Skip to contentSkip to conversation

forcedeth: hardirq lockdep warning

PATCH Applied Peter Zijlstra 5 participants 6 messages Sep 19, 2006 – Oct 5, 2006 Network drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 forcedeth: hardirq lockdep warning

BUG: warning at kernel/lockdep.c:1816/trace_hardirqs_on() (Not tainted)

Call Trace:
 show_trace
 dump_stack
 trace_hardirqs_on
 :forcedeth:nv_nic_irq_other
 handle_IRQ_event
 __do_IRQ
 do_IRQ
 ret_from_intr

DWARF2 barf
default_idle
cpu_idle
rest_init
start_kernel
_sinittext

These 3 functions nv_nic_irq_tx(), nv_nic_irq_rx() and nv_nic_irq_other() are reachable from IRQ context and process context. Make use of the irq-save/restore spinlock variant.

(Compile tested only, since I do not have the hardware)

Cc Jeff Garzik <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Arjan van de Ven <[email protected]>
Cc Dave Jones <[email protected]>
Cc Andrew Morton <[email protected]>
---
 drivers/net/forcedeth.c |   31 +++++++++++++++++--------------
 1 file changed, 17 insertions(+), 14 deletions(-)
drivers/net/forcedeth.c +17 -14
Index: linux-2.6-mm/drivers/net/forcedeth.c
===================================================================
@@ -2497,6 +2497,7 @@ static irqreturn_t nv_nic_irq_tx(int foo
 	u8 __iomem *base = get_hwbase(dev);
 	u32 events;
 	int i;
+	unsigned long flags;
 
 	dprintk(KERN_DEBUG "%s: nv_nic_irq_tx\n", dev->name);
 
@@ -2508,16 +2509,16 @@ static irqreturn_t nv_nic_irq_tx(int foo
 		if (!(events & np->irqmask))
 			break;
 
-		spin_lock_irq(&np->lock);
+		spin_lock_irqsave(&np->lock, flags);
 		nv_tx_done(dev);
-		spin_unlock_irq(&np->lock);
+		spin_unlock_irqrestore(&np->lock, flags);
 
 		if (events & (NVREG_IRQ_TX_ERR)) {
 			dprintk(KERN_DEBUG "%s: received irq with events 0x%x. Probably TX fail.\n",
 						dev->name, events);
 		}
 		if (i > max_interrupt_work) {
-			spin_lock_irq(&np->lock);
+			spin_lock_irqsave(&np->lock, flags);
 			/* disable interrupts on the nic */
 			writel(NVREG_IRQ_TX_ALL, base + NvRegIrqMask);
 			pci_push(base);
@@ -2527,7 +2528,7 @@ static irqreturn_t nv_nic_irq_tx(int foo
 				mod_timer(&np->nic_poll, jiffies + POLL_WAIT);
 			}
 			printk(KERN_DEBUG "%s: too many iterations (%d) in nv_nic_irq_tx.\n", dev->name, i);
-			spin_unlock_irq(&np->lock);
+			spin_unlock_irqrestore(&np->lock, flags);
 			break;
 		}
 
@@ -2601,6 +2602,7 @@ static irqreturn_t nv_nic_irq_rx(int foo
 	u8 __iomem *base = get_hwbase(dev);
 	u32 events;
 	int i;
+	unsigned long flags;
 
 	dprintk(KERN_DEBUG "%s: nv_nic_irq_rx\n", dev->name);
 
@@ -2614,14 +2616,14 @@ static irqreturn_t nv_nic_irq_rx(int foo
 
 		nv_rx_process(dev, dev->weight);
 		if (nv_alloc_rx(dev)) {
-			spin_lock_irq(&np->lock);
+			spin_lock_irqsave(&np->lock, flags);
 			if (!np->in_shutdown)
 				mod_timer(&np->oom_kick, jiffies + OOM_REFILL);
-			spin_unlock_irq(&np->lock);
+			spin_unlock_irqrestore(&np->lock, flags);
 		}
 
 		if (i > max_interrupt_work) {
-			spin_lock_irq(&np->lock);
+			spin_lock_irqsave(&np->lock, flags);
 			/* disable interrupts on the nic */
 			writel(NVREG_IRQ_RX_ALL, base + NvRegIrqMask);
 			pci_push(base);
@@ -2631,7 +2633,7 @@ static irqreturn_t nv_nic_irq_rx(int foo
 				mod_timer(&np->nic_poll, jiffies + POLL_WAIT);
 			}
 			printk(KERN_DEBUG "%s: too many iterations (%d) in nv_nic_irq_rx.\n", dev->name, i);
-			spin_unlock_irq(&np->lock);
+			spin_unlock_irqrestore(&np->lock, flags);
 			break;
 		}
 	}
@@ -2648,6 +2650,7 @@ static irqreturn_t nv_nic_irq_other(int 
 	u8 __iomem *base = get_hwbase(dev);
 	u32 events;
 	int i;
+	unsigned long flags;
 
 	dprintk(KERN_DEBUG "%s: nv_nic_irq_other\n", dev->name);
 
@@ -2660,14 +2663,14 @@ static irqreturn_t nv_nic_irq_other(int 
 			break;
 
 		if (events & NVREG_IRQ_LINK) {
-			spin_lock_irq(&np->lock);
+			spin_lock_irqsave(&np->lock, flags);
 			nv_link_irq(dev);
-			spin_unlock_irq(&np->lock);
+			spin_unlock_irqrestore(&np->lock, flags);
 		}
 		if (np->need_linktimer && time_after(jiffies, np->link_timeout)) {
-			spin_lock_irq(&np->lock);
+			spin_lock_irqsave(&np->lock, flags);
 			nv_linkchange(dev);
-			spin_unlock_irq(&np->lock);
+			spin_unlock_irqrestore(&np->lock, flags);
 			np->link_timeout = jiffies + LINK_TIMEOUT;
 		}
 		if (events & (NVREG_IRQ_UNKNOWN)) {
@@ -2675,7 +2678,7 @@ static irqreturn_t nv_nic_irq_other(int 
 						dev->name, events);
 		}
 		if (i > max_interrupt_work) {
-			spin_lock_irq(&np->lock);
+			spin_lock_irqsave(&np->lock, flags);
 			/* disable interrupts on the nic */
 			writel(NVREG_IRQ_OTHER, base + NvRegIrqMask);
 			pci_push(base);
@@ -2685,7 +2688,7 @@ static irqreturn_t nv_nic_irq_other(int 
 				mod_timer(&np->nic_poll, jiffies + POLL_WAIT);
 			}
 			printk(KERN_DEBUG "%s: too many iterations (%d) in nv_nic_irq_other.\n", dev->name, i);
-			spin_unlock_irq(&np->lock);
+			spin_unlock_irqrestore(&np->lock, flags);
 			break;
 		}
 
#1 lore raw
Quoted 1 line · On Tue, 19 Sep 2006 14:55:22 +0200 Peter Zijlstra <[email protected]>…
BUG: warning at kernel/lockdep.c:1816/trace_hardirqs_on() (Not tainted)

I wonder what line that was. DEBUG_LOCKS_WARN_ON(current->hardirq_context), I suppose.

Quoted 10 lines
Call Trace:
 show_trace
 dump_stack
 trace_hardirqs_on
 :forcedeth:nv_nic_irq_other
 handle_IRQ_event
 __do_IRQ
 do_IRQ
 ret_from_intr

DWARF2 barf

It's good, isn't it?

Quoted 8 lines

default_idle
cpu_idle
rest_init
start_kernel
_sinittext

These 3 functions nv_nic_irq_tx(), nv_nic_irq_rx() and nv_nic_irq_other() are reachable from IRQ context and process context.

That's "fix deadlock", not "fix lockdep warning". However it's often the case that it's not really deadlockable because (often) the card's IRQ line has been disabled. That appears to be the case in nv_do_nic_poll()'s call to nv_nic_irq_tx, for example.

Quoted 2 lines

Make use of the
irq-save/restore spinlock variant.

So this perhaps is another lockdep workaround..

#2 lore raw
Quoted 7 lines · Andrew Morton
Quoted 1 line · On Tue, 19 Sep 2006 14:55:22 +0200 Peter Zijlstra <[email protected]>…
BUG: warning at kernel/lockdep.c:1816/trace_hardirqs_on() (Not tainted)

I wonder what line that was. DEBUG_LOCKS_WARN_ON(current->hardirq_context), I suppose.

That's what it matches up to in the Fedora kernel. (We have a bunch of lockdep fixes scooped up from lkml over the last month or so, which may offset us).

Dave

#3 lore raw
In reply to Andrew Morton
Quoted 7 lines · Andrew Morton
Quoted 1 line · On Tue, 19 Sep 2006 14:55:22 +0200 Peter Zijlstra <[email protected]>…
BUG: warning at kernel/lockdep.c:1816/trace_hardirqs_on() (Not tainted)

I wonder what line that was. DEBUG_LOCKS_WARN_ON(current->hardirq_context), I suppose.

Correct indeed.

Quoted 11 lines
Quoted 10 lines
Call Trace:
 show_trace
 dump_stack
 trace_hardirqs_on
 :forcedeth:nv_nic_irq_other
 handle_IRQ_event
 __do_IRQ
 do_IRQ
 ret_from_intr

DWARF2 barf

Quoted 1 line

It's good, isn't it?

Yeah, I hope we'll get it sorted out quickly....

Quoted 6 lines
Quoted 6 lines

default_idle
cpu_idle
rest_init
start_kernel
_sinittext

Quoted 7 lines
Quoted 2 lines

These 3 functions nv_nic_irq_tx(), nv_nic_irq_rx() and nv_nic_irq_other() are reachable from IRQ context and process context.

That's "fix deadlock", not "fix lockdep warning". However it's often the case that it's not really deadlockable because (often) the card's IRQ line has been disabled. That appears to be the case in nv_do_nic_poll()'s call to nv_nic_irq_tx, for example.

Yeah, I saw some of that. But I'm not well versed in the various netdev irq receive paths nor in the driver. I couldn't find the actual maintainer in the MAINTAINERS file nor in the source, thanks for including him in the CC. Lets hope he can say if this is an actual fix or just a workaround.

#4 lore raw
In reply to Andrew Morton
Quoted 6 lines

That's "fix deadlock", not "fix lockdep warning". However it's often the case that it's not really deadlockable because (often) the card's IRQ line has been disabled. That appears to be the case in nv_do_nic_poll()'s call to nv_nic_irq_tx, for example.

except when it's a shared irq line, then the game rules changed ;)