[PATCH 0/2] Fix off-by-one in range_is_memory

[PATCH 0/2] Fix off-by-one in range_is_memory

[David Brazdil]

Hi, here is an off-by-one bug fix and a very minor improvement for
the range_is_memory function in hyp.

David Brazdil (2):
  KVM: arm64: Fix off-by-one in range_is_memory
  KVM: arm64: Minor optimization of range_is_memory

 arch/arm64/kvm/hyp/nvhe/mem_protect.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

-- 
2.32.0.432.gabb21c7263-goog

[PATCH 1/2] KVM: arm64: Fix off-by-one in range_is_memory

[David Brazdil]

Hyp checks whether an address range only covers RAM by checking the
start/endpoints against a list of memblock_region structs. However,
the endpoint here is exclusive but internally is treated as inclusive.
Fix the off-by-one error that caused valid address ranges to be
rejected.

Cc: Quentin Perret <qperret@google.com>
Fixes: 90134ac9cabb6 ("KVM: arm64: Protect the .hyp sections from the host")
Signed-off-by: David Brazdil <dbrazdil@google.com>
---
 arch/arm64/kvm/hyp/nvhe/mem_protect.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/mem_protect.c b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
index d938ce95d3bd..a6ce991b1467 100644
--- a/arch/arm64/kvm/hyp/nvhe/mem_protect.c
+++ b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
@@ -193,7 +193,7 @@ static bool range_is_memory(u64 start, u64 end)
 {
 	struct kvm_mem_range r1, r2;
 
-	if (!find_mem_range(start, &r1) || !find_mem_range(end, &r2))
+	if (!find_mem_range(start, &r1) || !find_mem_range(end - 1, &r2))
 		return false;
 	if (r1.start != r2.start)
 		return false;
-- 
2.32.0.432.gabb21c7263-goog

[PATCH 2/2] KVM: arm64: Minor optimization of range_is_memory

[David Brazdil]

Currently range_is_memory finds the corresponding struct memblock_region
for both the lower and upper bounds of the given address range with two
rounds of binary search, and then checks that the two memblocks are the
same. Simplify this by only doing binary search on the lower bound and
then checking that the upper bound is in the same memblock.

Signed-off-by: David Brazdil <dbrazdil@google.com>
---
 arch/arm64/kvm/hyp/nvhe/mem_protect.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/kvm/hyp/nvhe/mem_protect.c b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
index a6ce991b1467..37d73af69634 100644
--- a/arch/arm64/kvm/hyp/nvhe/mem_protect.c
+++ b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
@@ -189,13 +189,18 @@ static bool find_mem_range(phys_addr_t addr, struct kvm_mem_range *range)
 	return false;
 }
 
+static bool is_in_mem_range(phys_addr_t addr, struct kvm_mem_range *range)
+{
+	return range->start <= addr && addr < range->end;
+}
+
 static bool range_is_memory(u64 start, u64 end)
 {
-	struct kvm_mem_range r1, r2;
+	struct kvm_mem_range r;
 
-	if (!find_mem_range(start, &r1) || !find_mem_range(end - 1, &r2))
+	if (!find_mem_range(start, &r))
 		return false;
-	if (r1.start != r2.start)
+	if (!is_in_mem_range(end - 1, &r))
 		return false;
 
 	return true;
-- 
2.32.0.432.gabb21c7263-goog

Re: [PATCH 1/2] KVM: arm64: Fix off-by-one in range_is_memory

[Quentin Perret]

On Wednesday 28 Jul 2021 at 15:32:31 (+0000), David Brazdil wrote:
> Hyp checks whether an address range only covers RAM by checking the
> start/endpoints against a list of memblock_region structs. However,
> the endpoint here is exclusive but internally is treated as inclusive.
> Fix the off-by-one error that caused valid address ranges to be
> rejected.
> 
> Cc: Quentin Perret <qperret@google.com>
> Fixes: 90134ac9cabb6 ("KVM: arm64: Protect the .hyp sections from the host")
> Signed-off-by: David Brazdil <dbrazdil@google.com>
> ---
>  arch/arm64/kvm/hyp/nvhe/mem_protect.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/kvm/hyp/nvhe/mem_protect.c b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> index d938ce95d3bd..a6ce991b1467 100644
> --- a/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> +++ b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> @@ -193,7 +193,7 @@ static bool range_is_memory(u64 start, u64 end)
>  {
>  	struct kvm_mem_range r1, r2;
>  
> -	if (!find_mem_range(start, &r1) || !find_mem_range(end, &r2))
> +	if (!find_mem_range(start, &r1) || !find_mem_range(end - 1, &r2))
>  		return false;
>  	if (r1.start != r2.start)
>  		return false;

Looks good to me:

Reviewed-by: Quentin Perret <qperret@google.com>

Thanks,
Quentin

Re: [PATCH 2/2] KVM: arm64: Minor optimization of range_is_memory

[Quentin Perret]

On Wednesday 28 Jul 2021 at 15:32:32 (+0000), David Brazdil wrote:
> Currently range_is_memory finds the corresponding struct memblock_region
> for both the lower and upper bounds of the given address range with two
> rounds of binary search, and then checks that the two memblocks are the
> same. Simplify this by only doing binary search on the lower bound and
> then checking that the upper bound is in the same memblock.
> 
> Signed-off-by: David Brazdil <dbrazdil@google.com>
> ---
>  arch/arm64/kvm/hyp/nvhe/mem_protect.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/arm64/kvm/hyp/nvhe/mem_protect.c b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> index a6ce991b1467..37d73af69634 100644
> --- a/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> +++ b/arch/arm64/kvm/hyp/nvhe/mem_protect.c
> @@ -189,13 +189,18 @@ static bool find_mem_range(phys_addr_t addr, struct kvm_mem_range *range)
>  	return false;
>  }
>  
> +static bool is_in_mem_range(phys_addr_t addr, struct kvm_mem_range *range)
> +{

Nit: addr@ could be u64 for consistency -- struct kvm_mem_range holds
IPAs in general.

> +	return range->start <= addr && addr < range->end;
> +}
> +
>  static bool range_is_memory(u64 start, u64 end)
>  {
> -	struct kvm_mem_range r1, r2;
> +	struct kvm_mem_range r;
>  
> -	if (!find_mem_range(start, &r1) || !find_mem_range(end - 1, &r2))
> +	if (!find_mem_range(start, &r))
>  		return false;
> -	if (r1.start != r2.start)
> +	if (!is_in_mem_range(end - 1, &r))
>  		return false;
>  
>  	return true;

Nit: maybe drop the second if and simplify to:

	return is_in_mem_range(end - 1, &r);

With that:

Reviewed-by: Quentin Perret <qperret@google.com>

Thanks,
Quentin

Re: (subset) [PATCH 0/2] Fix off-by-one in range_is_memory

[Marc Zyngier]

On Wed, 28 Jul 2021 15:32:30 +0000, David Brazdil wrote:
> the range_is_memory function in hyp.
> 
> David Brazdil (2):
>   KVM: arm64: Fix off-by-one in range_is_memory
>   KVM: arm64: Minor optimization of range_is_memory
> 
>  arch/arm64/kvm/hyp/nvhe/mem_protect.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)

Applied to kvm-arm64/mmu/el2-tracking, thanks!

[2/2] KVM: arm64: Minor optimization of range_is_memory
      commit: 14ecf075fe5be01860927fdf3aa11d7b18023ab2

Cheers,

	M.
-- 
Without deviation from the norm, progress is not possible.