Merge tag 'hardening-v6.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux

Pull hardening updates from Kees Cook:

 - lkdtm/bugs: add test for hung smp_call_function_single() (Mark
   Rutland)

 - gcc-plugins: Remove duplicate included header file stringpool.h
   (Thorsten Blum)

 - ARM: Remove address checking for MMUless devices (Yanjun Yang)

 - randomize_kstack: Clean up per-arch entropy and codegen

 - KCFI: Make FineIBT mode Kconfig selectable

 - fortify: Do not special-case 0-sized destinations

* tag 'hardening-v6.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux:
  randomize_kstack: Improve stack alignment codegen
  ARM: Remove address checking for MMUless devices
  gcc-plugins: Remove duplicate included header file stringpool.h
  randomize_kstack: Remove non-functional per-arch entropy filtering
  fortify: Do not special-case 0-sized destinations
  x86/alternatives: Make FineIBT mode Kconfig selectable
  lkdtm/bugs: add test for hung smp_call_function_single()

2 files changed, 14 insertions, 12 deletions

diff --git a/include/linux/fortify-string.h b/include/linux/fortify-string.h
index 7e0f340bf363..0d99bf11d260 100644
--- a/include/linux/fortify-string.h
+++ b/include/linux/fortify-string.h
@@ -601,11 +601,7 @@ __FORTIFY_INLINE bool fortify_memcpy_chk(__kernel_size_t size,
 	/*
 	 * Warn when writing beyond destination field size.
 	 *
-	 * We must ignore p_size_field == 0 for existing 0-element
-	 * fake flexible arrays, until they are all converted to
-	 * proper flexible arrays.
-	 *
-	 * The implementation of __builtin_*object_size() behaves
+	 * Note the implementation of __builtin_*object_size() behaves
 	 * like sizeof() when not directly referencing a flexible
 	 * array member, which means there will be many bounds checks
 	 * that will appear at run-time, without a way for them to be
@@ -613,7 +609,7 @@ __FORTIFY_INLINE bool fortify_memcpy_chk(__kernel_size_t size,
 	 * is specifically the flexible array member).
 	 * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101832
 	 */
-	if (p_size_field != 0 && p_size_field != SIZE_MAX &&
+	if (p_size_field != SIZE_MAX &&
 	    p_size != p_size_field && p_size_field < size)
 		return true;
 
diff --git a/include/linux/randomize_kstack.h b/include/linux/randomize_kstack.h
index 6d92b68efbf6..1d982dbdd0d0 100644
--- a/include/linux/randomize_kstack.h
+++ b/include/linux/randomize_kstack.h
@@ -32,13 +32,19 @@ DECLARE_PER_CPU(u32, kstack_offset);
 #endif
 
 /*
- * Use, at most, 10 bits of entropy. We explicitly cap this to keep the
- * "VLA" from being unbounded (see above). 10 bits leaves enough room for
- * per-arch offset masks to reduce entropy (by removing higher bits, since
- * high entropy may overly constrain usable stack space), and for
- * compiler/arch-specific stack alignment to remove the lower bits.
+ * Use, at most, 6 bits of entropy (on 64-bit; 8 on 32-bit). This cap is
+ * to keep the "VLA" from being unbounded (see above). Additionally clear
+ * the bottom 4 bits (on 64-bit systems, 2 for 32-bit), since stack
+ * alignment will always be at least word size. This makes the compiler
+ * code gen better when it is applying the actual per-arch alignment to
+ * the final offset. The resulting randomness is reasonable without overly
+ * constraining usable stack space.
  */
-#define KSTACK_OFFSET_MAX(x)	((x) & 0x3FF)
+#ifdef CONFIG_64BIT
+#define KSTACK_OFFSET_MAX(x)	((x) & 0b1111110000)
+#else
+#define KSTACK_OFFSET_MAX(x)	((x) & 0b1111111100)
+#endif
 
 /**
  * add_random_kstack_offset - Increase stack utilization by previously