drivers/net/usb/usbnet.c | 38 +++++++++++++++++++------------------- include/linux/usb/usbnet.h | 2 +- 2 files changed, 20 insertions(+), 20 deletions(-)
Migrate tasklet APIs to the new bottom half workqueue mechanism. It
replaces all occurrences of tasklet usage with the appropriate workqueue
APIs throughout the usbnet driver. This transition ensures compatibility
with the latest design and enhances performance.
As suggested by Oliver and Sundeep, the usbnet_bh() function only be
called by usbnet_bh_workqueue(). It can be waited on in usbnet_stop(), which
in turn can be called for a device reset. Hence this must be GFP_NOIO
instead of GFP_ATOMIC in the rx_alloc_submit().
Suggested-by: Subbaraya Sundeep <sbhatta@marvell.com>
Suggested-by: Oliver Neukum <oneukum@suse.com>
Signed-off-by: Jun Miao <jun.miao@intel.com>
---
drivers/net/usb/usbnet.c | 38 +++++++++++++++++++-------------------
include/linux/usb/usbnet.h | 2 +-
2 files changed, 20 insertions(+), 20 deletions(-)
diff --git a/drivers/net/usb/usbnet.c b/drivers/net/usb/usbnet.c
index c04e715a4c2a..c2750fd5eb59 100644
--- a/drivers/net/usb/usbnet.c
+++ b/drivers/net/usb/usbnet.c
@@ -461,7 +461,7 @@ static enum skb_state defer_bh(struct usbnet *dev, struct sk_buff *skb,
__skb_queue_tail(&dev->done, skb);
if (dev->done.qlen == 1)
- tasklet_schedule(&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
spin_unlock(&dev->done.lock);
spin_unlock_irqrestore(&list->lock, flags);
return old_state;
@@ -549,7 +549,7 @@ static int rx_submit (struct usbnet *dev, struct urb *urb, gfp_t flags)
default:
netif_dbg(dev, rx_err, dev->net,
"rx submit, %d\n", retval);
- tasklet_schedule (&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
break;
case 0:
__usbnet_queue_skb(&dev->rxq, skb, rx_start);
@@ -709,7 +709,7 @@ void usbnet_resume_rx(struct usbnet *dev)
num++;
}
- tasklet_schedule(&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
netif_dbg(dev, rx_status, dev->net,
"paused rx queue disabled, %d skbs requeued\n", num);
@@ -778,7 +778,7 @@ void usbnet_unlink_rx_urbs(struct usbnet *dev)
{
if (netif_running(dev->net)) {
(void) unlink_urbs (dev, &dev->rxq);
- tasklet_schedule(&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
}
}
EXPORT_SYMBOL_GPL(usbnet_unlink_rx_urbs);
@@ -861,14 +861,14 @@ int usbnet_stop (struct net_device *net)
/* deferred work (timer, softirq, task) must also stop */
dev->flags = 0;
timer_delete_sync(&dev->delay);
- tasklet_kill(&dev->bh);
+ disable_work_sync(&dev->bh_work);
cancel_work_sync(&dev->kevent);
/* We have cyclic dependencies. Those calls are needed
* to break a cycle. We cannot fall into the gaps because
* we have a flag
*/
- tasklet_kill(&dev->bh);
+ disable_work_sync(&dev->bh_work);
timer_delete_sync(&dev->delay);
cancel_work_sync(&dev->kevent);
@@ -955,7 +955,7 @@ int usbnet_open (struct net_device *net)
clear_bit(EVENT_RX_KILL, &dev->flags);
// delay posting reads until we're fully open
- tasklet_schedule (&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
if (info->manage_power) {
retval = info->manage_power(dev, 1);
if (retval < 0) {
@@ -1123,7 +1123,7 @@ static void __handle_link_change(struct usbnet *dev)
*/
} else {
/* submitting URBs for reading packets */
- tasklet_schedule(&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
}
/* hard_mtu or rx_urb_size may change during link change */
@@ -1198,11 +1198,11 @@ usbnet_deferred_kevent (struct work_struct *work)
} else {
clear_bit (EVENT_RX_HALT, &dev->flags);
if (!usbnet_going_away(dev))
- tasklet_schedule(&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
}
}
- /* tasklet could resubmit itself forever if memory is tight */
+ /* workqueue could resubmit itself forever if memory is tight */
if (test_bit (EVENT_RX_MEMORY, &dev->flags)) {
struct urb *urb = NULL;
int resched = 1;
@@ -1224,7 +1224,7 @@ usbnet_deferred_kevent (struct work_struct *work)
fail_lowmem:
if (resched)
if (!usbnet_going_away(dev))
- tasklet_schedule(&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
}
}
@@ -1325,7 +1325,7 @@ void usbnet_tx_timeout (struct net_device *net, unsigned int txqueue)
struct usbnet *dev = netdev_priv(net);
unlink_urbs (dev, &dev->txq);
- tasklet_schedule (&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
/* this needs to be handled individually because the generic layer
* doesn't know what is sufficient and could not restore private
* information if a remedy of an unconditional reset were used.
