[PATCH v3 12/19] drivers/hv: run vp ioctl and isr

@@ -106,4 +106,18 @@ Get/set vp registers. See asm/hyperv-tlfs.h for the complete set of registers.
 Includes general purpose platform registers, MSRs, and virtual registers that
 are part of Microsoft Hypervisor platform and not directly exposed to the guest.
 
+3.6 MSHV_RUN_VP
+---------------
+:Type: vp ioctl
+:Parameters: struct hv_message
+:Returns: 0 on success
+
+Run the vp, returning when it triggers an intercept, or if the calling thread
+is interrupted by a signal. In this case errno will be set to EINTR.
+
+On return, the vp will be suspended.
+This ioctl will fail on any vp that's already running (not suspended).
+
+Information about the intercept is returned in the hv_message struct.
+
 

@@ -41,6 +41,7 @@ struct ms_hyperv_info ms_hyperv;
 EXPORT_SYMBOL_GPL(ms_hyperv);
 
 #if IS_ENABLED(CONFIG_HYPERV)
+static void (*mshv_handler)(void);
 static void (*vmbus_handler)(void);
 static void (*hv_stimer0_handler)(void);
 static void (*hv_kexec_handler)(void);

@@ -51,6 +52,9 @@ DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
 	struct pt_regs *old_regs = set_irq_regs(regs);
 
 	inc_irq_stat(irq_hv_callback_count);
+	if (mshv_handler)
+		mshv_handler();
+
 	if (vmbus_handler)
 		vmbus_handler();
 

@@ -60,6 +64,18 @@ DEFINE_IDTENTRY_SYSVEC(sysvec_hyperv_callback)
 	set_irq_regs(old_regs);
 }
 
+void hv_setup_mshv_irq(void (*handler)(void))
+{
+	mshv_handler = handler;
+}
+EXPORT_SYMBOL_GPL(hv_setup_mshv_irq);
+
+void hv_remove_mshv_irq(void)
+{
+	mshv_handler = NULL;
+}
+EXPORT_SYMBOL_GPL(hv_remove_mshv_irq);
+
 void hv_setup_vmbus_handler(void (*handler)(void))
 {
 	vmbus_handler = handler;

@@ -11,11 +11,107 @@
 #include <linux/kernel.h>
 #include <linux/mm.h>
 #include <linux/io.h>
+#include <linux/random.h>
 #include <linux/mshv.h>
 #include <asm/mshyperv.h>
 
 #include "mshv.h"
 
