summaryrefslogtreecommitdiff
path: root/drivers/xen
diff options
context:
space:
mode:
authorViresh Kumar <viresh.kumar@linaro.org>2023-10-16 12:41:27 +0530
committerJuergen Gross <jgross@suse.com>2023-10-16 15:18:33 +0200
commitf0d7db7b33243d2aeeff14dbdde4ccc0687ef257 (patch)
tree93d0eb8908eefb4733f56569717ae35c290eec95 /drivers/xen
parent9e90e58c11b74c2bddac4b2702cf79d36b981278 (diff)
xen: privcmd: Add support for ioeventfd
Virtio guests send VIRTIO_MMIO_QUEUE_NOTIFY notification when they need to notify the backend of an update to the status of the virtqueue. The backend or another entity, polls the MMIO address for updates to know when the notification is sent. It works well if the backend does this polling by itself. But as we move towards generic backend implementations, we end up implementing this in a separate user-space program. 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 do the polling and send an event via eventfd to the backend once the notification from guest is received. 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 address to poll and eventfd, which lets the kernel take care of polling and raise an event on the eventfd, instead of handling this in user space (which involves an extra context switch). This patch adds similar support for xen. Inspired by existing implementations for KVM, etc.. This also copies ioreq.h header file (only struct ioreq and related macros) from Xen's source tree (Top commit 5d84f07fe6bf ("xen/pci: drop remaining uses of bool_t")). Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org> Reviewed-by: Juergen Gross <jgross@suse.com> Link: https://lore.kernel.org/r/b20d83efba6453037d0c099912813c79c81f7714.1697439990.git.viresh.kumar@linaro.org Signed-off-by: Juergen Gross <jgross@suse.com>
Diffstat (limited to 'drivers/xen')
-rw-r--r--drivers/xen/Kconfig8
-rw-r--r--drivers/xen/privcmd.c405
2 files changed, 407 insertions, 6 deletions
diff --git a/drivers/xen/Kconfig b/drivers/xen/Kconfig
index d43153fec18e..d5989871dd5d 100644
--- a/drivers/xen/Kconfig
+++ b/drivers/xen/Kconfig
@@ -269,12 +269,12 @@ config XEN_PRIVCMD
disaggregated Xen setups this driver might be needed for other
domains, too.
-config XEN_PRIVCMD_IRQFD
- bool "Xen irqfd support"
+config XEN_PRIVCMD_EVENTFD
+ bool "Xen Ioeventfd and 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.
+ Using the ioeventfd / irqfd mechanism a virtio backend running in a
+ daemon can speed up interrupt delivery from / to a guest.
config XEN_ACPI_PROCESSOR
tristate "Xen ACPI processor"
diff --git a/drivers/xen/privcmd.c b/drivers/xen/privcmd.c
index da88173bac43..1ce7f3c7a950 100644
--- a/drivers/xen/privcmd.c
+++ b/drivers/xen/privcmd.c
@@ -29,15 +29,18 @@
#include <linux/seq_file.h>
#include <linux/miscdevice.h>
#include <linux/moduleparam.h>
+#include <linux/virtio_mmio.h>
#include <asm/xen/hypervisor.h>
#include <asm/xen/hypercall.h>
#include <xen/xen.h>
+#include <xen/events.h>
