[PATCH 1/2] KVM: arm64: Fix spurious warning for benign stage 2 teardown race
Lorenzo Stoakes (ARM)
ljs at kernel.org
Wed Aug 12 06:46:25 PDT 2026
The splats:
------------[ cut here ]------------
WARNING: arch/arm64/kvm/mmu.c:336 at __unmap_stage2_range+0x64/0x80, CPU#5: kcompactd0/66
...
Tainted: [S]=CPU_OUT_OF_SPEC
Hardware name: Apple Inc. Mac14,7/J493, BIOS 2026.04 04/01/2026
pstate: 61401009 (nZCv daif +PAN -UAO -TCO +DIT +SSBS BTYPE=--)
pc : __unmap_stage2_range (arch/arm64/kvm/mmu.c:335 (discriminator 3))
lr : __unmap_stage2_range (arch/arm64/kvm/mmu.c:336 (discriminator 5))
sp : ffff8000856e7450
x29: ffff8000856e7450 x28: ffff0000404b2000 x27: 0000000000000030
x26: ffffffffffffffe0 x25: 0000000000000001 x24: ffff00000eaa5920
x23: ffff0000404b2000 x22: 0000000000000000 x21: 0000000000000040
x20: 0000000000000001 x19: 0000000000000001 x18: 0000000000000006
x17: 0000000000000000 x16: 0000000000000002 x15: ffff800083adbed0
x14: ffff800083adbed0 x13: 0000000000000008 x12: 0000b00000000000
x11: 0000000000000001 x10: 0000006000000000 x9 : ffff8000817334fc
x8 : 0000000000200000 x7 : ffff0000c59ea000 x6 : 0000000000000800
x5 : 0000000038000000 x4 : 0000000000000000 x3 : ffff0000404b2000
x2 : 0000000000000001 x1 : 0000000038000000 x0 : 00000000ffffffea
Call trace:
__unmap_stage2_range (arch/arm64/kvm/mmu.c:335 (discriminator 3)) (P)
kvm_stage2_unmap_range (arch/arm64/kvm/mmu.c:346)
kvm_nested_s2_unmap (arch/arm64/kvm/nested.c:1168 (discriminator 14))
kvm_unmap_gfn_range (arch/arm64/kvm/mmu.c:2416)
kvm_mmu_notifier_invalidate_range_start (arch/arm64/kvm/../../../virt/kvm/kvm_main.c:718 (discriminator 32768) arch/arm64/kvm/../../../virt/kvm/kvm_main.c:633 (discriminator 32768) arch/arm64/kvm/../../../virt/kvm/kvm_main.c:765 (discriminator 32768))
mn_hlist_invalidate_range_start (mm/mmu_notifier.c:525)
__mmu_notifier_invalidate_range_start (mm/mmu_notifier.c:580)
try_to_migrate_one (./include/linux/mmu_notifier.h:478 ./include/linux/mmu_notifier.h:471 mm/rmap.c:2459)
rmap_walk_anon (mm/rmap.c:3001)
rmap_walk (mm/rmap.c:3106 mm/rmap.c:3101)
try_to_migrate (mm/rmap.c:2774)
migrate_folio_unmap (mm/migrate.c:1330 (discriminator 3))
migrate_pages_batch (mm/migrate.c:1909)
migrate_pages_sync (mm/migrate.c:2026)
migrate_pages (mm/migrate.c:2135)
compact_zone (mm/compaction.c:2663)
compact_node (mm/compaction.c:2932)
kcompactd (mm/compaction.c:3230)
kthread (kernel/kthread.c:436)
ret_from_fork (arch/arm64/kernel/entry.S:858)
---[ end trace 0000000000000000 ]---
------------[ cut here ]------------
WARNING: arch/arm64/kvm/mmu.c:336 at __unmap_stage2_range+0x64/0x80, CPU#5: kcompactd0/66
...
