x86/microcode: Improve late loading

index aa1b9a422f2b..63370651e376 100644
@@ -373,26 +373,23 @@ static int collect_cpu_info(int cpu)
 	return ret;
 }
 
-struct apply_microcode_ctx {
-	enum ucode_state err;
-};
-
 static void apply_microcode_local(void *arg)
 {
-	struct apply_microcode_ctx *ctx = arg;
+	enum ucode_state *err = arg;
 
-	ctx->err = microcode_ops->apply_microcode(smp_processor_id());
+	*err = microcode_ops->apply_microcode(smp_processor_id());
 }
 
 static int apply_microcode_on_target(int cpu)
 {
-	struct apply_microcode_ctx ctx = { .err = 0 };
+	enum ucode_state err;
 	int ret;
 
-	ret = smp_call_function_single(cpu, apply_microcode_local, &ctx, 1);
-	if (!ret)
-		ret = ctx.err;
-
+	ret = smp_call_function_single(cpu, apply_microcode_local, &err, 1);
+	if (!ret) {
+		if (err == UCODE_ERROR)
+			ret = 1;
+	}
 	return ret;
 }
 

index 923054a6b760..87bd6dc94081 100644
@@ -589,6 +589,17 @@ static int apply_microcode_early(struct ucode_cpu_info *uci, bool early)
 	if (!mc)
 		return 0;
 
+	/*
+	 * Save us the MSR write below - which is a particular expensive
+	 * operation - when the other hyperthread has updated the microcode
+	 * already.
+	 */
+	rev = intel_get_microcode_revision();
+	if (rev >= mc->hdr.rev) {
+		uci->cpu_sig.rev = rev;
+		return UCODE_OK;
+	}
+
 	/* write microcode via MSR 0x79 */
 	native_wrmsrl(MSR_IA32_UCODE_WRITE, (unsigned long)mc->bits);
 

@@ -776,7 +787,7 @@ static enum ucode_state apply_microcode_intel(int cpu)
 {
 	struct microcode_intel *mc;
 	struct ucode_cpu_info *uci;
-	struct cpuinfo_x86 *c;
+	struct cpuinfo_x86 *c = &cpu_data(cpu);
 	static int prev_rev;
 	u32 rev;
 

@@ -793,6 +804,18 @@ static enum ucode_state apply_microcode_intel(int cpu)
 			return UCODE_NFOUND;
 	}
 
+	/*
+	 * Save us the MSR write below - which is a particular expensive
+	 * operation - when the other hyperthread has updated the microcode
+	 * already.
+	 */
+	rev = intel_get_microcode_revision();
+	if (rev >= mc->hdr.rev) {
+		uci->cpu_sig.rev = rev;
+		c->microcode = rev;
+		return UCODE_OK;
+	}
+
 	/* write microcode via MSR 0x79 */
 	wrmsrl(MSR_IA32_UCODE_WRITE, (unsigned long)mc->bits);
 

@@ -813,8 +836,6 @@ static enum ucode_state apply_microcode_intel(int cpu)
 		prev_rev = rev;
 	}
 
-	c = &cpu_data(cpu);
-
 	uci->cpu_sig.rev = rev;
 	c->microcode = rev;
 

index 87bd6dc94081..e2864bc2d575 100644
@@ -600,6 +600,12 @@ static int apply_microcode_early(struct ucode_cpu_info *uci, bool early)
 		return UCODE_OK;
 	}
 
+	/*
+	 * Writeback and invalidate caches before updating microcode to avoid
+	 * internal issues depending on what the microcode is updating.
+	 */
+	native_wbinvd();
+
 	/* write microcode via MSR 0x79 */
 	native_wrmsrl(MSR_IA32_UCODE_WRITE, (unsigned long)mc->bits);
 

@@ -816,6 +822,12 @@ static enum ucode_state apply_microcode_intel(int cpu)
 		return UCODE_OK;
 	}
 
+	/*
+	 * Writeback and invalidate caches before updating microcode to avoid
+	 * internal issues depending on what the microcode is updating.
+	 */
+	native_wbinvd();
+
 	/* write microcode via MSR 0x79 */
 	wrmsrl(MSR_IA32_UCODE_WRITE, (unsigned long)mc->bits);
 

index 63370651e376..fa32cb3dcca5 100644
@@ -486,6 +486,16 @@ static void __exit microcode_dev_exit(void)
 /* fake device for request_firmware */
 static struct platform_device	*microcode_pdev;
 
