Skip to contentSkip to conversation

Fix SEV user-space mapping of unencrypted coherent memory

PATCH Superseded Thomas Hellström (VMware) 2 participants 5 messages Sep 10, 2019 – Sep 11, 2019

Subsystems (2)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

cover Fix SEV user-space mapping of unencrypted coherent memory

With SEV and sometimes with SME encryption, The dma api coherent memory is typically unencrypted, meaning the linear kernel map has the encryption bit cleared. However, default page protection returned from vm_get_page_prot() has the encryption bit set. So to compute the correct page protection we need to clear the encryption bit.

Also, in order for the encryption bit setting to survive across do_mmap() and mprotect_fixup(), We need to make pgprot_modify() aware of it and not touch it. Therefore make sme_me_mask part of _PAGE_CHG_MASK and make sure pgprot_modify() preserves also cleared bits that are part of _PAGE_CHG_MASK, not just set bits. The use of pgprot_modify() is currently quite limited and easy to audit.

(Note that the encryption status is not logically encoded in the pfn but in the page protection even if an address line in the physical address is used).

The patchset has seen some sanity testing by exporting dma_pgprot() and using it in the vmwgfx mmap handler with SEV enabled.

Changes since:
RFC:
- Make sme_me_mask port of _PAGE_CHG_MASK rather than using it by its own in pgprot_modify().

Cc Dave Hansen <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Christian König <[email protected]>
Cc Marek Szyprowski <[email protected]>
Cc Tom Lendacky <[email protected]>
#1 lore raw

PATCH 1/2 x86: Don't let pgprot_modify() change the page encryption bit

From: Thomas Hellstrom <[email protected]>

When SEV or SME is enabled and active, vm_get_page_prot() typically returns with the encryption bit set. This means that users of pgprot_modify(, vm_get_page_prot()) (mprotect_fixup, do_mmap) typically unintentionally sets encrypted page protection even on mmap'd coherent memory where the mmap callback has cleared the bit. Fix this by not allowing pgprot_modify() to change the encryption bit, similar to how it's done for PAT bits.

Cc Dave Hansen <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Christian König <[email protected]>
Cc Marek Szyprowski <[email protected]>
Cc Tom Lendacky <[email protected]>
---
 arch/x86/include/asm/pgtable.h       | 7 +++++--
 arch/x86/include/asm/pgtable_types.h | 2 +-
 2 files changed, 6 insertions(+), 3 deletions(-)
arch/x86/include/asm/pgtable.h +5 -2
index 0bc530c4eb13..1e6bb4c25334 100644
@@ -624,12 +624,15 @@ static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
 	return __pmd(val);
 }
 
-/* mprotect needs to preserve PAT bits when updating vm_page_prot */
+/*
+ * mprotect needs to preserve PAT and encryption bits when updating
+ * vm_page_prot
+ */
 #define pgprot_modify pgprot_modify
 static inline pgprot_t pgprot_modify(pgprot_t oldprot, pgprot_t newprot)
 {
 	pgprotval_t preservebits = pgprot_val(oldprot) & _PAGE_CHG_MASK;
-	pgprotval_t addbits = pgprot_val(newprot);
+	pgprotval_t addbits = pgprot_val(newprot) & ~_PAGE_CHG_MASK;
 	return __pgprot(preservebits | addbits);
 }
 
arch/x86/include/asm/pgtable_types.h +1 -1
index b5e49e6bac63..e13084b3d6cb 100644
@@ -123,7 +123,7 @@
  */
 #define _PAGE_CHG_MASK	(PTE_PFN_MASK | _PAGE_PCD | _PAGE_PWT |		\
 			 _PAGE_SPECIAL | _PAGE_ACCESSED | _PAGE_DIRTY |	\
-			 _PAGE_SOFT_DIRTY | _PAGE_DEVMAP)
+			 _PAGE_SOFT_DIRTY | _PAGE_DEVMAP | sme_me_mask)
 #define _HPAGE_CHG_MASK (_PAGE_CHG_MASK | _PAGE_PSE)
 
 /*
Signature · 1 line
2.20.1
#2 lore raw

PATCH 2/2 dma-mapping: Fix dma_pgprot() for unencrypted coherent pages

In reply to Thomas Hellström (VMware)

From: Thomas Hellstrom <[email protected]>

When force_dma_unencrypted() returns true, the linear kernel map of the coherent pages have had the encryption bit explicitly cleared and the page content is unencrypted. Make sure that any additional PTEs we set up to these pages also have the encryption bit cleared by having dma_pgprot() return a protection with the encryption bit cleared in this case.

