xen: privcmd: Add support for irqfd

Xen provides support for injecting interrupts to the guests via the
HYPERVISOR_dm_op() hypercall. The same is used by the Virtio based
device backend implementations, in an inefficient manner currently.

Generally, the Virtio backends are implemented to work with the Eventfd
based mechanism. In order to make such backends work with Xen, another
software layer needs to poll the Eventfds and raise an interrupt to the
guest using the Xen based mechanism. This results in an extra context
switch.

This is not a new problem in Linux though. It is present with other
hypervisors like KVM, etc. as well. The generic solution implemented in
the kernel for them is to provide an IOCTL call to pass the interrupt
details and eventfd, which lets the kernel take care of polling the
eventfd and raising of the interrupt, instead of handling this in user
space (which involves an extra context switch).

This patch adds support to inject a specific interrupt to guest using
the eventfd mechanism, by preventing the extra context switch.

Inspired by existing implementations for KVM, etc..

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Juergen Gross <jgross@suse.com>
Link: https://lore.kernel.org/r/8e724ac1f50c2bc1eb8da9b3ff6166f1372570aa.1692697321.git.viresh.kumar@linaro.org
Signed-off-by: Juergen Gross <jgross@suse.com>

2 files changed, 287 insertions, 2 deletions

diff --git a/drivers/xen/Kconfig b/drivers/xen/Kconfig
index d5d7c402b651..d43153fec18e 100644
--- a/drivers/xen/Kconfig
+++ b/drivers/xen/Kconfig
@@ -269,6 +269,13 @@ config XEN_PRIVCMD
 	  disaggregated Xen setups this driver might be needed for other
 	  domains, too.
 
+config XEN_PRIVCMD_IRQFD
+	bool "Xen irqfd support"
+	depends on XEN_PRIVCMD && XEN_VIRTIO && EVENTFD
+	help
+	  Using the irqfd mechanism a virtio backend running in a daemon can
+	  speed up interrupt injection into a guest.
+
 config XEN_ACPI_PROCESSOR
 	tristate "Xen ACPI processor"
 	depends on XEN && XEN_PV_DOM0 && X86 && ACPI_PROCESSOR && CPU_FREQ
diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
index f447cd37cc4c..f00ad5f5f1d4 100644
--- a/drivers/xen/privcmd.c
+++ b/drivers/xen/privcmd.c
@@ -9,11 +9,16 @@
 
 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
 
+#include <linux/eventfd.h>
+#include <linux/file.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/poll.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/string.h>
+#include <linux/workqueue.h>
 #include <linux/errno.h>
 #include <linux/mm.h>
 #include <linux/mman.h>
@@ -833,6 +838,263 @@ out:
 	return rc;
 }
 
