[ofa-general] RE: [ewg] OFED 1.2 and backwards binary compatibility

Steffen Persvold steffen.persvold at scali.com
Fri Jun 8 04:42:15 PDT 2007


Roland,
 
1.0 vs. 1.1 is all good. That works. I'm talking about 1.1/1.0 vs 1.2, that's not working. The diffset below is between 1.1 and 1.2.
 
What we're doing is using dlopen()/dlsym() to dynamically open the library so that we have no library dependencies (this allows us to runtime wise check if ofed is installed or other IB stacks). This apparently breaks.
 
I don't find any "post_send_wrapper_1_0" nor "post_send_wrapper_1_1" symbols in my libraries ?? :
 
[root at pe1850-1 lib]# nm libibverbs.so.1.0.0 |grep post_send
0000000000003aa0 T ibv_cmd_post_send

?
 
Cheers,
 
Steffen Persvold
Technical Director Americas
tel. 508-281-7100 x401
fax. 508-281-7171

http://www.scali.com/
Scaling the Linux datacenter

________________________________

From: Roland Dreier [mailto:rdreier at cisco.com]
Sent: Thu 6/7/2007 9:42 PM
To: Steffen Persvold
Cc: EWG; OpenFabrics General
Subject: Re: [ofa-general] RE: [ewg] OFED 1.2 and backwards binary compatibility



 > Just to follow up, I believe at least these changes (there are more) to verbs.h breaks the compatibility :
 > 
 > @@ -469,8 +502,8 @@
 >  };
 >  struct ibv_send_wr {
 > -       struct ibv_send_wr     *next;
 >         uint64_t                wr_id;
 > +       struct ibv_send_wr     *next;
 >         struct ibv_sge         *sg_list;
 >         int                     num_sge;
 >         enum ibv_wr_opcode      opcode;
 > @@ -496,12 +529,21 @@
 >  };
 >  struct ibv_recv_wr {
 > -       struct ibv_recv_wr     *next;
 >         uint64_t                wr_id;
 > +       struct ibv_recv_wr     *next;
 >         struct ibv_sge         *sg_list;
 >         int                     num_sge;
 >  };

These differences should be taken care of by the
post_send_wrapper_1_0() and post_recv_wrapper_1_0() functions in
src/compat-1_0.c in libibverbs 1.1.

 - R.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20070608/5a0a0794/attachment.html>


More information about the ewg mailing list