queue-4.19/powerpc-powernv-ioda-fix-locked_vm-counting-for-memo.patch

   1 From d547ae1ff1384b9d876237ae2ee3649098d921bd Mon Sep 17 00:00:00 2001
   2 From: Alexey Kardashevskiy <aik@ozlabs.ru>
   3 Date: Wed, 13 Feb 2019 14:38:18 +1100
   4 Subject: powerpc/powernv/ioda: Fix locked_vm counting for memory used by IOMMU
   5  tables
   6
   7 [ Upstream commit 11f5acce2fa43b015a8120fa7620fa4efd0a2952 ]
   8
   9 We store 2 multilevel tables in iommu_table - one for the hardware and
  10 one with the corresponding userspace addresses. Before allocating
  11 the tables, the iommu_table_group_ops::get_table_size() hook returns
  12 the combined size of the two and VFIO SPAPR TCE IOMMU driver adjusts
  13 the locked_vm counter correctly. When the table is actually allocated,
  14 the amount of allocated memory is stored in iommu_table::it_allocated_size
  15 and used to decrement the locked_vm counter when we release the memory
  16 used by the table; .get_table_size() and .create_table() calculate it
  17 independently but the result is expected to be the same.
  18
  19 However the allocator does not add the userspace table size to
  20 .it_allocated_size so when we destroy the table because of VFIO PCI
  21 unplug (i.e. VFIO container is gone but the userspace keeps running),
  22 we decrement locked_vm by just a half of size of memory we are
  23 releasing.
  24
  25 To make things worse, since we enabled on-demand allocation of
  26 indirect levels, it_allocated_size contains only the amount of memory
  27 actually allocated at the table creation time which can just be a
  28 fraction. It is not a problem with incrementing locked_vm (as
  29 get_table_size() value is used) but it is with decrementing.
  30
  31 As the result, we leak locked_vm and may not be able to allocate more
  32 IOMMU tables after few iterations of hotplug/unplug.
  33
  34 This sets it_allocated_size in the pnv_pci_ioda2_ops::create_table()
  35 hook to what pnv_pci_ioda2_get_table_size() returns so from now on we
  36 have a single place which calculates the maximum memory a table can
  37 occupy. The original meaning of it_allocated_size is somewhat lost now
  38 though.
  39
  40 We do not ditch it_allocated_size whatsoever here and we do not call
  41 get_table_size() from vfio_iommu_spapr_tce.c when decrementing
  42 locked_vm as we may have multiple IOMMU groups per container and even
  43 though they all are supposed to have the same get_table_size()
  44 implementation, there is a small chance for failure or confusion.
  45
  46 Fixes: 090bad39b237 ("powerpc/powernv: Add indirect levels to it_userspace")
  47 Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
  48 Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
  49 Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
  50 Signed-off-by: Sasha Levin <sashal@kernel.org>
  51 ---
  52  arch/powerpc/platforms/powernv/pci-ioda-tce.c | 1 -
  53  arch/powerpc/platforms/powernv/pci-ioda.c     | 7 ++++++-
  54  2 files changed, 6 insertions(+), 2 deletions(-)
  55
  56 diff --git a/arch/powerpc/platforms/powernv/pci-ioda-tce.c b/arch/powerpc/platforms/powernv/pci-ioda-tce.c
  57 index 7639b2168755..f5adb6b756f7 100644
  58 --- a/arch/powerpc/platforms/powernv/pci-ioda-tce.c
  59 +++ b/arch/powerpc/platforms/powernv/pci-ioda-tce.c
  60 @@ -313,7 +313,6 @@ long pnv_pci_ioda2_table_alloc_pages(int nid, __u64 bus_offset,
  61                         page_shift);
  62         tbl->it_level_size = 1ULL << (level_shift - 3);
  63         tbl->it_indirect_levels = levels - 1;
  64 -       tbl->it_allocated_size = total_allocated;
  65         tbl->it_userspace = uas;
  66         tbl->it_nid = nid;
  67
  68 diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c
  69 index cde710297a4e..326ca6288bb1 100644
  70 --- a/arch/powerpc/platforms/powernv/pci-ioda.c
  71 +++ b/arch/powerpc/platforms/powernv/pci-ioda.c
  72 @@ -2603,8 +2603,13 @@ static long pnv_pci_ioda2_create_table_userspace(
  73                 int num, __u32 page_shift, __u64 window_size, __u32 levels,
  74                 struct iommu_table **ptbl)
  75  {
  76 -       return pnv_pci_ioda2_create_table(table_group,
  77 +       long ret = pnv_pci_ioda2_create_table(table_group,
  78                         num, page_shift, window_size, levels, true, ptbl);
  79 +
  80 +       if (!ret)
  81 +               (*ptbl)->it_allocated_size = pnv_pci_ioda2_get_table_size(
  82 +                               page_shift, window_size, levels);
  83 +       return ret;
  84  }
  85
  86  static void pnv_ioda2_take_ownership(struct iommu_table_group *table_group)
  87 --
  88 2.19.1
  89