+#ifdef CONFIG_XEN_PRIVCMD_IRQFD
+/* Irqfd support */
+static struct workqueue_struct *irqfd_cleanup_wq;
+static DEFINE_MUTEX(irqfds_lock);
+static LIST_HEAD(irqfds_list);
+
+struct privcmd_kernel_irqfd {
+	struct xen_dm_op_buf xbufs;
+	domid_t dom;
+	bool error;
+	struct eventfd_ctx *eventfd;
+	struct work_struct shutdown;
+	wait_queue_entry_t wait;
+	struct list_head list;
+	poll_table pt;
+};
+
+static void irqfd_deactivate(struct privcmd_kernel_irqfd *kirqfd)
+{
+	lockdep_assert_held(&irqfds_lock);
+
+	list_del_init(&kirqfd->list);
+	queue_work(irqfd_cleanup_wq, &kirqfd->shutdown);
+}
+
+static void irqfd_shutdown(struct work_struct *work)
+{
+	struct privcmd_kernel_irqfd *kirqfd =
+		container_of(work, struct privcmd_kernel_irqfd, shutdown);
+	u64 cnt;
+
+	eventfd_ctx_remove_wait_queue(kirqfd->eventfd, &kirqfd->wait, &cnt);
+	eventfd_ctx_put(kirqfd->eventfd);
+	kfree(kirqfd);
+}
+
+static void irqfd_inject(struct privcmd_kernel_irqfd *kirqfd)
+{
+	u64 cnt;
+	long rc;
+
+	eventfd_ctx_do_read(kirqfd->eventfd, &cnt);
+
+	xen_preemptible_hcall_begin();
+	rc = HYPERVISOR_dm_op(kirqfd->dom, 1, &kirqfd->xbufs);
+	xen_preemptible_hcall_end();
+
+	/* Don't repeat the error message for consecutive failures */
+	if (rc && !kirqfd->error) {
+		pr_err("Failed to configure irq for guest domain: %d\n",
+		       kirqfd->dom);
+	}
+
+	kirqfd->error = rc;
+}
+
+static int
+irqfd_wakeup(wait_queue_entry_t *wait, unsigned int mode, int sync, void *key)
+{
+	struct privcmd_kernel_irqfd *kirqfd =
+		container_of(wait, struct privcmd_kernel_irqfd, wait);
+	__poll_t flags = key_to_poll(key);
+
+	if (flags & EPOLLIN)
+		irqfd_inject(kirqfd);
+
+	if (flags & EPOLLHUP) {
+		mutex_lock(&irqfds_lock);
+		irqfd_deactivate(kirqfd);
+		mutex_unlock(&irqfds_lock);
+	}
+
+	return 0;
+}
+
+static void
+irqfd_poll_func(struct file *file, wait_queue_head_t *wqh, poll_table *pt)
+{
+	struct privcmd_kernel_irqfd *kirqfd =
+		container_of(pt, struct privcmd_kernel_irqfd, pt);
+
+	add_wait_queue_priority(wqh, &kirqfd->wait);
+}
+
+static int privcmd_irqfd_assign(struct privcmd_irqfd *irqfd)
+{
+	struct privcmd_kernel_irqfd *kirqfd, *tmp;
+	__poll_t events;
+	struct fd f;
+	void *dm_op;
+	int ret;
+
+	kirqfd = kzalloc(sizeof(*kirqfd) + irqfd->size, GFP_KERNEL);
+	if (!kirqfd)
+		return -ENOMEM;
+	dm_op = kirqfd + 1;
+
+	if (copy_from_user(dm_op, irqfd->dm_op, irqfd->size)) {
+		ret = -EFAULT;
+		goto error_kfree;
+	}
+
+	kirqfd->xbufs.size = irqfd->size;
+	set_xen_guest_handle(kirqfd->xbufs.h, dm_op);
+	kirqfd->dom = irqfd->dom;
+	INIT_WORK(&kirqfd->shutdown, irqfd_shutdown);
+
+	f = fdget(irqfd->fd);
+	if (!f.file) {
+		ret = -EBADF;
+		goto error_kfree;
+	}
+
+	kirqfd->eventfd = eventfd_ctx_fileget(f.file);
+	if (IS_ERR(kirqfd->eventfd)) {
+		ret = PTR_ERR(kirqfd->eventfd);
+		goto error_fd_put;
+	}
+
+	/*
+	 * Install our own custom wake-up handling so we are notified via a
+	 * callback whenever someone signals the underlying eventfd.
+	 */
+	init_waitqueue_func_entry(&kirqfd->wait, irqfd_wakeup);
+	init_poll_funcptr(&kirqfd->pt, irqfd_poll_func);
+
+	mutex_lock(&irqfds_lock);
+
+	list_for_each_entry(tmp, &irqfds_list, list) {
+		if (kirqfd->eventfd == tmp->eventfd) {
+			ret = -EBUSY;
+			mutex_unlock(&irqfds_lock);
+			goto error_eventfd;
+		}
+	}
+
+	list_add_tail(&kirqfd->list, &irqfds_list);
+	mutex_unlock(&irqfds_lock);
+
+	/*
+	 * Check if there was an event already pending on the eventfd before we
