IB/hfi1: Fix panic with larger ipoib send_queue_size
authorMike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Sat, 15 Jan 2022 23:02:33 +0000 (18:02 -0500)
committerJason Gunthorpe <jgg@nvidia.com>
Fri, 28 Jan 2022 15:03:50 +0000 (11:03 -0400)
commit8c83d39cc730378bbac64d67a551897b203a606e
treea4ecfb280163a61266ce426b55863d7ee219f9b4
parente783362eb54cd99b2cac8b3a9aeac942e6f6ac07
IB/hfi1: Fix panic with larger ipoib send_queue_size

When the ipoib send_queue_size is increased from the default the following
panic happens:

  RIP: 0010:hfi1_ipoib_drain_tx_ring+0x45/0xf0 [hfi1]
  Code: 31 e4 eb 0f 8b 85 c8 02 00 00 41 83 c4 01 44 39 e0 76 60 8b 8d cc 02 00 00 44 89 e3 be 01 00 00 00 d3 e3 48 03 9d c0 02 00 00 <c7> 83 18 01 00 00 00 00 00 00 48 8b bb 30 01 00 00 e8 25 af a7 e0
  RSP: 0018:ffffc9000798f4a0 EFLAGS: 00010286
  RAX: 0000000000008000 RBX: ffffc9000aa0f000 RCX: 000000000000000f
  RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000000
  RBP: ffff88810ff08000 R08: ffff88889476d900 R09: 0000000000000101
  R10: 0000000000000000 R11: ffffc90006590ff8 R12: 0000000000000200
  R13: ffffc9000798fba8 R14: 0000000000000000 R15: 0000000000000001
  FS:  00007fd0f79cc3c0(0000) GS:ffff88885fb00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: ffffc9000aa0f118 CR3: 0000000889c84001 CR4: 00000000001706e0
  Call Trace:
   <TASK>
   hfi1_ipoib_napi_tx_disable+0x45/0x60 [hfi1]
   hfi1_ipoib_dev_stop+0x18/0x80 [hfi1]
   ipoib_ib_dev_stop+0x1d/0x40 [ib_ipoib]
   ipoib_stop+0x48/0xc0 [ib_ipoib]
   __dev_close_many+0x9e/0x110
   __dev_change_flags+0xd9/0x210
   dev_change_flags+0x21/0x60
   do_setlink+0x31c/0x10f0
   ? __nla_validate_parse+0x12d/0x1a0
   ? __nla_parse+0x21/0x30
   ? inet6_validate_link_af+0x5e/0xf0
   ? cpumask_next+0x1f/0x20
   ? __snmp6_fill_stats64.isra.53+0xbb/0x140
   ? __nla_validate_parse+0x47/0x1a0
   __rtnl_newlink+0x530/0x910
   ? pskb_expand_head+0x73/0x300
   ? __kmalloc_node_track_caller+0x109/0x280
   ? __nla_put+0xc/0x20
   ? cpumask_next_and+0x20/0x30
   ? update_sd_lb_stats.constprop.144+0xd3/0x820
   ? _raw_spin_unlock_irqrestore+0x25/0x37
   ? __wake_up_common_lock+0x87/0xc0
   ? kmem_cache_alloc_trace+0x3d/0x3d0
   rtnl_newlink+0x43/0x60

The issue happens when the shift that should have been a function of the
txq item size mistakenly used the ring size.

Fix by using the item size.

Cc: stable@vger.kernel.org
Fixes: d47dfc2b00e6 ("IB/hfi1: Remove cache and embed txreq in ring")
Link: https://lore.kernel.org/r/1642287756-182313-2-git-send-email-mike.marciniszyn@cornelisnetworks.com
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@cornelisnetworks.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@cornelisnetworks.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
drivers/infiniband/hw/hfi1/ipoib_tx.c