
> 
>> -----Original Message-----
>> From: dev [mailto:dev-bounces-VfR2kkLFssw@public.gmane.org] On Behalf Of jerry
>> Sent: Sunday, September 14, 2014 5:01 PM
>> To: dev-VfR2kkLFssw@public.gmane.org
>> Cc: luonengjun-hv44wF8Li93QT0dZR+AlfA@public.gmane.org
>> Subject: [dpdk-dev] l2fwd mmap memory failed
>>
>> Hi all,
>>
>> The l2fwd sample application starts failed in my environment with 90000 2M
>> hugepages set up.
>> It tells me that mmap failed: Cannot allocate memory.
>> Is there a limited max hugepages or memory size for dpdk to mmap EAL
>> memory?
>>
>> Some information as follows:
>>
>> 1. Environment:
>> Host OS: Suse11 Sp3 x86_64
>> NIC: intel 82599
>> DPDK: 1.6.0r2
>>
>> 2. Hugepage Configuration in Kernel command line:
>> hugepagesz=2M default_hugepagesz=2M hugepages=90000
>>
>> 3. l2fwd command line:
>> ./build/l2fwd -c 0x3 -n 3 --socket-mem 2048 -- -q 1 -p 3
>>
>> 4. cat /proc/meminfo:
>> HugePages_Total:   90000
>> HugePages_Free:    90000
>> HugePages_Rsvd:        0
>> HugePages_Surp:        0
>> Hugepagesize:       2048 kB
>>
>> 5. Error messages in dpdk log:
>> Sun Sep 14 08:37:28 2014:EAL: Detected lcore 0 as core 0 on socket 0 Sun Sep 14
>> 08:37:28 2014:EAL: Detected lcore 1 as core 1 on socket 0 Sun Sep 14 08:37:28
>> 2014:EAL: Setting up memory...
>> Sun Sep 14 08:38:44 2014:EAL: map_all_hugepages(): mmap failed: Cannot
>> allocate memory Sun Sep 14 08:38:44 2014:EAL: Failed to mmap 2 MB
>> hugepages Sun Sep 14 08:38:44 2014:EAL: Cannot init memory Sun Sep 14
>> 08:38:44 2014:EAL: Error - exiting with code: 1
>>  Cause: Sun Sep 14 08:38:44 2014:Invalid EAL parameters
>>
>>
>> Anyone know the issue and the corresponding fix? Thanks.
> 
> It's a system mmap failure. 
> As you have to map so many huge pages, I suggest that you check /proc/sys/vm/max_map_count to see if the map count exceeds the max value.
> 
> 
That works when echo 200000 > /proc/sys/vm/max_map_count.
I have tried the count 100000 and it still failed. This reason may be that dpdk calls map_all_hugepages() twice in my opinion.
Thanks for your help.