#include <xen/privcmd.h>
#include <xen/interface/xen.h>
#include <xen/interface/memory.h>
#include <xen/interface/hvm/dm_op.h>
+#include <xen/interface/hvm/ioreq.h>
#include <xen/features.h>
#include <xen/page.h>
#include <xen/xen-ops.h>
@@ -782,6 +785,7 @@ static long privcmd_ioctl_mmap_resource(struct file *file,
goto out;
pages = vma->vm_private_data;
+
for (i = 0; i < kdata.num; i++) {
xen_pfn_t pfn =
page_to_xen_pfn(pages[i / XEN_PFN_PER_PAGE]);
@@ -838,7 +842,7 @@ out:
return rc;
}
-#ifdef CONFIG_XEN_PRIVCMD_IRQFD
+#ifdef CONFIG_XEN_PRIVCMD_EVENTFD
/* Irqfd support */
static struct workqueue_struct *irqfd_cleanup_wq;
static DEFINE_MUTEX(irqfds_lock);
@@ -1079,6 +1083,389 @@ static void privcmd_irqfd_exit(void)
destroy_workqueue(irqfd_cleanup_wq);
}
+
+/* Ioeventfd Support */
+#define QUEUE_NOTIFY_VQ_MASK 0xFFFF
+
+static DEFINE_MUTEX(ioreq_lock);
+static LIST_HEAD(ioreq_list);
+
+/* per-eventfd structure */
+struct privcmd_kernel_ioeventfd {
+ struct eventfd_ctx *eventfd;
+ struct list_head list;
+ u64 addr;
+ unsigned int addr_len;
+ unsigned int vq;
+};
+
+/* per-guest CPU / port structure */
+struct ioreq_port {
+ int vcpu;
+ unsigned int port;
+ struct privcmd_kernel_ioreq *kioreq;
+};
+
+/* per-guest structure */
+struct privcmd_kernel_ioreq {
+ domid_t dom;
+ unsigned int vcpus;
+ u64 uioreq;
+ struct ioreq *ioreq;
+ spinlock_t lock; /* Protects ioeventfds list */
+ struct list_head ioeventfds;
+ struct list_head list;
+ struct ioreq_port ports[0];
+};
+
+static irqreturn_t ioeventfd_interrupt(int irq, void *dev_id)
+{
+ struct ioreq_port *port = dev_id;
+ struct privcmd_kernel_ioreq *kioreq = port->kioreq;
+ struct ioreq *ioreq = &kioreq->ioreq[port->vcpu];
+ struct privcmd_kernel_ioeventfd *kioeventfd;
+ unsigned int state = STATE_IOREQ_READY;
+
+ if (ioreq->state != STATE_IOREQ_READY ||
+ ioreq->type != IOREQ_TYPE_COPY || ioreq->dir != IOREQ_WRITE)
+ return IRQ_NONE;
+
+ /*
+ * We need a barrier, smp_mb(), here to ensure reads are finished before
+ * `state` is updated. Since the lock implementation ensures that
+ * appropriate barrier will be added anyway, we can avoid adding
+ * explicit barrier here.
+ *
+ * Ideally we don't need to update `state` within the locks, but we do
+ * that here to avoid adding explicit barrier.
+ */
+
+ spin_lock(&kioreq->lock);
+ ioreq->state = STATE_IOREQ_INPROCESS;
+
+ list_for_each_entry(kioeventfd, &kioreq->ioeventfds, list) {
+ if (ioreq->addr == kioeventfd->addr + VIRTIO_MMIO_QUEUE_NOTIFY &&
+ ioreq->size == kioeventfd->addr_len &&
+ (ioreq->data & QUEUE_NOTIFY_VQ_MASK) == kioeventfd->vq) {
+ eventfd_signal(kioeventfd->eventfd, 1);
+ state = STATE_IORESP_READY;
+ break;
+ }
+ }
+ spin_unlock(&kioreq->lock);
+
+ /*
+ * We need a barrier, smp_mb(), here to ensure writes are finished
+ * before `state` is updated. Since the lock implementation ensures that
+ * appropriate barrier will be added anyway, we can avoid adding
+ * explicit barrier here.
