[PATCH] x86/mm: Limit INVLPGB to VA in invlpgb_flush_addr_nosync()

[PATCH] x86/mm: Limit INVLPGB to VA in invlpgb_flush_addr_nosync()

[Jann Horn]

The intent of invlpgb_flush_addr_nosync() is to flush a specific virtual
address range, but INVLPGB_FLAG_VA is not set.
If I understand AMD's documentation correctly, this means this will flush
the entire TLB (except entries for guest ASIDs).
That's safe, but seems like an unintentionally broad flush.
Fix it by setting INVLPGB_FLAG_VA.

Fixes: b7aa05cbdc52 ("x86/mm: Add INVLPGB support code")
Signed-off-by: Jann Horn <jannh@google.com>
---
I am not entirely sure about this; Rik, can you confirm if this was an
oversight, or if there's actually a reason for not passing
INVLPGB_FLAG_VA here?
I feel a bit uncomfortable touching TLB flushing and narrowing a flush
there...
---
 arch/x86/include/asm/tlb.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/include/asm/tlb.h b/arch/x86/include/asm/tlb.h
index 866ea78ba156..e257201a141a 100644
--- a/arch/x86/include/asm/tlb.h
+++ b/arch/x86/include/asm/tlb.h
@@ -153,7 +153,8 @@ static inline void invlpgb_flush_all(void)
 /* Flush addr, including globals, for all PCIDs. */
 static inline void invlpgb_flush_addr_nosync(unsigned long addr, u16 nr)
 {
-	__invlpgb(0, 0, addr, nr, PTE_STRIDE, INVLPGB_FLAG_INCLUDE_GLOBAL);
+	__invlpgb(0, 0, addr, nr, PTE_STRIDE,
+			INVLPGB_FLAG_VA | INVLPGB_FLAG_INCLUDE_GLOBAL);
 }
 
 /* Flush all mappings for all PCIDs except globals. */

---
base-commit: b1456f6dc167f7f101746e495bede2bac3d0e19f
change-id: 20250527-x86-overbroad-invlpgb-34f903c9e2b3

-- 
Jann Horn <jannh@google.com>

Re: [PATCH] x86/mm: Limit INVLPGB to VA in invlpgb_flush_addr_nosync()

[Rik van Riel]

On Tue, 2025-05-27 at 23:56 +0200, Jann Horn wrote:
> The intent of invlpgb_flush_addr_nosync() is to flush a specific
> virtual
> address range, but INVLPGB_FLAG_VA is not set.
> If I understand AMD's documentation correctly, this means this will
> flush
> the entire TLB (except entries for guest ASIDs).
> That's safe, but seems like an unintentionally broad flush.
> Fix it by setting INVLPGB_FLAG_VA.
> 
> Fixes: b7aa05cbdc52 ("x86/mm: Add INVLPGB support code")
> Signed-off-by: Jann Horn <jannh@google.com>

Reviewed-by: Rik van Riel <riel@surriel.com>

> ---
> I am not entirely sure about this; Rik, can you confirm if this was
> an
> oversight, or if there's actually a reason for not passing
> INVLPGB_FLAG_VA here?

This was totally an oversight, and probably lost
in one of many rounds of cleanups.

Good catch!

-- 
All Rights Reversed.