[PATCH v2] net: tulip: use mod_timer() in t21142_lnk_change()

[Magnus Lindholm <linmag7@gmail.com>]

t21142_lnk_change() is called from tulip_interrupt(), i.e. in hardirq
context. On a link-fail or NWay renegotiation event it calls
timer_delete_sync(&tp->timer) before rescheduling the timer, which is
exactly what

  WARN_ON(in_hardirq() && !(timer->flags & TIMER_IRQSAFE));

in __timer_delete_sync() exists to catch, since tp->timer is not
TIMER_IRQSAFE:

  WARNING: kernel/time/timer.c:1611 at __timer_delete_sync+0x13c/0x150
  ...
  [<...>] t21142_lnk_change+...
  [<...>] tulip_interrupt+...

This isn't teardown, it's just rescheduling the media timer, which is
exactly what mod_timer() is for. mod_timer(timer, expires) is
documented as equivalent to timer_delete(); timer->expires = expires;
add_timer(), and as the only safe way to change the timeout when a
timer has multiple unserialized concurrent users. That is the case
here: t21142_media_task(), scheduled by this same timer's callback,
already ends with its own mod_timer() call on tp->timer, with a
comment noting it synchronizes against add_timer() calls from
interrupts.

Call mod_timer() before t21142_start_nway() rather than after, to keep
a property the old timer_delete_sync() had as a side effect: while the
timer was merely pending, deleting it first meant it could not fire
during the ~100us t21142_start_nway() takes to reprogram the NWay
state. Rearming first, before that state changes, preserves the same
property without the illegal wait. It does not cover a callback
already in flight: tulip_timer() only does
schedule_work(&tp->media_work), and neither the old
timer_delete_sync() nor mod_timer() waits for or blocks that work once
queued, tulip already uses a separate cancel_work_sync() for that at
shutdown, which is a different primitive for a different race.

Update the comment in tulip_interrupt() accordingly. pnic2_lnk_change()
still calls timer_delete_sync() from the same hardirq path, but its timer
callback re-arms the timer directly with mod_timer(), so fixing that path
requires separate consideration of the callback/reschedule race.

The warning was reproduced during a link-state change at boot on an
Alpha UP2000+ running v7.3-rc1 with:

  0001:02:08.0 Ethernet controller: Digital Equipment Corporation
  DECchip 21142/43 (rev 30)

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Magnus Lindholm <linmag7@gmail.com>
---
v2:
- Rearm the media timer with mod_timer() before restarting NWay
  rather than after, preserving the old timer_delete_sync()'s
  protection against a pending timer firing mid-restart. (Francois
  Romieu)
- Spell out in the commit message that this still doesn't serialize
  against an already-queued media_work, which the old code didn't
  cover either.

 drivers/net/ethernet/dec/tulip/21142.c     | 8 ++------
 drivers/net/ethernet/dec/tulip/interrupt.c | 5 ++---
 2 files changed, 4 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/dec/tulip/21142.c b/drivers/net/ethernet/dec/tulip/21142.c
index 76767dec216d..950abf4a8c14 100644
--- a/drivers/net/ethernet/dec/tulip/21142.c
+++ b/drivers/net/ethernet/dec/tulip/21142.c
@@ -216,20 +216,16 @@ void t21142_lnk_change(struct net_device *dev, int csr5)
 		    (csr12 & 2) == 2) ||
 		   (tp->nway && (csr5 & (TPLnkFail)))) {
 		/* Link blew? Maybe restart NWay. */
-		timer_delete_sync(&tp->timer);
+		mod_timer(&tp->timer, RUN_AT(3 * HZ));
 		t21142_start_nway(dev);
-		tp->timer.expires = RUN_AT(3*HZ);
-		add_timer(&tp->timer);
 	} else if (dev->if_port == 3  ||  dev->if_port == 5) {
 		if (tulip_debug > 1)
 			dev_info(&dev->dev, "21143 %s link beat %s\n",
 				 medianame[dev->if_port],
 				 (csr12 & 2) ? "failed" : "good");
 		if ((csr12 & 2)  &&  ! tp->medialock) {
-			timer_delete_sync(&tp->timer);
+			mod_timer(&tp->timer, RUN_AT(3 * HZ));
 			t21142_start_nway(dev);
-			tp->timer.expires = RUN_AT(3*HZ);
-			add_timer(&tp->timer);
 		} else if (dev->if_port == 5)
 			iowrite32(csr14 & ~0x080, ioaddr + CSR14);
 	} else if (dev->if_port == 0  ||  dev->if_port == 4) {
diff --git a/drivers/net/ethernet/dec/tulip/interrupt.c b/drivers/net/ethernet/dec/tulip/interrupt.c
index 0a12cb9b3ba7..6ed4b68ad86c 100644
--- a/drivers/net/ethernet/dec/tulip/interrupt.c
+++ b/drivers/net/ethernet/dec/tulip/interrupt.c
@@ -698,9 +698,8 @@ irqreturn_t tulip_interrupt(int irq, void *dev_instance)
 				dev->stats.rx_errors++;
 				tulip_start_rxtx(tp);
 			}
-			/*
-			 * NB: t21142_lnk_change() does a timer_delete_sync(), so be careful
-			 * if this call is ever done under the spinlock
+			/* NB: pnic2_lnk_change() does a timer_delete_sync(), so be careful
+			 * if this call is ever done under the spinlock.
 			 */
 			if (csr5 & (TPLnkPass | TPLnkFail | 0x08000000)) {
 				if (tp->link_change)

base-commit: 641d03105cc0d2437e32fdeec164f91a4ccef6c4
-- 
2.43.0