Re: [PATCH v3 13/20] kvm: arm64: Configure VTCR per VM

From: Suzuki K Poulose
Date: Tue Jul 03 2018 - 06:48:37 EST

On 02/07/18 13:16, Marc Zyngier wrote:
On 29/06/18 12:15, Suzuki K Poulose wrote:
We set VTCR_EL2 very early during the stage2 init and don't
touch it ever. This is fine as we had a fixed IPA size. This
patch changes the behavior to set the VTCR for a given VM,
depending on its stage2 table. The common configuration for
VTCR is still performed during the early init as we have to
retain the hardware access flag update bits (VTCR_EL2_HA)
per CPU (as they are only set for the CPUs which are capabile).

capable

The bits defining the number of levels in the page table (SL0)
and and the size of the Input address to the translation (T0SZ)
are programmed for each VM upon entry to the guest.

Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
Cc: Christoffer Dall <cdall@xxxxxxxxxx>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
---
Change since V2:
- Load VTCR for TLB operations
---
arch/arm64/include/asm/kvm_arm.h | 19 +++++++++----------
arch/arm64/include/asm/kvm_asm.h | 2 +-
arch/arm64/include/asm/kvm_host.h | 9 ++++++---
arch/arm64/include/asm/kvm_hyp.h | 11 +++++++++++
arch/arm64/kvm/hyp/s2-setup.c | 17 +----------------
5 files changed, 28 insertions(+), 30 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_arm.h b/arch/arm64/include/asm/kvm_arm.h
index 11a7db0..b02c316 100644
--- a/arch/arm64/include/asm/kvm_arm.h
+++ b/arch/arm64/include/asm/kvm_arm.h
@@ -120,9 +120,7 @@
#define VTCR_EL2_IRGN0_WBWA TCR_IRGN0_WBWA
#define VTCR_EL2_SL0_SHIFT 6
#define VTCR_EL2_SL0_MASK (3 << VTCR_EL2_SL0_SHIFT)
-#define VTCR_EL2_SL0_LVL1 (1 << VTCR_EL2_SL0_SHIFT)
#define VTCR_EL2_T0SZ_MASK 0x3f
-#define VTCR_EL2_T0SZ_40B 24
#define VTCR_EL2_VS_SHIFT 19
#define VTCR_EL2_VS_8BIT (0 << VTCR_EL2_VS_SHIFT)
#define VTCR_EL2_VS_16BIT (1 << VTCR_EL2_VS_SHIFT)
@@ -137,43 +135,44 @@
* VTCR_EL2.PS is extracted from ID_AA64MMFR0_EL1.PARange at boot time
* (see hyp-init.S).
*
+ * VTCR_EL2.SL0 and T0SZ are configured per VM at runtime before switching to
+ * the VM.
+ *
* Note that when using 4K pages, we concatenate two first level page tables
* together. With 16K pages, we concatenate 16 first level page tables.
*
*/
-#define VTCR_EL2_T0SZ_IPA VTCR_EL2_T0SZ_40B
#define VTCR_EL2_COMMON_BITS (VTCR_EL2_SH0_INNER | VTCR_EL2_ORGN0_WBWA | \
VTCR_EL2_IRGN0_WBWA | VTCR_EL2_RES1)
+#define VTCR_EL2_PRIVATE_MASK (VTCR_EL2_SL0_MASK | VTCR_EL2_T0SZ_MASK)

What does "private" mean here? It really is the IPA configuration, so
I'd rather have a naming that reflects that.

