x86/split_lock: Rework the initialization flow of split lock detection

Current initialization flow of split lock detection has following issues:

1. It assumes the initial value of MSR_TEST_CTRL.SPLIT_LOCK_DETECT to be
   zero. However, it's possible that BIOS/firmware has set it.

2. X86_FEATURE_SPLIT_LOCK_DETECT flag is unconditionally set even if
   there is a virtualization flaw that FMS indicates the existence while
   it's actually not supported.

Rework the initialization flow to solve above issues. In detail, explicitly
clear and set split_lock_detect bit to verify MSR_TEST_CTRL can be
accessed, and rdmsr after wrmsr to ensure bit is cleared/set successfully.

X86_FEATURE_SPLIT_LOCK_DETECT flag is set only when the feature does exist
and the feature is not disabled with kernel param "split_lock_detect=off"

On each processor, explicitly updating the SPLIT_LOCK_DETECT bit based on
sld_sate in split_lock_init() since BIOS/firmware may touch it.

Originally-by: Thomas Gleixner <[email protected]>
Signed-off-by: Xiaoyao Li <[email protected]>
Signed-off-by: Thomas Gleixner <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]

arch/x86/kernel/cpu/intel.c

@@ -44,7 +44,7 @@ enum split_lock_detect_state {
  * split_lock_setup() will switch this to sld_warn on systems that support
  * split lock detect, unless there is a command line override.
  */
-static enum split_lock_detect_state sld_state = sld_off;
+static enum split_lock_detect_state sld_state __ro_after_init = sld_off;
 
 /*
  * Processors which have self-snooping capability can handle conflicting
@@ -984,78 +984,87 @@ static inline bool match_option(const char *arg, int arglen, const char *opt)
 	return len == arglen && !strncmp(arg, opt, len);
 }
 
+static bool split_lock_verify_msr(bool on)
+{
+	u64 ctrl, tmp;
+
+	if (rdmsrl_safe(MSR_TEST_CTRL, &ctrl))
+		return false;
+	if (on)
+		ctrl |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
+	else
+		ctrl &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
+	if (wrmsrl_safe(MSR_TEST_CTRL, ctrl))
+		return false;
+	rdmsrl(MSR_TEST_CTRL, tmp);
+	return ctrl == tmp;
+}
+
 static void __init split_lock_setup(void)
 {
+	enum split_lock_detect_state state = sld_warn;
 	char arg[20];
 	int i, ret;
 
-	setup_force_cpu_cap(X86_FEATURE_SPLIT_LOCK_DETECT);
-	sld_state = sld_warn;
+	if (!split_lock_verify_msr(false)) {
+		pr_info("MSR access failed: Disabled\n");
+		return;
+	}
 
 	ret = cmdline_find_option(boot_command_line, "split_lock_detect",
 				  arg, sizeof(arg));
 	if (ret >= 0) {
 		for (i = 0; i < ARRAY_SIZE(sld_options); i++) {
 			if (match_option(arg, ret, sld_options[i].option)) {
-				sld_state = sld_options[i].state;
+				state = sld_options[i].state;
 				break;
 			}
 		}
 	}
 
-	switch (sld_state) {
+	switch (state) {
 	case sld_off:
 		pr_info("disabled\n");
-		break;
-
+		return;
 	case sld_warn:
 		pr_info("warning about user-space split_locks\n");
 		break;
-
 	case sld_fatal:
 		pr_info("sending SIGBUS on user-space split_locks\n");
 		break;
 	}
+
+	if (!split_lock_verify_msr(true)) {
+		pr_info("MSR access failed: Disabled\n");
+		return;
+	}
+
+	sld_state = state;
+	setup_force_cpu_cap(X86_FEATURE_SPLIT_LOCK_DETECT);
 }
 
 /*
- * Locking is not required at the moment because only bit 29 of this
- * MSR is implemented and locking would not prevent that the operation
- * of one thread is immediately undone by the sibling thread.
- * Use the "safe" versions of rdmsr/wrmsr here because although code
- * checks CPUID and MSR bits to make sure the TEST_CTRL MSR should
- * exist, there may be glitches in virtualization that leave a guest
- * with an incorrect view of real h/w capabilities.
+ * MSR_TEST_CTRL is per core, but we treat it like a per CPU MSR. Locking
+ * is not implemented as one thread could undo the setting of the other
+ * thread immediately after dropping the lock anyway.
  */
-static bool __sld_msr_set(bool on)
+static void sld_update_msr(bool on)
 {
 	u64 test_ctrl_val;
 
-	if (rdmsrl_safe(MSR_TEST_CTRL, &test_ctrl_val))
-		return false;
+	rdmsrl(MSR_TEST_CTRL, test_ctrl_val);
 
 	if (on)
 		test_ctrl_val |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
 	else
 		test_ctrl_val &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
 
-	return !wrmsrl_safe(MSR_TEST_CTRL, test_ctrl_val);
+	wrmsrl(MSR_TEST_CTRL, test_ctrl_val);
 }
 
 static void split_lock_init(void)
 {
-	if (sld_state == sld_off)
-		return;
-
-	if (__sld_msr_set(true))
-		return;
-
-	/*
-	 * If this is anything other than the boot-cpu, you've done
-	 * funny things and you get to keep whatever pieces.
-	 */
-	pr_warn("MSR fail -- disabled\n");
-	sld_state = sld_off;
+	split_lock_verify_msr(sld_state != sld_off);
 }
 
 bool handle_user_split_lock(struct pt_regs *regs, long error_code)
@@ -1071,7 +1080,7 @@ bool handle_user_split_lock(struct pt_regs *regs, long error_code)
 	 * progress and set TIF_SLD so the detection is re-enabled via
 	 * switch_to_sld() when the task is scheduled out.
 	 */
-	__sld_msr_set(false);
+	sld_update_msr(false);
 	set_tsk_thread_flag(current, TIF_SLD);
 	return true;
 }
@@ -1085,7 +1094,7 @@ bool handle_user_split_lock(struct pt_regs *regs, long error_code)
  */
 void switch_to_sld(unsigned long tifn)
 {
-	__sld_msr_set(!(tifn & _TIF_SLD));
+	sld_update_msr(!(tifn & _TIF_SLD));
 }
 
 #define SPLIT_LOCK_CPU(model) {X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY}