target/linux/generic/backport-5.15/020-v6.1-18-mm-introduce-arch_has_hw_nonleaf_pmd_young.patch

   1 From 46cbda7b65998a5af4493f745d94417af697bd68 Mon Sep 17 00:00:00 2001
   2 From: Juergen Gross <jgross@suse.com>
   3 Date: Wed, 23 Nov 2022 07:45:10 +0100
   4 Subject: [PATCH 18/29] mm: introduce arch_has_hw_nonleaf_pmd_young()
   5
   6 When running as a Xen PV guests commit eed9a328aa1a ("mm: x86: add
   7 CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG") can cause a protection violation in
   8 pmdp_test_and_clear_young():
   9
  10  BUG: unable to handle page fault for address: ffff8880083374d0
  11  #PF: supervisor write access in kernel mode
  12  #PF: error_code(0x0003) - permissions violation
  13  PGD 3026067 P4D 3026067 PUD 3027067 PMD 7fee5067 PTE 8010000008337065
  14  Oops: 0003 [#1] PREEMPT SMP NOPTI
  15  CPU: 7 PID: 158 Comm: kswapd0 Not tainted 6.1.0-rc5-20221118-doflr+ #1
  16  RIP: e030:pmdp_test_and_clear_young+0x25/0x40
  17
  18 This happens because the Xen hypervisor can't emulate direct writes to
  19 page table entries other than PTEs.
  20
  21 This can easily be fixed by introducing arch_has_hw_nonleaf_pmd_young()
  22 similar to arch_has_hw_pte_young() and test that instead of
  23 CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG.
  24
  25 Link: https://lkml.kernel.org/r/20221123064510.16225-1-jgross@suse.com
  26 Fixes: eed9a328aa1a ("mm: x86: add CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG")
  27 Signed-off-by: Juergen Gross <jgross@suse.com>
  28 Reported-by: Sander Eikelenboom <linux@eikelenboom.it>
  29 Acked-by: Yu Zhao <yuzhao@google.com>
  30 Tested-by: Sander Eikelenboom <linux@eikelenboom.it>
  31 Acked-by: David Hildenbrand <david@redhat.com>  [core changes]
  32 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  33 ---
  34  arch/x86/include/asm/pgtable.h |  8 ++++++++
  35  include/linux/pgtable.h        | 11 +++++++++++
  36  mm/vmscan.c                    | 10 +++++-----
  37  3 files changed, 24 insertions(+), 5 deletions(-)
  38
  39 --- a/arch/x86/include/asm/pgtable.h
  40 +++ b/arch/x86/include/asm/pgtable.h
  41 @@ -1405,6 +1405,14 @@ static inline bool arch_has_hw_pte_young
  42         return true;
  43  }
  44
  45 +#ifdef CONFIG_XEN_PV
  46 +#define arch_has_hw_nonleaf_pmd_young arch_has_hw_nonleaf_pmd_young
  47 +static inline bool arch_has_hw_nonleaf_pmd_young(void)
  48 +{
  49 +       return !cpu_feature_enabled(X86_FEATURE_XENPV);
  50 +}
  51 +#endif
  52 +
  53  #endif /* __ASSEMBLY__ */
  54
  55  #endif /* _ASM_X86_PGTABLE_H */
  56 --- a/include/linux/pgtable.h
  57 +++ b/include/linux/pgtable.h
  58 @@ -266,6 +266,17 @@ static inline int pmdp_clear_flush_young
  59  #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
  60  #endif
  61
  62 +#ifndef arch_has_hw_nonleaf_pmd_young
  63 +/*
  64 + * Return whether the accessed bit in non-leaf PMD entries is supported on the
  65 + * local CPU.
  66 + */
  67 +static inline bool arch_has_hw_nonleaf_pmd_young(void)
  68 +{
  69 +       return IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG);
  70 +}
  71 +#endif
  72 +
  73  #ifndef arch_has_hw_pte_young
  74  /*
  75   * Return whether the accessed bit is supported on the local CPU.
  76 --- a/mm/vmscan.c
  77 +++ b/mm/vmscan.c
  78 @@ -3727,7 +3727,7 @@ static void walk_pmd_range_locked(pud_t
  79                         goto next;
  80
  81                 if (!pmd_trans_huge(pmd[i])) {
  82 -                       if (IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG) &&
  83 +                       if (arch_has_hw_nonleaf_pmd_young() &&
  84                             get_cap(LRU_GEN_NONLEAF_YOUNG))
  85                                 pmdp_test_and_clear_young(vma, addr, pmd + i);
  86                         goto next;
  87 @@ -3825,14 +3825,14 @@ restart:
  88  #endif
  89                 walk->mm_stats[MM_NONLEAF_TOTAL]++;
  90
  91 -#ifdef CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG
  92 -               if (get_cap(LRU_GEN_NONLEAF_YOUNG)) {
  93 +               if (arch_has_hw_nonleaf_pmd_young() &&
  94 +                   get_cap(LRU_GEN_NONLEAF_YOUNG)) {
  95                         if (!pmd_young(val))
  96                                 continue;
  97
  98                         walk_pmd_range_locked(pud, addr, vma, args, bitmap, &pos);
  99                 }
 100 -#endif
 101 +
 102                 if (!walk->force_scan && !test_bloom_filter(walk->lruvec, walk->max_seq, pmd + i))
 103                         continue;
 104
 105 @@ -5132,7 +5132,7 @@ static ssize_t show_enabled(struct kobje
 106         if (arch_has_hw_pte_young() && get_cap(LRU_GEN_MM_WALK))
 107                 caps |= BIT(LRU_GEN_MM_WALK);
 108
 109 -       if (IS_ENABLED(CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG) && get_cap(LRU_GEN_NONLEAF_YOUNG))
 110 +       if (arch_has_hw_nonleaf_pmd_young() && get_cap(LRU_GEN_NONLEAF_YOUNG))
 111                 caps |= BIT(LRU_GEN_NONLEAF_YOUNG);
 112
 113         return snprintf(buf, PAGE_SIZE, "0x%04x\n", caps);