[libfabric-users] Failure of fi_getinfo with verbs backend
joern.schumacher at cern.ch
Fri Jun 24 06:04:13 PDT 2016
I found out that with rdma_client from the librdmacm distribution I have
a similar error:
> $ rdma_server -p 12345
> $ rdma_client -p 12345 -s 127.0.0.1
> rdma_client: start
> rdma_create_ep: Invalid argument
> rdma_client: end -1
So I believe there is a an issue with the system configuration, not with
libfabric. I will investigate further.
On 06/24/2016 09:30 AM, Jörn Schumacher wrote:
> Sorry, I was reading on mobile yesterday and missed your mail.
> Here is the output:
> So the verbs provider is registered:
>> libfabric:core:core:fi_register_provider():123<info> registering
>> provider: verbs (1.0)
> Looks like the issue could be here:
>> libfabric:verbs:fabric:fi_ibv_create_ep():181<info> rdma_create_ep:
>> Invalid argument(22)
> On 06/23/2016 07:40 PM, Ilango, Arun wrote:
>> Can you run with the environment variable FI_LOG_LEVEL set to "debug"
>> and share the output? That may offer some clues as to why this fails.
>> -----Original Message-----
>> From: Libfabric-users
>> [mailto:libfabric-users-bounces at lists.openfabrics.org] On Behalf Of
>> Hefty, Sean
>> Sent: Thursday, June 23, 2016 8:55 AM
>> To: Jörn Schumacher <joern.schumacher at cern.ch>;
>> libfabric-users at lists.openfabrics.org
>> Subject: Re: [libfabric-users] Failure of fi_getinfo with verbs backend
>>> Here is the relevant code snippet:
>>>> hints = fi_allocinfo();
>>>> hints->ep_attr->type = FI_EP_MSG;
>>>> hints->caps = FI_MSG;
>>>> hints->mode = FI_LOCAL_MR;
>>>> if(ret = fi_getinfo(FI_VERSION(1, 1), "127.0.0.1", "12345", 0,
>>>> ERROR("fi_getinfo failed: %d '%s'", ret, fi_strerror(-ret)); }
>>> Which fails with
>>>> error: fi_getinfo failed: -61 'No data available'
>> Hmmm... odd. Is the sockets provider installed as well? I'm
>> surprised that doesn't return anything either.
>>> The output of fi_info suggests the verbs provider has the capabilities
>>> I request. Oddly enough, it works on the server side, where the only
>>> difference is an added FI_SOURCE flag to the fi_getinfo call.
>> Is the server side running on a different system, or are you just
>> running as a server on the same system?
>> Libfabric-users mailing list
>> Libfabric-users at lists.openfabrics.org
> Libfabric-users mailing list
> Libfabric-users at lists.openfabrics.org
More information about the Libfabric-users