+void mshv_isr(void)
+{
+	struct hv_message_page **msg_page =
+			this_cpu_ptr(mshv.synic_message_page);
+	struct hv_message *msg;
+	u32 message_type;
+	struct mshv_partition *partition;
+	struct mshv_vp *vp;
+	u64 partition_id;
+	u32 vp_index;
+	int i;
+	unsigned long flags;
+	struct task_struct *task;
+
+	if (unlikely(!(*msg_page))) {
+		pr_err("%s: Missing synic page!\n", __func__);
+		return;
+	}
+
+	msg = &((*msg_page)->sint_message[HV_SYNIC_INTERCEPTION_SINT_INDEX]);
+
+	/*
+	 * If the type isn't set, there isn't really a message;
+	 * it may be some other hyperv interrupt
+	 */
+	message_type = msg->header.message_type;
+	if (message_type == HVMSG_NONE)
+		return;
+
+	/* Look for the partition */
+	partition_id = msg->header.sender;
+
+	/* Hold this lock for the rest of the isr, because the partition could
+	 * be released anytime.
+	 * e.g. the MSHV_RUN_VP thread could wake on another cpu; it could
+	 * release the partition unless we hold this!
+	 */
+	spin_lock_irqsave(&mshv.partitions.lock, flags);
+
+	for (i = 0; i < MSHV_MAX_PARTITIONS; i++) {
+		partition = mshv.partitions.array[i];
+		if (partition && partition->id == partition_id)
+			break;
+	}
+
+	if (unlikely(i == MSHV_MAX_PARTITIONS)) {
+		pr_err("%s: failed to find partition\n", __func__);
+		goto unlock_out;
+	}
+
+	/*
+	 * Since we directly index the vp, and it has to exist for us to be here
+	 * (because the vp is only deleted when the partition is), no additional
+	 * locking is needed here
+	 */
+	vp_index = ((struct hv_x64_intercept_message_header *)msg->u.payload)->vp_index;
+	vp = partition->vps.array[vp_index];
+	if (unlikely(!vp)) {
+		pr_err("%s: failed to find vp\n", __func__);
+		goto unlock_out;
+	}
+
+	memcpy(vp->run.intercept_message, msg, sizeof(struct hv_message));
+
+	if (unlikely(!vp->run.task)) {
+		pr_err("%s: vp run task not set\n", __func__);
+		goto unlock_out;
+	}
+
+	/* Save the task and reset it so we can wake without racing */
+	task = vp->run.task;
+	vp->run.task = NULL;
+
+	/*
+	 * up the semaphore before waking so that we don't race with
+	 * down_trylock
+	 */
+	up(&vp->run.sem);
+
+	/*
+	 * Finally, wake the process. If it wakes the vp and generates
+	 * another intercept then the message will be queued by the hypervisor
+	 */
+	wake_up_process(task);
+
+unlock_out:
+	spin_unlock_irqrestore(&mshv.partitions.lock, flags);
+
+	/* Acknowledge message with hypervisor */
+	msg->header.message_type = HVMSG_NONE;
+	wrmsrl(HV_X64_MSR_EOM, 0);
+
+	add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
+}
+
 int mshv_synic_init(unsigned int cpu)
 {
 	union hv_synic_simp simp;

@@ -29,6 +29,7 @@
 
 extern struct mshv mshv;
 
+void mshv_isr(void);
 int mshv_synic_init(unsigned int cpu);
 int mshv_synic_cleanup(unsigned int cpu);
 

@@ -17,6 +17,7 @@
 #include <linux/mm.h>
 #include <linux/io.h>
 #include <linux/cpuhotplug.h>
+#include <linux/random.h>
 #include <linux/mshv.h>
 #include <asm/mshyperv.h>
 

@@ -64,6 +65,141 @@ static struct miscdevice mshv_dev = {
 	.mode = 0600,
 };
 
