kvm tools: Add ioeventfd support
ioeventfd is way provided by KVM to receive notifications about
reads and writes to PIO and MMIO areas within the guest.
Such notifications are usefull if all we need to know is that
a specific area of the memory has been changed, and we don't need
a heavyweight exit to happen.
The implementation uses epoll to scale to large number of ioeventfds.
Benchmarks ran on a seperate (non boot) 1GB virtio-blk device, formatted
as ext4, using bonnie++.
cmd line:
# bonnie++ -d temp/ -c 2 -s 768 -u 0
Before:
Version 1.96 ------Sequential Output------ --Sequential Input- --Random-
Concurrency 2 -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
tux 768M 498 99 381127 74 269712 48 2871 99 717109 50 +++++ +++
Latency 18368us 31063us 21281us 3017us 6875us 251ms
Version 1.96 ------Sequential Create------ --------Random Create--------
tux -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++
Latency 148us 588us 2792us 1547us 1543us 218us
After:
Version 1.96 ------Sequential Output------ --Sequential Input- --Random-
Concurrency 2 -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
tux 768M 499 99 459779 73 350689 54 2997 99 860395 58 +++++ +++
Latency 17194us 14619us 26358us 4055us 7890us 44122us
Version 1.96 ------Sequential Create------ --------Random Create--------
tux -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++
Latency 135us 567us 2564us 134us 1500us 231us
Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
diff --git a/Makefile b/Makefile
index 2ebc86c..e7ceb5c 100644
--- a/Makefile
+++ b/Makefile
@@ -48,6 +48,7 @@
OBJS += rbtree.o
OBJS += util/rbtree-interval.o
OBJS += virtio/9p.o
+OBJS += ioeventfd.o
FLAGS_BFD=$(CFLAGS) -lbfd
diff --git a/include/kvm/ioeventfd.h b/include/kvm/ioeventfd.h
new file mode 100644
index 0000000..df01750
--- /dev/null
+++ b/include/kvm/ioeventfd.h
@@ -0,0 +1,27 @@
+#ifndef KVM__IOEVENTFD_H
+#define KVM__IOEVENTFD_H
+
+#include <linux/types.h>
+#include <linux/list.h>
+#include <sys/eventfd.h>
+
+struct kvm;
+
+struct ioevent {
+ u64 io_addr;
+ u8 io_len;
+ void (*fn)(struct kvm *kvm, void *ptr);
+ struct kvm *fn_kvm;
+ void *fn_ptr;
+ int fd;
+ u64 datamatch;
+
+ struct list_head list;
+};
+
+void ioeventfd__init(void);
+void ioeventfd__start(void);
+void ioeventfd__add_event(struct ioevent *ioevent);
+void ioeventfd__del_event(u64 addr, u64 datamatch);
+
+#endif
diff --git a/ioeventfd.c b/ioeventfd.c
new file mode 100644
index 0000000..3a240e4
--- /dev/null
+++ b/ioeventfd.c
@@ -0,0 +1,128 @@
+#include <sys/epoll.h>
+#include <sys/ioctl.h>
+#include <pthread.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <signal.h>
+
+#include <linux/kernel.h>
+#include <linux/kvm.h>
+#include <linux/types.h>
+
+#include "kvm/ioeventfd.h"
+#include "kvm/kvm.h"
+#include "kvm/util.h"
+
+#define IOEVENTFD_MAX_EVENTS 20
+
+static struct epoll_event events[IOEVENTFD_MAX_EVENTS];
+static int epoll_fd;
+static LIST_HEAD(used_ioevents);
+
+void ioeventfd__init(void)
+{
