[v2,4/7] KVM: selftests: Correctly initialize the VA space for TTBR0_EL1

Message ID 20221209015307.1781352-5-oliver.upton@linux.dev
State New
Headers
Series [v2,1/7] KVM: selftests: Fix build due to ucall_uninit() removal |

Commit Message

Oliver Upton Dec. 9, 2022, 1:53 a.m. UTC
  An interesting feature of the Arm architecture is that the stage-1 MMU
supports two distinct VA regions, controlled by TTBR{0,1}_EL1. As KVM
selftests on arm64 only uses TTBR0_EL1, the VA space is constrained to
[0, 2^(va_bits)). This is different from other architectures that
allow for addressing low and high regions of the VA space from a single
page table.

KVM selftests' VA space allocator presumes the valid address range is
split between low and high memory based the MSB, which of course is a
poor match for arm64's TTBR0 region.

Add a helper that correctly handles both addressing schemes with a
comment describing each.

Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
---
 .../selftests/kvm/include/kvm_util_base.h     |  1 +
 tools/testing/selftests/kvm/lib/kvm_util.c    | 49 ++++++++++++++++---
 2 files changed, 44 insertions(+), 6 deletions(-)
  

Comments

Sean Christopherson Dec. 9, 2022, 8:45 p.m. UTC | #1
On Fri, Dec 09, 2022, Oliver Upton wrote:
> An interesting feature of the Arm architecture is that the stage-1 MMU
> supports two distinct VA regions, controlled by TTBR{0,1}_EL1. As KVM
> selftests on arm64 only uses TTBR0_EL1, the VA space is constrained to
> [0, 2^(va_bits)). This is different from other architectures that
> allow for addressing low and high regions of the VA space from a single
> page table.
> 
> KVM selftests' VA space allocator presumes the valid address range is
> split between low and high memory based the MSB, which of course is a
> poor match for arm64's TTBR0 region.
> 
> Add a helper that correctly handles both addressing schemes with a
> comment describing each.
> 
> Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
> ---

Thanks much!  Looks awesome, especially the comment!

Reviewed-by: Sean Christopherson <seanjc@google.com>

>  .../selftests/kvm/include/kvm_util_base.h     |  1 +
>  tools/testing/selftests/kvm/lib/kvm_util.c    | 49 ++++++++++++++++---
>  2 files changed, 44 insertions(+), 6 deletions(-)
> 
> diff --git a/tools/testing/selftests/kvm/include/kvm_util_base.h b/tools/testing/selftests/kvm/include/kvm_util_base.h
> index 6cd86da698b3..b193863d754f 100644
> --- a/tools/testing/selftests/kvm/include/kvm_util_base.h
> +++ b/tools/testing/selftests/kvm/include/kvm_util_base.h
> @@ -103,6 +103,7 @@ struct kvm_vm {
>  	struct sparsebit *vpages_mapped;
>  	bool has_irqchip;
>  	bool pgd_created;
> +	bool has_split_va_space;
>  	vm_paddr_t ucall_mmio_addr;
>  	vm_paddr_t pgd;
>  	vm_vaddr_t gdt;
> diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
> index a256ec67aff6..53d15f32f220 100644
> --- a/tools/testing/selftests/kvm/lib/kvm_util.c
> +++ b/tools/testing/selftests/kvm/lib/kvm_util.c
> @@ -186,6 +186,43 @@ const struct vm_guest_mode_params vm_guest_mode_params[] = {
>  _Static_assert(sizeof(vm_guest_mode_params)/sizeof(struct vm_guest_mode_params) == NUM_VM_MODES,
>  	       "Missing new mode params?");
>  
> +/*
> + * Initializes vm->vpages_valid to match the canonical VA space of the
> + * architecture.
> + *
> + * Most architectures split the range addressed by a single page table into a
> + * low and high region based on the MSB of the VA. On architectures with this
> + * behavior the VA region spans [0, 2^(va_bits - 1)), [-(2^(va_bits - 1), -1].
> + *
> + * arm64 is a bit different from the rest of the crowd, as the low and high
> + * regions of the VA space are addressed by distinct paging structures
> + * (TTBR{0,1}_EL1).

Oooh, they're different CR3s in x86 terminology?
  
