This fixes a major bug in the Sun GEM Ether
driver's netpoll implementation. When both polled
and interrupt driven i/o are used simultaneously,
for example when using kgdb over Ether with active
NFS mounts, a condition easily arises where the bug
is hit.
The problem is that gem_poll() expects an rx softnet_data
structure to have been allocated (via __netif_rx_schedule())
prior to its being called. The existing gem driver code
uses gem_interrupt() to set things up for gem_poll(),
but unfortunately, gem_interrupt() doesn't know about
this, and so doesn't allocate the required structure
when it finds no bits set in the interrupt status
register. gem_poll() then move on its way and tries to
delete the non-existent softnet_data structure.
Signed-off-by: Geoff Levand <[EMAIL PROTECTED]>
--- orig/drivers/net/sungem.c
+++ mod/drivers/net/sungem.c
@@ -945,12 +945,12 @@
if (netif_rx_schedule_prep(dev)) {
u32 gem_status = readl(gp->regs + GREG_STAT);
+ gp->status = gem_status;
- if (gem_status == 0) {
+ if ((gem_status & ~GREG_STAT_TXNR) == 0) {
spin_unlock_irqrestore(&gp->lock, flags);
return IRQ_NONE;
}
- gp->status = gem_status;
gem_disable_ints(gp);
__netif_rx_schedule(dev);
}
@@ -970,7 +970,8 @@
/* gem_interrupt is safe to reentrance so no need
* to disable_irq here.
*/
- gem_interrupt(dev->irq, dev, NULL);
+ if(gem_interrupt(dev->irq, dev, NULL) == IRQ_NONE)
+ __netif_rx_schedule(dev);
}
#endif
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