+static int check_online_cpus(void)
+{
+	if (num_online_cpus() == num_present_cpus())
+		return 0;
+
+	pr_err("Not all CPUs online, aborting microcode update.\n");
+
+	return -EINVAL;
+}
+
 static enum ucode_state reload_for_cpu(int cpu)
 {
 	struct ucode_cpu_info *uci = ucode_cpu_info + cpu;

@@ -519,7 +529,13 @@ static ssize_t reload_store(struct device *dev,
 		return size;
 
 	get_online_cpus();
+
+	ret = check_online_cpus();
+	if (ret)
+		goto put;
+
 	mutex_lock(&microcode_mutex);
+
 	for_each_online_cpu(cpu) {
 		tmp_ret = reload_for_cpu(cpu);
 		if (tmp_ret > UCODE_NFOUND) {

@@ -538,6 +554,8 @@ static ssize_t reload_store(struct device *dev,
 		microcode_check();
 
 	mutex_unlock(&microcode_mutex);
+
+put:
 	put_online_cpus();
 
 	if (!ret)

index e2864bc2d575..2aded9db1d42 100644
@@ -791,9 +791,9 @@ static int collect_cpu_info(int cpu_num, struct cpu_signature *csig)
 
 static enum ucode_state apply_microcode_intel(int cpu)
 {
-	struct microcode_intel *mc;
-	struct ucode_cpu_info *uci;
+	struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
 	struct cpuinfo_x86 *c = &cpu_data(cpu);
+	struct microcode_intel *mc;
 	static int prev_rev;
 	u32 rev;
 

@@ -801,11 +801,10 @@ static enum ucode_state apply_microcode_intel(int cpu)
 	if (WARN_ON(raw_smp_processor_id() != cpu))
 		return UCODE_ERROR;
 
-	uci = ucode_cpu_info + cpu;
-	mc = uci->mc;
+	/* Look for a newer patch in our cache: */
+	mc = find_patch(uci);
 	if (!mc) {
-		/* Look for a newer patch in our cache: */
-		mc = find_patch(uci);
+		mc = uci->mc;
 		if (!mc)
 			return UCODE_NFOUND;
 	}

index fa32cb3dcca5..5dd157d48606 100644
@@ -499,15 +499,10 @@ static int check_online_cpus(void)
 static enum ucode_state reload_for_cpu(int cpu)
 {
 	struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
-	enum ucode_state ustate;
 
 	if (!uci->valid)
 		return UCODE_OK;
 
-	ustate = microcode_ops->request_microcode_fw(cpu, &microcode_pdev->dev, true);
-	if (ustate != UCODE_OK)
-		return ustate;
-
 	return apply_microcode_on_target(cpu);
 }
 

@@ -515,11 +510,11 @@ static ssize_t reload_store(struct device *dev,
 			    struct device_attribute *attr,
 			    const char *buf, size_t size)
 {
+	int cpu, bsp = boot_cpu_data.cpu_index;
 	enum ucode_state tmp_ret = UCODE_OK;
 	bool do_callback = false;
 	unsigned long val;
 	ssize_t ret = 0;
-	int cpu;
 
 	ret = kstrtoul(buf, 0, &val);
 	if (ret)

@@ -528,6 +523,10 @@ static ssize_t reload_store(struct device *dev,
 	if (val != 1)
 		return size;
 
+	tmp_ret = microcode_ops->request_microcode_fw(bsp, &microcode_pdev->dev, true);
+	if (tmp_ret != UCODE_OK)
+		return size;
+
 	get_online_cpus();
 
 	ret = check_online_cpus();

index 5dd157d48606..70ecbc8099c9 100644
@@ -22,13 +22,16 @@
 #define pr_fmt(fmt) "microcode: " fmt
 
 #include <linux/platform_device.h>
+#include <linux/stop_machine.h>
 #include <linux/syscore_ops.h>
 #include <linux/miscdevice.h>
 #include <linux/capability.h>
 #include <linux/firmware.h>
 #include <linux/kernel.h>
+#include <linux/delay.h>
 #include <linux/mutex.h>
 #include <linux/cpu.h>
+#include <linux/nmi.h>
 #include <linux/fs.h>
 #include <linux/mm.h>
 

@@ -64,6 +67,11 @@ LIST_HEAD(microcode_cache);
  */
 static DEFINE_MUTEX(microcode_mutex);
 