Oliver Upton Dec. 9, 2022, 9:22 p.m. UTC | #2
On Fri, Dec 09, 2022 at 08:45:01PM +0000, Sean Christopherson wrote:
> On Fri, Dec 09, 2022, Oliver Upton wrote:
> > An interesting feature of the Arm architecture is that the stage-1 MMU
> > supports two distinct VA regions, controlled by TTBR{0,1}_EL1. As KVM
> > selftests on arm64 only uses TTBR0_EL1, the VA space is constrained to
> > [0, 2^(va_bits)). This is different from other architectures that
> > allow for addressing low and high regions of the VA space from a single
> > page table.
> > 
> > KVM selftests' VA space allocator presumes the valid address range is
> > split between low and high memory based the MSB, which of course is a
> > poor match for arm64's TTBR0 region.
> > 
> > Add a helper that correctly handles both addressing schemes with a
> > comment describing each.
> > 
> > Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
> > ---
> 
> Thanks much!  Looks awesome, especially the comment!
> 
> Reviewed-by: Sean Christopherson <seanjc@google.com>

ty!

> >  .../selftests/kvm/include/kvm_util_base.h     |  1 +
> >  tools/testing/selftests/kvm/lib/kvm_util.c    | 49 ++++++++++++++++---
> >  2 files changed, 44 insertions(+), 6 deletions(-)
> > 
> > diff --git a/tools/testing/selftests/kvm/include/kvm_util_base.h b/tools/testing/selftests/kvm/include/kvm_util_base.h
> > index 6cd86da698b3..b193863d754f 100644
> > --- a/tools/testing/selftests/kvm/include/kvm_util_base.h
> > +++ b/tools/testing/selftests/kvm/include/kvm_util_base.h
> > @@ -103,6 +103,7 @@ struct kvm_vm {
> >  	struct sparsebit *vpages_mapped;
> >  	bool has_irqchip;
> >  	bool pgd_created;
> > +	bool has_split_va_space;
> >  	vm_paddr_t ucall_mmio_addr;
> >  	vm_paddr_t pgd;
> >  	vm_vaddr_t gdt;
> > diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
> > index a256ec67aff6..53d15f32f220 100644
> > --- a/tools/testing/selftests/kvm/lib/kvm_util.c
> > +++ b/tools/testing/selftests/kvm/lib/kvm_util.c
> > @@ -186,6 +186,43 @@ const struct vm_guest_mode_params vm_guest_mode_params[] = {
> >  _Static_assert(sizeof(vm_guest_mode_params)/sizeof(struct vm_guest_mode_params) == NUM_VM_MODES,
> >  	       "Missing new mode params?");
> >  
> > +/*
> > + * Initializes vm->vpages_valid to match the canonical VA space of the
> > + * architecture.
> > + *
> > + * Most architectures split the range addressed by a single page table into a
> > + * low and high region based on the MSB of the VA. On architectures with this
> > + * behavior the VA region spans [0, 2^(va_bits - 1)), [-(2^(va_bits - 1), -1].
> > + *
> > + * arm64 is a bit different from the rest of the crowd, as the low and high
> > + * regions of the VA space are addressed by distinct paging structures
> > + * (TTBR{0,1}_EL1).
> 
> Oooh, they're different CR3s in x86 terminology?

Right, we can have two active table roots at any given time, each
mapping their own portion of the address space.

--
Thanks,
Oliver
  
Paolo Bonzini Dec. 12, 2022, 10:34 a.m. UTC | #3
On 12/9/22 02:53, Oliver Upton wrote:
> @@ -268,17 +305,17 @@ struct kvm_vm *____vm_create(enum vm_guest_mode mode)
>   #ifdef __aarch64__
>   	if (vm->pa_bits != 40)
>   		vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);
> +
> +	/* selftests use TTBR0 only, meaning there is a single VA region. */
> +	vm->has_split_va_space = false;
> +#else
> +	vm->has_split_va_space = true;
>   #endif
>   

Even though there happens to be already a suitable #ifdef, I don't
really like them and don't think there should be a new bool unless
something else uses it.

However, the new comment is very useful, so I added it to kvm_util.c as
follows:

/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
index 759a45540108..56d5ea949cbb 100644
--- a/tools/testing/selftests/kvm/lib/kvm_util.c
+++ b/tools/testing/selftests/kvm/lib/kvm_util.c
@@ -186,6 +186,15 @@ const struct vm_guest_mode_params vm_guest_mode_params[] = {
  _Static_assert(sizeof(vm_guest_mode_params)/sizeof(struct vm_guest_mode_params) == NUM_VM_MODES,
  	       "Missing new mode params?");
  
