From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Wed, 3 May 2023 16:38:58 +0000 (-0700)
Subject: x86: mm: remove 'sign' games from LAM untagged_addr*() macros
X-Git-Tag: v6.6.17~4911^2~1
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=1dbc0a9515fdf1f0b9d6c9b1954a347c94e5f5f9;p=platform%2Fkernel%2Flinux-rpi.git

x86: mm: remove 'sign' games from LAM untagged_addr*() macros

The intent of the sign games was to not modify kernel addresses when
untagging them.  However, that had two issues:

 (a) it didn't actually work as intended, since the mask was calculated
     as 'addr >> 63' on an _unsigned_ address. So instead of getting a
     mask of all ones for kernel addresses, you just got '1'.

 (b) untagging a kernel address isn't actually a valid operation anyway.

Now, (a) had originally been true for both 'untagged_addr()' and the
remote version of it, but had accidentally been fixed for the regular
version of untagged_addr() by commit e0bddc19ba95 ("x86/mm: Reduce
untagged_addr() overhead for systems without LAM").  That one rewrote
the shift to be part of the alternative asm code, and in the process
changed the unsigned shift into a signed 'sar' instruction.

And while it is true that we don't want to turn what looks like a kernel
address into a user address by masking off the high bit, that doesn't
need these sign masking games - all it needs is that the mm context
'untag_mask' value has the high bit set.

Which it always does.

So simplify the code by just removing the superfluous (and in the case
of untagged_addr_remote(), still buggy) sign bit games in the address
masking.

Acked-by: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
---

diff --git a/arch/x86/include/asm/uaccess_64.h b/arch/x86/include/asm/uaccess_64.h
index 20411e6..e5b23e9 100644
--- a/arch/x86/include/asm/uaccess_64.h
+++ b/arch/x86/include/asm/uaccess_64.h
@@ -15,25 +15,17 @@
 #ifdef CONFIG_ADDRESS_MASKING
 /*
  * Mask out tag bits from the address.
- *
- * Magic with the 'sign' allows to untag userspace pointer without any branches
- * while leaving kernel addresses intact.
  */
 static inline unsigned long __untagged_addr(unsigned long addr)
 {
-	long sign;
-
 	/*
 	 * Refer tlbstate_untag_mask directly to avoid RIP-relative relocation
 	 * in alternative instructions. The relocation gets wrong when gets
 	 * copied to the target place.
 	 */
 	asm (ALTERNATIVE("",
-			 "sar $63, %[sign]\n\t" /* user_ptr ? 0 : -1UL */
-			 "or %%gs:tlbstate_untag_mask, %[sign]\n\t"
-			 "and %[sign], %[addr]\n\t", X86_FEATURE_LAM)
-	     : [addr] "+r" (addr), [sign] "=r" (sign)
-	     : "m" (tlbstate_untag_mask), "[sign]" (addr));
+			 "and %%gs:tlbstate_untag_mask, %[addr]\n\t", X86_FEATURE_LAM)
+	     : [addr] "+r" (addr) : "m" (tlbstate_untag_mask));
 
 	return addr;
 }
@@ -46,12 +38,8 @@ static inline unsigned long __untagged_addr(unsigned long addr)
 static inline unsigned long __untagged_addr_remote(struct mm_struct *mm,
 						   unsigned long addr)
 {
-	long sign = addr >> 63;
-
 	mmap_assert_locked(mm);
-	addr &= (mm)->context.untag_mask | sign;
-
-	return addr;
+	return addr & (mm)->context.untag_mask;
 }
 
 #define untagged_addr_remote(mm, addr)	({				\