Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1757389AbYBXU6f (ORCPT ); Sun, 24 Feb 2008 15:58:35 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1753240AbYBXU61 (ORCPT ); Sun, 24 Feb 2008 15:58:27 -0500 Received: from ogre.sisk.pl ([217.79.144.158]:54412 "EHLO ogre.sisk.pl" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753217AbYBXU60 (ORCPT ); Sun, 24 Feb 2008 15:58:26 -0500 From: "Rafael J. Wysocki" To: Alan Stern Subject: Re: [Bug 10030] Suspend doesn't work when SD card is inserted Date: Sun, 24 Feb 2008 21:56:55 +0100 User-Agent: KMail/1.9.6 (enterprise 20070904.708012) Cc: Pierre Ossman , Zdenek Kabelac , Kernel development list , pm list , Pavel Machek References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <200802242156.55948.rjw@sisk.pl> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 10107 Lines: 339 On Sunday, 24 of February 2008, Alan Stern wrote: > On Sun, 24 Feb 2008, Rafael J. Wysocki wrote: > > > From: Rafael J. Wysocki > > > > Remove the code that acquires all device semaphores from the suspend > > code path as it causes multiple problems to appear (most notably, > > http://bugzilla.kernel.org/show_bug.cgi?id=10030) and revert the > > change introduced by commit 4145ed6dc597a9bea5f6ae8c574653b2de10620f > > depending on the code being removed. > > > > Remove pm_sleep_lock()/pm_sleep_unlock() from device add to avoid > > the issue reported at http://bugzilla.kernel.org/show_bug.cgi?id=9874. > > Do you actually know whether removing that lock fixes the reported bug? In fact I'm waiting for the confirmation, but given the symptoms described by Lukas I'm quite confident that the problem shouldn't be reproducible with the lock removed. > I agree, the lock should be removed for now. But I'd sure like to get > some feedback about what's going wrong. No question about that. > It's starting to look as though we'd be a lot better off blocking device > registration during sleep instead of failing it. Agreed. > Shouldn't resume_device() and suspend_device() now acquire the device > semaphore before calling the various methods? That's the way they > used to work. Yes, thanks for reminding me about that. Updated patch follows. Thanks, Rafael --- From: Rafael J. Wysocki Remove the code that acquires all device semaphores from the suspend code path as it causes multiple problems to appear (most notably, http://bugzilla.kernel.org/show_bug.cgi?id=10030) and revert the change introduced by commit 4145ed6dc597a9bea5f6ae8c574653b2de10620f depending on the code being removed. Remove pm_sleep_lock()/pm_sleep_unlock() from device_add() to avoid the issue reported at http://bugzilla.kernel.org/show_bug.cgi?id=9874. Signed-off-by: Rafael J. Wysocki --- drivers/base/core.c | 8 --- drivers/base/power/main.c | 106 ++++++---------------------------------------- drivers/usb/core/usb.c | 2 3 files changed, 17 insertions(+), 99 deletions(-) Index: linux-2.6/drivers/base/power/main.c =================================================================== --- linux-2.6.orig/drivers/base/power/main.c +++ linux-2.6/drivers/base/power/main.c @@ -48,7 +48,6 @@ */ LIST_HEAD(dpm_active); -static LIST_HEAD(dpm_locked); static LIST_HEAD(dpm_off); static LIST_HEAD(dpm_off_irq); static LIST_HEAD(dpm_destroy); @@ -81,28 +80,6 @@ void device_pm_add(struct device *dev) */ void device_pm_remove(struct device *dev) { - /* - * If this function is called during a suspend, it will be blocked, - * because we're holding the device's semaphore at that time, which may - * lead to a deadlock. In that case we want to print a warning. - * However, it may also be called by unregister_dropped_devices() with - * the device's semaphore released, in which case the warning should - * not be printed. - */ - if (down_trylock(&dev->sem)) { - if (down_read_trylock(&pm_sleep_rwsem)) { - /* No suspend in progress, wait on dev->sem */ - down(&dev->sem); - up_read(&pm_sleep_rwsem); - } else { - /* Suspend in progress, we may deadlock */ - dev_warn(dev, "Suspicious %s during suspend\n", - __FUNCTION__); - dump_stack(); - /* The user has been warned ... */ - down(&dev->sem); - } - } pr_debug("PM: Removing info for %s:%s\n", dev->bus ? dev->bus->name : "No Bus", kobject_name(&dev->kobj)); @@ -110,7 +87,6 @@ void device_pm_remove(struct device *dev dpm_sysfs_remove(dev); list_del_init(&dev->power.entry); mutex_unlock(&dpm_list_mtx); - up(&dev->sem); } /** @@ -230,6 +206,8 @@ static int resume_device(struct device * TRACE_DEVICE(dev); TRACE_RESUME(0); + down(&dev->sem); + if (dev->bus && dev->bus->resume) { dev_dbg(dev,"resuming\n"); error = dev->bus->resume(dev); @@ -245,6 +223,8 @@ static int resume_device(struct device * error = dev->class->resume(dev); } + up(&dev->sem); + TRACE_RESUME(error); return error; } @@ -266,7 +246,7 @@ static void dpm_resume(void) struct list_head *entry = dpm_off.next; struct device *dev = to_device(entry); - list_move_tail(entry, &dpm_locked); + list_move_tail(entry, &dpm_active); mutex_unlock(&dpm_list_mtx); resume_device(dev); mutex_lock(&dpm_list_mtx); @@ -275,25 +255,6 @@ static void dpm_resume(void) } /** - * unlock_all_devices - Release each