2008-08-21 11:37:16

by Alex Nixon

[permalink] [raw]
Subject: [PATCH] Xen: Fix warning when hot-unplugging virtual block devices

From: Alex Nixon <[email protected]>
Date: Thu, 21 Aug 2008 12:30:54 +0100
Subject: [PATCH] Xen: Fix warning when hot-unplugging virtual block devices

WARNING: at kernel/softirq.c:136 local_bh_enable+0x3f/0x8a()

del_gendisk needs IRQs enabled, as the code path eventually reaches local_bh_enable.
blk_stop_queue needs them disabled - so do so temporarily.

It shouldn't matter if an interrupt comes in whilst blkif_io_lock is held, as it will block on the lock. Upon acquisition, it'll realise the device is down and exit cleanly.

Signed-off-by: Alex Nixon <[email protected]>
---
drivers/block/xen-blkfront.c | 6 ++++--
1 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index d5e7532..95cee2c 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -899,16 +899,18 @@ static void blkfront_closing(struct xenbus_device *dev)
if (info->rq == NULL)
goto out;

- spin_lock_irqsave(&blkif_io_lock, flags);
+ spin_lock(&blkif_io_lock);

del_gendisk(info->gd);

+ local_irq_save(flags);
/* No more blkif_request(). */
blk_stop_queue(info->rq);
+ local_irq_restore(flags);

/* No more gnttab callback work. */
gnttab_cancel_free_callback(&info->callback);
- spin_unlock_irqrestore(&blkif_io_lock, flags);
+ spin_unlock(&blkif_io_lock);

/* Flush gnttab callback work. Must be done with no locks held. */
flush_scheduled_work();
--
1.5.4.3


2008-08-21 20:13:24

by Jeremy Fitzhardinge

[permalink] [raw]
Subject: Re: [PATCH] Xen: Fix warning when hot-unplugging virtual block devices

Alex Nixon wrote:
> From: Alex Nixon <[email protected]>
> Date: Thu, 21 Aug 2008 12:30:54 +0100
> Subject: [PATCH] Xen: Fix warning when hot-unplugging virtual block devices
>
> WARNING: at kernel/softirq.c:136 local_bh_enable+0x3f/0x8a()
>
> del_gendisk needs IRQs enabled, as the code path eventually reaches local_bh_enable.
> blk_stop_queue needs them disabled - so do so temporarily.
>
> It shouldn't matter if an interrupt comes in whilst blkif_io_lock is held, as it will block on the lock. Upon acquisition, it'll realise the device is down and exit cleanly.
>
> Signed-off-by: Alex Nixon <[email protected]>
>
Acked-by: Jeremy Fitzhardinge <[email protected]>

Send a clean copy to Jens with my ack.

J
> ---
> drivers/block/xen-blkfront.c | 6 ++++--
> 1 files changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
> index d5e7532..95cee2c 100644
> --- a/drivers/block/xen-blkfront.c
> +++ b/drivers/block/xen-blkfront.c
> @@ -899,16 +899,18 @@ static void blkfront_closing(struct xenbus_device *dev)
> if (info->rq == NULL)
> goto out;
>
> - spin_lock_irqsave(&blkif_io_lock, flags);
> + spin_lock(&blkif_io_lock);
>
> del_gendisk(info->gd);
>
> + local_irq_save(flags);
> /* No more blkif_request(). */
> blk_stop_queue(info->rq);
> + local_irq_restore(flags);
>
> /* No more gnttab callback work. */
> gnttab_cancel_free_callback(&info->callback);
> - spin_unlock_irqrestore(&blkif_io_lock, flags);
> + spin_unlock(&blkif_io_lock);
>
> /* Flush gnttab callback work. Must be done with no locks held. */
> flush_scheduled_work();
>