x86/microcode: Handle "offline" CPUs correctly

Offline CPUs need to be parked in a safe loop when microcode update is
in progress on the primary CPU. Currently, offline CPUs are parked in
mwait_play_dead(), and for Intel CPUs, its not a safe instruction,
because the MWAIT instruction can be patched in the new microcode update
that can cause instability.

  - Add a new microcode state 'UCODE_OFFLINE' to report status on per-CPU
  basis.
  - Force NMI on the offline CPUs.

Wake up offline CPUs while the update is in progress and then return
them back to mwait_play_dead() after microcode update is complete.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20231002115903.660850472@linutronix.de

arch/x86/include/asm/microcode.h

@@ -73,6 +73,7 @@ static inline u32 intel_get_microcode_revision(void)
 #endif /* !CONFIG_CPU_SUP_INTEL */
 
 bool microcode_nmi_handler(void);
+void microcode_offline_nmi_handler(void);
 
 #ifdef CONFIG_MICROCODE_LATE_LOADING
 DECLARE_STATIC_KEY_FALSE(microcode_nmi_handler_enable);

arch/x86/kernel/cpu/microcode/core.c

@@ -272,8 +272,9 @@ struct microcode_ctrl {
 
 DEFINE_STATIC_KEY_FALSE(microcode_nmi_handler_enable);
 static DEFINE_PER_CPU(struct microcode_ctrl, ucode_ctrl);
+static atomic_t late_cpus_in, offline_in_nmi;
 static unsigned int loops_per_usec;
-static atomic_t late_cpus_in;
+static cpumask_t cpu_offline_mask;
 
 static noinstr bool wait_for_cpus(atomic_t *cnt)
 {
@@ -381,7 +382,7 @@ static noinstr void load_secondary(unsigned int cpu)
 	instrumentation_end();
 }
 
-static void load_primary(unsigned int cpu)
+static void __load_primary(unsigned int cpu)
 {
 	struct cpumask *secondaries = topology_sibling_cpumask(cpu);
 	enum sibling_ctrl ctrl;
@@ -416,6 +417,67 @@ static void load_primary(unsigned int cpu)
 	}
 }
 
+static bool kick_offline_cpus(unsigned int nr_offl)
+{
+	unsigned int cpu, timeout;
+
+	for_each_cpu(cpu, &cpu_offline_mask) {
+		/* Enable the rendezvous handler and send NMI */
+		per_cpu(ucode_ctrl.nmi_enabled, cpu) = true;
+		apic_send_nmi_to_offline_cpu(cpu);
+	}
+
+	/* Wait for them to arrive */
+	for (timeout = 0; timeout < (USEC_PER_SEC / 2); timeout++) {
+		if (atomic_read(&offline_in_nmi) == nr_offl)
+			return true;
+		udelay(1);
+	}
+	/* Let the others time out */
+	return false;
+}
+
+static void release_offline_cpus(void)
+{
+	unsigned int cpu;
+
+	for_each_cpu(cpu, &cpu_offline_mask)
+		per_cpu(ucode_ctrl.ctrl, cpu) = SCTRL_DONE;
+}
+
+static void load_primary(unsigned int cpu)
+{
+	unsigned int nr_offl = cpumask_weight(&cpu_offline_mask);
+	bool proceed = true;
+
+	/* Kick soft-offlined SMT siblings if required */
+	if (!cpu && nr_offl)
+		proceed = kick_offline_cpus(nr_offl);
+
+	/* If the soft-offlined CPUs did not respond, abort */
+	if (proceed)
+		__load_primary(cpu);
+
+	/* Unconditionally release soft-offlined SMT siblings if required */
+	if (!cpu && nr_offl)
+		release_offline_cpus();
+}
+
+/*
+ * Minimal stub rendezvous handler for soft-offlined CPUs which participate
+ * in the NMI rendezvous to protect against a concurrent NMI on affected
+ * CPUs.
+ */
+void noinstr microcode_offline_nmi_handler(void)
+{
+	if (!raw_cpu_read(ucode_ctrl.nmi_enabled))
+		return;
+	raw_cpu_write(ucode_ctrl.nmi_enabled, false);
+	raw_cpu_write(ucode_ctrl.result, UCODE_OFFLINE);
+	raw_atomic_inc(&offline_in_nmi);
+	wait_for_ctrl();
+}
+
 static noinstr bool microcode_update_handler(void)
 {
 	unsigned int cpu = raw_smp_processor_id();
@@ -472,6 +534,7 @@ static int load_cpus_stopped(void *unused)
 static int load_late_stop_cpus(void)
 {
 	unsigned int cpu, updated = 0, failed = 0, timedout = 0, siblings = 0;
+	unsigned int nr_offl, offline = 0;
 	int old_rev = boot_cpu_data.microcode;
 	struct cpuinfo_x86 prev_info;
 