+/*
+ * Serialize late loading so that CPUs get updated one-by-one.
+ */
+static DEFINE_SPINLOCK(update_lock);
+
 struct ucode_cpu_info		ucode_cpu_info[NR_CPUS];
 
 struct cpu_info_ctx {

@@ -486,6 +494,19 @@ static void __exit microcode_dev_exit(void)
 /* fake device for request_firmware */
 static struct platform_device	*microcode_pdev;
 
+/*
+ * Late loading dance. Why the heavy-handed stomp_machine effort?
+ *
+ * - HT siblings must be idle and not execute other code while the other sibling
+ *   is loading microcode in order to avoid any negative interactions caused by
+ *   the loading.
+ *
+ * - In addition, microcode update on the cores must be serialized until this
+ *   requirement can be relaxed in the future. Right now, this is conservative
+ *   and good.
+ */
+#define SPINUNIT 100 /* 100 nsec */
+
 static int check_online_cpus(void)
 {
 	if (num_online_cpus() == num_present_cpus())

@@ -496,23 +517,85 @@ static int check_online_cpus(void)
 	return -EINVAL;
 }
 
-static enum ucode_state reload_for_cpu(int cpu)
+static atomic_t late_cpus;
+
+/*
+ * Returns:
+ * < 0 - on error
+ *   0 - no update done
+ *   1 - microcode was updated
+ */
+static int __reload_late(void *info)
 {
-	struct ucode_cpu_info *uci = ucode_cpu_info + cpu;
+	unsigned int timeout = NSEC_PER_SEC;
+	int all_cpus = num_online_cpus();
+	int cpu = smp_processor_id();
+	enum ucode_state err;
+	int ret = 0;
 
-	if (!uci->valid)
-		return UCODE_OK;
+	atomic_dec(&late_cpus);
+
+	/*
+	 * Wait for all CPUs to arrive. A load will not be attempted unless all
+	 * CPUs show up.
+	 * */
+	while (atomic_read(&late_cpus)) {
+		if (timeout < SPINUNIT) {
+			pr_err("Timeout while waiting for CPUs rendezvous, remaining: %d\n",
+				atomic_read(&late_cpus));
+			return -1;
+		}
+
+		ndelay(SPINUNIT);
+		timeout -= SPINUNIT;
+
+		touch_nmi_watchdog();
+	}
+
+	spin_lock(&update_lock);
+	apply_microcode_local(&err);
+	spin_unlock(&update_lock);
+
+	if (err > UCODE_NFOUND) {
+		pr_warn("Error reloading microcode on CPU %d\n", cpu);
+		ret = -1;
+	} else if (err == UCODE_UPDATED) {
+		ret = 1;
+	}
 
-	return apply_microcode_on_target(cpu);
+	atomic_inc(&late_cpus);
+
+	while (atomic_read(&late_cpus) != all_cpus)
+		cpu_relax();
+
+	return ret;
+}
+
+/*
+ * Reload microcode late on all CPUs. Wait for a sec until they
+ * all gather together.
+ */
+static int microcode_reload_late(void)
+{
+	int ret;
+
+	atomic_set(&late_cpus, num_online_cpus());
+
+	ret = stop_machine_cpuslocked(__reload_late, NULL, cpu_online_mask);
+	if (ret < 0)
+		return ret;
+	else if (ret > 0)
+		microcode_check();
+
+	return ret;
 }
 
 static ssize_t reload_store(struct device *dev,
 			    struct device_attribute *attr,
 			    const char *buf, size_t size)
 {
-	int cpu, bsp = boot_cpu_data.cpu_index;
 	enum ucode_state tmp_ret = UCODE_OK;
-	bool do_callback = false;
+	int bsp = boot_cpu_data.cpu_index;
 	unsigned long val;
 	ssize_t ret = 0;
 

@@ -534,30 +617,13 @@ static ssize_t reload_store(struct device *dev,
 		goto put;
 
 	mutex_lock(&microcode_mutex);
-
-	for_each_online_cpu(cpu) {
-		tmp_ret = reload_for_cpu(cpu);
-		if (tmp_ret > UCODE_NFOUND) {
-			pr_warn("Error reloading microcode on CPU %d\n", cpu);
-
-			/* set retval for the first encountered reload error */
-			if (!ret)
-				ret = -EINVAL;
-		}
-
-		if (tmp_ret == UCODE_UPDATED)
-			do_callback = true;
-	}
-
-	if (!ret && do_callback)
-		microcode_check();
-
+	ret = microcode_reload_late();
 	mutex_unlock(&microcode_mutex);
 
 put:
 	put_online_cpus();
 
-	if (!ret)
+	if (ret >= 0)
 		ret = size;
 
 	return ret;