+ */
+
+ ioreq->state = state;
+
+ if (state == STATE_IORESP_READY) {
+ notify_remote_via_evtchn(port->port);
+ return IRQ_HANDLED;
+ }
+
+ return IRQ_NONE;
+}
+
+static void ioreq_free(struct privcmd_kernel_ioreq *kioreq)
+{
+ struct ioreq_port *ports = kioreq->ports;
+ int i;
+
+ lockdep_assert_held(&ioreq_lock);
+
+ list_del(&kioreq->list);
+
+ for (i = kioreq->vcpus - 1; i >= 0; i--)
+ unbind_from_irqhandler(irq_from_evtchn(ports[i].port), &ports[i]);
+
+ kfree(kioreq);
+}
+
+static
+struct privcmd_kernel_ioreq *alloc_ioreq(struct privcmd_ioeventfd *ioeventfd)
+{
+ struct privcmd_kernel_ioreq *kioreq;
+ struct mm_struct *mm = current->mm;
+ struct vm_area_struct *vma;
+ struct page **pages;
+ unsigned int *ports;
+ int ret, size, i;
+
+ lockdep_assert_held(&ioreq_lock);
+
+ size = struct_size(kioreq, ports, ioeventfd->vcpus);
+ kioreq = kzalloc(size, GFP_KERNEL);
+ if (!kioreq)
+ return ERR_PTR(-ENOMEM);
+
+ kioreq->dom = ioeventfd->dom;
+ kioreq->vcpus = ioeventfd->vcpus;
+ kioreq->uioreq = ioeventfd->ioreq;
+ spin_lock_init(&kioreq->lock);
+ INIT_LIST_HEAD(&kioreq->ioeventfds);
+
+ /* The memory for ioreq server must have been mapped earlier */
+ mmap_write_lock(mm);
+ vma = find_vma(mm, (unsigned long)ioeventfd->ioreq);
+ if (!vma) {
+ pr_err("Failed to find vma for ioreq page!\n");
+ mmap_write_unlock(mm);
+ ret = -EFAULT;
+ goto error_kfree;
+ }
+
+ pages = vma->vm_private_data;
+ kioreq->ioreq = (struct ioreq *)(page_to_virt(pages[0]));
+ mmap_write_unlock(mm);
+
+ size = sizeof(*ports) * kioreq->vcpus;
+ ports = kzalloc(size, GFP_KERNEL);
+ if (!ports) {
+ ret = -ENOMEM;
+ goto error_kfree;
+ }
+
+ if (copy_from_user(ports, u64_to_user_ptr(ioeventfd->ports), size)) {
+ ret = -EFAULT;
+ goto error_kfree_ports;
+ }
+
+ for (i = 0; i < kioreq->vcpus; i++) {
+ kioreq->ports[i].vcpu = i;
+ kioreq->ports[i].port = ports[i];
+ kioreq->ports[i].kioreq = kioreq;
+
+ ret = bind_evtchn_to_irqhandler_lateeoi(ports[i],
+ ioeventfd_interrupt, IRQF_SHARED, "ioeventfd",
+ &kioreq->ports[i]);
+ if (ret < 0)
+ goto error_unbind;
+ }
+
+ kfree(ports);
+
+ list_add_tail(&kioreq->list, &ioreq_list);
+
+ return kioreq;
+
+error_unbind:
+ while (--i >= 0)
+ unbind_from_irqhandler(irq_from_evtchn(ports[i]), &kioreq->ports[i]);
+error_kfree_ports:
+ kfree(ports);
+error_kfree:
+ kfree(kioreq);
+ return ERR_PTR(ret);
+}
+
+static struct privcmd_kernel_ioreq *
+get_ioreq(struct privcmd_ioeventfd *ioeventfd, struct eventfd_ctx *eventfd)
+{
+ struct privcmd_kernel_ioreq *kioreq;
+ unsigned long flags;
+
+ list_for_each_entry(kioreq, &ioreq_list, list) {
+ struct privcmd_kernel_ioeventfd *kioeventfd;
+
+ /*
+ * kioreq fields can be accessed here without a lock as they are
+ * never updated after being added to the ioreq_list.
