Merge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux

Pull arm64 fixes from Catalin Marinas:

 - It turns out that the optimised string routines merged in 5.14 are
   not safe with in-kernel MTE (KASAN_HW_TAGS) because of reading beyond
   the end of a string (strcmp, strncmp). Such reading may go across a
   16 byte tag granule and cause a tag check fault. When KASAN_HW_TAGS
   is enabled, use the generic strcmp/strncmp C implementation.

 - An errata workaround for ThunderX relied on the CPU capabilities
   being enabled in a specific order. This disappeared with the
   automatic generation of the cpucaps.h file (sorted alphabetically).
   Fix it by checking the current CPU only rather than the system-wide
   capability.

 - Add system_supports_mte() checks on the kernel entry/exit path and
   thread switching to avoid unnecessary barriers and function calls on
   systems where MTE is not supported.

 - kselftests: skip arm64 tests if the required features are missing.

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: Restore forced disabling of KPTI on ThunderX
  kselftest/arm64: signal: Skip tests if required features are missing
  arm64: Mitigate MTE issues with str{n}cmp()
  arm64: add MTE supported check to thread switching and syscall entry/exit

arch/arm64/include/asm/assembler.h

@@ -525,6 +525,11 @@ alternative_endif
 #define EXPORT_SYMBOL_NOKASAN(name)	EXPORT_SYMBOL(name)
 #endif
 
+#ifdef CONFIG_KASAN_HW_TAGS
+#define EXPORT_SYMBOL_NOHWKASAN(name)
+#else
+#define EXPORT_SYMBOL_NOHWKASAN(name)	EXPORT_SYMBOL_NOKASAN(name)
+#endif
 	/*
 	 * Emit a 64-bit absolute little endian symbol reference in a way that
 	 * ensures that it will be resolved at build time, even when building a

arch/arm64/include/asm/mte.h

@@ -99,11 +99,17 @@ void mte_check_tfsr_el1(void);
 
 static inline void mte_check_tfsr_entry(void)
 {
+	if (!system_supports_mte())
+		return;
+
 	mte_check_tfsr_el1();
 }
 
 static inline void mte_check_tfsr_exit(void)
 {
+	if (!system_supports_mte())
+		return;
+
 	/*
 	 * The asynchronous faults are sync'ed automatically with
 	 * TFSR_EL1 on kernel entry but for exit an explicit dsb()

arch/arm64/include/asm/string.h

@@ -12,11 +12,13 @@ extern char *strrchr(const char *, int c);
 #define __HAVE_ARCH_STRCHR
 extern char *strchr(const char *, int c);
 
+#ifndef CONFIG_KASAN_HW_TAGS
 #define __HAVE_ARCH_STRCMP
 extern int strcmp(const char *, const char *);
 
 #define __HAVE_ARCH_STRNCMP
 extern int strncmp(const char *, const char *, __kernel_size_t);
+#endif
 
 #define __HAVE_ARCH_STRLEN
 extern __kernel_size_t strlen(const char *);

arch/arm64/kernel/cpufeature.c

@@ -1526,9 +1526,13 @@ static bool unmap_kernel_at_el0(const struct arm64_cpu_capabilities *entry,
 	/*
 	 * For reasons that aren't entirely clear, enabling KPTI on Cavium
 	 * ThunderX leads to apparent I-cache corruption of kernel text, which
-	 * ends as well as you might imagine. Don't even try.
+	 * ends as well as you might imagine. Don't even try. We cannot rely
+	 * on the cpus_have_*cap() helpers here to detect the CPU erratum
+	 * because cpucap detection order may change. However, since we know
+	 * affected CPUs are always in a homogeneous configuration, it is
+	 * safe to rely on this_cpu_has_cap() here.
 	 */
-	if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_27456)) {
+	if (this_cpu_has_cap(ARM64_WORKAROUND_CAVIUM_27456)) {
 		str = "ARM64_WORKAROUND_CAVIUM_27456";
 		__kpti_forced = -1;
 	}

arch/arm64/kernel/mte.c

@@ -142,12 +142,7 @@ void mte_enable_kernel_async(void)
 #ifdef CONFIG_KASAN_HW_TAGS
 void mte_check_tfsr_el1(void)
 {
-	u64 tfsr_el1;
-
-	if (!system_supports_mte())
-		return;
-
-	tfsr_el1 = read_sysreg_s(SYS_TFSR_EL1);
+	u64 tfsr_el1 = read_sysreg_s(SYS_TFSR_EL1);
 
 	if (unlikely(tfsr_el1 & SYS_TFSR_EL1_TF1)) {
 		/*
@@ -199,6 +194,9 @@ void mte_thread_init_user(void)
 
 void mte_thread_switch(struct task_struct *next)
 {
+	if (!system_supports_mte())
+		return;
+
 	mte_update_sctlr_user(next);
 
 	/*

arch/arm64/lib/strcmp.S

@@ -173,4 +173,4 @@ L(done):
 	ret
 
 SYM_FUNC_END_PI(strcmp)
-EXPORT_SYMBOL_NOKASAN(strcmp)
+EXPORT_SYMBOL_NOHWKASAN(strcmp)

arch/arm64/lib/strncmp.S

@@ -258,4 +258,4 @@ L(ret0):
 	ret
 
 SYM_FUNC_END_PI(strncmp)
-EXPORT_SYMBOL_NOKASAN(strncmp)
+EXPORT_SYMBOL_NOHWKASAN(strncmp)

tools/testing/selftests/arm64/signal/test_signals_utils.c

@@ -266,16 +266,19 @@ int test_init(struct tdescr *td)
 			td->feats_supported |= FEAT_SSBS;
 		if (getauxval(AT_HWCAP) & HWCAP_SVE)
 			td->feats_supported |= FEAT_SVE;
-		if (feats_ok(td))
+		if (feats_ok(td)) {
 			fprintf(stderr,
 				"Required Features: [%s] supported\n",
 				feats_to_string(td->feats_required &
 						td->feats_supported));
-		else
+		} else {
 			fprintf(stderr,
 				"Required Features: [%s] NOT supported\n",
 				feats_to_string(td->feats_required &
 						~td->feats_supported));
+			td->result = KSFT_SKIP;
+			return 0;
+		}
 	}
 
 	/* Perform test specific additional initialization */