Tainted: [S]=CPU_OUT_OF_SPEC, [W]=WARN
Hardware name: Apple Inc. Mac14,7/J493, BIOS 2026.04 04/01/2026
pstate: 61401009 (nZCv daif +PAN -UAO -TCO +DIT +SSBS BTYPE=--)
pc : __unmap_stage2_range (arch/arm64/kvm/mmu.c:335 (discriminator 3))
lr : __unmap_stage2_range (arch/arm64/kvm/mmu.c:336 (discriminator 5))
sp : ffff8000856e7450
x29: ffff8000856e7450 x28: ffff0000404b2000 x27: 0000000000000030
x26: ffffffffffffffe0 x25: 0000000000000001 x24: ffff00000eaa5920
x23: ffff0000404b2000 x22: 0000000000000001 x21: 0000000000000040
x20: 0000000000000001 x19: 0000000000000001 x18: 0000000000000006
x17: 0000000000000000 x16: 0000000000000002 x15: ffff800083adbed0
x14: ffff800083adbed0 x13: 0000000000000008 x12: 0000b00000000000
x11: 0000000000000001 x10: 0000006000000000 x9 : ffff8000817334fc
x8 : 0000000000200000 x7 : ffff0000c59ea078 x6 : 0000000000000800
x5 : 0000000002000000 x4 : 0000000000000000 x3 : ffff0000404b2000
x2 : 0000000000000001 x1 : 0000000002000000 x0 : 00000000ffffffea
Call trace:
__unmap_stage2_range (arch/arm64/kvm/mmu.c:335 (discriminator 3)) (P)
kvm_stage2_unmap_range (arch/arm64/kvm/mmu.c:346)
kvm_nested_s2_unmap (arch/arm64/kvm/nested.c:1168 (discriminator 14))
kvm_unmap_gfn_range (arch/arm64/kvm/mmu.c:2416)
kvm_mmu_notifier_invalidate_range_start (arch/arm64/kvm/../../../virt/kvm/kvm_main.c:718 (discriminator 32768) arch/arm64/kvm/../../../virt/kvm/kvm_main.c:633 (discriminator 32768) arch/arm64/kvm/../../../virt/kvm/kvm_main.c:765 (discriminator 32768))
mn_hlist_invalidate_range_start (mm/mmu_notifier.c:525)
__mmu_notifier_invalidate_range_start (mm/mmu_notifier.c:580)
try_to_migrate_one (./include/linux/mmu_notifier.h:478 ./include/linux/mmu_notifier.h:471 mm/rmap.c:2459)
rmap_walk_anon (mm/rmap.c:3001)
rmap_walk (mm/rmap.c:3106 mm/rmap.c:3101)
try_to_migrate (mm/rmap.c:2774)
migrate_folio_unmap (mm/migrate.c:1330 (discriminator 3))
migrate_pages_batch (mm/migrate.c:1909)
migrate_pages_sync (mm/migrate.c:2026)
migrate_pages (mm/migrate.c:2135)
compact_zone (mm/compaction.c:2663)
compact_node (mm/compaction.c:2932)
kcompactd (mm/compaction.c:3230)
kthread (kernel/kthread.c:436)
ret_from_fork (arch/arm64/kernel/entry.S:858)
---[ end trace 0000000000000000 ]---
------------[ cut here ]------------
WARNING: arch/arm64/kvm/mmu.c:336 at __unmap_stage2_range+0x64/0x80, CPU#5: kcompactd0/66
...
Tainted: [S]=CPU_OUT_OF_SPEC, [W]=WARN
Hardware name: Apple Inc. Mac14,7/J493, BIOS 2026.04 04/01/2026
pstate: 61401009 (nZCv daif +PAN -UAO -TCO +DIT +SSBS BTYPE=--)
pc : __unmap_stage2_range (arch/arm64/kvm/mmu.c:335 (discriminator 3))
lr : __unmap_stage2_range (arch/arm64/kvm/mmu.c:336 (discriminator 5))
sp : ffff8000856e7450
x29: ffff8000856e7450 x28: ffff0000404b2000 x27: 0000000000000030
x26: ffffffffffffffe0 x25: 0000000000000001 x24: ffff00000eaa5920
x23: ffff0000404b2000 x22: 0000000000000002 x21: 0000000000000040
x20: 0000000000000001 x19: 0000000000000001 x18: 0000000000000006
x17: 0000000000000000 x16: 0000000000000002 x15: ffff800083adbed0
x14: ffff800083adbed0 x13: 0000000000000008 x12: 0000b00000000000
x11: 0000000000000001 x10: 0000006000000000 x9 : ffff8000817334fc
x8 : 0000000000200000 x7 : ffff0000c59ea0f0 x6 : 0000000000000800
x5 : 0000000002000000 x4 : 0000000000000000 x3 : ffff0000404b2000
x2 : 0000000000000001 x1 : 0000000002000000 x0 : 00000000ffffffea
Call trace:
__unmap_stage2_range (arch/arm64/kvm/mmu.c:335 (discriminator 3)) (P)
kvm_stage2_unmap_range (arch/arm64/kvm/mmu.c:346)
