> On Mar 19, 2018, at 2:06 PM, Chuck Lever <chuck.lever@xxxxxxxxxx> wrote: > > > >> On Mar 19, 2018, at 1:26 PM, Kalderon, Michal <Michal.Kalderon@xxxxxxxxxx> wrote: >> >>> From: Kalderon, Michal >>> Sent: Monday, March 19, 2018 7:22 PM >>> To: 'Chuck Lever' <chuck.lever@xxxxxxxxxx> >>> Cc: linux-rdma@xxxxxxxxxxxxxxx >>> Subject: RE: [PATCH v1] xprtrdma: Fix corner cases when handling device >>> removal >>> >>>> From: Chuck Lever [mailto:chuck.lever@xxxxxxxxxx] >>>> Sent: Monday, March 19, 2018 4:59 PM >>>> To: Kalderon, Michal <Michal.Kalderon@xxxxxxxxxx> >>>> Cc: linux-rdma@xxxxxxxxxxxxxxx >>>> Subject: Re: [PATCH v1] xprtrdma: Fix corner cases when handling >>>> device removal >>>> >>>> Hi Michal- >>>> >>>> Have you tried this one? If we can nail this down now, I can easily >>>> get it into the v4.17 merge window. >>> >>> Yes, looks good thanks. >> >> Chuck, while testing changes, I ran into a different issue though. >> If I perform mount to a server that does not have the nfs rdma port in portlist using mount -o rdma, >> And then perform rmmod qedr, it hangs with following stack : >> >> root@lb-tlvb-pcie37 ~]# cat /proc/5981/stack >> [<0>] rpcrdma_conn_upcall+0x26a/0x310 [rpcrdma] >> [<0>] cma_remove_one+0x26d/0x2a0 [rdma_cm] >> [<0>] ib_unregister_device+0xcc/0x180 [ib_core] >> [<0>] qedr_remove+0x37/0x80 [qedr] >> [<0>] qede_rdma_unregister_driver+0x4b/0x90 [qede] >> [<0>] SyS_delete_module+0x1c2/0x240 >> [<0>] do_syscall_64+0x6e/0x190 >> [<0>] entry_SYSCALL_64_after_hwframe+0x3d/0xa2 >> >> (wait_for_completion(&ia->ri_remove_done) > > I will try to reproduce it here. I can't reproduce with my IB or my RoCE rig, but I might not have understood how you are getting into this state. I'm trying this: # mount.nfs -v -o vers=4.0,proto=rdma,port=12345,sec=sys nfs-roce:/export/tmp /mnt and while it is looping trying to connect, I do this in another window: # rmmod mlx4_en On it's next attempt to connect, the connect worker fails with RDMA_CM_EVENT_ADDR_ERROR. I don't ever see RDMA_CM_EVENT_DEVICE_REMOVAL. >> Thanks, >> Michal >>>> >>>> >>>>> On Mar 14, 2018, at 10:42 AM, Chuck Lever <chuck.lever@xxxxxxxxxx> >>>> wrote: >>>>> >>>>> Michal Kalderon has found some corner cases around device unload >>>>> with active NFS mounts that I didn't have the imagination to test >>>>> when xprtrdma device removal was added last year. >>>>> >>>>> - The ULP device removal handler is responsible for deallocating >>>>> the PD. That wasn't clear to me initially, and my own testing >>>>> suggested it was not necessary, but that is incorrect. >>>>> >>>>> - The transport destruction path can no longer assume that there is >>>>> a valid ID. >>>>> >>>>> - When destroying a transport, ensure that ib_free_cq() is not >>>>> invoked on a CQ that was already released. >>>>> >>>>> Reported-by: Michal Kalderon <Michal.Kalderon@xxxxxxxxxx> >>>>> Fixes: bebd031866ca ("xprtrdma: Support unplugging an HCA from ...") >>>>> Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx> >>>>> Cc: stable@xxxxxxxxxxxxxxx >>>>> --- >>>>> net/sunrpc/xprtrdma/verbs.c | 13 +++++++++---- >>>>> 1 file changed, 9 insertions(+), 4 deletions(-) >>>>> >>>>> diff --git a/net/sunrpc/xprtrdma/verbs.c >>>>> b/net/sunrpc/xprtrdma/verbs.c index d19ea02..c284ee7 100644 >>>>> --- a/net/sunrpc/xprtrdma/verbs.c >>>>> +++ b/net/sunrpc/xprtrdma/verbs.c >>>>> @@ -251,7 +251,6 @@ >>>>> wait_for_completion(&ia->ri_remove_done); >>>>> >>>>> ia->ri_id = NULL; >>>>> - ia->ri_pd = NULL; >>>>> ia->ri_device = NULL; >>>>> /* Return 1 to ensure the core destroys the id. */ >>>>> return 1; >>>>> @@ -449,7 +448,9 @@ >>>>> ia->ri_id->qp = NULL; >>>>> } >>>>> ib_free_cq(ep->rep_attr.recv_cq); >>>>> + ep->rep_attr.recv_cq = NULL; >>>>> ib_free_cq(ep->rep_attr.send_cq); >>>>> + ep->rep_attr.send_cq = NULL; >>>>> >>>>> /* The ULP is responsible for ensuring all DMA >>>>> * mappings and MRs are gone. >>>>> @@ -462,6 +463,8 @@ >>>>> rpcrdma_dma_unmap_regbuf(req->rl_recvbuf); >>>>> } >>>>> rpcrdma_mrs_destroy(buf); >>>>> + ib_dealloc_pd(ia->ri_pd); >>>>> + ia->ri_pd = NULL; >>>>> >>>>> /* Allow waiters to continue */ >>>>> complete(&ia->ri_remove_done); >>>>> @@ -629,14 +632,16 @@ >>>>> { >>>>> cancel_delayed_work_sync(&ep->rep_connect_worker); >>>>> >>>>> - if (ia->ri_id->qp) { >>>>> + if (ia->ri_id && ia->ri_id->qp) { >>>>> rpcrdma_ep_disconnect(ep, ia); >>>>> rdma_destroy_qp(ia->ri_id); >>>>> ia->ri_id->qp = NULL; >>>>> } >>>>> >>>>> - ib_free_cq(ep->rep_attr.recv_cq); >>>>> - ib_free_cq(ep->rep_attr.send_cq); >>>>> + if (ep->rep_attr.recv_cq) >>>>> + ib_free_cq(ep->rep_attr.recv_cq); >>>>> + if (ep->rep_attr.send_cq) >>>>> + ib_free_cq(ep->rep_attr.send_cq); >>>>> } >>>>> >>>>> /* Re-establish a connection after a device removal event. >>>>> >>>>> -- >>>>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" >>>>> in the body of a message to majordomo@xxxxxxxxxxxxxxx More >>>> majordomo >>>>> info at http://vger.kernel.org/majordomo-info.html >>>> >>>> -- >>>> Chuck Lever >>>> >>>> >> >> -- >> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in >> the body of a message to majordomo@xxxxxxxxxxxxxxx >> More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- > Chuck Lever > > > > -- > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- Chuck Lever -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
