Nab, The driver/package is downloaded from MLNX(MLNX_OFED_LINUX-3.2-2.0.0.0-ubuntu14.04-x86_64): http://www.mellanox.com/page/mlnx_ofed_eula?mtag=linux_sw_drivers&mrequest=downloads&mtype=ofed&mver=MLNX_OFED-3.2-2.0.0.0&mname=MLNX_OFED_LINUX-3.2-2.0.0.0-ubuntu14.04-x86_64.tgz Its default install script does not include ib_isert.ko(isert-dkms_1.0-OFED.3.2.2.0.0.1.g378ff02_all.deb). We as the end user have no idea which version of the kernel source should be recompiled with. It seems we'd better go back to MLNX to find the support. Thanks, Patrick -----Original Message----- From: Nicholas A. Bellinger [mailto:nab@xxxxxxxxxxxxxxx] Sent: Monday, April 04, 2016 5:33 PM To: Patrick Li-SSI Cc: sagig; Tameesh Suri-SSI; Vijay Balakrishnan-SSI; Harry (Huan) Li-SSI; target-devel Subject: Re: ib_isert.c On Mon, 2016-04-04 at 22:13 +0000, Patrick Li-SSI wrote: > Thanks Sagig/Nic for your following up. Btw, please avoid top posting responses to kernel lists. It makes the thread more difficult for others to follow. :) > isted below. When we ran 'sudo targetcli', there is no issue. > The only issue is that when the initiator tries to login the target > with ISER interface, it gets time-out error, can not login to the > target, even though its discovery with ISER interface is ok. > > Sorry NIc for the deplay/slow response. > > The reason we did not use the original ib_isert.ko is because if > ib_iser.ko is loaded, the ib_isert.ko loading will fail; if > ib_isert.ko is loaded, the ib_iser.ko loading will fail. > The null iscsi-login shows only once(first time login request). Then > this issue never happens again. I'm still confused wrt your use of upstream and dkms out-of-tree modules. Are you saying that using the stock upstream modules for 4.2.0-040200-generic (eg: not dkms built ones), that ib_iser + ib_isert can't be loaded at the same time..? Or that is just when mixing upstream + dkms built modules..? If your doing dkms, you need to re-build ALL modules with ib_core dependencies, and not mix-and-match some ib_* modules using the original upstream build of ib_core, vs. your dkms build of ib_core. Mix-and-matching like this will certainly produced undefined results. --nab ��.n��������+%������w��{.n����j� ����{ay� ʇڙ���f���h������_�(�階�ݢj"��������G����?���&��
