queue-5.1/ib-hfi1-correct-tid-qp-rcd-to-match-verbs-context.patch

   1 From cc78076af14e1478c1a8fb18997674b5f8cbe3c8 Mon Sep 17 00:00:00 2001
   2 From: Mike Marciniszyn <mike.marciniszyn@intel.com>
   3 Date: Mon, 10 Jun 2019 12:28:18 -0400
   4 Subject: IB/hfi1: Correct tid qp rcd to match verbs context
   5
   6 From: Mike Marciniszyn <mike.marciniszyn@intel.com>
   7
   8 commit cc78076af14e1478c1a8fb18997674b5f8cbe3c8 upstream.
   9
  10 The qp priv rcd pointer doesn't match the context being used for verbs
  11 causing issues when 9B and kdeth packets are processed by different
  12 receive contexts and hence different CPUs.
  13
  14 When running on different CPUs the following panic can occur:
  15
  16  WARNING: CPU: 3 PID: 2584 at lib/list_debug.c:59 __list_del_entry+0xa1/0xd0
  17  list_del corruption. prev->next should be ffff9a7ac31f7a30, but was ffff9a7c3bc89230
  18  CPU: 3 PID: 2584 Comm: z_wr_iss Kdump: loaded Tainted: P           OE  ------------   3.10.0-862.2.3.el7_lustre.x86_64 #1
  19  Call Trace:
  20   <IRQ>  [<ffffffffb7b0d78e>] dump_stack+0x19/0x1b
  21   [<ffffffffb74916d8>] __warn+0xd8/0x100
  22   [<ffffffffb749175f>] warn_slowpath_fmt+0x5f/0x80
  23   [<ffffffffb7768671>] __list_del_entry+0xa1/0xd0
  24   [<ffffffffc0c7a945>] process_rcv_qp_work+0xb5/0x160 [hfi1]
  25   [<ffffffffc0c7bc2b>] handle_receive_interrupt_nodma_rtail+0x20b/0x2b0 [hfi1]
  26   [<ffffffffc0c70683>] receive_context_interrupt+0x23/0x40 [hfi1]
  27   [<ffffffffb7540a94>] __handle_irq_event_percpu+0x44/0x1c0
  28   [<ffffffffb7540c42>] handle_irq_event_percpu+0x32/0x80
  29   [<ffffffffb7540ccc>] handle_irq_event+0x3c/0x60
  30   [<ffffffffb7543a1f>] handle_edge_irq+0x7f/0x150
  31   [<ffffffffb742d504>] handle_irq+0xe4/0x1a0
  32   [<ffffffffb7b23f7d>] do_IRQ+0x4d/0xf0
  33   [<ffffffffb7b16362>] common_interrupt+0x162/0x162
  34   <EOI>  [<ffffffffb775a326>] ? memcpy+0x6/0x110
  35   [<ffffffffc109210d>] ? abd_copy_from_buf_off_cb+0x1d/0x30 [zfs]
  36   [<ffffffffc10920f0>] ? abd_copy_to_buf_off_cb+0x30/0x30 [zfs]
  37   [<ffffffffc1093257>] abd_iterate_func+0x97/0x120 [zfs]
  38   [<ffffffffc10934d9>] abd_copy_from_buf_off+0x39/0x60 [zfs]
  39   [<ffffffffc109b828>] arc_write_ready+0x178/0x300 [zfs]
  40   [<ffffffffb7b11032>] ? mutex_lock+0x12/0x2f
  41   [<ffffffffb7b11032>] ? mutex_lock+0x12/0x2f
  42   [<ffffffffc1164d05>] zio_ready+0x65/0x3d0 [zfs]
  43   [<ffffffffc04d725e>] ? tsd_get_by_thread+0x2e/0x50 [spl]
  44   [<ffffffffc04d1318>] ? taskq_member+0x18/0x30 [spl]
  45   [<ffffffffc115ef22>] zio_execute+0xa2/0x100 [zfs]
  46   [<ffffffffc04d1d2c>] taskq_thread+0x2ac/0x4f0 [spl]
  47   [<ffffffffb74cee80>] ? wake_up_state+0x20/0x20
  48   [<ffffffffc115ee80>] ? zio_taskq_member.isra.7.constprop.10+0x80/0x80 [zfs]
  49   [<ffffffffc04d1a80>] ? taskq_thread_spawn+0x60/0x60 [spl]
  50   [<ffffffffb74bae31>] kthread+0xd1/0xe0
  51   [<ffffffffb74bad60>] ? insert_kthread_work+0x40/0x40
  52   [<ffffffffb7b1f5f7>] ret_from_fork_nospec_begin+0x21/0x21
  53   [<ffffffffb74bad60>] ? insert_kthread_work+0x40/0x40
  54
  55 Fix by reading the map entry in the same manner as the hardware so that
  56 the kdeth and verbs contexts match.
  57
  58 Cc: <stable@vger.kernel.org>
  59 Fixes: 5190f052a365 ("IB/hfi1: Allow the driver to initialize QP priv struct")
  60 Reviewed-by: Kaike Wan <kaike.wan@intel.com>
  61 Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
  62 Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
  63 Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
  64 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  65
  66 ---
  67  drivers/infiniband/hw/hfi1/chip.c     |   13 +++++++++++++
  68  drivers/infiniband/hw/hfi1/chip.h     |    1 +
  69  drivers/infiniband/hw/hfi1/tid_rdma.c |    4 +---
  70  3 files changed, 15 insertions(+), 3 deletions(-)
  71
  72 --- a/drivers/infiniband/hw/hfi1/chip.c
  73 +++ b/drivers/infiniband/hw/hfi1/chip.c
  74 @@ -14028,6 +14028,19 @@ static void init_kdeth_qp(struct hfi1_de
  75  }
  76
  77  /**
  78 + * hfi1_get_qp_map
  79 + * @dd: device data
  80 + * @idx: index to read
  81 + */
  82 +u8 hfi1_get_qp_map(struct hfi1_devdata *dd, u8 idx)
  83 +{
  84 +       u64 reg = read_csr(dd, RCV_QP_MAP_TABLE + (idx / 8) * 8);
  85 +
  86 +       reg >>= (idx % 8) * 8;
  87 +       return reg;
  88 +}
  89 +
  90 +/**
  91   * init_qpmap_table
  92   * @dd - device data
  93   * @first_ctxt - first context
  94 --- a/drivers/infiniband/hw/hfi1/chip.h
  95 +++ b/drivers/infiniband/hw/hfi1/chip.h
  96 @@ -1442,6 +1442,7 @@ void clear_all_interrupts(struct hfi1_de
  97  void remap_intr(struct hfi1_devdata *dd, int isrc, int msix_intr);
  98  void remap_sdma_interrupts(struct hfi1_devdata *dd, int engine, int msix_intr);
  99  void reset_interrupts(struct hfi1_devdata *dd);
 100 +u8 hfi1_get_qp_map(struct hfi1_devdata *dd, u8 idx);
 101
 102  /*
 103   * Interrupt source table.
 104 --- a/drivers/infiniband/hw/hfi1/tid_rdma.c
 105 +++ b/drivers/infiniband/hw/hfi1/tid_rdma.c
 106 @@ -305,9 +305,7 @@ static struct hfi1_ctxtdata *qp_to_rcd(s
 107         if (qp->ibqp.qp_num == 0)
 108                 ctxt = 0;
 109         else
 110 -               ctxt = ((qp->ibqp.qp_num >> dd->qos_shift) %
 111 -                       (dd->n_krcv_queues - 1)) + 1;
 112 -
 113 +               ctxt = hfi1_get_qp_map(dd, qp->ibqp.qp_num >> dd->qos_shift);
 114         return dd->rcd[ctxt];
 115  }
 116