[dpdk-dev] Problem with MBUF pool allocation

[dpdk-dev] Problem with MBUF pool allocation

[Marco Varlese]

Hi,
I'm seeing an error reported by the DPDK EAL when allocating the MBUF pools.

The error is:
"MBUF: error setting mempool handler"

I have a multi-node NUMA system and I get the exact same error when trying to
allocate the memory from the different nodes. 

The bootstrap parameters I use to reserve huge-pages are:
"default_hugepagesz=1G hugepagesz=1G hugepages=32"

If I check the available huge-pages in my systems they look Ok (see below).
fromm:~/:[130]# cat /proc/meminfo 
MemTotal:       263470432 kB
MemFree:        228408660 kB
MemAvailable:   227792692 kB
Buffers:            2148 kB
Cached:           229132 kB
SwapCached:            0 kB
Active:           113424 kB
Inactive:         150148 kB
Active(anon):      33508 kB
Inactive(anon):     1912 kB
Active(file):      79916 kB
Inactive(file):   148236 kB
Unevictable:          80 kB
Mlocked:              80 kB
SwapTotal:       2104316 kB
SwapFree:        2104316 kB
Dirty:               624 kB
Writeback:             0 kB
AnonPages:         33216 kB
Mapped:            60232 kB
Shmem:              2388 kB
Slab:             185240 kB
SReclaimable:      46320 kB
SUnreclaim:       138920 kB
KernelStack:       22592 kB
PageTables:         3000 kB
NFS_Unstable:          0 kB
Bounce:                0 kB
WritebackTmp:          0 kB
CommitLimit:    117062316 kB
Committed_AS:     241992 kB
VmallocTotal:   34359738367 kB
VmallocUsed:           0 kB
VmallocChunk:          0 kB
HardwareCorrupted:     0 kB
HugePages_Total:      32
HugePages_Free:       32
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:    1048576 kB
DirectMap4k:      317756 kB
DirectMap2M:    11014144 kB
DirectMap1G:    258998272 kB

Any idea / help would be much appreciated.


Thanks,
Marco

Re: [dpdk-dev] Problem with MBUF pool allocation

[Bruce Richardson]

On Mon, Jun 12, 2017 at 04:15:08PM +0200, Marco Varlese wrote:
> Hi,
> I'm seeing an error reported by the DPDK EAL when allocating the MBUF pools.
> 
> The error is:
> "MBUF: error setting mempool handler"
> 
> I have a multi-node NUMA system and I get the exact same error when trying to
> allocate the memory from the different nodes. 
> 
> The bootstrap parameters I use to reserve huge-pages are:
> "default_hugepagesz=1G hugepagesz=1G hugepages=32"
> 
> If I check the available huge-pages in my systems they look Ok (see below).
> fromm:~/:[130]# cat /proc/meminfo 
<snip>
> 
> Any idea / help would be much appreciated.
> 
> 
> Thanks,
> Marco
> 
Hi Marco,

in recent releases the mempools are now based on different mempool
drivers, rather than just being a single mempool library. So if this is
a shared library build, you may need to pass in "-d /path/to/driver"
parameter to make sure the default mempool driver is picked up by the
EAL.

Regards,
/Bruce