| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <200710022111.l92LBg8w022504@imap1.linux-foundation.org>
Date: Tue, 02 Oct 2007 14:11:42 -0700
From: akpm@...ux-foundation.org
To: jeff@...zik.org
Cc: netdev@...r.kernel.org, akpm@...ux-foundation.org,
macro@...ux-mips.org, afleming@...escale.com, jgarzik@...ox.com
Subject: [patch 10/13] PHYLIB: IRQ event workqueue handling fixes
From: "Maciej W. Rozycki" <macro@...ux-mips.org>
Keep track of disable_irq_nosync() invocations and call enable_irq() the
right number of times if work has been cancelled that would include them.
Now that the call to flush_work_keventd() (problematic because of
rtnl_mutex being held) has been replaced by cancel_work_sync() another
issue has arisen and been left unresolved. As the MDIO bus cannot be
accessed from the interrupt context the PHY interrupt handler uses
disable_irq_nosync() to prevent from looping and schedules some work to be
done as a softirq, which, apart from handling the state change of the
originating PHY, is responsible for reenabling the interrupt. Now if the
interrupt line is shared by another device and a call to the softirq
handler has been cancelled, that call to enable_irq() never happens and the
other device cannot use its interrupt anymore as its stuck disabled.
I decided to use a counter rather than a flag because there may be more
than one call to phy_change() cancelled in the queue -- a real one and a
fake one triggered by free_irq() if DEBUG_SHIRQ is used, if nothing else.
Therefore because of its nesting property enable_irq() has to be called the
right number of times to match the number disable_irq_nosync() was called
and restore the original state. This DEBUG_SHIRQ feature is also the
reason why free_irq() has to be called before cancel_work_sync().
While at it I updated the comment about phy_stop_interrupts() being called
from `keventd' -- this is no longer relevant as the use of
cancel_work_sync() makes such an approach unnecessary. OTOH a similar
comment referring to flush_scheduled_work() in phy_stop() still applies as
using cancel_work_sync() there would be dangerous.
Checked with checkpatch.pl and at the run time (with and without
DEBUG_SHIRQ).
Signed-off-by: Maciej W. Rozycki <macro@...ux-mips.org>
Cc: Andy Fleming <afleming@...escale.com>
Cc: Jeff Garzik <jgarzik@...ox.com>
Signed-off-by: Andrew Morton <akpm@...ux-foundation.org>
---
drivers/net/phy/phy.c | 24 +++++++++++++++++++-----
include/linux/phy.h | 3 +++
2 files changed, 22 insertions(+), 5 deletions(-)
diff -puN drivers/net/phy/phy.c~phylib-irq-event-workqueue-handling-fixes drivers/net/phy/phy.c
--- a/drivers/net/phy/phy.c~phylib-irq-event-workqueue-handling-fixes
+++ a/drivers/net/phy/phy.c
@@ -7,7 +7,7 @@
* Author: Andy Fleming
*
* Copyright (c) 2004 Freescale Semiconductor, Inc.
- * Copyright (c) 2006 Maciej W. Rozycki
+ * Copyright (c) 2006, 2007 Maciej W. Rozycki
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the
@@ -35,6 +35,7 @@
#include <linux/timer.h>
#include <linux/workqueue.h>
+#include <asm/atomic.h>
#include <asm/io.h>
#include <asm/irq.h>
#include <asm/uaccess.h>
@@ -562,6 +563,7 @@ static irqreturn_t phy_interrupt(int irq
* queue will write the PHY to disable and clear the
* interrupt, and then reenable the irq line. */
disable_irq_nosync(irq);
+ atomic_inc(&phydev->irq_disable);
schedule_work(&phydev->phy_queue);
@@ -632,6 +634,7 @@ int phy_start_interrupts(struct phy_devi
INIT_WORK(&phydev->phy_queue, phy_change);
+ atomic_set(&phydev->irq_disable, 0);
if (request_irq(phydev->irq, phy_interrupt,
IRQF_SHARED,
"phy_interrupt",
@@ -662,13 +665,22 @@ int phy_stop_interrupts(struct phy_devic
if (err)
phy_error(phydev);
+ free_irq(phydev->irq, phydev);
+
/*
- * Finish any pending work; we might have been scheduled to be called
- * from keventd ourselves, but cancel_work_sync() handles that.
+ * Cannot call flush_scheduled_work() here as desired because
+ * of rtnl_lock(), but we do not really care about what would
+ * be done, except from enable_irq(), so cancel any work
+ * possibly pending and take care of the matter below.
*/
cancel_work_sync(&phydev->phy_queue);
-
- free_irq(phydev->irq, phydev);
+ /*
+ * If work indeed has been cancelled, disable_irq() will have
+ * been left unbalanced from phy_interrupt() and enable_irq()
+ * has to be called so that other devices on the line work.
+ */
+ while (atomic_dec_return(&phydev->irq_disable) >= 0)
+ enable_irq(phydev->irq);
return err;
}
@@ -695,6 +707,7 @@ static void phy_change(struct work_struc
phydev->state = PHY_CHANGELINK;
spin_unlock_bh(&phydev->lock);
+ atomic_dec(&phydev->irq_disable);
enable_irq(phydev->irq);
/* Reenable interrupts */
@@ -708,6 +721,7 @@ static void phy_change(struct work_struc
irq_enable_err:
disable_irq(phydev->irq);
+ atomic_inc(&phydev->irq_disable);
phy_err:
phy_error(phydev);
}
diff -puN include/linux/phy.h~phylib-irq-event-workqueue-handling-fixes include/linux/phy.h
--- a/include/linux/phy.h~phylib-irq-event-workqueue-handling-fixes
+++ a/include/linux/phy.h
@@ -25,6 +25,8 @@
#include <linux/timer.h>
#include <linux/workqueue.h>
+#include <asm/atomic.h>
+
#define PHY_BASIC_FEATURES (SUPPORTED_10baseT_Half | \
SUPPORTED_10baseT_Full | \
SUPPORTED_100baseT_Half | \
@@ -281,6 +283,7 @@ struct phy_device {
/* Interrupt and Polling infrastructure */
struct work_struct phy_queue;
struct timer_list phy_timer;
+ atomic_t irq_disable;
spinlock_t lock;
_
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists