anolis: uio: Replace mutex info_lock with percpu_ref to improve performance
ANBZ: #3289
The mutex info_lock was introduced to fix crash after the device is
unregistered in
commit 57c5f4df0a ("uio: fix crash after the device is unregistered"),
we can replace it with more powerful percpu-ref to improve performance.
Use tcm_loop and tcmu(backstore is file) to evaluate performance,
fio job: fio -filename=/dev/sdb -ioengine=libaio -direct=1 -size=2G
-name=1 -thread -runtime=60 -time_based -rw=randread -numjobs=16
-iodepth=16 -bs=128k
Without this patch:
READ: bw=2965MiB/s (3110MB/s), 183MiB/s-188MiB/s (192MB/s-197MB/s),
io=174GiB (187GB), run=60005-60007msec
With this patch:
READ: bw=5823MiB/s (6106MB/s), 338MiB/s-379MiB/s (354MB/s-397MB/s),
io=341GiB (366GB), run=60002-60005msec
There is about a 100% performance improvement in this case.
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Reviewed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Link: https://gitee.com/anolis/cloud-kernel/pulls/1255
This commit is contained in:
Guixin Liu
小龙
parent
af0054ed40
commit
ed32aa5dc7
|
|
@ -12,6 +12,8 @@
|
|||
* Base Functions
|
||||
*/
|
||||
|
||||
#include <linux/completion.h>
|
||||
#include <linux/percpu-refcount.h>
|
||||
#include <linux/module.h>
|
||||
#include <linux/init.h>
|
||||
#include <linux/poll.h>
|
||||
|
|
@ -216,7 +218,9 @@ static ssize_t name_show(struct device *dev,
|
|||
struct uio_device *idev = dev_get_drvdata(dev);
|
||||
int ret;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return -EINVAL;
|
||||
|
||||
if (!idev->info) {
|
||||
ret = -EINVAL;
|
||||
dev_err(dev, "the device has been unregistered\n");
|
||||
|
|
@ -226,7 +230,7 @@ static ssize_t name_show(struct device *dev,
|
|||
ret = sprintf(buf, "%s\n", idev->info->name);
|
||||
|
||||
out:
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
return ret;
|
||||
}
|
||||
static DEVICE_ATTR_RO(name);
|
||||
|
|
@ -237,7 +241,9 @@ static ssize_t version_show(struct device *dev,
|
|||
struct uio_device *idev = dev_get_drvdata(dev);
|
||||
int ret;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return -EINVAL;
|
||||
|
||||
if (!idev->info) {
|
||||
ret = -EINVAL;
|
||||
dev_err(dev, "the device has been unregistered\n");
|
||||
|
|
@ -247,7 +253,7 @@ static ssize_t version_show(struct device *dev,
|
|||
ret = sprintf(buf, "%s\n", idev->info->version);
|
||||
|
||||
out:
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
return ret;
|
||||
}
|
||||
static DEVICE_ATTR_RO(version);
|
||||
|
|
@ -487,16 +493,20 @@ static int uio_open(struct inode *inode, struct file *filep)
|
|||
listener->event_count = atomic_read(&idev->event);
|
||||
filep->private_data = listener;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref)) {
|
||||
ret = -EINVAL;
|
||||
goto err_infoopen;
|
||||
}
|
||||
|
||||
if (!idev->info) {
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
ret = -EINVAL;
|
||||
goto err_infoopen;
|
||||
}
|
||||
|
||||
if (idev->info->open)
|
||||
ret = idev->info->open(idev->info, inode);
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
if (ret)
|
||||
goto err_infoopen;
|
||||
|
||||
|
|
@ -529,10 +539,12 @@ static int uio_release(struct inode *inode, struct file *filep)
|
|||
struct uio_listener *listener = filep->private_data;
|
||||
struct uio_device *idev = listener->dev;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return -EINVAL;
|
||||
|
||||
if (idev->info && idev->info->release)
|
||||
ret = idev->info->release(idev->info, inode);
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
|
||||
module_put(idev->owner);
|
||||
kfree(listener);
|
||||
|
|
@ -546,10 +558,12 @@ static __poll_t uio_poll(struct file *filep, poll_table *wait)
|
|||
struct uio_device *idev = listener->dev;
|
||||
__poll_t ret = 0;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return -EINVAL;
|
||||
|
||||
if (!idev->info || !idev->info->irq)
|
||||
ret = -EIO;
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
|
||||
if (ret)
|
||||
return ret;
|
||||
|
|
@ -575,13 +589,17 @@ static ssize_t uio_read(struct file *filep, char __user *buf,
|
|||
add_wait_queue(&idev->wait, &wait);
|
||||
|
||||
do {
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!idev->info || !idev->info->irq) {
|
||||
retval = -EIO;
|
||||
mutex_unlock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref)) {
|
||||
retval = -EINVAL;
|
||||
break;
|
||||
}
|
||||
mutex_unlock(&idev->info_lock);
|
||||
|
||||
if (!idev->info || !idev->info->irq) {
|
||||
retval = -EIO;
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
break;
|
||||
}
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
|
||||
set_current_state(TASK_INTERRUPTIBLE);
|
||||
|
||||
|
|
@ -629,7 +647,9 @@ static ssize_t uio_write(struct file *filep, const char __user *buf,
|
|||
if (copy_from_user(&irq_on, buf, count))
|
||||
return -EFAULT;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return -EINVAL;
|
||||
|
||||
if (!idev->info) {
|
||||
retval = -EINVAL;
|
||||
goto out;
|
||||
|
|
@ -648,7 +668,7 @@ static ssize_t uio_write(struct file *filep, const char __user *buf,
|
|||
retval = idev->info->irqcontrol(idev->info, irq_on);
|
||||
|
||||
out:
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
return retval ? retval : sizeof(s32);
|
||||
}
|
||||
|
||||
|
|
@ -673,7 +693,9 @@ static vm_fault_t uio_vma_fault(struct vm_fault *vmf)
|
|||
vm_fault_t ret = 0;
|
||||
int mi;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return VM_FAULT_SIGBUS;
|
||||
|
||||
if (!idev->info) {
|
||||
ret = VM_FAULT_SIGBUS;
|
||||
goto out;
|
||||
|
|
@ -700,8 +722,7 @@ static vm_fault_t uio_vma_fault(struct vm_fault *vmf)
|
|||
vmf->page = page;
|
||||
|
||||
out:
|
||||
mutex_unlock(&idev->info_lock);
|
||||
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
return ret;
|
||||
}
|
||||
|
||||
|
|
@ -770,7 +791,9 @@ static int uio_mmap(struct file *filep, struct vm_area_struct *vma)
|
|||
|
||||
vma->vm_private_data = idev;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return -EINVAL;
|
||||
|
||||
if (!idev->info) {
|
||||
ret = -EINVAL;
|
||||
goto out;
|
||||
|
|
@ -809,7 +832,7 @@ static int uio_mmap(struct file *filep, struct vm_area_struct *vma)
|
|||
}
|
||||
|
||||
out:
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
return ret;
|
||||
}
|
||||
|
||||
|
|
@ -819,7 +842,8 @@ static long uio_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
|
|||
struct uio_device *idev = listener->dev;
|
||||
long retval = 0;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
if (!percpu_ref_tryget_live(&idev->info_ref))
|
||||
return -EINVAL;
|
||||
|
||||
if (!idev->info || !idev->info->ioctl) {
|
||||
retval = -EINVAL;
|
||||
|
|
@ -828,7 +852,7 @@ static long uio_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
|
|||
|
||||
retval = idev->info->ioctl(idev->info, cmd, arg);
|
||||
out:
|
||||
mutex_unlock(&idev->info_lock);
|
||||
percpu_ref_put(&idev->info_ref);
|
||||
return retval;
|
||||
}
|
||||
|
||||
|
|
@ -926,6 +950,13 @@ static void uio_device_release(struct device *dev)
|
|||
kfree(idev);
|
||||
}
|
||||
|
||||
static void uio_info_free(struct percpu_ref *ref)
|
||||
{
|
||||
struct uio_device *idev = container_of(ref, struct uio_device, info_ref);
|
||||
|
||||
complete(&idev->free_done);
|
||||
}
|
||||
|
||||
/**
|
||||
* uio_register_device - register a new userspace IO device
|
||||
* @owner: module that creates the new device
|
||||
|
|
@ -956,10 +987,18 @@ int __uio_register_device(struct module *owner,
|
|||
|
||||
idev->owner = owner;
|
||||
idev->info = info;
|
||||
mutex_init(&idev->info_lock);
|
||||
init_waitqueue_head(&idev->wait);
|
||||
atomic_set(&idev->event, 0);
|
||||
|
||||
ret = percpu_ref_init(&idev->info_ref, uio_info_free, 0, GFP_KERNEL);
|
||||
if (ret) {
|
||||
pr_err("percpu_ref init failed!\n");
|
||||
kfree(idev);
|
||||
return ret;
|
||||
}
|
||||
init_completion(&idev->confirm_done);
|
||||
init_completion(&idev->free_done);
|
||||
|
||||
ret = uio_get_minor(idev);
|
||||
if (ret) {
|
||||
kfree(idev);
|
||||
|
|
@ -1055,6 +1094,13 @@ int __devm_uio_register_device(struct module *owner,
|
|||
}
|
||||
EXPORT_SYMBOL_GPL(__devm_uio_register_device);
|
||||
|
||||
static void uio_confirm_info(struct percpu_ref *ref)
|
||||
{
|
||||
struct uio_device *idev = container_of(ref, struct uio_device, info_ref);
|
||||
|
||||
complete(&idev->confirm_done);
|
||||
}
|
||||
|
||||
/**
|
||||
* uio_unregister_device - unregister a industrial IO device
|
||||
* @info: UIO device capabilities
|
||||
|
|
@ -1071,14 +1117,16 @@ void uio_unregister_device(struct uio_info *info)
|
|||
idev = info->uio_dev;
|
||||
minor = idev->minor;
|
||||
|
||||
mutex_lock(&idev->info_lock);
|
||||
percpu_ref_kill_and_confirm(&idev->info_ref, uio_confirm_info);
|
||||
wait_for_completion(&idev->confirm_done);
|
||||
wait_for_completion(&idev->free_done);
|
||||
/* now, we can set info to NULL */
|
||||
uio_dev_del_attributes(idev);
|
||||
|
||||
if (info->irq && info->irq != UIO_IRQ_CUSTOM)
|
||||
free_irq(info->irq, idev);
|
||||
|
||||
idev->info = NULL;
|
||||
mutex_unlock(&idev->info_lock);
|
||||
|
||||
wake_up_interruptible(&idev->wait);
|
||||
kill_fasync(&idev->async_queue, SIGIO, POLL_HUP);
|
||||
|
|
|
|||
|
|
@ -16,6 +16,7 @@
|
|||
#include <linux/device.h>
|
||||
#include <linux/fs.h>
|
||||
#include <linux/interrupt.h>
|
||||
#include <linux/percpu-refcount.h>
|
||||
|
||||
struct module;
|
||||
struct uio_map;
|
||||
|
|
@ -74,9 +75,11 @@ struct uio_device {
|
|||
struct fasync_struct *async_queue;
|
||||
wait_queue_head_t wait;
|
||||
struct uio_info *info;
|
||||
struct mutex info_lock;
|
||||
struct kobject *map_dir;
|
||||
struct kobject *portio_dir;
|
||||
struct percpu_ref info_ref;
|
||||
struct completion confirm_done;
|
||||
struct completion free_done;
|
||||
};
|
||||
|
||||
/**
|
||||
|
|
|
|||
Loading…
Reference in New Issue