@@ -1547,7 +1547,7 @@ static inline void usb_free_skb(struct sk_buff *skb)
/*-------------------------------------------------------------------------*/
-// tasklet (work deferred from completions, in_irq) or timer
+// workqueue (work deferred from completions, in_irq) or timer
static void usbnet_bh (struct timer_list *t)
{
@@ -1594,23 +1594,23 @@ static void usbnet_bh (struct timer_list *t)
int temp = dev->rxq.qlen;
if (temp < RX_QLEN(dev)) {
- if (rx_alloc_submit(dev, GFP_ATOMIC) == -ENOLINK)
+ if (rx_alloc_submit(dev, GFP_NOIO) == -ENOLINK)
return;
if (temp != dev->rxq.qlen)
netif_dbg(dev, link, dev->net,
"rxqlen %d --> %d\n",
temp, dev->rxq.qlen);
if (dev->rxq.qlen < RX_QLEN(dev))
- tasklet_schedule (&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
}
if (dev->txq.qlen < TX_QLEN (dev))
netif_wake_queue (dev->net);
}
}
-static void usbnet_bh_tasklet(struct tasklet_struct *t)
+static void usbnet_bh_workqueue(struct work_struct *work)
{
- struct usbnet *dev = from_tasklet(dev, t, bh);
+ struct usbnet *dev = from_work(dev, work, bh_work);
usbnet_bh(&dev->delay);
}
@@ -1742,7 +1742,7 @@ usbnet_probe (struct usb_interface *udev, const struct usb_device_id *prod)
skb_queue_head_init (&dev->txq);
skb_queue_head_init (&dev->done);
skb_queue_head_init(&dev->rxq_pause);
- tasklet_setup(&dev->bh, usbnet_bh_tasklet);
+ INIT_WORK (&dev->bh_work, usbnet_bh_workqueue);
INIT_WORK (&dev->kevent, usbnet_deferred_kevent);
init_usb_anchor(&dev->deferred);
timer_setup(&dev->delay, usbnet_bh, 0);
@@ -1971,7 +1971,7 @@ int usbnet_resume (struct usb_interface *intf)
if (!(dev->txq.qlen >= TX_QLEN(dev)))
netif_tx_wake_all_queues(dev->net);
- tasklet_schedule (&dev->bh);
+ queue_work(system_bh_wq, &dev->bh_work);
}
}
diff --git a/include/linux/usb/usbnet.h b/include/linux/usb/usbnet.h
index 0b9f1e598e3a..208682f77179 100644
--- a/include/linux/usb/usbnet.h
+++ b/include/linux/usb/usbnet.h
@@ -58,7 +58,7 @@ struct usbnet {
unsigned interrupt_count;
struct mutex interrupt_mutex;
struct usb_anchor deferred;
- struct tasklet_struct bh;
+ struct work_struct bh_work;
struct work_struct kevent;
unsigned long flags;
--
2.32.0On Tue, 10 Jun 2025 22:54:03 +0800 Jun Miao wrote: > - if (rx_alloc_submit(dev, GFP_ATOMIC) == -ENOLINK) > + if (rx_alloc_submit(dev, GFP_NOIO) == -ENOLINK) Sorry, I think Subbaraya mislead you. v1 was fine. If we want to change the flags (which Im not sure is correct) it should be a separate commit. Could you repost v1? There is no need to attribute reviewers with Suggested-by tags. They can send their review tags if they so wish. -- pw-bot: cr
> >On Tue, 10 Jun 2025 22:54:03 +0800 Jun Miao wrote: >> - if (rx_alloc_submit(dev, GFP_ATOMIC) == -ENOLINK) >> + if (rx_alloc_submit(dev, GFP_NOIO) == -ENOLINK) > >Sorry, I think Subbaraya mislead you. v1 was fine. >If we want to change the flags (which Im not sure is correct) it should be a >separate commit. Could you repost v1? There is no need to attribute reviewers No problem, repost v1 again without Suggested-by tags. Separate commit can be in other patch when changing the flags. Thanks Jun Miao >with Suggested-by tags. They can send their review tags if they so wish. >-- >pw-bot: cr
On 2025-06-13 at 01:51:31, Jakub Kicinski (kuba@kernel.org) wrote: > On Tue, 10 Jun 2025 22:54:03 +0800 Jun Miao wrote: > > - if (rx_alloc_submit(dev, GFP_ATOMIC) == -ENOLINK) > > + if (rx_alloc_submit(dev, GFP_NOIO) == -ENOLINK) > > Sorry, I think Subbaraya mislead you. v1 was fine. > If we want to change the flags (which Im not sure is correct) it should > be a separate commit. Could you repost v1? There is no need to attribute Yeah hence asked to correct me if wrong. GFP_ATOMIC has to be there - "this patchset implements BH workqueues which are like regular workqueues but executes work items in the BH (softirq) context" from: https://lwn.net/Articles/960020/ Thanks, Sundeep > reviewers with Suggested-by tags. They can send their review tags if > they so wish. > -- > pw-bot: cr
© 2016 - 2025 Red Hat, Inc.