#ifdef CONFIG_ARM64_64K_PAGES
/*
* Stage2 translation configuration:
* 64kB pages (TG0 = 1)
- * 2 level page tables (SL = 1)
*/
-#define VTCR_EL2_TGRAN_FLAGS (VTCR_EL2_TG0_64K | VTCR_EL2_SL0_LVL1)
+#define VTCR_EL2_TGRAN VTCR_EL2_TG0_64K
#define VTCR_EL2_TGRAN_SL0_BASE 3UL
#elif defined(CONFIG_ARM64_16K_PAGES)
/*
* Stage2 translation configuration:
* 16kB pages (TG0 = 2)
- * 2 level page tables (SL = 1)
*/
-#define VTCR_EL2_TGRAN_FLAGS (VTCR_EL2_TG0_16K | VTCR_EL2_SL0_LVL1)
+#define VTCR_EL2_TGRAN VTCR_EL2_TG0_16K
#define VTCR_EL2_TGRAN_SL0_BASE 3UL
#else /* 4K */
/*
* Stage2 translation configuration:
* 4kB pages (TG0 = 0)
- * 3 level page tables (SL = 1)
*/
-#define VTCR_EL2_TGRAN_FLAGS (VTCR_EL2_TG0_4K | VTCR_EL2_SL0_LVL1)
+#define VTCR_EL2_TGRAN VTCR_EL2_TG0_4K
#define VTCR_EL2_TGRAN_SL0_BASE 2UL
#endif
-#define VTCR_EL2_FLAGS (VTCR_EL2_COMMON_BITS | VTCR_EL2_TGRAN_FLAGS)
+#define VTCR_EL2_FLAGS (VTCR_EL2_COMMON_BITS | VTCR_EL2_TGRAN)
+
/*
* VTCR_EL2:SL0 indicates the entry level for Stage2 translation.
* Interestingly, it depends on the page size.
diff --git a/arch/arm64/include/asm/kvm_asm.h b/arch/arm64/include/asm/kvm_asm.h
index 102b5a5..91372eb 100644
--- a/arch/arm64/include/asm/kvm_asm.h
+++ b/arch/arm64/include/asm/kvm_asm.h
@@ -72,7 +72,7 @@ extern void __vgic_v3_init_lrs(void);
extern u32 __kvm_get_mdcr_el2(void);
-extern u32 __init_stage2_translation(void);
+extern void __init_stage2_translation(void);
/* Home-grown __this_cpu_{ptr,read} variants that always work at HYP */
#define __hyp_this_cpu_ptr(sym) \
diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index fe8777b..328f472 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -442,10 +442,13 @@ int kvm_arm_vcpu_arch_has_attr(struct kvm_vcpu *vcpu,
static inline void __cpu_init_stage2(void)
{
- u32 parange = kvm_call_hyp(__init_stage2_translation);
+ u32 ps;
- WARN_ONCE(parange < 40,
- "PARange is %d bits, unsupported configuration!", parange);
+ kvm_call_hyp(__init_stage2_translation);
+ /* Sanity check for minimum IPA size support */
+ ps = id_aa64mmfr0_parange_to_phys_shift(read_sysreg(id_aa64mmfr0_el1) & 0x7);
+ WARN_ONCE(ps < 40,
+ "PARange is %d bits, unsupported configuration!", ps);
}
/* Guest/host FPSIMD coordination helpers */
diff --git a/arch/arm64/include/asm/kvm_hyp.h b/arch/arm64/include/asm/kvm_hyp.h
index 82f9994..3e8052d1 100644
--- a/arch/arm64/include/asm/kvm_hyp.h
+++ b/arch/arm64/include/asm/kvm_hyp.h
@@ -20,6 +20,7 @@
#include <linux/compiler.h>
#include <linux/kvm_host.h>
+#include <asm/kvm_mmu.h>
#include <asm/sysreg.h>
#define __hyp_text __section(.hyp.text) notrace
@@ -158,6 +159,16 @@ void __noreturn __hyp_do_panic(unsigned long, ...);
/* Must be called from hyp code running at EL2 */

Marc,

static __always_inline void __hyp_text __load_guest_stage2(struct kvm *kvm)
{
+ /*
+ * Configure the VTCR translation control bits
+ * for this VM.
+ */
+ u64 vtcr = read_sysreg(vtcr_el2);
+
+ vtcr &= ~VTCR_EL2_PRIVATE_MASK;
+ vtcr |= VTCR_EL2_SL0(kvm_stage2_levels(kvm)) |
+ VTCR_EL2_T0SZ(kvm_phys_shift(kvm));
+ write_sysreg(vtcr, vtcr_el2);

Can't we generate the whole vtcr value in one go, without reading it
back? Specially given that on patch 16, you're actually switching to a
per-VM variable, and it would make a lot of sense to start with that here.

...

-u32 __hyp_text __init_stage2_translation(void)
+void __hyp_text __init_stage2_translation(void)
..


And then most of the code here could run on a per-VM basis.

There is one problem with generating the entire vtcr for a VM.
On a system with mismatched CPU features, we need to have either :

- Per CPU VTCR fixed bits
OR
- Track system wide safe VTCR bits. (Not ideal with dirty bit and access
flag updates, if and when we support them ).

So far the only fields of interest are HA & HD, which may be turned on
for CPUs that can support the feature. Rest can be filled in from the
sanitised EL1 system registers and IPA limit and the others would need
to be filled as RES0. This could potentially have some issues on
newer versions of the architecture running on older kernels.

What do you think ?

Suzuki