@@ -479,6 +542,7 @@ static int load_late_stop_cpus(void)
 	pr_err("You should switch to early loading, if possible.\n");
 
 	atomic_set(&late_cpus_in, num_online_cpus());
+	atomic_set(&offline_in_nmi, 0);
 	loops_per_usec = loops_per_jiffy / (TICK_NSEC / 1000);
 
 	/*
@@ -501,6 +565,7 @@ static int load_late_stop_cpus(void)
 		case UCODE_UPDATED:	updated++; break;
 		case UCODE_TIMEOUT:	timedout++; break;
 		case UCODE_OK:		siblings++; break;
+		case UCODE_OFFLINE:	offline++; break;
 		default:		failed++; break;
 		}
 	}
@@ -512,6 +577,13 @@ static int load_late_stop_cpus(void)
 		/* Nothing changed. */
 		if (!failed && !timedout)
 			return 0;
+
+		nr_offl = cpumask_weight(&cpu_offline_mask);
+		if (offline < nr_offl) {
+			pr_warn("%u offline siblings did not respond.\n",
+				nr_offl - atomic_read(&offline_in_nmi));
+			return -EIO;
+		}
 		pr_err("update failed: %u CPUs failed %u CPUs timed out\n",
 		       failed, timedout);
 		return -EIO;
@@ -545,19 +617,49 @@ static int load_late_stop_cpus(void)
  *    modern CPUs uses MWAIT, which is also not guaranteed to be safe
  *    against a microcode update which affects MWAIT.
  *
- * 2) Initialize the per CPU control structure
+ *    As soft-offlined CPUs still react on NMIs, the SMT sibling
+ *    restriction can be lifted when the vendor driver signals to use NMI
+ *    for rendezvous and the APIC provides a mechanism to send an NMI to a
+ *    soft-offlined CPU. The soft-offlined CPUs are then able to
+ *    participate in the rendezvous in a trivial stub handler.
+ *
+ * 2) Initialize the per CPU control structure and create a cpumask
+ *    which contains "offline"; secondary threads, so they can be handled
+ *    correctly by a control CPU.
  */
 static bool setup_cpus(void)
 {
 	struct microcode_ctrl ctrl = { .ctrl = SCTRL_WAIT, .result = -1, };
+	bool allow_smt_offline;
 	unsigned int cpu;
 
+	allow_smt_offline = microcode_ops->nmi_safe ||
+		(microcode_ops->use_nmi && apic->nmi_to_offline_cpu);
+
+	cpumask_clear(&cpu_offline_mask);
+
 	for_each_cpu_and(cpu, cpu_present_mask, &cpus_booted_once_mask) {
+		/*
+		 * Offline CPUs sit in one of the play_dead() functions
+		 * with interrupts disabled, but they still react on NMIs
+		 * and execute arbitrary code. Also MWAIT being updated
+		 * while the offline CPU sits there is not necessarily safe
+		 * on all CPU variants.
+		 *
+		 * Mark them in the offline_cpus mask which will be handled
+		 * by CPU0 later in the update process.
+		 *
+		 * Ensure that the primary thread is online so that it is
+		 * guaranteed that all cores are updated.
+		 */
 		if (!cpu_online(cpu)) {
-			if (topology_is_primary_thread(cpu) || !microcode_ops->nmi_safe) {
-				pr_err("CPU %u not online\n", cpu);
+			if (topology_is_primary_thread(cpu) || !allow_smt_offline) {
+				pr_err("CPU %u not online, loading aborted\n", cpu);
 				return false;
 			}
+			cpumask_set_cpu(cpu, &cpu_offline_mask);
+			per_cpu(ucode_ctrl, cpu) = ctrl;
+			continue;
 		}
 
 		/*

arch/x86/kernel/cpu/microcode/internal.h

@@ -17,6 +17,7 @@ enum ucode_state {
 	UCODE_NFOUND,
 	UCODE_ERROR,
 	UCODE_TIMEOUT,
+	UCODE_OFFLINE,
 };
 
 struct microcode_ops {

arch/x86/kernel/nmi.c

@@ -502,8 +502,11 @@ DEFINE_IDTENTRY_RAW(exc_nmi)
 	if (IS_ENABLED(CONFIG_NMI_CHECK_CPU))
 		raw_atomic_long_inc(&nsp->idt_calls);
 
-	if (IS_ENABLED(CONFIG_SMP) && arch_cpu_is_offline(smp_processor_id()))
+	if (IS_ENABLED(CONFIG_SMP) && arch_cpu_is_offline(smp_processor_id())) {
+		if (microcode_nmi_handler_enabled())
+			microcode_offline_nmi_handler();
 		return;
+	}
 
 	if (this_cpu_read(nmi_state) != NMI_NOT_RUNNING) {
 		this_cpu_write(nmi_state, NMI_LATCHED);