[dpdk-users] Secondary process try to map hugepage even with --no-huge

[dpdk-users] Secondary process try to map hugepage even with --no-huge

[tom.barbette]

Hi list,

I've got a primary/secondary application that works with huge pages enabled but not with "--no-huge". The secondary application try to access .rte_hugepage_info no matter the "--no-huge" being passed to both master and slave.

This is especially important for functional regression test where I cannot use huge pages or sudo (like Travis, Gitlab CI, ...)

Trying with the simple_mp example :
 ./simple_mp --no-huge -m 128MB --proc-type=secondary --
EAL: Detected 8 lcore(s)
EAL: Probing VFIO support...
EAL: Could not mmap /home/tom/.rte_hugepage_info
PANIC in rte_eal_init():
Cannot init memory
6: [./simple_mp(_start+0x2a) [0x43512a]]
5: [/lib64/libc.so.6(__libc_start_main+0xf1) [0x7ffff6ed4401]]
4: [./simple_mp(main+0x6) [0x4347b6]]
3: [./simple_mp(rte_eal_init+0xe90) [0x46e2d0]]
2: [./simple_mp(__rte_panic+0xbe) [0x4301ab]]
1: [./simple_mp(rte_dump_stack+0x16) [0x475c06]]
zsh: abort (core dumped)  ./simple_mp --no-huge --proc-type=secondary --


With sudo, I get a little further, I guess this is not the problem.
EAL: Could not mmap /dev/hugepages/rtemap_53
PANIC in rte_eal_init():

Thanks,

Tom

Re: [dpdk-users] Secondary process try to map hugepage even with --no-huge

[Tan, Jianfeng]

Hi,

> -----Original Message-----
> From: users [mailto:users-bounces@dpdk.org] On Behalf Of
> tom.barbette@ulg.ac.be
> Sent: Wednesday, July 5, 2017 9:58 PM
> To: users@dpdk.org
> Cc: Gonzalez Monroy, Sergio
> Subject: [dpdk-users] Secondary process try to map hugepage even with --
> no-huge
> 
> Hi list,
> 
> I've got a primary/secondary application that works with huge pages enabled
> but not with "--no-huge". The secondary application try to
> access .rte_hugepage_info no matter the "--no-huge" being passed to both
> master and slave.

So far if --no-huge is used, we are mmap anonymous memory, which makes it not possible to share the memory with secondary (always refer to a config file, open and mmap to the same addr).

And we did not write anything into config file if we use --no-huge. This is a limitation for now.

> 
> This is especially important for functional regression test where I cannot use
> huge pages or sudo (like Travis, Gitlab CI, ...)
> 
> Trying with the simple_mp example :
>  ./simple_mp --no-huge -m 128MB --proc-type=secondary --
> EAL: Detected 8 lcore(s)
> EAL: Probing VFIO support...
> EAL: Could not mmap /home/tom/.rte_hugepage_info
> PANIC in rte_eal_init():
> Cannot init memory
> 6: [./simple_mp(_start+0x2a) [0x43512a]]
> 5: [/lib64/libc.so.6(__libc_start_main+0xf1) [0x7ffff6ed4401]]
> 4: [./simple_mp(main+0x6) [0x4347b6]]
> 3: [./simple_mp(rte_eal_init+0xe90) [0x46e2d0]]
> 2: [./simple_mp(__rte_panic+0xbe) [0x4301ab]]
> 1: [./simple_mp(rte_dump_stack+0x16) [0x475c06]]
> zsh: abort (core dumped)  ./simple_mp --no-huge --proc-type=secondary --
> 
> 
> With sudo, I get a little further, I guess this is not the problem.
> EAL: Could not mmap /dev/hugepages/rtemap_53
> PANIC in rte_eal_init():
> 
> Thanks,
> 
> Tom

Re: [dpdk-users] Secondary process try to map hugepage even with --no-huge

[tom.barbette]

Thanks for your prompt answer. Are there any plan to change that behavior, like using shm memory if no-huge is there? As I said, this would be very useful for testing.

Thanks,
Tom

----- Mail original -----
> De: "Tan, Jianfeng" <jianfeng.tan@intel.com>
> À: "tom barbette" <tom.barbette@ulg.ac.be>, users@dpdk.org
> Cc: "Gonzalez Monroy, Sergio" <sergio.gonzalez.monroy@intel.com>
> Envoyé: Jeudi 6 Juillet 2017 04:46:02
> Objet: RE: [dpdk-users] Secondary process try to map hugepage even with	--no-huge

> Hi,
> 
>> -----Original Message-----
>> From: users [mailto:users-bounces@dpdk.org] On Behalf Of
>> tom.barbette@ulg.ac.be
>> Sent: Wednesday, July 5, 2017 9:58 PM
>> To: users@dpdk.org
>> Cc: Gonzalez Monroy, Sergio
>> Subject: [dpdk-users] Secondary process try to map hugepage even with --
>> no-huge
>> 
>> Hi list,
>> 
>> I've got a primary/secondary application that works with huge pages enabled
>> but not with "--no-huge". The secondary application try to
>> access .rte_hugepage_info no matter the "--no-huge" being passed to both
>> master and slave.
> 
> So far if --no-huge is used, we are mmap anonymous memory, which makes it not
> possible to share the memory with secondary (always refer to a config file,
> open and mmap to the same addr).
> 
> And we did not write anything into config file if we use --no-huge. This is a
> limitation for now.
> 
>> 
>> This is especially important for functional regression test where I cannot use
>> huge pages or sudo (like Travis, Gitlab CI, ...)
>> 
>> Trying with the simple_mp example :
>>  ./simple_mp --no-huge -m 128MB --proc-type=secondary --
>> EAL: Detected 8 lcore(s)
>> EAL: Probing VFIO support...
>> EAL: Could not mmap /home/tom/.rte_hugepage_info
>> PANIC in rte_eal_init():
>> Cannot init memory
>> 6: [./simple_mp(_start+0x2a) [0x43512a]]
>> 5: [/lib64/libc.so.6(__libc_start_main+0xf1) [0x7ffff6ed4401]]
>> 4: [./simple_mp(main+0x6) [0x4347b6]]
>> 3: [./simple_mp(rte_eal_init+0xe90) [0x46e2d0]]
>> 2: [./simple_mp(__rte_panic+0xbe) [0x4301ab]]
>> 1: [./simple_mp(rte_dump_stack+0x16) [0x475c06]]
>> zsh: abort (core dumped)  ./simple_mp --no-huge --proc-type=secondary --
>> 
>> 
>> With sudo, I get a little further, I guess this is not the problem.
>> EAL: Could not mmap /dev/hugepages/rtemap_53
>> PANIC in rte_eal_init():
>> 
>> Thanks,
>> 
> > Tom