kvm_nested_s2_unmap (arch/arm64/kvm/nested.c:1168 (discriminator 14))
kvm_unmap_gfn_range (arch/arm64/kvm/mmu.c:2416)
kvm_mmu_notifier_invalidate_range_start (arch/arm64/kvm/../../../virt/kvm/kvm_main.c:718 (discriminator 32768) arch/arm64/kvm/../../../virt/kvm/kvm_main.c:633 (discriminator 32768) arch/arm64/kvm/../../../virt/kvm/kvm_main.c:765 (discriminator 32768))
mn_hlist_invalidate_range_start (mm/mmu_notifier.c:525)
__mmu_notifier_invalidate_range_start (mm/mmu_notifier.c:580)
try_to_migrate_one (./include/linux/mmu_notifier.h:478 ./include/linux/mmu_notifier.h:471 mm/rmap.c:2459)
rmap_walk_anon (mm/rmap.c:3001)
rmap_walk (mm/rmap.c:3106 mm/rmap.c:3101)
try_to_migrate (mm/rmap.c:2774)
migrate_folio_unmap (mm/migrate.c:1330 (discriminator 3))
migrate_pages_batch (mm/migrate.c:1909)
migrate_pages_sync (mm/migrate.c:2026)
migrate_pages (mm/migrate.c:2135)
compact_zone (mm/compaction.c:2663)
compact_node (mm/compaction.c:2932)
kcompactd (mm/compaction.c:3230)
kthread (kernel/kthread.c:436)
ret_from_fork (arch/arm64/kernel/entry.S:858)
---[ end trace 0000000000000000 ]---
--
Also I wrote something extra but dropped from the already-far-too-long commit
msg but the registers match the diagnosis, basically:
--
The observed assembly from the splats is consistent with this analysis:
Examining the assembly for stage2_apply_range() loop and pgt check,
conveniently x1 contains the current IPA being examined (addr) and is not
clobbered on return:
mov x1, x19 addr = next
ldr x0, [x26, #16] struct kvm_pgtable *pgt = mmu->pgt;
cbnz x0, ... if (!pgt)...
ldp x19, x20, [sp, #16] }
mov w0, #0xffffffea } return -EINVAL
ret }
Splat 1 has x1 of 0x38000000 which tells us the splat occurred 27 * 32 MiB
or 864 MiB into the nested guest physical address space (27 iterations).
This is consistent with an an interleaving stage 2 tear down occurring
part-way through this operation.
There being 2 other splats is consistent with there being a total of 3
nested mmus, as expected, and that the mmu is invalidated before the
warning can occur.
The remaining 2 have x1 equal to 0x2000000 or exactly 32 MiB, which is what
you would expect with the stage2_apply_range() and
kvm_arch_flush_shadow_all() running in lockstep across mmus:
kvm_unmap_gfn_range() | kvm_arch_flush_shadow_all()
---------------------------------|----------------------------
< kvm->mmu_lock held > |
stage2_apply_range() |
.. 26 successful iterations .. | < contend kvm->mmu_lock >
< releases kvm->mmu lock > | < acquires kvm->mmu_lock >
| mmu->pgt = NULL
| < releases kvm->mmu_lock >
Then 3 occurrences of:
< reacquires kvm->mmu_lock > |
return to __unmap_stage2_range() |
WARN_ON(-EINVAL); |
mmu = &kvm->arch.nested_mmus[1]; |
stage2_apply_range() |
.. 1 successful iteration .. |
< releases kvm->mmu lock > | < contend kvm->mmu_lock >
| < acquires kvm->mmu_lock >
| mmu->pgt = NULL
| < releases kvm->mmu_lock >
One single iteration sets addr = next = 0x2000000 which is reflected in the
splats.
Finally, in kvm_nested_s2_unmap() the disassembly indicates that w22
maintains the i counter, which happens not to be clobbered over the
operation either:
mov w22, #0x0 for (i = 0;
add w22, w22, #0x1 ; i++)
sxtw x2, w22 }
cmp x2, x1 } i < nested_mmus_size
The splats have w22 equal to 0, 1 and 2 respectively which further
corroborates this analysis.
--
Cheers, Lorenzo
More information about the linux-arm-kernel
mailing list