+ epoll_fd = epoll_create(IOEVENTFD_MAX_EVENTS);
+ if (epoll_fd < 0)
+ die("Failed creating epoll fd");
+}
+
+void ioeventfd__add_event(struct ioevent *ioevent)
+{
+ struct kvm_ioeventfd kvm_ioevent;
+ struct epoll_event epoll_event;
+ struct ioevent *new_ioevent;
+ int event;
+
+ new_ioevent = malloc(sizeof(*new_ioevent));
+ if (new_ioevent == NULL)
+ die("Failed allocating memory for new ioevent");
+
+ *new_ioevent = *ioevent;
+ event = new_ioevent->fd;
+
+ kvm_ioevent = (struct kvm_ioeventfd) {
+ .addr = ioevent->io_addr,
+ .len = ioevent->io_len,
+ .datamatch = ioevent->datamatch,
+ .fd = event,
+ .flags = KVM_IOEVENTFD_FLAG_PIO | KVM_IOEVENTFD_FLAG_DATAMATCH,
+ };
+
+ if (ioctl(ioevent->fn_kvm->vm_fd, KVM_IOEVENTFD, &kvm_ioevent) != 0)
+ die("Failed creating new ioeventfd");
+
+ epoll_event = (struct epoll_event) {
+ .events = EPOLLIN,
+ .data.ptr = new_ioevent,
+ };
+
+ if (epoll_ctl(epoll_fd, EPOLL_CTL_ADD, event, &epoll_event) != 0)
+ die("Failed assigning new event to the epoll fd");
+
+ list_add_tail(&new_ioevent->list, &used_ioevents);
+}
+
+void ioeventfd__del_event(u64 addr, u64 datamatch)
+{
+ struct kvm_ioeventfd kvm_ioevent;
+ struct ioevent *ioevent;
+ u8 found = 0;
+
+ list_for_each_entry(ioevent, &used_ioevents, list) {
+ if (ioevent->io_addr == addr) {
+ found = 1;
+ break;
+ }
+ }
+
+ if (found == 0 || ioevent == NULL)
+ return;
+
+ kvm_ioevent = (struct kvm_ioeventfd) {
+ .addr = ioevent->io_addr,
+ .len = ioevent->io_len,
+ .datamatch = ioevent->datamatch,
+ .flags = KVM_IOEVENTFD_FLAG_PIO
+ | KVM_IOEVENTFD_FLAG_DEASSIGN
+ | KVM_IOEVENTFD_FLAG_DATAMATCH,
+ };
+
+ ioctl(ioevent->fn_kvm->vm_fd, KVM_IOEVENTFD, &kvm_ioevent);
+
+ epoll_ctl(epoll_fd, EPOLL_CTL_DEL, ioevent->fd, NULL);
+
+ list_del(&ioevent->list);
+
+ close(ioevent->fd);
+ free(ioevent);
+}
+
+static void *ioeventfd__thread(void *param)
+{
+ for (;;) {
+ int nfds, i;
+
+ nfds = epoll_wait(epoll_fd, events, IOEVENTFD_MAX_EVENTS, -1);
+ for (i = 0; i < nfds; i++) {
+ u64 tmp;
+ struct ioevent *ioevent;
+
+ ioevent = events[i].data.ptr;
+
+ if (read(ioevent->fd, &tmp, sizeof(tmp)) < 0)
+ die("Failed reading event");
+
+ ioevent->fn(ioevent->fn_kvm, ioevent->fn_ptr);
+ }
+ }
+
+ return NULL;
+}
+
+void ioeventfd__start(void)
+{
+ pthread_t thread;
+
+ if (pthread_create(&thread, NULL, ioeventfd__thread, NULL) != 0)
+ die("Failed starting ioeventfd thread");
+}
diff --git a/kvm-run.c b/kvm-run.c
index f384ddd..48b8e70 100644
--- a/kvm-run.c
+++ b/kvm-run.c
@@ -29,6 +29,7 @@
#include <kvm/symbol.h>
#include <kvm/virtio-9p.h>
#include <kvm/vesa.h>
+#include <kvm/ioeventfd.h>
/* header files for gitish interface */
#include <kvm/kvm-run.h>
@@ -505,6 +506,8 @@
kvm = kvm__init(kvm_dev, ram_size);
+ ioeventfd__init();
+
max_cpus = kvm__max_cpus(kvm);
if (nrcpus > max_cpus) {
@@ -612,6 +615,7 @@
vesa__init(kvm);
thread_pool__init(nr_online_cpus);
+ ioeventfd__start();
for (i = 0; i < nrcpus; i++) {
if (pthread_create(&kvm_cpus[i]->thread, NULL, kvm_cpu_thread, kvm_cpus[i]) != 0)
