From: Marek Majtyka <marek.majtyka@xxxxxxxxx>
A critical bug has been found in device memory stage1 translation for machines
with more then 4GB of RAM. Once vm_pgoff size is smaller then pa (which is true
for LPAE case, u32 and u64 respectively) some more significant bits of pa may
be lost as a shift operation is performed on u32 and later cast onto u64.
Example: vm_pgoff(u32)=0x00210030, PAGE_SHIFT=12
expected pa(u64): 0x0000002010030000
produced pa(u64): 0x0000000010030000
A suggested fix is changing the order of operations (casting first onto
phys_addr_t and then shifting), which works for both alternatives
(with and without LPAE).
A bug has been found, fixed and tested on kernel v3.19. However, it seems to be
valid for most, if not all, newer version starting from v3.18.
Signed-off-by: Marek Majtyka <marek.majtyka@xxxxxxxxx>
---
arch/arm/kvm/mmu.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/arch/arm/kvm/mmu.c b/arch/arm/kvm/mmu.c
index 1366625..602bc63 100644
--- a/arch/arm/kvm/mmu.c
+++ b/arch/arm/kvm/mmu.c
@@ -1408,7 +1408,8 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
if (vma->vm_flags & VM_PFNMAP) {
gpa_t gpa = mem->guest_phys_addr +
(vm_start - mem->userspace_addr);
- phys_addr_t pa = (vma->vm_pgoff << PAGE_SHIFT) +
+ phys_addr_t pa =
+ ((phys_addr_t) vma->vm_pgoff << PAGE_SHIFT) +
vm_start - vma->vm_start;
ret = kvm_phys_addr_ioremap(kvm, gpa, pa,
--
1.9.1
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
