ChangeSet 1.1500.8.23, 2004/02/05 16:26:16-08:00, [EMAIL PROTECTED]
[PATCH] USB Gadget: ethernet gadget locking tweaks
[USB] ethernet gadget, locking tweaks
This problem showed pretty quickly on an SMP system. Basically,
access to the freelist (tx more than rx) needs a spinlock.
Stop repeating some alloc_etherdev() work. Disable DEBUG messages.
drivers/usb/gadget/ether.c | 38 +++++++++++++++++++++++++++++---------
1 files changed, 29 insertions(+), 9 deletions(-)
diff -Nru a/drivers/usb/gadget/ether.c b/drivers/usb/gadget/ether.c
--- a/drivers/usb/gadget/ether.c Mon Feb 9 14:38:25 2004
+++ b/drivers/usb/gadget/ether.c Mon Feb 9 14:38:25 2004
@@ -19,7 +19,7 @@
*/
-#define DEBUG 1
+// #define DEBUG 1
// #define VERBOSE
#include <linux/config.h>
@@ -885,8 +885,11 @@
#ifndef DEV_CONFIG_CDC
if (result == 0) {
netif_carrier_on (dev->net);
- if (netif_running (dev->net))
+ if (netif_running (dev->net)) {
+ spin_unlock (&dev->lock);
eth_start (dev, GFP_ATOMIC);
+ spin_lock (&dev->lock);
+ }
} else {
(void) usb_ep_disable (dev->in_ep);
dev->in_ep = 0;
@@ -1246,8 +1249,11 @@
#ifdef EP_STATUS_NUM
issue_start_status (dev);
#endif
- if (netif_running (dev->net))
+ if (netif_running (dev->net)) {
+ spin_unlock (&dev->lock);
eth_start (dev, GFP_ATOMIC);
+ spin_lock (&dev->lock);
+ }
} else {
netif_stop_queue (dev->net);
netif_carrier_off (dev->net);
@@ -1414,16 +1420,14 @@
rx_submit (struct eth_dev *dev, struct usb_request *req, int gfp_flags)
{
struct sk_buff *skb;
- int retval = 0;
+ int retval = -ENOMEM;
size_t size;
size = (sizeof (struct ethhdr) + dev->net->mtu + RX_EXTRA);
if ((skb = alloc_skb (size, gfp_flags)) == 0) {
DEBUG (dev, "no rx skb\n");
- defer_kevent (dev, WORK_RX_MEMORY);
- list_add (&req->list, &dev->rx_reqs);
- return -ENOMEM;
+ goto enomem;
}
req->buf = skb->data;
@@ -1433,11 +1437,14 @@
retval = usb_ep_queue (dev->out_ep, req, gfp_flags);
if (retval == -ENOMEM)
+enomem:
defer_kevent (dev, WORK_RX_MEMORY);
if (retval) {
DEBUG (dev, "rx submit --> %d\n", retval);
dev_kfree_skb_any (skb);
+ spin_lock (&dev->lock);
list_add (&req->list, &dev->rx_reqs);
+ spin_unlock (&dev->lock);
}
return retval;
}
@@ -1502,6 +1509,7 @@
dev_kfree_skb_any (skb);
if (!netif_running (dev->net)) {
clean:
+ /* nobody reading rx_reqs, so no dev->lock */
list_add (&req->list, &dev->rx_reqs);
req = 0;
}
@@ -1568,19 +1576,26 @@
static void rx_fill (struct eth_dev *dev, int gfp_flags)
{
struct usb_request *req;
+ unsigned long flags;
clear_bit (WORK_RX_MEMORY, &dev->todo);
/* fill unused rxq slots with some skb */
+ spin_lock_irqsave (&dev->lock, flags);
while (!list_empty (&dev->rx_reqs)) {
req = container_of (dev->rx_reqs.next,
struct usb_request, list);
list_del_init (&req->list);
+ spin_unlock_irqrestore (&dev->lock, flags);
+
if (rx_submit (dev, req, gfp_flags) < 0) {
defer_kevent (dev, WORK_RX_MEMORY);
return;
}
+
+ spin_lock_irqsave (&dev->lock, flags);
}
+ spin_unlock_irqrestore (&dev->lock, flags);
}
static void eth_work (void *_dev)
@@ -1616,7 +1631,9 @@
}
dev->stats.tx_packets++;
+ spin_lock (&dev->lock);
list_add (&req->list, &dev->tx_reqs);
+ spin_unlock (&dev->lock);
dev_kfree_skb_any (skb);
atomic_dec (&dev->tx_qlen);
@@ -1630,11 +1647,14 @@
int length = skb->len;
int retval;
struct usb_request *req = 0;
+ unsigned long flags;
+ spin_lock_irqsave (&dev->lock, flags);
req = container_of (dev->tx_reqs.next, struct usb_request, list);
list_del (&req->list);
if (list_empty (&dev->tx_reqs))
netif_stop_queue (net);
+ spin_unlock_irqrestore (&dev->lock, flags);
/* no buffer copies needed, unless the network stack did it
* or the hardware can't use skb buffers.
@@ -1675,9 +1695,11 @@
if (retval) {
dev->stats.tx_dropped++;
dev_kfree_skb_any (skb);
+ spin_lock_irqsave (&dev->lock, flags);
if (list_empty (&dev->tx_reqs))
netif_start_queue (net);
list_add (&req->list, &dev->tx_reqs);
+ spin_unlock_irqrestore (&dev->lock, flags);
}
return 0;
}
@@ -1798,9 +1820,7 @@
/* network device setup */
dev->net = net;
SET_MODULE_OWNER (net);
- net->priv = dev;
strcpy (net->name, "usb%d");
- ether_setup (net);
/* one random address for the gadget device ... both of these could
* reasonably come from an id prom or a module parameter.
-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