Cc Dave Hansen <[email protected]>
Cc Andy Lutomirski <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Christoph Hellwig <[email protected]>
Cc Christian König <[email protected]>
Cc Marek Szyprowski <[email protected]>
Cc Tom Lendacky <[email protected]>
---
 kernel/dma/mapping.c | 2 ++
 1 file changed, 2 insertions(+)
kernel/dma/mapping.c +2 -0
index b0038ca3aa92..2b499dcae74f 100644
@@ -157,6 +157,8 @@ EXPORT_SYMBOL(dma_get_sgtable_attrs);
  */
 pgprot_t dma_pgprot(struct device *dev, pgprot_t prot, unsigned long attrs)
 {
+	if (force_dma_unencrypted(dev))
+		prot = pgprot_decrypted(prot);
 	if (dev_is_dma_coherent(dev) ||
 	    (IS_ENABLED(CONFIG_DMA_NONCOHERENT_CACHE_SYNC) &&
              (attrs & DMA_ATTR_NON_CONSISTENT)))
Signature · 1 line
2.20.1
#3 lore raw
In reply to Thomas Hellström (VMware)
Quoted 16 lines · Thomas Hellström (VMware)

With SEV and sometimes with SME encryption, The dma api coherent memory is typically unencrypted, meaning the linear kernel map has the encryption bit cleared. However, default page protection returned from vm_get_page_prot() has the encryption bit set. So to compute the correct page protection we need to clear the encryption bit.

Also, in order for the encryption bit setting to survive across do_mmap() and mprotect_fixup(), We need to make pgprot_modify() aware of it and not touch it. Therefore make sme_me_mask part of _PAGE_CHG_MASK and make sure pgprot_modify() preserves also cleared bits that are part of _PAGE_CHG_MASK, not just set bits. The use of pgprot_modify() is currently quite limited and easy to audit.

(Note that the encryption status is not logically encoded in the pfn but in the page protection even if an address line in the physical address is used).

Quoted 7 lines

The patchset has seen some sanity testing by exporting dma_pgprot() and using it in the vmwgfx mmap handler with SEV enabled.

Changes since:
RFC:
- Make sme_me_mask port of _PAGE_CHG_MASK rather than using it by its own in pgprot_modify().

Could you please add a "why is this patch-set needed", not just describe the "what does this patch set do"? I've seen zero discussion in the three changelogs of exactly why we'd want this, which drivers and features are affected and in what way, etc.

It's called a "fix" but doesn't explain what bad behavior it fixes.

Thanks,

Ingo

#4 lore raw
Quoted 30 lines · Ingo Molnar
Quoted 16 lines · Thomas Hellström (VMware)

With SEV and sometimes with SME encryption, The dma api coherent memory is typically unencrypted, meaning the linear kernel map has the encryption bit cleared. However, default page protection returned from vm_get_page_prot() has the encryption bit set. So to compute the correct page protection we need to clear the encryption bit.

Also, in order for the encryption bit setting to survive across do_mmap() and mprotect_fixup(), We need to make pgprot_modify() aware of it and not touch it. Therefore make sme_me_mask part of _PAGE_CHG_MASK and make sure pgprot_modify() preserves also cleared bits that are part of _PAGE_CHG_MASK, not just set bits. The use of pgprot_modify() is currently quite limited and easy to audit.

(Note that the encryption status is not logically encoded in the pfn but in the page protection even if an address line in the physical address is used).

Quoted 7 lines

The patchset has seen some sanity testing by exporting dma_pgprot() and using it in the vmwgfx mmap handler with SEV enabled.

Changes since:
RFC:
- Make sme_me_mask port of _PAGE_CHG_MASK rather than using it by its own in pgprot_modify().

Could you please add a "why is this patch-set needed", not just describe the "what does this patch set do"? I've seen zero discussion in the three changelogs of exactly why we'd want this, which drivers and features are affected and in what way, etc.

Quoted 5 lines

It's called a "fix" but doesn't explain what bad behavior it fixes.

Thanks,

Ingo

I'll update the changelog to be more clear about that.

Thanks,

Thomas