+	 * registered, and trigger it as if we didn't miss it.
+	 */
+	events = vfs_poll(f.file, &kirqfd->pt);
+	if (events & EPOLLIN)
+		irqfd_inject(kirqfd);
+
+	/*
+	 * Do not drop the file until the kirqfd is fully initialized, otherwise
+	 * we might race against the EPOLLHUP.
+	 */
+	fdput(f);
+	return 0;
+
+error_eventfd:
+	eventfd_ctx_put(kirqfd->eventfd);
+
+error_fd_put:
+	fdput(f);
+
+error_kfree:
+	kfree(kirqfd);
+	return ret;
+}
+
+static int privcmd_irqfd_deassign(struct privcmd_irqfd *irqfd)
+{
+	struct privcmd_kernel_irqfd *kirqfd;
+	struct eventfd_ctx *eventfd;
+
+	eventfd = eventfd_ctx_fdget(irqfd->fd);
+	if (IS_ERR(eventfd))
+		return PTR_ERR(eventfd);
+
+	mutex_lock(&irqfds_lock);
+
+	list_for_each_entry(kirqfd, &irqfds_list, list) {
+		if (kirqfd->eventfd == eventfd) {
+			irqfd_deactivate(kirqfd);
+			break;
+		}
+	}
+
+	mutex_unlock(&irqfds_lock);
+
+	eventfd_ctx_put(eventfd);
+
+	/*
+	 * Block until we know all outstanding shutdown jobs have completed so
+	 * that we guarantee there will not be any more interrupts once this
+	 * deassign function returns.
+	 */
+	flush_workqueue(irqfd_cleanup_wq);
+
+	return 0;
+}
+
+static long privcmd_ioctl_irqfd(struct file *file, void __user *udata)
+{
+	struct privcmd_data *data = file->private_data;
+	struct privcmd_irqfd irqfd;
+
+	if (copy_from_user(&irqfd, udata, sizeof(irqfd)))
+		return -EFAULT;
+
+	/* No other flags should be set */
+	if (irqfd.flags & ~PRIVCMD_IRQFD_FLAG_DEASSIGN)
+		return -EINVAL;
+
+	/* If restriction is in place, check the domid matches */
+	if (data->domid != DOMID_INVALID && data->domid != irqfd.dom)
+		return -EPERM;
+
+	if (irqfd.flags & PRIVCMD_IRQFD_FLAG_DEASSIGN)
+		return privcmd_irqfd_deassign(&irqfd);
+
+	return privcmd_irqfd_assign(&irqfd);
+}
+
+static int privcmd_irqfd_init(void)
+{
+	irqfd_cleanup_wq = alloc_workqueue("privcmd-irqfd-cleanup", 0, 0);
+	if (!irqfd_cleanup_wq)
+		return -ENOMEM;
+
+	return 0;
+}
+
+static void privcmd_irqfd_exit(void)
+{
+	struct privcmd_kernel_irqfd *kirqfd, *tmp;
+
+	mutex_lock(&irqfds_lock);
+
+	list_for_each_entry_safe(kirqfd, tmp, &irqfds_list, list)
+		irqfd_deactivate(kirqfd);
+
+	mutex_unlock(&irqfds_lock);
+
+	destroy_workqueue(irqfd_cleanup_wq);
+}
+#else
+static inline long privcmd_ioctl_irqfd(struct file *file, void __user *udata)
+{
+	return -EOPNOTSUPP;
+}
+
+static inline int privcmd_irqfd_init(void)
+{
+	return 0;
+}
+
+static inline void privcmd_irqfd_exit(void)
+{
+}
+#endif /* CONFIG_XEN_PRIVCMD_IRQFD */
+
 static long privcmd_ioctl(struct file *file,
 			  unsigned int cmd, unsigned long data)
 {
@@ -868,6 +1130,10 @@ static long privcmd_ioctl(struct file *file,
 		ret = privcmd_ioctl_mmap_resource(file, udata);
 		break;
 
+	case IOCTL_PRIVCMD_IRQFD:
+		ret = privcmd_ioctl_irqfd(file, udata);
+		break;
+
 	default:
 		break;
 	}
@@ -992,15 +1258,27 @@ static int __init privcmd_init(void)
 	err = misc_register(&xen_privcmdbuf_dev);
 	if (err != 0) {
 		pr_err("Could not register Xen hypercall-buf device\n");
-		misc_deregister(&privcmd_dev);
-		return err;
+		goto err_privcmdbuf;
+	}
+
+	err = privcmd_irqfd_init();
+	if (err != 0) {
+		pr_err("irqfd init failed\n");
+		goto err_irqfd;
 	}
 
 	return 0;
+
+err_irqfd:
+	misc_deregister(&xen_privcmdbuf_dev);
+err_privcmdbuf:
+	misc_deregister(&privcmd_dev);
+	return err;
 }
 
 static void __exit privcmd_exit(void)
 {
+	privcmd_irqfd_exit();
 	misc_deregister(&privcmd_dev);
 	misc_deregister(&xen_privcmdbuf_dev);
 }