+static long
+mshv_vp_ioctl_run_vp(struct mshv_vp *vp, void __user *ret_message)
+{
+	long ret;
+	u32 msg_type;
+	struct hv_register_assoc suspend_registers[2] = {
+		{ .name = HV_REGISTER_EXPLICIT_SUSPEND },
+		{ .name = HV_REGISTER_INTERCEPT_SUSPEND }
+	};
+	/* Pointers to values for convenience */
+	union hv_explicit_suspend_register *explicit_suspend =
+				&suspend_registers[0].value.explicit_suspend;
+	union hv_intercept_suspend_register *intercept_suspend =
+				&suspend_registers[1].value.intercept_suspend;
+
+	/* Check that the VP is suspended */
+	ret = hv_call_get_vp_registers(
+			vp->index,
+			vp->partition->id,
+			2,
+			suspend_registers);
+	if (ret)
+		return ret;
+
+	if (!explicit_suspend->suspended &&
+	    !intercept_suspend->suspended) {
+		pr_err("%s: vp not suspended!\n", __func__);
+		return -EBADFD;
+	}
+
+	/*
+	 * If intercept_suspend is set, we missed a message and need to
+	 * wait for mshv_isr to complete
+	 */
+	if (intercept_suspend->suspended) {
+		if (down_interruptible(&vp->run.sem))
+			return -EINTR;
+		if (copy_to_user(ret_message, vp->run.intercept_message,
+				 sizeof(struct hv_message)))
+			return -EFAULT;
+		intercept_suspend->suspended = 0;
+		explicit_suspend->suspended = 1;
+		ret = hv_call_set_vp_registers(
+				vp->index,
+				vp->partition->id,
+				2,
+				suspend_registers);
+		if (ret) {
+			pr_err("%s: failed to set suspend bits\n", __func__);
+			return ret;
+		}
+		return 0;
+	}
+
+	/*
+	 * At this point the semaphore ensures that mshv_isr is done,
+	 * and the mutex ensures that no other threads are touching this vp
+	 */
+	vp->run.task = current;
+	set_current_state(TASK_INTERRUPTIBLE);
+
+	/* Now actually start the vp running */
+	explicit_suspend->suspended = 0;
+	intercept_suspend->suspended = 0;
+	ret = hv_call_set_vp_registers(
+			vp->index,
+			vp->partition->id,
+			2,
+			suspend_registers);
+	if (ret) {
+		pr_err("%s: failed to clear suspend bits\n", __func__);
+		set_current_state(TASK_RUNNING);
+		vp->run.task = NULL;
+		return ret;
+	}
+
+	schedule();
+
+	/* Explicitly suspend the vp to make sure it's stopped */
+	explicit_suspend->suspended = 1;
+	ret = hv_call_set_vp_registers(
+		vp->index,
+		vp->partition->id,
+		1,
+		&suspend_registers[0]);
+	if (ret) {
+		pr_err("%s: failed to set explicit suspend bit\n", __func__);
+		return -EBADFD;
+	}
+
+	/*
+	 * Check if woken up by a signal
+	 * Note that if the signal came after being woken by mshv_isr(),
+	 * we will still get the message correctly on re-entry
+	 */
+	if (signal_pending(current)) {
+		pr_debug("%s: woke up, received signal\n", __func__);
+		return -EINTR;
+	}
+
+	/*
+	 * No signal pending, so we were woken by hv_host_isr()
+	 * The isr can't be running now, and the intercept_suspend bit is set
+	 * We use it as a flag to tell if we missed a message due to a signal,
+	 * so we must clear it here and reset the semaphore
+	 */
+	intercept_suspend->suspended = 0;
+	ret = hv_call_set_vp_registers(
+		vp->index,
+		vp->partition->id,
+		1,
+		&suspend_registers[1]);
+	if (ret) {
+		pr_err("%s: failed to clear intercept suspend bit\n", __func__);
+		return -EBADFD;
+	}
+	if (down_trylock(&vp->run.sem)) {
+		pr_err("%s: semaphore in unexpected state\n", __func__);
+		return -EBADFD;
+	}
+
+	msg_type = vp->run.intercept_message->header.message_type;
+
+	if (msg_type == HVMSG_NONE) {
+		pr_err("%s: woke up, but no message\n", __func__);
+		return -ENOMSG;
+	}
+
+	if (copy_to_user(ret_message, vp->run.intercept_message,
+			 sizeof(struct hv_message)))
+		return -EFAULT;
+
+	return 0;
+}
+
 static long
 mshv_vp_ioctl_get_regs(struct mshv_vp *vp, void __user *user_args)
 {

@@ -110,6 +246,7 @@ mshv_vp_ioctl_set_regs(struct mshv_vp *vp, void __user *user_args)
 	struct mshv_vp_registers args;
 	struct hv_register_assoc *registers;
 	long ret;
+	int i;
 
 	if (copy_from_user(&args, user_args, sizeof(args)))
 		return -EFAULT;

@@ -129,6 +266,20 @@ mshv_vp_ioctl_set_regs(struct mshv_vp *vp, void __user *user_args)
 		goto free_return;
 	}
 
+	for (i = 0; i < args.count; i++) {
+		/*
+		 * Disallow setting suspend registers to ensure run vp state
+		 * is consistent
+		 */
+		if (registers[i].name == HV_REGISTER_EXPLICIT_SUSPEND ||
+		    registers[i].name == HV_REGISTER_INTERCEPT_SUSPEND) {
+			pr_err("%s: not allowed to set suspend registers\n",
+			       __func__);
+			ret = -EINVAL;
+			goto free_return;
+		}
+	}
+
 	ret = hv_call_set_vp_registers(vp->index, vp->partition->id,
 				       args.count, registers);
 