+ */
+ if (kioreq->uioreq != ioeventfd->ioreq) {
+ continue;
+ } else if (kioreq->dom != ioeventfd->dom ||
+ kioreq->vcpus != ioeventfd->vcpus) {
+ pr_err("Invalid ioeventfd configuration mismatch, dom (%u vs %u), vcpus (%u vs %u)\n",
+ kioreq->dom, ioeventfd->dom, kioreq->vcpus,
+ ioeventfd->vcpus);
+ return ERR_PTR(-EINVAL);
+ }
+
+ /* Look for a duplicate eventfd for the same guest */
+ spin_lock_irqsave(&kioreq->lock, flags);
+ list_for_each_entry(kioeventfd, &kioreq->ioeventfds, list) {
+ if (eventfd == kioeventfd->eventfd) {
+ spin_unlock_irqrestore(&kioreq->lock, flags);
+ return ERR_PTR(-EBUSY);
+ }
+ }
+ spin_unlock_irqrestore(&kioreq->lock, flags);
+
+ return kioreq;
+ }
+
+ /* Matching kioreq isn't found, allocate a new one */
+ return alloc_ioreq(ioeventfd);
+}
+
+static void ioeventfd_free(struct privcmd_kernel_ioeventfd *kioeventfd)
+{
+ list_del(&kioeventfd->list);
+ eventfd_ctx_put(kioeventfd->eventfd);
+ kfree(kioeventfd);
+}
+
+static int privcmd_ioeventfd_assign(struct privcmd_ioeventfd *ioeventfd)
+{
+ struct privcmd_kernel_ioeventfd *kioeventfd;
+ struct privcmd_kernel_ioreq *kioreq;
+ unsigned long flags;
+ struct fd f;
+ int ret;
+
+ /* Check for range overflow */
+ if (ioeventfd->addr + ioeventfd->addr_len < ioeventfd->addr)
+ return -EINVAL;
+
+ /* Vhost requires us to support length 1, 2, 4, and 8 */
+ if (!(ioeventfd->addr_len == 1 || ioeventfd->addr_len == 2 ||
+ ioeventfd->addr_len == 4 || ioeventfd->addr_len == 8))
+ return -EINVAL;
+
+ /* 4096 vcpus limit enough ? */
+ if (!ioeventfd->vcpus || ioeventfd->vcpus > 4096)
+ return -EINVAL;
+
+ kioeventfd = kzalloc(sizeof(*kioeventfd), GFP_KERNEL);
+ if (!kioeventfd)
+ return -ENOMEM;
+
+ f = fdget(ioeventfd->event_fd);
+ if (!f.file) {
+ ret = -EBADF;
+ goto error_kfree;
+ }
+
+ kioeventfd->eventfd = eventfd_ctx_fileget(f.file);
+ fdput(f);
+
+ if (IS_ERR(kioeventfd->eventfd)) {
+ ret = PTR_ERR(kioeventfd->eventfd);
+ goto error_kfree;
+ }
+
+ kioeventfd->addr = ioeventfd->addr;
+ kioeventfd->addr_len = ioeventfd->addr_len;
+ kioeventfd->vq = ioeventfd->vq;
+
+ mutex_lock(&ioreq_lock);
+ kioreq = get_ioreq(ioeventfd, kioeventfd->eventfd);
+ if (IS_ERR(kioreq)) {
+ mutex_unlock(&ioreq_lock);
+ ret = PTR_ERR(kioreq);
+ goto error_eventfd;
+ }
+
+ spin_lock_irqsave(&kioreq->lock, flags);
+ list_add_tail(&kioeventfd->list, &kioreq->ioeventfds);
+ spin_unlock_irqrestore(&kioreq->lock, flags);
+
+ mutex_unlock(&ioreq_lock);
+
+ return 0;
+
+error_eventfd:
+ eventfd_ctx_put(kioeventfd->eventfd);
+
+error_kfree:
+ kfree(kioeventfd);
+ return ret;
+}
+
+static int privcmd_ioeventfd_deassign(struct privcmd_ioeventfd *ioeventfd)
+{
+ struct privcmd_kernel_ioreq *kioreq, *tkioreq;
+ struct eventfd_ctx *eventfd;
+ unsigned long flags;
+ int ret = 0;
+
+ eventfd = eventfd_ctx_fdget(ioeventfd->event_fd);
+ if (IS_ERR(eventfd))
+ return PTR_ERR(eventfd);
+
+ mutex_lock(&ioreq_lock);
+ list_for_each_entry_safe(kioreq, tkioreq, &ioreq_list, list) {
+ struct privcmd_kernel_ioeventfd *kioeventfd, *tmp;
+ /*
+ * kioreq fields can be accessed here without a lock as they are
+ * never updated after being added to the ioreq_list.
