[ewg] Need urgent help : Soft-RoCE in Linux-3.19

Dinesh Kb dinesh.kb at vvdntech.in
Tue Jun 9 01:36:56 PDT 2015


​One more thing we have modified is in the file
"drivers/infiniband/hw/rxe/rxe_net.c

we have commented the *line 426*

//struct net_device *ndev = arg;

and added the following

struct netdev_notifier_info *info =arg;
struct net_device *ndev = info->dev;

moreover we have copied "ib_pack.h" from the vanilla(3.0.0+) to latest
kernel (4.0.4)

please help me... to achieve SoftRoCE in latest kernel

​


*with warm regards*
*Dinesh.K.B *
Software Engineer


*Cell : *+91 9944456867 | *Skype :* dinesh_kb93

On Tue, Jun 9, 2015 at 1:53 PM, Dinesh Kb <dinesh.kb at vvdntech.in> wrote:

> Hi,
>
>                                                          *"I am new to
> OFED"*
>
> I have downloaded the "*Soft-RoCE*" which includes the *kernel (3.0.0+) *and
> *rxe support.*
> I have Installed all the packages included in the *OFED-1.5.2-rxe *except
> the* kernel-ib. *
>
> *In kernel 3.0.0+, *I am able to run the "rping" test successfully and
> able to perform the benchmark tests, I am comfort with benchmark output.
>
> Now I am in need to perform the RoCE operation in the* higher kernel
> version...say 4.0.4*
>
> I have downloaded source file for  the kernel version 4.0.4 and i have
> copied "driver/infiniband/hw/rxe" directory from kernel 3.0.0+ and pasted
> in the 4.0.4 kernel source and I have applied some patch for rxe to work
> smoothly...
>
> I am facing *Kernel crash* while running rping...
> and regarding the *benchmark...*
>
> when running rping as server in the higher kernel version and client in
> the lower version is hold fine....
> but its worse(crash) when running vice-versa
>
> Now i am facing problems, when kernel 4.04 acts as a client...
> while taking the *wireshark *output.. the server constantly sends the "read
> request" and its not getting any response from the client...
>
> the virtual address and dma length fields are "zero".
>
>
> Kindly help me.....
>
> *Is there any possibilities/ perfect patch to make SoftRoCE work fine with
> higher kernel version (say kernel 3.19+)*
>
> I hereby attached the wireshark output for* ib_read_bw* (when higher
> version acts as a client and 3.0.0+ acts as a server) and the patch i have
> applied as an attachment
>
> the crash in the server is as follows
>
> root at ls2085aqds:/etc/libibverbs.d# rping -s
> ------------[ cut here ]------------
> WARNING: CPU: 0 PID: 810 at kernel/softirq.c:146 __local_bh_enable_ip+0x84/0xc0()
> Modules linked in: ib_rxe_net ib_rxe rdma_ucm ib_ucm ib_uverbs ib_umad rdma_cm iw_cm ib_cm ib_sa ib_mad ib_core ib_addr ipv6
> CPU: 0 PID: 810 Comm: kworker/0:1 Not tainted 3.16.0-Layerscape2-SDK+g330dd4b #20
> Workqueue: ib_cm cm_work_handler [ib_cm]
> Call trace:
> [<ffffffc000088084>] dump_backtrace+0x0/0x12c
> [<ffffffc0000881c0>] show_stack+0x10/0x1c
> [<ffffffc000552fa4>] dump_stack+0x74/0xc4
> [<ffffffc0000a9b8c>] warn_slowpath_common+0x84/0xac
> [<ffffffc0000a9c78>] warn_slowpath_null+0x14/0x20
> [<ffffffc0000ae3d0>] __local_bh_enable_ip+0x80/0xc0
> [<ffffffc00049fccc>] __dev_queue_xmit+0x1cc/0x408
> [<ffffffc00049ff14>] dev_queue_xmit+0xc/0x18
> [<ffffffbffc11b620>] send_finish+0x34/0x40 [ib_rxe_net]
> [<ffffffbffc11b6b8>] send+0x8c/0xec [ib_rxe_net]
> [<ffffffbffc10933c>] $x+0x318/0x334 [ib_rxe]
> [<ffffffbffc109be0>] $x+0xa8/0x120 [ib_rxe]
> [<ffffffbffc109d54>] rxe_run_task+0x4c/0x90 [ib_rxe]
> [<ffffffbffc1093b4>] arbiter_skb_queue+0x5c/0x8c [ib_rxe]
> [<ffffffbffc0fffd4>] rxe_requester+0x83c/0xddc [ib_rxe]
> [<ffffffbffc109be0>] $x+0xa8/0x120 [ib_rxe]
> [<ffffffbffc109d54>] rxe_run_task+0x4c/0x90 [ib_rxe]
> [<ffffffbffc104e0c>] rxe_post_send+0x80/0x40c [ib_rxe]
> [<ffffffbffc097068>] ib_send_mad+0x288/0x454 [ib_mad]
> [<ffffffbffc09759c>] ib_post_send_mad+0x190/0x544 [ib_mad]
> [<ffffffbffc0b075c>] ib_send_cm_rej+0xd0/0x194 [ib_cm]
> [<ffffffbffc0b19b8>] cm_destroy_id+0x188/0x300 [ib_cm]
> [<ffffffbffc0b1e80>] cm_process_work+0x154/0x17c [ib_cm]
> [<ffffffbffc0b2600>] cm_req_handler+0x758/0x978 [ib_cm]
> [<ffffffbffc0b28ec>] cm_work_handler+0xcc/0x1584 [ib_cm]
> [<ffffffc0000c0918>] process_one_work+0x114/0x354
> [<ffffffc0000c12f8>] worker_thread+0x13c/0x500
> [<ffffffc0000c7034>] kthread+0xd0/0xe8
> ---[ end trace be7a1b95934c8f03 ]---
>
>
> *with warm regards*
> *Dinesh.K.B *
> Software Engineer
>
>
> *Cell : *+91 9944456867 | *Skype :* dinesh_kb93
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20150609/62ef5d63/attachment.html>


More information about the ewg mailing list