Skip to contentSkip to conversation

x86: add additional EPT bit definitions

PATCH In review Ayush Ranjan 2 participants 3 messages Jan 23, 2022 – Jan 25, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86: add additional EPT bit definitions

From: Michael Pratt <[email protected]>

Used in gvisor for EPT support.

Tested: Builds cleanly

---
 arch/x86/include/asm/vmx.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/include/asm/vmx.h +3 -1
index 0ffaa3156a4e..c77ad687cdf7 100644
@@ -496,7 +496,9 @@ enum vmcs_field {
 #define VMX_EPT_WRITABLE_MASK			0x2ull
 #define VMX_EPT_EXECUTABLE_MASK			0x4ull
 #define VMX_EPT_IPAT_BIT    			(1ull << 6)
-#define VMX_EPT_ACCESS_BIT			(1ull << 8)
+#define VMX_EPT_PSE_BIT				(1ull << 7)
+#define VMX_EPT_ACCESS_SHIFT			8
+#define VMX_EPT_ACCESS_BIT			(1ull << VMX_EPT_ACCESS_SHIFT)
 #define VMX_EPT_DIRTY_BIT			(1ull << 9)
 #define VMX_EPT_RWX_MASK                        (VMX_EPT_READABLE_MASK |       \
 						 VMX_EPT_WRITABLE_MASK |       \
Signature · 1 line
2.35.0.rc0.227.g00780c9af4-goog
#1 lore raw
Quoted 3 lines · Ayush Ranjan

From: Michael Pratt <[email protected]>

Used in gvisor for EPT support.

As you may have surmised from the other patch, the changelogs from patches carried in our internal kernels rarely meet the criteria for acceptance upstream. E.g. this doesn't provide sufficient justification since there's obviously no in-kernel gvisor that's consuming this.

Submitting patches that we carry internally is perfectly ok, but there needs to be sufficient justfication, and the patch needs to follow the rules laid out by Documentation/process/submitting-patches.rst.

Quoted 12 lines

Tested: Builds cleanly

---
 arch/x86/include/asm/vmx.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/include/asm/vmx.h +0 -0
index 0ffaa3156a4e..c77ad687cdf7 100644

arch/x86/include/asm/vmx.h enum vmcs_field { · line 499

Quoted 5 lines
arch/x86/include/asm/vmx.h +1 -1
@@ -496,7 +496,9 @@ enum vmcs_field {
 #define VMX_EPT_WRITABLE_MASK			0x2ull
 #define VMX_EPT_EXECUTABLE_MASK			0x4ull
 #define VMX_EPT_IPAT_BIT    			(1ull << 6)
-#define VMX_EPT_ACCESS_BIT			(1ull << 8)
+#define VMX_EPT_PSE_BIT				(1ull << 7)

I'm not a fan of "PSE", it's unnecessarily terse and "PSE" has different meaning in IA32 paging. VMX_EPT_PAGE_SIZE_BIT would be choice.

As for justification, something that has been mentioned once or thrice is the lack of build-time assertions that the PT_* bits in mmu.h that are reused for EPT entries do indeed match the EPT definitions. I can throw together a patch/series to add that and do the below cleanup.

arch/x86/include/asm/vmx.h enum vmcs_field { · line 500

Quoted 1 line
arch/x86/include/asm/vmx.h +1 -0
+#define VMX_EPT_ACCESS_SHIFT			8

I'd prefer we don't define the "shifts" for EPT (or PTE) bits, they really shouldn't be used as doing things like test_and_clear_bit() via a shift value can generate unnecessary lock instructions. arch/x86/kvm/mmu.h could use a bit of spring cleaning in this regard.

Quoted 7 lines
arch/x86/include/asm/vmx.h +1 -0
+#define VMX_EPT_ACCESS_BIT			(1ull << VMX_EPT_ACCESS_SHIFT)
 #define VMX_EPT_DIRTY_BIT			(1ull << 9)
 #define VMX_EPT_RWX_MASK                        (VMX_EPT_READABLE_MASK |       \
 						 VMX_EPT_WRITABLE_MASK |       \
Signature · 1 line
2.35.0.rc0.227.g00780c9af4-goog
#2 lore raw

Abandoning this patch.

Quoted 55 lines · Sean Christopherson
Quoted 3 lines · Ayush Ranjan

From: Michael Pratt <[email protected]>

Used in gvisor for EPT support.

As you may have surmised from the other patch, the changelogs from patches carried in our internal kernels rarely meet the criteria for acceptance upstream. E.g. this doesn't provide sufficient justification since there's obviously no in-kernel gvisor that's consuming this.

Submitting patches that we carry internally is perfectly ok, but there needs to be sufficient justfication, and the patch needs to follow the rules laid out by Documentation/process/submitting-patches.rst.

Quoted 12 lines

Tested: Builds cleanly

---
 arch/x86/include/asm/vmx.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/include/asm/vmx.h +0 -0
index 0ffaa3156a4e..c77ad687cdf7 100644

arch/x86/include/asm/vmx.h enum vmcs_field { · line 499

Quoted 5 lines
arch/x86/include/asm/vmx.h +1 -1
@@ -496,7 +496,9 @@ enum vmcs_field {
 #define VMX_EPT_WRITABLE_MASK                        0x2ull
 #define VMX_EPT_EXECUTABLE_MASK                      0x4ull
 #define VMX_EPT_IPAT_BIT                     (1ull << 6)
-#define VMX_EPT_ACCESS_BIT                   (1ull << 8)
+#define VMX_EPT_PSE_BIT                              (1ull << 7)

I'm not a fan of "PSE", it's unnecessarily terse and "PSE" has different meaning in IA32 paging. VMX_EPT_PAGE_SIZE_BIT would be choice.

As for justification, something that has been mentioned once or thrice is the lack of build-time assertions that the PT_* bits in mmu.h that are reused for EPT entries do indeed match the EPT definitions. I can throw together a patch/series to add that and do the below cleanup.

arch/x86/include/asm/vmx.h enum vmcs_field { · line 500

Quoted 1 line
arch/x86/include/asm/vmx.h +1 -0
+#define VMX_EPT_ACCESS_SHIFT                 8

I'd prefer we don't define the "shifts" for EPT (or PTE) bits, they really shouldn't be used as doing things like test_and_clear_bit() via a shift value can generate unnecessary lock instructions. arch/x86/kvm/mmu.h could use a bit of spring cleaning in this regard.

Quoted 7 lines
arch/x86/include/asm/vmx.h +1 -1
+#define VMX_EPT_ACCESS_BIT                   (1ull << VMX_EPT_ACCESS_SHIFT)
 #define VMX_EPT_DIRTY_BIT                    (1ull << 9)
 #define VMX_EPT_RWX_MASK                        (VMX_EPT_READABLE_MASK |       \
                                               VMX_EPT_WRITABLE_MASK |       \
--

2.35.0.rc0.227.g00780c9af4-goog