+/*
+ * Initializes vm->vpages_valid to match the canonical VA space of the
+ * architecture.
+ *
+ * The default implementation is valid for architectures which split the
+ * range addressed by a single page table into a low and high region
+ * based on the MSB of the VA. On architectures with this behavior
+ * the VA region spans [0, 2^(va_bits - 1)), [-(2^(va_bits - 1), -1].
+ */
  __weak void vm_vaddr_populate_bitmap(struct kvm_vm *vm)
  {
  	sparsebit_set_num(vm->vpages_valid,
  

Patch

diff --git a/tools/testing/selftests/kvm/include/kvm_util_base.h b/tools/testing/selftests/kvm/include/kvm_util_base.h
index 6cd86da698b3..b193863d754f 100644
--- a/tools/testing/selftests/kvm/include/kvm_util_base.h
+++ b/tools/testing/selftests/kvm/include/kvm_util_base.h
@@ -103,6 +103,7 @@  struct kvm_vm {
 	struct sparsebit *vpages_mapped;
 	bool has_irqchip;
 	bool pgd_created;
+	bool has_split_va_space;
 	vm_paddr_t ucall_mmio_addr;
 	vm_paddr_t pgd;
 	vm_vaddr_t gdt;
diff --git a/tools/testing/selftests/kvm/lib/kvm_util.c b/tools/testing/selftests/kvm/lib/kvm_util.c
index a256ec67aff6..53d15f32f220 100644
--- a/tools/testing/selftests/kvm/lib/kvm_util.c
+++ b/tools/testing/selftests/kvm/lib/kvm_util.c
@@ -186,6 +186,43 @@  const struct vm_guest_mode_params vm_guest_mode_params[] = {
 _Static_assert(sizeof(vm_guest_mode_params)/sizeof(struct vm_guest_mode_params) == NUM_VM_MODES,
 	       "Missing new mode params?");
 
+/*
+ * Initializes vm->vpages_valid to match the canonical VA space of the
+ * architecture.
+ *
+ * Most architectures split the range addressed by a single page table into a
+ * low and high region based on the MSB of the VA. On architectures with this
+ * behavior the VA region spans [0, 2^(va_bits - 1)), [-(2^(va_bits - 1), -1].
+ *
+ * arm64 is a bit different from the rest of the crowd, as the low and high
+ * regions of the VA space are addressed by distinct paging structures
+ * (TTBR{0,1}_EL1). KVM selftests on arm64 only uses TTBR0_EL1, meaning that we
+ * only have a low VA region. As there is no VA split based on the MSB, the VA
+ * region spans [0, 2^va_bits).
+ */
+static void vm_vaddr_populate_bitmap(struct kvm_vm *vm)
+{
+	sparsebit_num_t contig_va_bits = vm->va_bits;
+	sparsebit_num_t nr_contig_pages;
+
+	/*
+	 * Depending on the architecture, the MSB of the VA could split between
+	 * low and high regions. When that is the case each region has
+	 * va_bits - 1 of address.
+	 */
+	if (vm->has_split_va_space)
+		contig_va_bits--;
+
+	nr_contig_pages = (1ULL << contig_va_bits) >> vm->page_shift;
+
+	sparsebit_set_num(vm->vpages_valid, 0, nr_contig_pages);
+
+	if (vm->has_split_va_space)
+		sparsebit_set_num(vm->vpages_valid,
+				  -(1ULL << contig_va_bits),
+				  nr_contig_pages);
+}
+
 struct kvm_vm *____vm_create(enum vm_guest_mode mode)
 {
 	struct kvm_vm *vm;
@@ -268,17 +305,17 @@  struct kvm_vm *____vm_create(enum vm_guest_mode mode)
 #ifdef __aarch64__
 	if (vm->pa_bits != 40)
 		vm->type = KVM_VM_TYPE_ARM_IPA_SIZE(vm->pa_bits);
+
+	/* selftests use TTBR0 only, meaning there is a single VA region. */
+	vm->has_split_va_space = false;
+#else
+	vm->has_split_va_space = true;
 #endif
 
 	vm_open(vm);
 
-	/* Limit to VA-bit canonical virtual addresses. */
 	vm->vpages_valid = sparsebit_alloc();
-	sparsebit_set_num(vm->vpages_valid,
-		0, (1ULL << (vm->va_bits - 1)) >> vm->page_shift);
-	sparsebit_set_num(vm->vpages_valid,
-		(~((1ULL << (vm->va_bits - 1)) - 1)) >> vm->page_shift,
-		(1ULL << (vm->va_bits - 1)) >> vm->page_shift);
+	vm_vaddr_populate_bitmap(vm);
 
 	/* Limit physical addresses to PA-bits. */
 	vm->max_gfn = vm_compute_max_gfn(vm);