@@ -147,6 +298,9 @@ mshv_vp_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
 		return -EINTR;
 
 	switch (ioctl) {
+	case MSHV_RUN_VP:
+		r = mshv_vp_ioctl_run_vp(vp, (void __user *)arg);
+		break;
 	case MSHV_GET_VP_REGISTERS:
 		r = mshv_vp_ioctl_get_regs(vp, (void __user *)arg);
 		break;

@@ -197,12 +351,20 @@ mshv_partition_ioctl_create_vp(struct mshv_partition *partition,
 		return -ENOMEM;
 
 	mutex_init(&vp->mutex);
+	sema_init(&vp->run.sem, 0);
+
+	vp->run.intercept_message =
+		(struct hv_message *)get_zeroed_page(GFP_KERNEL);
+	if (!vp->run.intercept_message) {
+		ret = -ENOMEM;
+		goto free_vp;
+	}
 
 	vp->index = args.vp_index;
 	vp->partition = mshv_partition_get(partition);
 	if (!vp->partition) {
 		ret = -EBADF;
-		goto free_vp;
+		goto free_message;
 	}
 
 	fd = get_unused_fd_flags(O_CLOEXEC);

@@ -240,6 +402,8 @@ mshv_partition_ioctl_create_vp(struct mshv_partition *partition,
 	put_unused_fd(fd);
 put_partition:
 	mshv_partition_put(partition);
+free_message:
+	free_page((unsigned long)vp->run.intercept_message);
 free_vp:
 	kfree(vp);
 

@@ -456,6 +620,9 @@ destroy_partition(struct mshv_partition *partition)
 		mshv.partitions.array[i] = NULL;
 	}
 
+	if (!mshv.partitions.count)
+		hv_remove_mshv_irq();
+
 	spin_unlock_irqrestore(&mshv.partitions.lock, flags);
 
 	/*

@@ -474,6 +641,7 @@ destroy_partition(struct mshv_partition *partition)
 		vp = partition->vps.array[i];
 		if (!vp)
 			continue;
+		free_page((unsigned long)vp->run.intercept_message);
 		kfree(vp);
 	}
 

@@ -537,6 +705,9 @@ add_partition(struct mshv_partition *partition)
 	mshv.partitions.count++;
 	mshv.partitions.array[i] = partition;
 
+	if (mshv.partitions.count == 1)
+		hv_setup_mshv_irq(mshv_isr);
+
 out_unlock:
 	spin_unlock_irqrestore(&mshv.partitions.lock, flags);
 

@@ -143,6 +143,9 @@ void hv_remove_vmbus_handler(void);
 void hv_setup_stimer0_handler(void (*handler)(void));
 void hv_remove_stimer0_handler(void);
 
+void hv_setup_mshv_irq(void (*handler)(void));
+void hv_remove_mshv_irq(void);
+
 void hv_setup_kexec_handler(void (*handler)(void));
 void hv_remove_kexec_handler(void);
 void hv_setup_crash_handler(void (*handler)(struct pt_regs *regs));

@@ -8,6 +8,8 @@
 
 #include <linux/spinlock.h>
 #include <linux/mutex.h>
+#include <linux/semaphore.h>
+#include <linux/sched.h>
 #include <uapi/linux/mshv.h>
 
 #define MSHV_MAX_PARTITIONS		128

@@ -18,6 +20,11 @@ struct mshv_vp {
 	u32 index;
 	struct mshv_partition *partition;
 	struct mutex mutex;
+	struct {
+		struct semaphore sem;
+		struct task_struct *task;
+		struct hv_message *intercept_message;
+	} run;
 };
 
 struct mshv_mem_region {

@@ -55,5 +55,6 @@ struct mshv_vp_registers {
 /* vp device */
 #define MSHV_GET_VP_REGISTERS   _IOWR(MSHV_IOCTL, 0x05, struct mshv_vp_registers)
 #define MSHV_SET_VP_REGISTERS   _IOW(MSHV_IOCTL, 0x06, struct mshv_vp_registers)
+#define MSHV_RUN_VP		_IOR(MSHV_IOCTL, 0x07, struct hv_message)
 
 #endif