+ */
+ if (kioreq->dom != ioeventfd->dom ||
+ kioreq->uioreq != ioeventfd->ioreq ||
+ kioreq->vcpus != ioeventfd->vcpus)
+ continue;
+
+ spin_lock_irqsave(&kioreq->lock, flags);
+ list_for_each_entry_safe(kioeventfd, tmp, &kioreq->ioeventfds, list) {
+ if (eventfd == kioeventfd->eventfd) {
+ ioeventfd_free(kioeventfd);
+ spin_unlock_irqrestore(&kioreq->lock, flags);
+
+ if (list_empty(&kioreq->ioeventfds))
+ ioreq_free(kioreq);
+ goto unlock;
+ }
+ }
+ spin_unlock_irqrestore(&kioreq->lock, flags);
+ break;
+ }
+
+ pr_err("Ioeventfd isn't already assigned, dom: %u, addr: %llu\n",
+ ioeventfd->dom, ioeventfd->addr);
+ ret = -ENODEV;
+
+unlock:
+ mutex_unlock(&ioreq_lock);
+ eventfd_ctx_put(eventfd);
+
+ return ret;
+}
+
+static long privcmd_ioctl_ioeventfd(struct file *file, void __user *udata)
+{
+ struct privcmd_data *data = file->private_data;
+ struct privcmd_ioeventfd ioeventfd;
+
+ if (copy_from_user(&ioeventfd, udata, sizeof(ioeventfd)))
+ return -EFAULT;
+
+ /* No other flags should be set */
+ if (ioeventfd.flags & ~PRIVCMD_IOEVENTFD_FLAG_DEASSIGN)
+ return -EINVAL;
+
+ /* If restriction is in place, check the domid matches */
+ if (data->domid != DOMID_INVALID && data->domid != ioeventfd.dom)
+ return -EPERM;
+
+ if (ioeventfd.flags & PRIVCMD_IOEVENTFD_FLAG_DEASSIGN)
+ return privcmd_ioeventfd_deassign(&ioeventfd);
+
+ return privcmd_ioeventfd_assign(&ioeventfd);
+}
+
+static void privcmd_ioeventfd_exit(void)
+{
+ struct privcmd_kernel_ioreq *kioreq, *tmp;
+ unsigned long flags;
+
+ mutex_lock(&ioreq_lock);
+ list_for_each_entry_safe(kioreq, tmp, &ioreq_list, list) {
+ struct privcmd_kernel_ioeventfd *kioeventfd, *tmp;
+
+ spin_lock_irqsave(&kioreq->lock, flags);
+ list_for_each_entry_safe(kioeventfd, tmp, &kioreq->ioeventfds, list)
+ ioeventfd_free(kioeventfd);
+ spin_unlock_irqrestore(&kioreq->lock, flags);
+
+ ioreq_free(kioreq);
+ }
+ mutex_unlock(&ioreq_lock);
+}
#else
static inline long privcmd_ioctl_irqfd(struct file *file, void __user *udata)
{
@@ -1093,7 +1480,16 @@ static inline int privcmd_irqfd_init(void)
static inline void privcmd_irqfd_exit(void)
{
}
-#endif /* CONFIG_XEN_PRIVCMD_IRQFD */
+
+static inline long privcmd_ioctl_ioeventfd(struct file *file, void __user *udata)
+{
+ return -EOPNOTSUPP;
+}
+
+static inline void privcmd_ioeventfd_exit(void)
+{
+}
+#endif /* CONFIG_XEN_PRIVCMD_EVENTFD */
static long privcmd_ioctl(struct file *file,
unsigned int cmd, unsigned long data)
@@ -1134,6 +1530,10 @@ static long privcmd_ioctl(struct file *file,
ret = privcmd_ioctl_irqfd(file, udata);
break;
+ case IOCTL_PRIVCMD_IOEVENTFD:
+ ret = privcmd_ioctl_ioeventfd(file, udata);
+ break;
+
default:
break;
}
@@ -1278,6 +1678,7 @@ err_privcmdbuf:
static void __exit privcmd_exit(void)
{
+ privcmd_ioeventfd_exit();
privcmd_irqfd_exit();
misc_deregister(&privcmd_dev);
misc_deregister(&xen_privcmdbuf_dev);