device's semaphore - * - * Go through the dpm_off list. Put each device on the dpm_active - * list and unlock it. - */ -static void unlock_all_devices(void) -{ - mutex_lock(&dpm_list_mtx); - while (!list_empty(&dpm_locked)) { - struct list_head *entry = dpm_locked.prev; - struct device *dev = to_device(entry); - - list_move(entry, &dpm_active); - up(&dev->sem); - } - mutex_unlock(&dpm_list_mtx); -} - -/** * unregister_dropped_devices - Unregister devices scheduled for removal * * Unregister all devices on the dpm_destroy list. @@ -305,7 +266,6 @@ static void unregister_dropped_devices(v struct list_head *entry = dpm_destroy.next; struct device *dev = to_device(entry); - up(&dev->sem); mutex_unlock(&dpm_list_mtx); /* This also removes the device from the list */ device_unregister(dev); @@ -324,7 +284,6 @@ void device_resume(void) { might_sleep(); dpm_resume(); - unlock_all_devices(); unregister_dropped_devices(); up_write(&pm_sleep_rwsem); } @@ -388,18 +347,15 @@ int device_power_down(pm_message_t state struct list_head *entry = dpm_off.prev; struct device *dev = to_device(entry); - list_del_init(&dev->power.entry); error = suspend_device_late(dev, state); if (error) { printk(KERN_ERR "Could not power down device %s: " "error %d\n", kobject_name(&dev->kobj), error); - if (list_empty(&dev->power.entry)) - list_add(&dev->power.entry, &dpm_off); break; } - if (list_empty(&dev->power.entry)) - list_add(&dev->power.entry, &dpm_off_irq); + if (!list_empty(&dev->power.entry)) + list_move(&dev->power.entry, &dpm_off_irq); } if (!error) @@ -419,6 +375,8 @@ static int suspend_device(struct device { int error = 0; + down(&dev->sem); + if (dev->power.power_state.event) { dev_dbg(dev, "PM: suspend %d-->%d\n", dev->power.power_state.event, state.event); @@ -441,6 +399,9 @@ static int suspend_device(struct device error = dev->bus->suspend(dev, state); suspend_report_result(dev->bus->suspend, error); } + + up(&dev->sem); + return error; } @@ -461,11 +422,10 @@ static int dpm_suspend(pm_message_t stat int error = 0; mutex_lock(&dpm_list_mtx); - while (!list_empty(&dpm_locked)) { - struct list_head *entry = dpm_locked.prev; + while (!list_empty(&dpm_active)) { + struct list_head *entry = dpm_active.prev; struct device *dev = to_device(entry); - list_del_init(&dev->power.entry); mutex_unlock(&dpm_list_mtx); error = suspend_device(dev, state); if (error) { @@ -476,14 +436,11 @@ static int dpm_suspend(pm_message_t stat (error == -EAGAIN ? " (please convert to suspend_late)" : "")); - mutex_lock(&dpm_list_mtx); - if (list_empty(&dev->power.entry)) - list_add(&dev->power.entry, &dpm_locked); break; } mutex_lock(&dpm_list_mtx); - if (list_empty(&dev->power.entry)) - list_add(&dev->power.entry, &dpm_off); + if (!list_empty(&dev->power.entry)) + list_move(&dev->power.entry, &dpm_off); } mutex_unlock(&dpm_list_mtx); @@ -491,36 +448,6 @@ static int dpm_suspend(pm_message_t stat } /** - * lock_all_devices - Acquire every device's semaphore - * - * Go through the dpm_active list. Carefully lock each device's - * semaphore and put it in on the dpm_locked list. - */ -static void lock_all_devices(void) -{ - mutex_lock(&dpm_list_mtx); - while (!list_empty(&dpm_active)) { - struct list_head *entry = dpm_active.next; - struct device *dev = to_device(entry); - - /* Required locking order is dev->sem first, - * then dpm_list_mutex. Hence this awkward code. - */ - get_device(dev); - mutex_unlock(&dpm_list_mtx); - down(&dev->sem); - mutex_lock(&dpm_list_mtx); - - if (list_empty(entry)) - up(&dev->sem); /* Device was removed */ - else - list_move_tail(entry, &dpm_locked); - put_device(dev); - } - mutex_unlock(&dpm_list_mtx); -} - -/** * device_suspend - Save state and stop all devices in system. * @state: new power management state * @@ -533,7 +460,6 @@ int device_suspend(pm_message_t state) might_sleep(); down_write(&pm_sleep_rwsem); - lock_all_devices(); error = dpm_suspend(state); if (error) device_resume(); Index: linux-2.6/drivers/usb/core/usb.c =================================================================== --- linux-2.6.orig/drivers/usb/core/usb.c +++ linux-2.6/drivers/usb/core/usb.c @@ -234,7 +234,7 @@ static int ksuspend_usb_init(void) * singlethreaded. Its job doesn't justify running on more * than one CPU. */ - ksuspend_usb_wq = create_singlethread_workqueue("ksuspend_usbd"); + ksuspend_usb_wq = create_freezeable_workqueue("ksuspend_usbd"); if (!ksuspend_usb_wq) return -ENOMEM; return 0; Index: linux-2.6/drivers/base/core.c =================================================================== --- linux-2.6.orig/drivers/base/core.c +++ linux-2.6/drivers/base/core.c @@ -770,13 +770,6 @@ int device_add(struct device *dev) struct class_interface *class_intf; int error; - error = pm_sleep_lock(); - if (error) { - dev_warn(dev, "Suspicious %s during suspend\n", __FUNCTION__); - dump_stack(); - return error; - } - dev = get_device(dev); if (!dev || !strlen(dev->bus_id)) { error = -EINVAL; @@ -843,7 +836,6 @@ int device_add(struct device *dev) } Done: put_device(dev); - pm_sleep_unlock(); return error; BusError: device_pm_remove(dev); -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/